{ "schema": "aner.distributed-neural-public-results/1", "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "protocol": { "rows": 65536, "epochs": 10, "batch": 1536, "rate": 0.05, "seed": 42, "optimizer": "sgd", "model": [ 64, 128, 64, 4 ], "parameters": 16836, "training_rows": 45876, "validation_rows": 9830, "test_rows": 9830, "updates": 300, "tail_rows": 1332, "partition_seed": 42, "host_accumulation": "Float64", "canonical_parameters_and_optimizer": "selected group dtype on CPU", "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "predeclared_tolerances": { "Float32": { "parameter_max_abs": 0.001, "loss_max_abs": 0.0005, "accuracy_max_abs": 0.005 }, "Float64": { "parameter_max_abs": 1e-08, "loss_max_abs": 1e-08, "accuracy_max_abs": 1e-12 } } }, "trial_count": 57, "case_count": 19, "excluded_warmup_count": 19, "case_inventory": [ { "case": "atlas01-cpu32", "label": "Atlas-01 CPU \u00b7 Float32", "dtype": "Float32", "kind": "standalone", "replicas": [ { "configuration_id": "atlas-01/C03", "platform": "atlas-01", "device": "cpu", "relative_batch_weight": 1 } ], "coordinator_configuration_id": "atlas-01/C03", "numerical_reference_case": "atlas01-cpu32" }, { "case": "atlas01-cpu64", "label": "Atlas-01 CPU \u00b7 Float64", "dtype": "Float64", "kind": "standalone", "replicas": [ { "configuration_id": "atlas-01/C03", "platform": "atlas-01", "device": "cpu", "relative_batch_weight": 1 } ], "coordinator_configuration_id": "atlas-01/C03", "numerical_reference_case": "atlas01-cpu64" }, { "case": "atlas02-cpu32", "label": "Atlas-02 CPU \u00b7 Float32", "dtype": "Float32", "kind": "standalone", "replicas": [ { "configuration_id": "atlas-02/C03", "platform": "atlas-02", "device": "cpu", "relative_batch_weight": 1 } ], "coordinator_configuration_id": "atlas-02/C03", "numerical_reference_case": "atlas01-cpu32" }, { "case": "atlas02-cpu64", "label": "Atlas-02 CPU \u00b7 Float64", "dtype": "Float64", "kind": "standalone", "replicas": [ { "configuration_id": "atlas-02/C03", "platform": "atlas-02", "device": "cpu", "relative_batch_weight": 1 } ], "coordinator_configuration_id": "atlas-02/C03", "numerical_reference_case": "atlas01-cpu64" }, { "case": "atlas03-cpu32", "label": "Atlas-03 CPU \u00b7 Float32", "dtype": "Float32", "kind": "standalone", "replicas": [ { "configuration_id": "atlas-03/C03", "platform": "atlas-03", "device": "cpu", "relative_batch_weight": 1 } ], "coordinator_configuration_id": "atlas-03/C03", "numerical_reference_case": "atlas01-cpu32" }, { "case": "atlas03-cpu64", "label": "Atlas-03 CPU \u00b7 Float64", "dtype": "Float64", "kind": "standalone", "replicas": [ { "configuration_id": "atlas-03/C03", "platform": "atlas-03", "device": "cpu", "relative_batch_weight": 1 } ], "coordinator_configuration_id": "atlas-03/C03", "numerical_reference_case": "atlas01-cpu64" }, { "case": "atlas01-gpu32", "label": "Atlas-01 MPS \u00b7 Float32", "dtype": "Float32", "kind": "standalone", "replicas": [ { "configuration_id": "atlas-01/C03", "platform": "atlas-01", "device": "mps", "relative_batch_weight": 1 } ], "coordinator_configuration_id": "atlas-01/C03", "numerical_reference_case": "atlas01-cpu32" }, { "case": "atlas02-gpu32", "label": "Atlas-02 CUDA \u00b7 Float32", "dtype": "Float32", "kind": "standalone", "replicas": [ { "configuration_id": "atlas-02/C03", "platform": "atlas-02", "device": "cuda", "relative_batch_weight": 1 } ], "coordinator_configuration_id": "atlas-02/C03", "numerical_reference_case": "atlas01-cpu32" }, { "case": "atlas02-gpu64", "label": "Atlas-02 CUDA \u00b7 Float64", "dtype": "Float64", "kind": "standalone", "replicas": [ { "configuration_id": "atlas-02/C03", "platform": "atlas-02", "device": "cuda", "relative_batch_weight": 1 } ], "coordinator_configuration_id": "atlas-02/C03", "numerical_reference_case": "atlas01-cpu64" }, { "case": "atlas03-gpu32", "label": "Atlas-03 CUDA \u00b7 Float32", "dtype": "Float32", "kind": "standalone", "replicas": [ { "configuration_id": "atlas-03/C03", "platform": "atlas-03", "device": "cuda", "relative_batch_weight": 1 } ], "coordinator_configuration_id": "atlas-03/C03", "numerical_reference_case": "atlas01-cpu32" }, { "case": "atlas03-gpu64", "label": "Atlas-03 CUDA \u00b7 Float64", "dtype": "Float64", "kind": "standalone", "replicas": [ { "configuration_id": "atlas-03/C03", "platform": "atlas-03", "device": "cuda", "relative_batch_weight": 1 } ], "coordinator_configuration_id": "atlas-03/C03", "numerical_reference_case": "atlas01-cpu64" }, { "case": "01-02-gpu32", "label": "Atlas-01 MPS + Atlas-02 CUDA \u00b7 Float32", "dtype": "Float32", "kind": "group", "replicas": [ { "configuration_id": "atlas-01/C03", "platform": "atlas-01", "device": "mps", "relative_batch_weight": 1 }, { "configuration_id": "atlas-02/C03", "platform": "atlas-02", "device": "cuda", "relative_batch_weight": 1 } ], "coordinator_configuration_id": "atlas-01/C03", "numerical_reference_case": "atlas01-cpu32" }, { "case": "01-03-gpu32", "label": "Atlas-01 MPS + Atlas-03 CUDA \u00b7 Float32", "dtype": "Float32", "kind": "group", "replicas": [ { "configuration_id": "atlas-01/C03", "platform": "atlas-01", "device": "mps", "relative_batch_weight": 1 }, { "configuration_id": "atlas-03/C03", "platform": "atlas-03", "device": "cuda", "relative_batch_weight": 1 } ], "coordinator_configuration_id": "atlas-01/C03", "numerical_reference_case": "atlas01-cpu32" }, { "case": "02-03-gpu32", "label": "Atlas-02 CUDA + Atlas-03 CUDA \u00b7 Float32", "dtype": "Float32", "kind": "group", "replicas": [ { "configuration_id": "atlas-02/C03", "platform": "atlas-02", "device": "cuda", "relative_batch_weight": 1 }, { "configuration_id": "atlas-03/C03", "platform": "atlas-03", "device": "cuda", "relative_batch_weight": 1 } ], "coordinator_configuration_id": "atlas-01/C03", "numerical_reference_case": "atlas01-cpu32" }, { "case": "01-02-03-gpu32", "label": "Atlas-01 MPS + Atlas-02 CUDA + Atlas-03 CUDA \u00b7 Float32", "dtype": "Float32", "kind": "group", "replicas": [ { "configuration_id": "atlas-01/C03", "platform": "atlas-01", "device": "mps", "relative_batch_weight": 1 }, { "configuration_id": "atlas-02/C03", "platform": "atlas-02", "device": "cuda", "relative_batch_weight": 1 }, { "configuration_id": "atlas-03/C03", "platform": "atlas-03", "device": "cuda", "relative_batch_weight": 1 } ], "coordinator_configuration_id": "atlas-01/C03", "numerical_reference_case": "atlas01-cpu32" }, { "case": "02-03-gpu64", "label": "Atlas-02 CUDA + Atlas-03 CUDA \u00b7 Float64", "dtype": "Float64", "kind": "group", "replicas": [ { "configuration_id": "atlas-02/C03", "platform": "atlas-02", "device": "cuda", "relative_batch_weight": 1 }, { "configuration_id": "atlas-03/C03", "platform": "atlas-03", "device": "cuda", "relative_batch_weight": 1 } ], "coordinator_configuration_id": "atlas-01/C03", "numerical_reference_case": "atlas01-cpu64" }, { "case": "01-02-03-cpu32", "label": "Atlas-01 CPU + Atlas-02 CPU + Atlas-03 CPU \u00b7 Float32", "dtype": "Float32", "kind": "group", "replicas": [ { "configuration_id": "atlas-01/C03", "platform": "atlas-01", "device": "cpu", "relative_batch_weight": 1 }, { "configuration_id": "atlas-02/C03", "platform": "atlas-02", "device": "cpu", "relative_batch_weight": 1 }, { "configuration_id": "atlas-03/C03", "platform": "atlas-03", "device": "cpu", "relative_batch_weight": 1 } ], "coordinator_configuration_id": "atlas-01/C03", "numerical_reference_case": "atlas01-cpu32" }, { "case": "01-02-03-cpu64", "label": "Atlas-01 CPU + Atlas-02 CPU + Atlas-03 CPU \u00b7 Float64", "dtype": "Float64", "kind": "group", "replicas": [ { "configuration_id": "atlas-01/C03", "platform": "atlas-01", "device": "cpu", "relative_batch_weight": 1 }, { "configuration_id": "atlas-02/C03", "platform": "atlas-02", "device": "cpu", "relative_batch_weight": 1 }, { "configuration_id": "atlas-03/C03", "platform": "atlas-03", "device": "cpu", "relative_batch_weight": 1 } ], "coordinator_configuration_id": "atlas-01/C03", "numerical_reference_case": "atlas01-cpu64" }, { "case": "02-03-gpu32-w1-7", "label": "Atlas-02 CUDA + Atlas-03 CUDA \u00b7 Float32 \u00b7 allocation 1:7", "dtype": "Float32", "kind": "group", "replicas": [ { "configuration_id": "atlas-02/C03", "platform": "atlas-02", "device": "cuda", "relative_batch_weight": 1 }, { "configuration_id": "atlas-03/C03", "platform": "atlas-03", "device": "cuda", "relative_batch_weight": 7 } ], "coordinator_configuration_id": "atlas-01/C03", "numerical_reference_case": "atlas01-cpu32" } ], "timing_scope": "fresh command includes local or SSH startup, source generation/partition/preprocessing, replica initialization, updates, validation/final test and native shutdown; excludes checkpoint/report disk I/O", "memory_scopes": { "native_process_peak_rss_bytes": "POSIX process high-water mark up to native report query; later parameter JSON serialization/shutdown are outside that query", "tensor_owned_host_peak_bytes": "tracked Tensor capacity; excludes Dataset raw source, allocator overhead, provider scratch and driver context", "tensor_owned_device_peak_bytes": "tracked owned CUDA/MPS buffers; provider/context allocations excluded; MPS shared memory overlaps host RAM", "external_process_gpu_peak_bytes": "optional sampled exact local child PID lower bound including driver/provider allocations; remote/group field unavailable" }, "parameter_vectors": "Retained privately; parameter hashes and independently recomputed same-dtype maximum errors are published.", "records": [ { "case": "atlas01-cpu32", "repetition": 1, "started_utc": "2026-10-12T20:57:48.516549+00:00", "finished_utc": "2026-10-12T20:57:51.570465+00:00", "outer_command_wall_seconds": 3.0537536250049016, "warmup_excluded_from_statistics": false, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 104071168, "test_accuracy": 0.8955239064089522, "parameter_sha256": "a8aeb21461c306c7ff0e8ef9fdd8aacd26058a8de7a405d931627793679b2c75", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8858596134282808, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.3770934789242497, "validation_loss": 1.2565156643897568, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 0.251266875 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.168070280486428, "validation_loss": 1.072663564672305, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 0.503438292 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784492905028291, "validation_loss": 0.8820205004872529, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 0.755206208 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923334447027282, "validation_loss": 0.7103084927900398, "validation_accuracy": 0.7501525940996948, "elapsed_seconds": 1.006717417 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370661036836642, "validation_loss": 0.577175882655752, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 1.263244833 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222784697760033, "validation_loss": 0.48197512238581086, "validation_accuracy": 0.8319430315361139, "elapsed_seconds": 1.517656792 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.44097673220999917, "validation_loss": 0.4151257340626702, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 1.7674555 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.3831498027939214, "validation_loss": 0.36713235249111875, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 2.013257875 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.3406089502044476, "validation_loss": 0.3310881936901826, "validation_accuracy": 0.8787385554425229, "elapsed_seconds": 2.25733 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.3078001272045378, "validation_loss": 0.3033744259622793, "validation_accuracy": 0.8858596134282808, "elapsed_seconds": 2.501655458 } ], "validation_loss": 0.3033744259622793, "coordinator_cpu_threads": null, "rate": 0.05, "coordinator_cpu_provider": "apple_accelerate", "test_loss": 0.29705392923262935, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332174875397, "fit_seconds": 2.670434875, "training_loss": 0.2923475363134184, "tensor_owned_device_peak_bytes": 0, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 21086868, "dtype": "Float32", "command_seconds": 2.961245625, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float32", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 1.931893879000001, "initialization_seconds": 0, "failed": false, "nodes": [ { "response_read_wait_seconds": 0, "requests": 300, "coordinator_exchange_wall_seconds": 0, "inner_threads": null, "backend": "apple_accelerate", "target": "macos-arm64", "owned_host_tensor_peak_bytes": 21086868, "request_write_seconds": 0, "request_payload_bytes": 0, "process_peak_rss_bytes": 104038400, "samples": 458760, "upload_preparation_seconds": 0.16247137999999994, "worker_total_seconds": 1.9309346270000005, "zero_sample_steps": 0, "response_payload_bytes": 0, "compute_seconds": 1.7451295810000005, "owned_device_tensor_peak_bytes": 0, "device": "cpu", "readback_seconds": 0.012026081000000004, "configuration_id": "atlas-01/C03", "platform": "atlas-01", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu32", "reference_trial": 1, "dtype": "Float32", "predeclared_tolerances": { "parameter_max_abs": 0.001, "loss_max_abs": 0.0005, "accuracy_max_abs": 0.005 }, "parameter_max_abs_error": 0.0, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 0.0, "validation_loss": 0.0, "test_loss": 0.0 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "atlas01-cpu32", "repetition": 2, "started_utc": "2026-10-12T20:57:51.626165+00:00", "finished_utc": "2026-10-12T20:57:54.743830+00:00", "outer_command_wall_seconds": 3.1175427089910954, "warmup_excluded_from_statistics": false, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 121389056, "test_accuracy": 0.8955239064089522, "parameter_sha256": "a8aeb21461c306c7ff0e8ef9fdd8aacd26058a8de7a405d931627793679b2c75", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8858596134282808, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.3770934789242497, "validation_loss": 1.2565156643897568, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 0.25295875 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.168070280486428, "validation_loss": 1.072663564672305, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 0.5334325 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784492905028291, "validation_loss": 0.8820205004872529, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 0.793179291 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923334447027282, "validation_loss": 0.7103084927900398, "validation_accuracy": 0.7501525940996948, "elapsed_seconds": 1.047631333 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370661036836642, "validation_loss": 0.577175882655752, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 1.306455458 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222784697760033, "validation_loss": 0.48197512238581086, "validation_accuracy": 0.8319430315361139, "elapsed_seconds": 1.563790166 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.44097673220999917, "validation_loss": 0.4151257340626702, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 1.818500125 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.3831498027939214, "validation_loss": 0.36713235249111875, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 2.075435833 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.3406089502044476, "validation_loss": 0.3310881936901826, "validation_accuracy": 0.8787385554425229, "elapsed_seconds": 2.325380083 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.3078001272045378, "validation_loss": 0.3033744259622793, "validation_accuracy": 0.8858596134282808, "elapsed_seconds": 2.571033625 } ], "validation_loss": 0.3033744259622793, "coordinator_cpu_threads": null, "rate": 0.05, "coordinator_cpu_provider": "apple_accelerate", "test_loss": 0.29705392923262935, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332174875397, "fit_seconds": 2.743213375, "training_loss": 0.2923475363134184, "tensor_owned_device_peak_bytes": 0, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 21086868, "dtype": "Float32", "command_seconds": 3.033698916, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float32", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 1.9872317799999997, "initialization_seconds": 0, "failed": false, "nodes": [ { "response_read_wait_seconds": 0, "requests": 300, "coordinator_exchange_wall_seconds": 0, "inner_threads": null, "backend": "apple_accelerate", "target": "macos-arm64", "owned_host_tensor_peak_bytes": 21086868, "request_write_seconds": 0, "request_payload_bytes": 0, "process_peak_rss_bytes": 121356288, "samples": 458760, "upload_preparation_seconds": 0.16654621, "worker_total_seconds": 1.986230381, "zero_sample_steps": 0, "response_payload_bytes": 0, "compute_seconds": 1.7956941709999998, "owned_device_tensor_peak_bytes": 0, "device": "cpu", "readback_seconds": 0.012358166999999995, "configuration_id": "atlas-01/C03", "platform": "atlas-01", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu32", "reference_trial": 1, "dtype": "Float32", "predeclared_tolerances": { "parameter_max_abs": 0.001, "loss_max_abs": 0.0005, "accuracy_max_abs": 0.005 }, "parameter_max_abs_error": 0.0, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 0.0, "validation_loss": 0.0, "test_loss": 0.0 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "atlas01-cpu32", "repetition": 3, "started_utc": "2026-10-12T20:57:54.819332+00:00", "finished_utc": "2026-10-12T20:57:57.882414+00:00", "outer_command_wall_seconds": 3.0628960830072174, "warmup_excluded_from_statistics": false, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 127565824, "test_accuracy": 0.8955239064089522, "parameter_sha256": "a8aeb21461c306c7ff0e8ef9fdd8aacd26058a8de7a405d931627793679b2c75", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8858596134282808, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.3770934789242497, "validation_loss": 1.2565156643897568, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 0.252058417 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.168070280486428, "validation_loss": 1.072663564672305, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 0.505863458 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784492905028291, "validation_loss": 0.8820205004872529, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 0.768089417 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923334447027282, "validation_loss": 0.7103084927900398, "validation_accuracy": 0.7501525940996948, "elapsed_seconds": 1.04269275 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370661036836642, "validation_loss": 0.577175882655752, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 1.298298167 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222784697760033, "validation_loss": 0.48197512238581086, "validation_accuracy": 0.8319430315361139, "elapsed_seconds": 1.554874292 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.44097673220999917, "validation_loss": 0.4151257340626702, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 1.806143917 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.3831498027939214, "validation_loss": 0.36713235249111875, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 2.05350025 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.3406089502044476, "validation_loss": 0.3310881936901826, "validation_accuracy": 0.8787385554425229, "elapsed_seconds": 2.30058925 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.3078001272045378, "validation_loss": 0.3033744259622793, "validation_accuracy": 0.8858596134282808, "elapsed_seconds": 2.5477595 } ], "validation_loss": 0.3033744259622793, "coordinator_cpu_threads": null, "rate": 0.05, "coordinator_cpu_provider": "apple_accelerate", "test_loss": 0.29705392923262935, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332174875397, "fit_seconds": 2.721057125, "training_loss": 0.2923475363134184, "tensor_owned_device_peak_bytes": 0, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 21086868, "dtype": "Float32", "command_seconds": 3.0174825, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float32", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 1.9603461270000004, "initialization_seconds": 0, "failed": false, "nodes": [ { "response_read_wait_seconds": 0, "requests": 300, "coordinator_exchange_wall_seconds": 0, "inner_threads": null, "backend": "apple_accelerate", "target": "macos-arm64", "owned_host_tensor_peak_bytes": 21086868, "request_write_seconds": 0, "request_payload_bytes": 0, "process_peak_rss_bytes": 127533056, "samples": 458760, "upload_preparation_seconds": 0.16428429600000008, "worker_total_seconds": 1.959291791, "zero_sample_steps": 0, "response_payload_bytes": 0, "compute_seconds": 1.7714715769999996, "owned_device_tensor_peak_bytes": 0, "device": "cpu", "readback_seconds": 0.012126252999999997, "configuration_id": "atlas-01/C03", "platform": "atlas-01", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu32", "reference_trial": 1, "dtype": "Float32", "predeclared_tolerances": { "parameter_max_abs": 0.001, "loss_max_abs": 0.0005, "accuracy_max_abs": 0.005 }, "parameter_max_abs_error": 0.0, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 0.0, "validation_loss": 0.0, "test_loss": 0.0 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "atlas01-cpu64", "repetition": 1, "started_utc": "2026-10-12T20:58:01.516343+00:00", "finished_utc": "2026-10-12T20:58:04.994540+00:00", "outer_command_wall_seconds": 3.477942041005008, "warmup_excluded_from_statistics": false, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 112525312, "test_accuracy": 0.8955239064089522, "parameter_sha256": "97b5b013f99579d2b33679af40a39c73e0cd9e28faee9d9df1ac1ad46da08f4c", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8857578840284842, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.3770934749530683, "validation_loss": 1.2565156669823923, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 0.291825041 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.1680702765782571, "validation_loss": 1.0726634152606465, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 0.582371833 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784489967242787, "validation_loss": 0.8820201711707012, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 0.872390041 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923336749857169, "validation_loss": 0.710308352958468, "validation_accuracy": 0.7502543234994914, "elapsed_seconds": 1.164520125 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.637065594943179, "validation_loss": 0.5771756065192747, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 1.45405225 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222779060476093, "validation_loss": 0.4819741321230154, "validation_accuracy": 0.8320447609359105, "elapsed_seconds": 1.742971916 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.4409758238828523, "validation_loss": 0.4151243786499954, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 2.021978208 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.38314860589065225, "validation_loss": 0.36713147013723113, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 2.299024208 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.3406074100312188, "validation_loss": 0.3310860565229031, "validation_accuracy": 0.8786368260427263, "elapsed_seconds": 2.614082166 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.30779926433719895, "validation_loss": 0.30337316499510086, "validation_accuracy": 0.8857578840284842, "elapsed_seconds": 2.901377 } ], "validation_loss": 0.30337316499510086, "coordinator_cpu_threads": null, "rate": 0.05, "coordinator_cpu_provider": "apple_accelerate", "test_loss": 0.2970531787102064, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332128378295, "fit_seconds": 3.097414333, "training_loss": 0.29234691523654444, "tensor_owned_device_peak_bytes": 0, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 16495216, "dtype": "Float64", "command_seconds": 3.39695175, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float64", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 2.362035244999999, "initialization_seconds": 0, "failed": false, "nodes": [ { "response_read_wait_seconds": 0, "requests": 300, "coordinator_exchange_wall_seconds": 0, "inner_threads": null, "backend": "apple_accelerate", "target": "macos-arm64", "owned_host_tensor_peak_bytes": 16495216, "request_write_seconds": 0, "request_payload_bytes": 0, "process_peak_rss_bytes": 112492544, "samples": 458760, "upload_preparation_seconds": 0.10974542300000004, "worker_total_seconds": 2.3609744629999994, "zero_sample_steps": 0, "response_payload_bytes": 0, "compute_seconds": 2.2300271619999994, "owned_device_tensor_peak_bytes": 0, "device": "cpu", "readback_seconds": 0.009962339000000002, "configuration_id": "atlas-01/C03", "platform": "atlas-01", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu64", "reference_trial": 1, "dtype": "Float64", "predeclared_tolerances": { "parameter_max_abs": 1e-08, "loss_max_abs": 1e-08, "accuracy_max_abs": 1e-12 }, "parameter_max_abs_error": 0.0, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 0.0, "validation_loss": 0.0, "test_loss": 0.0 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "atlas01-cpu64", "repetition": 2, "started_utc": "2026-10-12T20:58:05.108202+00:00", "finished_utc": "2026-10-12T20:58:08.539301+00:00", "outer_command_wall_seconds": 3.430949834000785, "warmup_excluded_from_statistics": false, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 94257152, "test_accuracy": 0.8955239064089522, "parameter_sha256": "97b5b013f99579d2b33679af40a39c73e0cd9e28faee9d9df1ac1ad46da08f4c", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8857578840284842, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.3770934749530683, "validation_loss": 1.2565156669823923, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 0.29172725 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.1680702765782571, "validation_loss": 1.0726634152606465, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 0.584370625 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784489967242787, "validation_loss": 0.8820201711707012, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 0.87386875 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923336749857169, "validation_loss": 0.710308352958468, "validation_accuracy": 0.7502543234994914, "elapsed_seconds": 1.167723167 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.637065594943179, "validation_loss": 0.5771756065192747, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 1.456975083 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222779060476093, "validation_loss": 0.4819741321230154, "validation_accuracy": 0.8320447609359105, "elapsed_seconds": 1.742532167 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.4409758238828523, "validation_loss": 0.4151243786499954, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 2.026784625 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.38314860589065225, "validation_loss": 0.36713147013723113, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 2.312651625 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.3406074100312188, "validation_loss": 0.3310860565229031, "validation_accuracy": 0.8786368260427263, "elapsed_seconds": 2.5962145 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.30779926433719895, "validation_loss": 0.30337316499510086, "validation_accuracy": 0.8857578840284842, "elapsed_seconds": 2.879356542 } ], "validation_loss": 0.30337316499510086, "coordinator_cpu_threads": null, "rate": 0.05, "coordinator_cpu_provider": "apple_accelerate", "test_loss": 0.2970531787102064, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332128378295, "fit_seconds": 3.077315833, "training_loss": 0.29234691523654444, "tensor_owned_device_peak_bytes": 0, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 16495216, "dtype": "Float64", "command_seconds": 3.37957175, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float64", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 2.3432101309999984, "initialization_seconds": 0, "failed": false, "nodes": [ { "response_read_wait_seconds": 0, "requests": 300, "coordinator_exchange_wall_seconds": 0, "inner_threads": null, "backend": "apple_accelerate", "target": "macos-arm64", "owned_host_tensor_peak_bytes": 16495216, "request_write_seconds": 0, "request_payload_bytes": 0, "process_peak_rss_bytes": 94224384, "samples": 458760, "upload_preparation_seconds": 0.107875924, "worker_total_seconds": 2.3420942970000005, "zero_sample_steps": 0, "response_payload_bytes": 0, "compute_seconds": 2.213592166000002, "owned_device_tensor_peak_bytes": 0, "device": "cpu", "readback_seconds": 0.009549411, "configuration_id": "atlas-01/C03", "platform": "atlas-01", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu64", "reference_trial": 1, "dtype": "Float64", "predeclared_tolerances": { "parameter_max_abs": 1e-08, "loss_max_abs": 1e-08, "accuracy_max_abs": 1e-12 }, "parameter_max_abs_error": 0.0, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 0.0, "validation_loss": 0.0, "test_loss": 0.0 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "atlas01-cpu64", "repetition": 3, "started_utc": "2026-10-12T20:58:08.656392+00:00", "finished_utc": "2026-10-12T20:58:12.111148+00:00", "outer_command_wall_seconds": 3.4516792920039734, "warmup_excluded_from_statistics": false, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 94224384, "test_accuracy": 0.8955239064089522, "parameter_sha256": "97b5b013f99579d2b33679af40a39c73e0cd9e28faee9d9df1ac1ad46da08f4c", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8857578840284842, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.3770934749530683, "validation_loss": 1.2565156669823923, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 0.291042833 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.1680702765782571, "validation_loss": 1.0726634152606465, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 0.583339208 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784489967242787, "validation_loss": 0.8820201711707012, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 0.876527417 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923336749857169, "validation_loss": 0.710308352958468, "validation_accuracy": 0.7502543234994914, "elapsed_seconds": 1.175279208 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.637065594943179, "validation_loss": 0.5771756065192747, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 1.464054125 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222779060476093, "validation_loss": 0.4819741321230154, "validation_accuracy": 0.8320447609359105, "elapsed_seconds": 1.754142125 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.4409758238828523, "validation_loss": 0.4151243786499954, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 2.04105575 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.38314860589065225, "validation_loss": 0.36713147013723113, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 2.325979083 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.3406074100312188, "validation_loss": 0.3310860565229031, "validation_accuracy": 0.8786368260427263, "elapsed_seconds": 2.610150083 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.30779926433719895, "validation_loss": 0.30337316499510086, "validation_accuracy": 0.8857578840284842, "elapsed_seconds": 2.891530417 } ], "validation_loss": 0.30337316499510086, "coordinator_cpu_threads": null, "rate": 0.05, "coordinator_cpu_provider": "apple_accelerate", "test_loss": 0.2970531787102064, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332128378295, "fit_seconds": 3.086051291, "training_loss": 0.29234691523654444, "tensor_owned_device_peak_bytes": 0, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 16495216, "dtype": "Float64", "command_seconds": 3.387386792, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float64", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 2.3520534969999995, "initialization_seconds": 0, "failed": false, "nodes": [ { "response_read_wait_seconds": 0, "requests": 300, "coordinator_exchange_wall_seconds": 0, "inner_threads": null, "backend": "apple_accelerate", "target": "macos-arm64", "owned_host_tensor_peak_bytes": 16495216, "request_write_seconds": 0, "request_payload_bytes": 0, "process_peak_rss_bytes": 94191616, "samples": 458760, "upload_preparation_seconds": 0.10784828699999997, "worker_total_seconds": 2.3510604120000003, "zero_sample_steps": 0, "response_payload_bytes": 0, "compute_seconds": 2.2227794629999997, "owned_device_tensor_peak_bytes": 0, "device": "cpu", "readback_seconds": 0.009374091000000001, "configuration_id": "atlas-01/C03", "platform": "atlas-01", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu64", "reference_trial": 1, "dtype": "Float64", "predeclared_tolerances": { "parameter_max_abs": 1e-08, "loss_max_abs": 1e-08, "accuracy_max_abs": 1e-12 }, "parameter_max_abs_error": 0.0, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 0.0, "validation_loss": 0.0, "test_loss": 0.0 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "atlas02-cpu32", "repetition": 1, "started_utc": "2026-10-12T20:58:35.896245+00:00", "finished_utc": "2026-10-12T20:58:58.751987+00:00", "outer_command_wall_seconds": 22.852109415995074, "warmup_excluded_from_statistics": false, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 177995776, "test_accuracy": 0.8955239064089522, "parameter_sha256": "44166434f82221f11d83b4836f9b584b873bc9705b804eb25b3e2186648823a0", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8857578840284842, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.3770934829155626, "validation_loss": 1.2565156643897568, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 1.916842175 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.168070276495115, "validation_loss": 1.072663564672305, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 3.746751497 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784492964897984, "validation_loss": 0.8820205098008576, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 5.547711343 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923334504246493, "validation_loss": 0.7103084927900398, "validation_accuracy": 0.7501525940996948, "elapsed_seconds": 7.398923343 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370661056793206, "validation_loss": 0.577175882655752, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 9.202397074 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222785087614635, "validation_loss": 0.481975248113411, "validation_accuracy": 0.8319430315361139, "elapsed_seconds": 11.703904209 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.4409768225446347, "validation_loss": 0.41512579553488677, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 13.481117395 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.3831498494190038, "validation_loss": 0.36713215224861734, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 15.270185209 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.3406088343784194, "validation_loss": 0.33108737317130865, "validation_accuracy": 0.8787385554425229, "elapsed_seconds": 17.067763051 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.30779997538653403, "validation_loss": 0.3033732850214472, "validation_accuracy": 0.8857578840284842, "elapsed_seconds": 18.851808952 } ], "validation_loss": 0.3033732850214472, "coordinator_cpu_threads": 1, "rate": 0.05, "coordinator_cpu_provider": "openblas", "test_loss": 0.29705295057650866, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.599133218017636, "fit_seconds": 20.186910193, "training_loss": 0.29234654818284644, "tensor_owned_device_peak_bytes": 0, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 21086868, "dtype": "Float32", "command_seconds": 21.367686295, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float32", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 15.42905153300001, "initialization_seconds": 0, "failed": false, "nodes": [ { "response_read_wait_seconds": 0, "requests": 300, "coordinator_exchange_wall_seconds": 0, "inner_threads": 1, "backend": "openblas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 21086868, "request_write_seconds": 0, "request_payload_bytes": 0, "process_peak_rss_bytes": 177995776, "samples": 458760, "upload_preparation_seconds": 1.15576639, "worker_total_seconds": 15.424223340000001, "zero_sample_steps": 0, "response_payload_bytes": 0, "compute_seconds": 13.792235244999995, "owned_device_tensor_peak_bytes": 0, "device": "cpu", "readback_seconds": 0.07318645099999999, "configuration_id": "atlas-02/C03", "platform": "atlas-02", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu32", "reference_trial": 1, "dtype": "Float32", "predeclared_tolerances": { "parameter_max_abs": 0.001, "loss_max_abs": 0.0005, "accuracy_max_abs": 0.005 }, "parameter_max_abs_error": 4.6350061893463135e-05, "loss_abs_errors": { "initial_loss": 5.300964112109341e-10, "training_loss": 9.881305719772193e-07, "validation_loss": 1.1409408320894876e-06, "test_loss": 9.786561206981403e-07 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.00010172939979657958, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "atlas02-cpu32", "repetition": 2, "started_utc": "2026-10-12T20:58:58.925554+00:00", "finished_utc": "2026-10-12T20:59:19.814589+00:00", "outer_command_wall_seconds": 20.88762570900144, "warmup_excluded_from_statistics": false, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 177876992, "test_accuracy": 0.8955239064089522, "parameter_sha256": "44166434f82221f11d83b4836f9b584b873bc9705b804eb25b3e2186648823a0", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8857578840284842, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.3770934829155626, "validation_loss": 1.2565156643897568, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 1.680628831 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.168070276495115, "validation_loss": 1.072663564672305, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 3.350606413 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784492964897984, "validation_loss": 0.8820205098008576, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 5.014084914 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923334504246493, "validation_loss": 0.7103084927900398, "validation_accuracy": 0.7501525940996948, "elapsed_seconds": 6.68740022 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370661056793206, "validation_loss": 0.577175882655752, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 8.329519959 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222785087614635, "validation_loss": 0.481975248113411, "validation_accuracy": 0.8319430315361139, "elapsed_seconds": 9.979196962 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.4409768225446347, "validation_loss": 0.41512579553488677, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 11.625521629 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.3831498494190038, "validation_loss": 0.36713215224861734, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 13.256943309 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.3406088343784194, "validation_loss": 0.33108737317130865, "validation_accuracy": 0.8787385554425229, "elapsed_seconds": 14.922962854 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.30779997538653403, "validation_loss": 0.3033732850214472, "validation_accuracy": 0.8857578840284842, "elapsed_seconds": 16.552055189 } ], "validation_loss": 0.3033732850214472, "coordinator_cpu_threads": 1, "rate": 0.05, "coordinator_cpu_provider": "openblas", "test_loss": 0.29705295057650866, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.599133218017636, "fit_seconds": 17.864245818, "training_loss": 0.29234654818284644, "tensor_owned_device_peak_bytes": 0, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 21086868, "dtype": "Float32", "command_seconds": 18.971624858, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float32", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 13.447510067999996, "initialization_seconds": 0, "failed": false, "nodes": [ { "response_read_wait_seconds": 0, "requests": 300, "coordinator_exchange_wall_seconds": 0, "inner_threads": 1, "backend": "openblas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 21086868, "request_write_seconds": 0, "request_payload_bytes": 0, "process_peak_rss_bytes": 177876992, "samples": 458760, "upload_preparation_seconds": 1.0238003210000002, "worker_total_seconds": 13.444564214000014, "zero_sample_steps": 0, "response_payload_bytes": 0, "compute_seconds": 12.017443807999992, "owned_device_tensor_peak_bytes": 0, "device": "cpu", "readback_seconds": 0.060066732000000025, "configuration_id": "atlas-02/C03", "platform": "atlas-02", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu32", "reference_trial": 1, "dtype": "Float32", "predeclared_tolerances": { "parameter_max_abs": 0.001, "loss_max_abs": 0.0005, "accuracy_max_abs": 0.005 }, "parameter_max_abs_error": 4.6350061893463135e-05, "loss_abs_errors": { "initial_loss": 5.300964112109341e-10, "training_loss": 9.881305719772193e-07, "validation_loss": 1.1409408320894876e-06, "test_loss": 9.786561206981403e-07 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.00010172939979657958, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "atlas02-cpu32", "repetition": 3, "started_utc": "2026-10-12T20:59:20.002838+00:00", "finished_utc": "2026-10-12T20:59:42.107932+00:00", "outer_command_wall_seconds": 22.104401999997208, "warmup_excluded_from_statistics": false, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 177963008, "test_accuracy": 0.8955239064089522, "parameter_sha256": "44166434f82221f11d83b4836f9b584b873bc9705b804eb25b3e2186648823a0", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8857578840284842, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.3770934829155626, "validation_loss": 1.2565156643897568, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 1.658516051 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.168070276495115, "validation_loss": 1.072663564672305, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 3.342155078 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784492964897984, "validation_loss": 0.8820205098008576, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 4.996571026 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923334504246493, "validation_loss": 0.7103084927900398, "validation_accuracy": 0.7501525940996948, "elapsed_seconds": 6.914418799 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370661056793206, "validation_loss": 0.577175882655752, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 8.974540751 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222785087614635, "validation_loss": 0.481975248113411, "validation_accuracy": 0.8319430315361139, "elapsed_seconds": 10.648191438 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.4409768225446347, "validation_loss": 0.41512579553488677, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 12.608641335 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.3831498494190038, "validation_loss": 0.36713215224861734, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 14.603775616 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.3406088343784194, "validation_loss": 0.33108737317130865, "validation_accuracy": 0.8787385554425229, "elapsed_seconds": 16.266912465 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.30779997538653403, "validation_loss": 0.3033732850214472, "validation_accuracy": 0.8857578840284842, "elapsed_seconds": 17.99823922 } ], "validation_loss": 0.3033732850214472, "coordinator_cpu_threads": 1, "rate": 0.05, "coordinator_cpu_provider": "openblas", "test_loss": 0.29705295057650866, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.599133218017636, "fit_seconds": 19.335128503, "training_loss": 0.29234654818284644, "tensor_owned_device_peak_bytes": 0, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 21086868, "dtype": "Float32", "command_seconds": 20.064694395, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float32", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 14.509458632999994, "initialization_seconds": 0, "failed": false, "nodes": [ { "response_read_wait_seconds": 0, "requests": 300, "coordinator_exchange_wall_seconds": 0, "inner_threads": 1, "backend": "openblas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 21086868, "request_write_seconds": 0, "request_payload_bytes": 0, "process_peak_rss_bytes": 177963008, "samples": 458760, "upload_preparation_seconds": 1.0965687200000003, "worker_total_seconds": 14.506156379000004, "zero_sample_steps": 0, "response_payload_bytes": 0, "compute_seconds": 12.980122186000001, "owned_device_tensor_peak_bytes": 0, "device": "cpu", "readback_seconds": 0.064397052, "configuration_id": "atlas-02/C03", "platform": "atlas-02", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu32", "reference_trial": 1, "dtype": "Float32", "predeclared_tolerances": { "parameter_max_abs": 0.001, "loss_max_abs": 0.0005, "accuracy_max_abs": 0.005 }, "parameter_max_abs_error": 4.6350061893463135e-05, "loss_abs_errors": { "initial_loss": 5.300964112109341e-10, "training_loss": 9.881305719772193e-07, "validation_loss": 1.1409408320894876e-06, "test_loss": 9.786561206981403e-07 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.00010172939979657958, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "atlas02-cpu64", "repetition": 1, "started_utc": "2026-10-12T21:00:02.126749+00:00", "finished_utc": "2026-10-12T21:00:21.828759+00:00", "outer_command_wall_seconds": 19.701527957993676, "warmup_excluded_from_statistics": false, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 172453888, "test_accuracy": 0.8955239064089522, "parameter_sha256": "012115e11d05b6a7def740254a06068ea009e1acd63dc6799684cada7ccf2b63", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8857578840284842, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.3770934749530683, "validation_loss": 1.2565156669823923, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 1.528564053 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.1680702765782573, "validation_loss": 1.0726634152606465, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 3.076269583 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784489967242787, "validation_loss": 0.8820201711707012, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 4.618010927 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923336749857169, "validation_loss": 0.7103083529584681, "validation_accuracy": 0.7502543234994914, "elapsed_seconds": 6.158337029 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.637065594943179, "validation_loss": 0.5771756065192747, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 7.814241478 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222779060476093, "validation_loss": 0.48197413212301554, "validation_accuracy": 0.8320447609359105, "elapsed_seconds": 9.344051095 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.4409758238828523, "validation_loss": 0.41512437864999546, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 10.877230246 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.38314860589065225, "validation_loss": 0.367131470137231, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 12.42975424 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.3406074100312188, "validation_loss": 0.3310860565229031, "validation_accuracy": 0.8786368260427263, "elapsed_seconds": 13.948203727 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.30779926433719895, "validation_loss": 0.30337316499510086, "validation_accuracy": 0.8857578840284842, "elapsed_seconds": 15.510041722 } ], "validation_loss": 0.30337316499510086, "coordinator_cpu_threads": 1, "rate": 0.05, "coordinator_cpu_provider": "openblas", "test_loss": 0.29705317871020637, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332128378295, "fit_seconds": 16.659829462, "training_loss": 0.29234691523654444, "tensor_owned_device_peak_bytes": 0, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 16495216, "dtype": "Float64", "command_seconds": 17.671156699, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float64", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 12.881922573000002, "initialization_seconds": 0, "failed": false, "nodes": [ { "response_read_wait_seconds": 0, "requests": 300, "coordinator_exchange_wall_seconds": 0, "inner_threads": 1, "backend": "openblas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 16495216, "request_write_seconds": 0, "request_payload_bytes": 0, "process_peak_rss_bytes": 172453888, "samples": 458760, "upload_preparation_seconds": 0.5563793830000007, "worker_total_seconds": 12.879265018999998, "zero_sample_steps": 0, "response_payload_bytes": 0, "compute_seconds": 12.006613135999997, "owned_device_tensor_peak_bytes": 0, "device": "cpu", "readback_seconds": 0.048369314, "configuration_id": "atlas-02/C03", "platform": "atlas-02", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu64", "reference_trial": 1, "dtype": "Float64", "predeclared_tolerances": { "parameter_max_abs": 1e-08, "loss_max_abs": 1e-08, "accuracy_max_abs": 1e-12 }, "parameter_max_abs_error": 3.3306690738754696e-16, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 0.0, "validation_loss": 0.0, "test_loss": 5.551115123125783e-17 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "atlas02-cpu64", "repetition": 2, "started_utc": "2026-10-12T21:00:22.046672+00:00", "finished_utc": "2026-10-12T21:00:42.858283+00:00", "outer_command_wall_seconds": 20.810743416994228, "warmup_excluded_from_statistics": false, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 172425216, "test_accuracy": 0.8955239064089522, "parameter_sha256": "012115e11d05b6a7def740254a06068ea009e1acd63dc6799684cada7ccf2b63", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8857578840284842, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.3770934749530683, "validation_loss": 1.2565156669823923, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 1.54360592 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.1680702765782573, "validation_loss": 1.0726634152606465, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 3.086305431 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784489967242787, "validation_loss": 0.8820201711707012, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 4.62439226 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923336749857169, "validation_loss": 0.7103083529584681, "validation_accuracy": 0.7502543234994914, "elapsed_seconds": 7.143762471 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.637065594943179, "validation_loss": 0.5771756065192747, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 8.681466571 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222779060476093, "validation_loss": 0.48197413212301554, "validation_accuracy": 0.8320447609359105, "elapsed_seconds": 10.20736747 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.4409758238828523, "validation_loss": 0.41512437864999546, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 11.759152488 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.38314860589065225, "validation_loss": 0.367131470137231, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 13.316223205 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.3406074100312188, "validation_loss": 0.3310860565229031, "validation_accuracy": 0.8786368260427263, "elapsed_seconds": 14.848088214 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.30779926433719895, "validation_loss": 0.30337316499510086, "validation_accuracy": 0.8857578840284842, "elapsed_seconds": 16.834203657 } ], "validation_loss": 0.30337316499510086, "coordinator_cpu_threads": 1, "rate": 0.05, "coordinator_cpu_provider": "openblas", "test_loss": 0.29705317871020637, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332128378295, "fit_seconds": 18.071835351, "training_loss": 0.29234691523654444, "tensor_owned_device_peak_bytes": 0, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 16495216, "dtype": "Float64", "command_seconds": 18.795863525, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float64", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 14.069346064999998, "initialization_seconds": 0, "failed": false, "nodes": [ { "response_read_wait_seconds": 0, "requests": 300, "coordinator_exchange_wall_seconds": 0, "inner_threads": 1, "backend": "openblas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 16495216, "request_write_seconds": 0, "request_payload_bytes": 0, "process_peak_rss_bytes": 172425216, "samples": 458760, "upload_preparation_seconds": 0.6052092589999998, "worker_total_seconds": 14.06670137800001, "zero_sample_steps": 0, "response_payload_bytes": 0, "compute_seconds": 13.120414492000009, "owned_device_tensor_peak_bytes": 0, "device": "cpu", "readback_seconds": 0.05226979300000003, "configuration_id": "atlas-02/C03", "platform": "atlas-02", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu64", "reference_trial": 1, "dtype": "Float64", "predeclared_tolerances": { "parameter_max_abs": 1e-08, "loss_max_abs": 1e-08, "accuracy_max_abs": 1e-12 }, "parameter_max_abs_error": 3.3306690738754696e-16, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 0.0, "validation_loss": 0.0, "test_loss": 5.551115123125783e-17 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "atlas02-cpu64", "repetition": 3, "started_utc": "2026-10-12T21:00:43.111319+00:00", "finished_utc": "2026-10-12T21:01:03.763169+00:00", "outer_command_wall_seconds": 20.65155120800773, "warmup_excluded_from_statistics": false, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 172531712, "test_accuracy": 0.8955239064089522, "parameter_sha256": "012115e11d05b6a7def740254a06068ea009e1acd63dc6799684cada7ccf2b63", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8857578840284842, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.3770934749530683, "validation_loss": 1.2565156669823923, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 1.692689018 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.1680702765782573, "validation_loss": 1.0726634152606465, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 3.298529676 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784489967242787, "validation_loss": 0.8820201711707012, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 4.879398147 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923336749857169, "validation_loss": 0.7103083529584681, "validation_accuracy": 0.7502543234994914, "elapsed_seconds": 6.464196574 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.637065594943179, "validation_loss": 0.5771756065192747, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 8.365800856 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222779060476093, "validation_loss": 0.48197413212301554, "validation_accuracy": 0.8320447609359105, "elapsed_seconds": 9.928804721 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.4409758238828523, "validation_loss": 0.41512437864999546, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 11.48230761 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.38314860589065225, "validation_loss": 0.367131470137231, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 12.999113711 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.3406074100312188, "validation_loss": 0.3310860565229031, "validation_accuracy": 0.8786368260427263, "elapsed_seconds": 14.5997002 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.30779926433719895, "validation_loss": 0.30337316499510086, "validation_accuracy": 0.8857578840284842, "elapsed_seconds": 16.172547773 } ], "validation_loss": 0.30337316499510086, "coordinator_cpu_threads": 1, "rate": 0.05, "coordinator_cpu_provider": "openblas", "test_loss": 0.29705317871020637, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332128378295, "fit_seconds": 18.017975232, "training_loss": 0.29234691523654444, "tensor_owned_device_peak_bytes": 0, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 16495216, "dtype": "Float64", "command_seconds": 19.072891984, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float64", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 13.41574941000001, "initialization_seconds": 0, "failed": false, "nodes": [ { "response_read_wait_seconds": 0, "requests": 300, "coordinator_exchange_wall_seconds": 0, "inner_threads": 1, "backend": "openblas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 16495216, "request_write_seconds": 0, "request_payload_bytes": 0, "process_peak_rss_bytes": 172531712, "samples": 458760, "upload_preparation_seconds": 0.5774402320000003, "worker_total_seconds": 13.412996769999998, "zero_sample_steps": 0, "response_payload_bytes": 0, "compute_seconds": 12.507452861000004, "owned_device_tensor_peak_bytes": 0, "device": "cpu", "readback_seconds": 0.050057721999999985, "configuration_id": "atlas-02/C03", "platform": "atlas-02", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu64", "reference_trial": 1, "dtype": "Float64", "predeclared_tolerances": { "parameter_max_abs": 1e-08, "loss_max_abs": 1e-08, "accuracy_max_abs": 1e-12 }, "parameter_max_abs_error": 3.3306690738754696e-16, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 0.0, "validation_loss": 0.0, "test_loss": 5.551115123125783e-17 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "atlas03-cpu32", "repetition": 1, "started_utc": "2026-10-12T21:01:19.640665+00:00", "finished_utc": "2026-10-12T21:01:34.470506+00:00", "outer_command_wall_seconds": 14.828693917006603, "warmup_excluded_from_statistics": false, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 177975296, "test_accuracy": 0.8955239064089522, "parameter_sha256": "8dcc7cf64ee5fc2adf55c88ca37831aad3c93a301687a1cf8e1174efd2007af8", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8857578840284842, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.3770934789242497, "validation_loss": 1.2565156643897568, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 1.19257484 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.168070280486428, "validation_loss": 1.072663564672305, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 2.430111094 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784492885071727, "validation_loss": 0.8820205004872529, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 3.61881574 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923334484289929, "validation_loss": 0.7103084927900398, "validation_accuracy": 0.7501525940996948, "elapsed_seconds": 4.83639166 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370661056793206, "validation_loss": 0.577175882655752, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 6.009877282 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.522278505767979, "validation_loss": 0.48197524997491925, "validation_accuracy": 0.8319430315361139, "elapsed_seconds": 7.190735894 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.44097682454029113, "validation_loss": 0.41512579553488677, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 8.37001418 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.3831498552734491, "validation_loss": 0.3671321569054197, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 9.561839384 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.3406088323827629, "validation_loss": 0.3310873796896192, "validation_accuracy": 0.8787385554425229, "elapsed_seconds": 10.765313695 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.30779997638436224, "validation_loss": 0.3033732850214472, "validation_accuracy": 0.8857578840284842, "elapsed_seconds": 11.995835291 } ], "validation_loss": 0.3033732850214472, "coordinator_cpu_threads": 1, "rate": 0.05, "coordinator_cpu_provider": "openblas", "test_loss": 0.29705295057650866, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332140263232, "fit_seconds": 12.860272355, "training_loss": 0.2923465511763311, "tensor_owned_device_peak_bytes": 0, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 21086868, "dtype": "Float32", "command_seconds": 13.380385378, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float32", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 9.770812434999993, "initialization_seconds": 0, "failed": false, "nodes": [ { "response_read_wait_seconds": 0, "requests": 300, "coordinator_exchange_wall_seconds": 0, "inner_threads": 1, "backend": "openblas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 21086868, "request_write_seconds": 0, "request_payload_bytes": 0, "process_peak_rss_bytes": 177975296, "samples": 458760, "upload_preparation_seconds": 0.7007955530000008, "worker_total_seconds": 9.768997478999998, "zero_sample_steps": 0, "response_payload_bytes": 0, "compute_seconds": 8.709238542999996, "owned_device_tensor_peak_bytes": 0, "device": "cpu", "readback_seconds": 0.04580889400000001, "configuration_id": "atlas-03/C03", "platform": "atlas-03", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu32", "reference_trial": 1, "dtype": "Float32", "predeclared_tolerances": { "parameter_max_abs": 0.001, "loss_max_abs": 0.0005, "accuracy_max_abs": 0.005 }, "parameter_max_abs_error": 4.635751247406006e-05, "loss_abs_errors": { "initial_loss": 3.461216424582858e-09, "training_loss": 9.851370872948628e-07, "validation_loss": 1.1409408320894876e-06, "test_loss": 9.786561206981403e-07 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.00010172939979657958, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "atlas03-cpu32", "repetition": 2, "started_utc": "2026-10-12T21:01:34.753128+00:00", "finished_utc": "2026-10-12T21:01:49.345072+00:00", "outer_command_wall_seconds": 14.590928958001314, "warmup_excluded_from_statistics": false, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 177893376, "test_accuracy": 0.8955239064089522, "parameter_sha256": "8dcc7cf64ee5fc2adf55c88ca37831aad3c93a301687a1cf8e1174efd2007af8", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8857578840284842, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.3770934789242497, "validation_loss": 1.2565156643897568, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 1.157535387 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.168070280486428, "validation_loss": 1.072663564672305, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 2.329299818 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784492885071727, "validation_loss": 0.8820205004872529, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 3.497608525 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923334484289929, "validation_loss": 0.7103084927900398, "validation_accuracy": 0.7501525940996948, "elapsed_seconds": 4.693747966 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370661056793206, "validation_loss": 0.577175882655752, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 5.917045173 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.522278505767979, "validation_loss": 0.48197524997491925, "validation_accuracy": 0.8319430315361139, "elapsed_seconds": 7.081272963 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.44097682454029113, "validation_loss": 0.41512579553488677, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 8.24132724 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.3831498552734491, "validation_loss": 0.3671321569054197, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 9.411947563 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.3406088323827629, "validation_loss": 0.3310873796896192, "validation_accuracy": 0.8787385554425229, "elapsed_seconds": 10.564080953 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.30779997638436224, "validation_loss": 0.3033732850214472, "validation_accuracy": 0.8857578840284842, "elapsed_seconds": 11.725574235 } ], "validation_loss": 0.3033732850214472, "coordinator_cpu_threads": 1, "rate": 0.05, "coordinator_cpu_provider": "openblas", "test_loss": 0.29705295057650866, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332140263232, "fit_seconds": 12.587794052, "training_loss": 0.2923465511763311, "tensor_owned_device_peak_bytes": 0, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 21086868, "dtype": "Float32", "command_seconds": 13.101270396, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float32", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 9.585833630999998, "initialization_seconds": 0, "failed": false, "nodes": [ { "response_read_wait_seconds": 0, "requests": 300, "coordinator_exchange_wall_seconds": 0, "inner_threads": 1, "backend": "openblas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 21086868, "request_write_seconds": 0, "request_payload_bytes": 0, "process_peak_rss_bytes": 177893376, "samples": 458760, "upload_preparation_seconds": 0.6872004380000003, "worker_total_seconds": 9.584013947999999, "zero_sample_steps": 0, "response_payload_bytes": 0, "compute_seconds": 8.543419429, "owned_device_tensor_peak_bytes": 0, "device": "cpu", "readback_seconds": 0.045808335000000054, "configuration_id": "atlas-03/C03", "platform": "atlas-03", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu32", "reference_trial": 1, "dtype": "Float32", "predeclared_tolerances": { "parameter_max_abs": 0.001, "loss_max_abs": 0.0005, "accuracy_max_abs": 0.005 }, "parameter_max_abs_error": 4.635751247406006e-05, "loss_abs_errors": { "initial_loss": 3.461216424582858e-09, "training_loss": 9.851370872948628e-07, "validation_loss": 1.1409408320894876e-06, "test_loss": 9.786561206981403e-07 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.00010172939979657958, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "atlas03-cpu32", "repetition": 3, "started_utc": "2026-10-12T21:01:49.594072+00:00", "finished_utc": "2026-10-12T21:02:04.755153+00:00", "outer_command_wall_seconds": 15.160592833010014, "warmup_excluded_from_statistics": false, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 177852416, "test_accuracy": 0.8955239064089522, "parameter_sha256": "8dcc7cf64ee5fc2adf55c88ca37831aad3c93a301687a1cf8e1174efd2007af8", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8857578840284842, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.3770934789242497, "validation_loss": 1.2565156643897568, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 1.178724314 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.168070280486428, "validation_loss": 1.072663564672305, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 2.346690802 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784492885071727, "validation_loss": 0.8820205004872529, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 3.513784151 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923334484289929, "validation_loss": 0.7103084927900398, "validation_accuracy": 0.7501525940996948, "elapsed_seconds": 4.690979418 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370661056793206, "validation_loss": 0.577175882655752, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 5.865474319 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.522278505767979, "validation_loss": 0.48197524997491925, "validation_accuracy": 0.8319430315361139, "elapsed_seconds": 7.062231205 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.44097682454029113, "validation_loss": 0.41512579553488677, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 8.253556711 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.3831498552734491, "validation_loss": 0.3671321569054197, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 9.468810048 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.3406088323827629, "validation_loss": 0.3310873796896192, "validation_accuracy": 0.8787385554425229, "elapsed_seconds": 10.650494257 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.30779997638436224, "validation_loss": 0.3033732850214472, "validation_accuracy": 0.8857578840284842, "elapsed_seconds": 11.832188201 } ], "validation_loss": 0.3033732850214472, "coordinator_cpu_threads": 1, "rate": 0.05, "coordinator_cpu_provider": "openblas", "test_loss": 0.29705295057650866, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332140263232, "fit_seconds": 12.708668223, "training_loss": 0.2923465511763311, "tensor_owned_device_peak_bytes": 0, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 21086868, "dtype": "Float32", "command_seconds": 13.231820473, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float32", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 9.688264808000008, "initialization_seconds": 0, "failed": false, "nodes": [ { "response_read_wait_seconds": 0, "requests": 300, "coordinator_exchange_wall_seconds": 0, "inner_threads": 1, "backend": "openblas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 21086868, "request_write_seconds": 0, "request_payload_bytes": 0, "process_peak_rss_bytes": 177852416, "samples": 458760, "upload_preparation_seconds": 0.7012716800000004, "worker_total_seconds": 9.686538782000001, "zero_sample_steps": 0, "response_payload_bytes": 0, "compute_seconds": 8.630125725999997, "owned_device_tensor_peak_bytes": 0, "device": "cpu", "readback_seconds": 0.045342648, "configuration_id": "atlas-03/C03", "platform": "atlas-03", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu32", "reference_trial": 1, "dtype": "Float32", "predeclared_tolerances": { "parameter_max_abs": 0.001, "loss_max_abs": 0.0005, "accuracy_max_abs": 0.005 }, "parameter_max_abs_error": 4.635751247406006e-05, "loss_abs_errors": { "initial_loss": 3.461216424582858e-09, "training_loss": 9.851370872948628e-07, "validation_loss": 1.1409408320894876e-06, "test_loss": 9.786561206981403e-07 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.00010172939979657958, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "atlas03-cpu64", "repetition": 1, "started_utc": "2026-10-12T21:02:19.542634+00:00", "finished_utc": "2026-10-12T21:02:33.061990+00:00", "outer_command_wall_seconds": 13.518542040998, "warmup_excluded_from_statistics": false, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 172683264, "test_accuracy": 0.8955239064089522, "parameter_sha256": "0b9cb6964e88ea01eaaf5fc901715b34b184b5c0942cb2e7439c70e1603c1b4c", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8857578840284842, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.3770934749530683, "validation_loss": 1.2565156669823923, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 1.069735406 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.1680702765782573, "validation_loss": 1.0726634152606465, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 2.133997006 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784489967242784, "validation_loss": 0.8820201711707012, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 3.202123316 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923336749857169, "validation_loss": 0.7103083529584678, "validation_accuracy": 0.7502543234994914, "elapsed_seconds": 4.260359645 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370655949431789, "validation_loss": 0.5771756065192747, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 5.311174947 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222779060476093, "validation_loss": 0.4819741321230154, "validation_accuracy": 0.8320447609359105, "elapsed_seconds": 6.400595531 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.4409758238828523, "validation_loss": 0.41512437864999546, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 7.489501221 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.38314860589065225, "validation_loss": 0.36713147013723113, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 8.549072739 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.3406074100312188, "validation_loss": 0.3310860565229031, "validation_accuracy": 0.8786368260427263, "elapsed_seconds": 9.609312337 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.307799264337199, "validation_loss": 0.3033731649951008, "validation_accuracy": 0.8857578840284842, "elapsed_seconds": 10.666615396 } ], "validation_loss": 0.3033731649951008, "coordinator_cpu_threads": 1, "rate": 0.05, "coordinator_cpu_provider": "openblas", "test_loss": 0.29705317871020637, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332128378295, "fit_seconds": 11.45661501, "training_loss": 0.29234691523654444, "tensor_owned_device_peak_bytes": 0, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 16495216, "dtype": "Float64", "command_seconds": 11.956218479, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float64", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 8.896443166999994, "initialization_seconds": 0, "failed": false, "nodes": [ { "response_read_wait_seconds": 0, "requests": 300, "coordinator_exchange_wall_seconds": 0, "inner_threads": 1, "backend": "openblas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 16495216, "request_write_seconds": 0, "request_payload_bytes": 0, "process_peak_rss_bytes": 172683264, "samples": 458760, "upload_preparation_seconds": 0.35629507399999993, "worker_total_seconds": 8.894932573000004, "zero_sample_steps": 0, "response_payload_bytes": 0, "compute_seconds": 8.261263486, "owned_device_tensor_peak_bytes": 0, "device": "cpu", "readback_seconds": 0.03217572499999999, "configuration_id": "atlas-03/C03", "platform": "atlas-03", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu64", "reference_trial": 1, "dtype": "Float64", "predeclared_tolerances": { "parameter_max_abs": 1e-08, "loss_max_abs": 1e-08, "accuracy_max_abs": 1e-12 }, "parameter_max_abs_error": 2.220446049250313e-16, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 0.0, "validation_loss": 5.551115123125783e-17, "test_loss": 5.551115123125783e-17 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "atlas03-cpu64", "repetition": 2, "started_utc": "2026-10-12T21:02:33.365294+00:00", "finished_utc": "2026-10-12T21:02:47.308096+00:00", "outer_command_wall_seconds": 13.942002208001213, "warmup_excluded_from_statistics": false, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 172462080, "test_accuracy": 0.8955239064089522, "parameter_sha256": "0b9cb6964e88ea01eaaf5fc901715b34b184b5c0942cb2e7439c70e1603c1b4c", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8857578840284842, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.3770934749530683, "validation_loss": 1.2565156669823923, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 1.065248287 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.1680702765782573, "validation_loss": 1.0726634152606465, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 2.172973117 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784489967242784, "validation_loss": 0.8820201711707012, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 3.249977956 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923336749857169, "validation_loss": 0.7103083529584678, "validation_accuracy": 0.7502543234994914, "elapsed_seconds": 4.310421541 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370655949431789, "validation_loss": 0.5771756065192747, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 5.383047381 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222779060476093, "validation_loss": 0.4819741321230154, "validation_accuracy": 0.8320447609359105, "elapsed_seconds": 6.43190579 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.4409758238828523, "validation_loss": 0.41512437864999546, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 7.491124869 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.38314860589065225, "validation_loss": 0.36713147013723113, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 8.549162978 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.3406074100312188, "validation_loss": 0.3310860565229031, "validation_accuracy": 0.8786368260427263, "elapsed_seconds": 9.610155136 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.307799264337199, "validation_loss": 0.3033731649951008, "validation_accuracy": 0.8857578840284842, "elapsed_seconds": 10.689751811 } ], "validation_loss": 0.3033731649951008, "coordinator_cpu_threads": 1, "rate": 0.05, "coordinator_cpu_provider": "openblas", "test_loss": 0.29705317871020637, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332128378295, "fit_seconds": 11.479060799, "training_loss": 0.29234691523654444, "tensor_owned_device_peak_bytes": 0, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 16495216, "dtype": "Float64", "command_seconds": 11.984865174, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float64", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 8.899523592999994, "initialization_seconds": 0, "failed": false, "nodes": [ { "response_read_wait_seconds": 0, "requests": 300, "coordinator_exchange_wall_seconds": 0, "inner_threads": 1, "backend": "openblas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 16495216, "request_write_seconds": 0, "request_payload_bytes": 0, "process_peak_rss_bytes": 172462080, "samples": 458760, "upload_preparation_seconds": 0.3585980489999996, "worker_total_seconds": 8.897904400999998, "zero_sample_steps": 0, "response_payload_bytes": 0, "compute_seconds": 8.265567833999997, "owned_device_tensor_peak_bytes": 0, "device": "cpu", "readback_seconds": 0.032498655, "configuration_id": "atlas-03/C03", "platform": "atlas-03", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu64", "reference_trial": 1, "dtype": "Float64", "predeclared_tolerances": { "parameter_max_abs": 1e-08, "loss_max_abs": 1e-08, "accuracy_max_abs": 1e-12 }, "parameter_max_abs_error": 2.220446049250313e-16, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 0.0, "validation_loss": 5.551115123125783e-17, "test_loss": 5.551115123125783e-17 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "atlas03-cpu64", "repetition": 3, "started_utc": "2026-10-12T21:02:47.627987+00:00", "finished_utc": "2026-10-12T21:03:01.035074+00:00", "outer_command_wall_seconds": 13.406571915998938, "warmup_excluded_from_statistics": false, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 172457984, "test_accuracy": 0.8955239064089522, "parameter_sha256": "0b9cb6964e88ea01eaaf5fc901715b34b184b5c0942cb2e7439c70e1603c1b4c", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8857578840284842, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.3770934749530683, "validation_loss": 1.2565156669823923, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 1.071794846 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.1680702765782573, "validation_loss": 1.0726634152606465, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 2.131054386 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784489967242784, "validation_loss": 0.8820201711707012, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 3.193843265 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923336749857169, "validation_loss": 0.7103083529584678, "validation_accuracy": 0.7502543234994914, "elapsed_seconds": 4.259404454 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370655949431789, "validation_loss": 0.5771756065192747, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 5.337634736 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222779060476093, "validation_loss": 0.4819741321230154, "validation_accuracy": 0.8320447609359105, "elapsed_seconds": 6.428035341 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.4409758238828523, "validation_loss": 0.41512437864999546, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 7.478465969 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.38314860589065225, "validation_loss": 0.36713147013723113, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 8.521499833 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.3406074100312188, "validation_loss": 0.3310860565229031, "validation_accuracy": 0.8786368260427263, "elapsed_seconds": 9.600578049 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.307799264337199, "validation_loss": 0.3033731649951008, "validation_accuracy": 0.8857578840284842, "elapsed_seconds": 10.660213843 } ], "validation_loss": 0.3033731649951008, "coordinator_cpu_threads": 1, "rate": 0.05, "coordinator_cpu_provider": "openblas", "test_loss": 0.29705317871020637, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332128378295, "fit_seconds": 11.436723592, "training_loss": 0.29234691523654444, "tensor_owned_device_peak_bytes": 0, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 16495216, "dtype": "Float64", "command_seconds": 11.936004619, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float64", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 8.887670797000004, "initialization_seconds": 0, "failed": false, "nodes": [ { "response_read_wait_seconds": 0, "requests": 300, "coordinator_exchange_wall_seconds": 0, "inner_threads": 1, "backend": "openblas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 16495216, "request_write_seconds": 0, "request_payload_bytes": 0, "process_peak_rss_bytes": 172457984, "samples": 458760, "upload_preparation_seconds": 0.3564620449999999, "worker_total_seconds": 8.886005835000004, "zero_sample_steps": 0, "response_payload_bytes": 0, "compute_seconds": 8.256238315999996, "owned_device_tensor_peak_bytes": 0, "device": "cpu", "readback_seconds": 0.034002631000000005, "configuration_id": "atlas-03/C03", "platform": "atlas-03", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu64", "reference_trial": 1, "dtype": "Float64", "predeclared_tolerances": { "parameter_max_abs": 1e-08, "loss_max_abs": 1e-08, "accuracy_max_abs": 1e-12 }, "parameter_max_abs_error": 2.220446049250313e-16, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 0.0, "validation_loss": 5.551115123125783e-17, "test_loss": 5.551115123125783e-17 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "atlas01-gpu32", "repetition": 1, "started_utc": "2026-10-12T21:03:05.084207+00:00", "finished_utc": "2026-10-12T21:03:08.121254+00:00", "outer_command_wall_seconds": 3.0367226250091335, "warmup_excluded_from_statistics": false, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 85131264, "test_accuracy": 0.8955239064089522, "parameter_sha256": "e6e47d12ca151780c09565d005880563cd9bf9b214e322e63a6dc6d1df72288a", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8857578840284842, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.3770934714717202, "validation_loss": 1.2565156457625473, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 0.252886875 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.1680702799563316, "validation_loss": 1.072663564672305, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 0.485764708 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784492745375777, "validation_loss": 0.8820205004872529, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 0.739177375 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923334404463672, "validation_loss": 0.7103084890670233, "validation_accuracy": 0.7501525940996948, "elapsed_seconds": 0.988996041 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370660917097257, "validation_loss": 0.577175882655752, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 1.244876583 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222784927962121, "validation_loss": 0.4819752387998063, "validation_accuracy": 0.8319430315361139, "elapsed_seconds": 1.480733541 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.4409768185533219, "validation_loss": 0.41512578622128204, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 1.723208166 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.3831498435645586, "validation_loss": 0.3671321541161891, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 1.971611708 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.34060883298301897, "validation_loss": 0.33108755570825943, "validation_accuracy": 0.8787385554425229, "elapsed_seconds": 2.213722916 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.3078000803066117, "validation_loss": 0.3033739387867773, "validation_accuracy": 0.8857578840284842, "elapsed_seconds": 2.440051666 } ], "validation_loss": 0.3033739387867773, "coordinator_cpu_threads": null, "rate": 0.05, "coordinator_cpu_provider": "apple_accelerate", "test_loss": 0.2970536025106482, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332174875397, "fit_seconds": 2.632902875, "training_loss": 0.29234693766326264, "tensor_owned_device_peak_bytes": 6524456, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 6303328, "dtype": "Float32", "command_seconds": 2.984413792, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float32", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 1.859700832, "initialization_seconds": 0, "failed": false, "nodes": [ { "response_read_wait_seconds": 0, "requests": 300, "coordinator_exchange_wall_seconds": 0, "inner_threads": null, "backend": "metal_mps", "target": "macos-arm64", "owned_host_tensor_peak_bytes": 6303328, "request_write_seconds": 0, "request_payload_bytes": 0, "process_peak_rss_bytes": 85131264, "samples": 458760, "upload_preparation_seconds": 0.23252137699999997, "worker_total_seconds": 1.858832952, "zero_sample_steps": 0, "response_payload_bytes": 0, "compute_seconds": 1.5886115010000001, "owned_device_tensor_peak_bytes": 6524456, "device": "mps", "readback_seconds": 0.015358756000000006, "configuration_id": "atlas-01/C03", "platform": "atlas-01", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu32", "reference_trial": 1, "dtype": "Float32", "predeclared_tolerances": { "parameter_max_abs": 0.001, "loss_max_abs": 0.0005, "accuracy_max_abs": 0.005 }, "parameter_max_abs_error": 4.6253204345703125e-05, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 5.986501557808666e-07, "validation_loss": 4.87175501995285e-07, "test_loss": 3.2672198113647966e-07 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.00010172939979657958, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "atlas01-gpu32", "repetition": 2, "started_utc": "2026-10-12T21:03:08.446101+00:00", "finished_utc": "2026-10-12T21:03:11.663475+00:00", "outer_command_wall_seconds": 3.2152430829883087, "warmup_excluded_from_statistics": false, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 87965696, "test_accuracy": 0.8955239064089522, "parameter_sha256": "e6e47d12ca151780c09565d005880563cd9bf9b214e322e63a6dc6d1df72288a", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8857578840284842, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.3770934714717202, "validation_loss": 1.2565156457625473, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 0.256816875 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.1680702799563316, "validation_loss": 1.072663564672305, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 0.491161166 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784492745375777, "validation_loss": 0.8820205004872529, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 0.765338833 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923334404463672, "validation_loss": 0.7103084890670233, "validation_accuracy": 0.7501525940996948, "elapsed_seconds": 1.00178475 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370660917097257, "validation_loss": 0.577175882655752, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 1.27094675 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222784927962121, "validation_loss": 0.4819752387998063, "validation_accuracy": 0.8319430315361139, "elapsed_seconds": 1.557315958 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.4409768185533219, "validation_loss": 0.41512578622128204, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 1.826810833 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.3831498435645586, "validation_loss": 0.3671321541161891, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 2.094058375 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.34060883298301897, "validation_loss": 0.33108755570825943, "validation_accuracy": 0.8787385554425229, "elapsed_seconds": 2.353146583 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.3078000803066117, "validation_loss": 0.3033739387867773, "validation_accuracy": 0.8857578840284842, "elapsed_seconds": 2.621128416 } ], "validation_loss": 0.3033739387867773, "coordinator_cpu_threads": null, "rate": 0.05, "coordinator_cpu_provider": "apple_accelerate", "test_loss": 0.2970536025106482, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332174875397, "fit_seconds": 2.804163125, "training_loss": 0.29234693766326264, "tensor_owned_device_peak_bytes": 6524456, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 6303328, "dtype": "Float32", "command_seconds": 3.161081833, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float32", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 1.9930527430000007, "initialization_seconds": 0, "failed": false, "nodes": [ { "response_read_wait_seconds": 0, "requests": 300, "coordinator_exchange_wall_seconds": 0, "inner_threads": null, "backend": "metal_mps", "target": "macos-arm64", "owned_host_tensor_peak_bytes": 6303328, "request_write_seconds": 0, "request_payload_bytes": 0, "process_peak_rss_bytes": 87965696, "samples": 458760, "upload_preparation_seconds": 0.24405004799999996, "worker_total_seconds": 1.9918613789999977, "zero_sample_steps": 0, "response_payload_bytes": 0, "compute_seconds": 1.7054683319999997, "owned_device_tensor_peak_bytes": 6524456, "device": "mps", "readback_seconds": 0.017307832999999998, "configuration_id": "atlas-01/C03", "platform": "atlas-01", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu32", "reference_trial": 1, "dtype": "Float32", "predeclared_tolerances": { "parameter_max_abs": 0.001, "loss_max_abs": 0.0005, "accuracy_max_abs": 0.005 }, "parameter_max_abs_error": 4.6253204345703125e-05, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 5.986501557808666e-07, "validation_loss": 4.87175501995285e-07, "test_loss": 3.2672198113647966e-07 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.00010172939979657958, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "atlas01-gpu32", "repetition": 3, "started_utc": "2026-10-12T21:03:12.056928+00:00", "finished_utc": "2026-10-12T21:03:15.312411+00:00", "outer_command_wall_seconds": 3.255333624998457, "warmup_excluded_from_statistics": false, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 84983808, "test_accuracy": 0.8955239064089522, "parameter_sha256": "e6e47d12ca151780c09565d005880563cd9bf9b214e322e63a6dc6d1df72288a", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8857578840284842, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.3770934714717202, "validation_loss": 1.2565156457625473, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 0.289857 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.1680702799563316, "validation_loss": 1.072663564672305, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 0.573041 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784492745375777, "validation_loss": 0.8820205004872529, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 0.8536545 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923334404463672, "validation_loss": 0.7103084890670233, "validation_accuracy": 0.7501525940996948, "elapsed_seconds": 1.1317945 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370660917097257, "validation_loss": 0.577175882655752, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 1.378268416 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222784927962121, "validation_loss": 0.4819752387998063, "validation_accuracy": 0.8319430315361139, "elapsed_seconds": 1.623989833 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.4409768185533219, "validation_loss": 0.41512578622128204, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 1.878510375 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.3831498435645586, "validation_loss": 0.3671321541161891, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 2.118405083 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.34060883298301897, "validation_loss": 0.33108755570825943, "validation_accuracy": 0.8787385554425229, "elapsed_seconds": 2.354885708 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.3078000803066117, "validation_loss": 0.3033739387867773, "validation_accuracy": 0.8857578840284842, "elapsed_seconds": 2.623815541 } ], "validation_loss": 0.3033739387867773, "coordinator_cpu_threads": null, "rate": 0.05, "coordinator_cpu_provider": "apple_accelerate", "test_loss": 0.2970536025106482, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332174875397, "fit_seconds": 2.816833625, "training_loss": 0.29234693766326264, "tensor_owned_device_peak_bytes": 6524456, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 6303328, "dtype": "Float32", "command_seconds": 3.189324208, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float32", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 2.0105638719999988, "initialization_seconds": 0, "failed": false, "nodes": [ { "response_read_wait_seconds": 0, "requests": 300, "coordinator_exchange_wall_seconds": 0, "inner_threads": null, "backend": "metal_mps", "target": "macos-arm64", "owned_host_tensor_peak_bytes": 6303328, "request_write_seconds": 0, "request_payload_bytes": 0, "process_peak_rss_bytes": 84983808, "samples": 458760, "upload_preparation_seconds": 0.250555169, "worker_total_seconds": 2.0093898709999993, "zero_sample_steps": 0, "response_payload_bytes": 0, "compute_seconds": 1.7175838759999995, "owned_device_tensor_peak_bytes": 6524456, "device": "mps", "readback_seconds": 0.016662950999999995, "configuration_id": "atlas-01/C03", "platform": "atlas-01", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu32", "reference_trial": 1, "dtype": "Float32", "predeclared_tolerances": { "parameter_max_abs": 0.001, "loss_max_abs": 0.0005, "accuracy_max_abs": 0.005 }, "parameter_max_abs_error": 4.6253204345703125e-05, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 5.986501557808666e-07, "validation_loss": 4.87175501995285e-07, "test_loss": 3.2672198113647966e-07 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.00010172939979657958, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "atlas02-gpu32", "repetition": 1, "started_utc": "2026-10-12T21:03:25.622379+00:00", "finished_utc": "2026-10-12T21:03:34.393508+00:00", "outer_command_wall_seconds": 8.770283082994865, "warmup_excluded_from_statistics": false, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 296456192, "test_accuracy": 0.8955239064089522, "parameter_sha256": "ed951a02e2b2e52170c840ab714347f81580d9334a40d7bbb4d5bc5024e36067", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8857578840284842, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.377093479454346, "validation_loss": 1.2565156643897568, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 0.465571376 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.1680702799563316, "validation_loss": 1.072663564672305, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 0.934999141 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784492688156565, "validation_loss": 0.8820204688234222, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 1.40522902 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923333773649147, "validation_loss": 0.7103079675415404, "validation_accuracy": 0.7500508646998982, "elapsed_seconds": 1.881653461 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370662267595382, "validation_loss": 0.5771763893740587, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 2.351299712 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222790034815743, "validation_loss": 0.4819753710275264, "validation_accuracy": 0.8318413021363174, "elapsed_seconds": 2.805284206 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.4409766082142524, "validation_loss": 0.4151265396652241, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 3.256341595 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.38315067391354407, "validation_loss": 0.3671332475067034, "validation_accuracy": 0.868056968463886, "elapsed_seconds": 3.712884242 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.3406097975944681, "validation_loss": 0.3310883492989312, "validation_accuracy": 0.8787385554425229, "elapsed_seconds": 4.175576616 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.3078013175200518, "validation_loss": 0.30337606328913785, "validation_accuracy": 0.8857578840284842, "elapsed_seconds": 4.668493368 } ], "validation_loss": 0.30337606328913785, "coordinator_cpu_threads": 1, "rate": 0.05, "coordinator_cpu_provider": "openblas", "test_loss": 0.29705589171101093, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.599133218017636, "fit_seconds": 5.904053987, "training_loss": 0.2923495870296973, "tensor_owned_device_peak_bytes": 6524456, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 6303328, "dtype": "Float32", "command_seconds": 7.097920784, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float32", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 1.6630517060000007, "initialization_seconds": 0, "failed": false, "nodes": [ { "response_read_wait_seconds": 0, "requests": 300, "coordinator_exchange_wall_seconds": 0, "inner_threads": 1, "backend": "cuda_cublas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 6303328, "request_write_seconds": 0, "request_payload_bytes": 0, "process_peak_rss_bytes": 296456192, "samples": 458760, "upload_preparation_seconds": 0.9662449480000002, "worker_total_seconds": 1.6604763869999994, "zero_sample_steps": 0, "response_payload_bytes": 0, "compute_seconds": 0.48920735600000026, "owned_device_tensor_peak_bytes": 6524456, "device": "cuda", "readback_seconds": 0.06735880900000002, "configuration_id": "atlas-02/C03", "platform": "atlas-02", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu32", "reference_trial": 1, "dtype": "Float32", "predeclared_tolerances": { "parameter_max_abs": 0.001, "loss_max_abs": 0.0005, "accuracy_max_abs": 0.005 }, "parameter_max_abs_error": 5.9798359870910645e-05, "loss_abs_errors": { "initial_loss": 5.300964112109341e-10, "training_loss": 2.050716278900744e-06, "validation_loss": 1.6373268585390122e-06, "test_loss": 1.9624783815785918e-06 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.00010172939979657958, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "atlas02-gpu32", "repetition": 2, "started_utc": "2026-10-12T21:03:34.789131+00:00", "finished_utc": "2026-10-12T21:03:43.304632+00:00", "outer_command_wall_seconds": 8.515092125002411, "warmup_excluded_from_statistics": false, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 296271872, "test_accuracy": 0.8955239064089522, "parameter_sha256": "ed951a02e2b2e52170c840ab714347f81580d9334a40d7bbb4d5bc5024e36067", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8857578840284842, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.377093479454346, "validation_loss": 1.2565156643897568, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 0.473458608 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.1680702799563316, "validation_loss": 1.072663564672305, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 0.963795649 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784492688156565, "validation_loss": 0.8820204688234222, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 1.419911856 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923333773649147, "validation_loss": 0.7103079675415404, "validation_accuracy": 0.7500508646998982, "elapsed_seconds": 1.904136549 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370662267595382, "validation_loss": 0.5771763893740587, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 2.382232027 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222790034815743, "validation_loss": 0.4819753710275264, "validation_accuracy": 0.8318413021363174, "elapsed_seconds": 2.849556786 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.4409766082142524, "validation_loss": 0.4151265396652241, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 3.318392737 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.38315067391354407, "validation_loss": 0.3671332475067034, "validation_accuracy": 0.868056968463886, "elapsed_seconds": 3.787251413 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.3406097975944681, "validation_loss": 0.3310883492989312, "validation_accuracy": 0.8787385554425229, "elapsed_seconds": 4.26051947 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.3078013175200518, "validation_loss": 0.30337606328913785, "validation_accuracy": 0.8857578840284842, "elapsed_seconds": 4.727385903 } ], "validation_loss": 0.30337606328913785, "coordinator_cpu_threads": 1, "rate": 0.05, "coordinator_cpu_provider": "openblas", "test_loss": 0.29705589171101093, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.599133218017636, "fit_seconds": 5.944281978, "training_loss": 0.2923495870296973, "tensor_owned_device_peak_bytes": 6524456, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 6303328, "dtype": "Float32", "command_seconds": 7.032981553, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float32", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 1.6869914419999994, "initialization_seconds": 0, "failed": false, "nodes": [ { "response_read_wait_seconds": 0, "requests": 300, "coordinator_exchange_wall_seconds": 0, "inner_threads": 1, "backend": "cuda_cublas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 6303328, "request_write_seconds": 0, "request_payload_bytes": 0, "process_peak_rss_bytes": 296271872, "samples": 458760, "upload_preparation_seconds": 0.9739203370000002, "worker_total_seconds": 1.6844404309999996, "zero_sample_steps": 0, "response_payload_bytes": 0, "compute_seconds": 0.5040637470000001, "owned_device_tensor_peak_bytes": 6524456, "device": "cuda", "readback_seconds": 0.06745282600000004, "configuration_id": "atlas-02/C03", "platform": "atlas-02", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu32", "reference_trial": 1, "dtype": "Float32", "predeclared_tolerances": { "parameter_max_abs": 0.001, "loss_max_abs": 0.0005, "accuracy_max_abs": 0.005 }, "parameter_max_abs_error": 5.9798359870910645e-05, "loss_abs_errors": { "initial_loss": 5.300964112109341e-10, "training_loss": 2.050716278900744e-06, "validation_loss": 1.6373268585390122e-06, "test_loss": 1.9624783815785918e-06 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.00010172939979657958, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "atlas02-gpu32", "repetition": 3, "started_utc": "2026-10-12T21:03:43.705667+00:00", "finished_utc": "2026-10-12T21:03:52.398508+00:00", "outer_command_wall_seconds": 8.69213504200161, "warmup_excluded_from_statistics": false, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 296333312, "test_accuracy": 0.8955239064089522, "parameter_sha256": "ed951a02e2b2e52170c840ab714347f81580d9334a40d7bbb4d5bc5024e36067", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8857578840284842, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.377093479454346, "validation_loss": 1.2565156643897568, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 0.465863937 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.1680702799563316, "validation_loss": 1.072663564672305, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 0.935114345 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784492688156565, "validation_loss": 0.8820204688234222, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 1.39877092 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923333773649147, "validation_loss": 0.7103079675415404, "validation_accuracy": 0.7500508646998982, "elapsed_seconds": 1.888328347 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370662267595382, "validation_loss": 0.5771763893740587, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 2.40171737 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222790034815743, "validation_loss": 0.4819753710275264, "validation_accuracy": 0.8318413021363174, "elapsed_seconds": 2.918954876 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.4409766082142524, "validation_loss": 0.4151265396652241, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 3.409875012 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.38315067391354407, "validation_loss": 0.3671332475067034, "validation_accuracy": 0.868056968463886, "elapsed_seconds": 3.88229768 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.3406097975944681, "validation_loss": 0.3310883492989312, "validation_accuracy": 0.8787385554425229, "elapsed_seconds": 4.347639132 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.3078013175200518, "validation_loss": 0.30337606328913785, "validation_accuracy": 0.8857578840284842, "elapsed_seconds": 4.814185974 } ], "validation_loss": 0.30337606328913785, "coordinator_cpu_threads": 1, "rate": 0.05, "coordinator_cpu_provider": "openblas", "test_loss": 0.29705589171101093, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.599133218017636, "fit_seconds": 6.029027588, "training_loss": 0.2923495870296973, "tensor_owned_device_peak_bytes": 6524456, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 6303328, "dtype": "Float32", "command_seconds": 7.155840428, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float32", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 1.7054734730000012, "initialization_seconds": 0, "failed": false, "nodes": [ { "response_read_wait_seconds": 0, "requests": 300, "coordinator_exchange_wall_seconds": 0, "inner_threads": 1, "backend": "cuda_cublas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 6303328, "request_write_seconds": 0, "request_payload_bytes": 0, "process_peak_rss_bytes": 296333312, "samples": 458760, "upload_preparation_seconds": 0.9876225599999999, "worker_total_seconds": 1.703028964, "zero_sample_steps": 0, "response_payload_bytes": 0, "compute_seconds": 0.5048317030000001, "owned_device_tensor_peak_bytes": 6524456, "device": "cuda", "readback_seconds": 0.07029516799999998, "configuration_id": "atlas-02/C03", "platform": "atlas-02", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu32", "reference_trial": 1, "dtype": "Float32", "predeclared_tolerances": { "parameter_max_abs": 0.001, "loss_max_abs": 0.0005, "accuracy_max_abs": 0.005 }, "parameter_max_abs_error": 5.9798359870910645e-05, "loss_abs_errors": { "initial_loss": 5.300964112109341e-10, "training_loss": 2.050716278900744e-06, "validation_loss": 1.6373268585390122e-06, "test_loss": 1.9624783815785918e-06 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.00010172939979657958, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "atlas02-gpu64", "repetition": 1, "started_utc": "2026-10-12T21:04:01.702944+00:00", "finished_utc": "2026-10-12T21:04:11.429968+00:00", "outer_command_wall_seconds": 9.726846584002487, "warmup_excluded_from_statistics": false, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 300347392, "test_accuracy": 0.8955239064089522, "parameter_sha256": "a238a137db9084d9bbc52207f21c90e2e4fd813de1533ee1fecc17d7f66c0b55", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8857578840284842, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.3770934749530683, "validation_loss": 1.2565156669823923, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 0.507822061 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.1680702765782571, "validation_loss": 1.0726634152606465, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 1.008190422 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784489967242784, "validation_loss": 0.8820201711707012, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 1.524027071 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923336749857173, "validation_loss": 0.7103083529584678, "validation_accuracy": 0.7502543234994914, "elapsed_seconds": 2.294527743 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370655949431792, "validation_loss": 0.5771756065192747, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 2.81338678 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222779060476093, "validation_loss": 0.4819741321230154, "validation_accuracy": 0.8320447609359105, "elapsed_seconds": 3.324302517 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.4409758238828523, "validation_loss": 0.4151243786499954, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 3.845401352 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.38314860589065225, "validation_loss": 0.36713147013723113, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 4.371046139 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.3406074100312188, "validation_loss": 0.3310860565229031, "validation_accuracy": 0.8786368260427263, "elapsed_seconds": 4.924747742 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.3077992643371988, "validation_loss": 0.3033731649951009, "validation_accuracy": 0.8857578840284842, "elapsed_seconds": 5.427835988 } ], "validation_loss": 0.3033731649951009, "coordinator_cpu_threads": 1, "rate": 0.05, "coordinator_cpu_provider": "openblas", "test_loss": 0.29705317871020637, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332128378295, "fit_seconds": 6.5129025, "training_loss": 0.29234691523654444, "tensor_owned_device_peak_bytes": 13048912, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 4202560, "dtype": "Float64", "command_seconds": 7.993388733, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float64", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 2.578302544, "initialization_seconds": 0, "failed": false, "nodes": [ { "response_read_wait_seconds": 0, "requests": 300, "coordinator_exchange_wall_seconds": 0, "inner_threads": 1, "backend": "cuda_cublas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 4202560, "request_write_seconds": 0, "request_payload_bytes": 0, "process_peak_rss_bytes": 300347392, "samples": 458760, "upload_preparation_seconds": 0.8554335600000007, "worker_total_seconds": 2.575395199, "zero_sample_steps": 0, "response_payload_bytes": 0, "compute_seconds": 1.3179493970000007, "owned_device_tensor_peak_bytes": 13048912, "device": "cuda", "readback_seconds": 0.11018652099999998, "configuration_id": "atlas-02/C03", "platform": "atlas-02", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu64", "reference_trial": 1, "dtype": "Float64", "predeclared_tolerances": { "parameter_max_abs": 1e-08, "loss_max_abs": 1e-08, "accuracy_max_abs": 1e-12 }, "parameter_max_abs_error": 3.3306690738754696e-16, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 0.0, "validation_loss": 5.551115123125783e-17, "test_loss": 5.551115123125783e-17 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "atlas02-gpu64", "repetition": 2, "started_utc": "2026-10-12T21:04:11.821580+00:00", "finished_utc": "2026-10-12T21:04:20.618716+00:00", "outer_command_wall_seconds": 8.796772208006587, "warmup_excluded_from_statistics": false, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 300224512, "test_accuracy": 0.8955239064089522, "parameter_sha256": "a238a137db9084d9bbc52207f21c90e2e4fd813de1533ee1fecc17d7f66c0b55", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8857578840284842, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.3770934749530683, "validation_loss": 1.2565156669823923, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 0.481565866 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.1680702765782571, "validation_loss": 1.0726634152606465, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 0.94919121 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784489967242784, "validation_loss": 0.8820201711707012, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 1.431449564 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923336749857173, "validation_loss": 0.7103083529584678, "validation_accuracy": 0.7502543234994914, "elapsed_seconds": 1.902681174 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370655949431792, "validation_loss": 0.5771756065192747, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 2.38230669 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222779060476093, "validation_loss": 0.4819741321230154, "validation_accuracy": 0.8320447609359105, "elapsed_seconds": 2.850148787 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.4409758238828523, "validation_loss": 0.4151243786499954, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 3.327765272 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.38314860589065225, "validation_loss": 0.36713147013723113, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 3.793254255 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.3406074100312188, "validation_loss": 0.3310860565229031, "validation_accuracy": 0.8786368260427263, "elapsed_seconds": 4.29353039 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.3077992643371988, "validation_loss": 0.3033731649951009, "validation_accuracy": 0.8857578840284842, "elapsed_seconds": 4.76968277 } ], "validation_loss": 0.3033731649951009, "coordinator_cpu_threads": 1, "rate": 0.05, "coordinator_cpu_provider": "openblas", "test_loss": 0.29705317871020637, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332128378295, "fit_seconds": 5.843198569, "training_loss": 0.29234691523654444, "tensor_owned_device_peak_bytes": 13048912, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 4202560, "dtype": "Float64", "command_seconds": 7.217335922, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float64", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 2.260612975, "initialization_seconds": 0, "failed": false, "nodes": [ { "response_read_wait_seconds": 0, "requests": 300, "coordinator_exchange_wall_seconds": 0, "inner_threads": 1, "backend": "cuda_cublas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 4202560, "request_write_seconds": 0, "request_payload_bytes": 0, "process_peak_rss_bytes": 300224512, "samples": 458760, "upload_preparation_seconds": 0.6621139690000001, "worker_total_seconds": 2.258230271999999, "zero_sample_steps": 0, "response_payload_bytes": 0, "compute_seconds": 1.2547439029999992, "owned_device_tensor_peak_bytes": 13048912, "device": "cuda", "readback_seconds": 0.07557949400000004, "configuration_id": "atlas-02/C03", "platform": "atlas-02", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu64", "reference_trial": 1, "dtype": "Float64", "predeclared_tolerances": { "parameter_max_abs": 1e-08, "loss_max_abs": 1e-08, "accuracy_max_abs": 1e-12 }, "parameter_max_abs_error": 3.3306690738754696e-16, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 0.0, "validation_loss": 5.551115123125783e-17, "test_loss": 5.551115123125783e-17 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "atlas02-gpu64", "repetition": 3, "started_utc": "2026-10-12T21:04:21.060961+00:00", "finished_utc": "2026-10-12T21:04:30.570077+00:00", "outer_command_wall_seconds": 9.508705583997653, "warmup_excluded_from_statistics": false, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 300429312, "test_accuracy": 0.8955239064089522, "parameter_sha256": "a238a137db9084d9bbc52207f21c90e2e4fd813de1533ee1fecc17d7f66c0b55", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8857578840284842, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.3770934749530683, "validation_loss": 1.2565156669823923, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 0.4754791 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.1680702765782571, "validation_loss": 1.0726634152606465, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 0.946100071 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784489967242784, "validation_loss": 0.8820201711707012, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 1.423539712 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923336749857173, "validation_loss": 0.7103083529584678, "validation_accuracy": 0.7502543234994914, "elapsed_seconds": 1.886998771 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370655949431792, "validation_loss": 0.5771756065192747, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 2.396571495 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222779060476093, "validation_loss": 0.4819741321230154, "validation_accuracy": 0.8320447609359105, "elapsed_seconds": 2.904540216 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.4409758238828523, "validation_loss": 0.4151243786499954, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 3.383123618 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.38314860589065225, "validation_loss": 0.36713147013723113, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 3.847796099 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.3406074100312188, "validation_loss": 0.3310860565229031, "validation_accuracy": 0.8786368260427263, "elapsed_seconds": 4.324883319 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.3077992643371988, "validation_loss": 0.3033731649951009, "validation_accuracy": 0.8857578840284842, "elapsed_seconds": 4.834376932 } ], "validation_loss": 0.3033731649951009, "coordinator_cpu_threads": 1, "rate": 0.05, "coordinator_cpu_provider": "openblas", "test_loss": 0.29705317871020637, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332128378295, "fit_seconds": 5.958052482, "training_loss": 0.29234691523654444, "tensor_owned_device_peak_bytes": 13048912, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 4202560, "dtype": "Float64", "command_seconds": 7.141771782, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float64", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 2.2630139600000003, "initialization_seconds": 0, "failed": false, "nodes": [ { "response_read_wait_seconds": 0, "requests": 300, "coordinator_exchange_wall_seconds": 0, "inner_threads": 1, "backend": "cuda_cublas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 4202560, "request_write_seconds": 0, "request_payload_bytes": 0, "process_peak_rss_bytes": 300429312, "samples": 458760, "upload_preparation_seconds": 0.6689899059999999, "worker_total_seconds": 2.260629614, "zero_sample_steps": 0, "response_payload_bytes": 0, "compute_seconds": 1.2478117600000003, "owned_device_tensor_peak_bytes": 13048912, "device": "cuda", "readback_seconds": 0.075353733, "configuration_id": "atlas-02/C03", "platform": "atlas-02", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu64", "reference_trial": 1, "dtype": "Float64", "predeclared_tolerances": { "parameter_max_abs": 1e-08, "loss_max_abs": 1e-08, "accuracy_max_abs": 1e-12 }, "parameter_max_abs_error": 3.3306690738754696e-16, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 0.0, "validation_loss": 5.551115123125783e-17, "test_loss": 5.551115123125783e-17 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "atlas03-gpu32", "repetition": 1, "started_utc": "2026-10-12T21:04:38.092456+00:00", "finished_utc": "2026-10-12T21:04:44.114762+00:00", "outer_command_wall_seconds": 6.021835834006197, "warmup_excluded_from_statistics": false, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 331165696, "test_accuracy": 0.8955239064089522, "parameter_sha256": "3492d0395ffb53158e162126537ba6af8c4f47d5180ead405ea211d6f07652de", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8858596134282808, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.3770934749329367, "validation_loss": 1.2565156643897568, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 0.317831869 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.1680702879389575, "validation_loss": 1.072663564672305, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 0.639270692 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784492964897984, "validation_loss": 0.8820205004872529, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 0.955844729 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923334564116187, "validation_loss": 0.7103084927900398, "validation_accuracy": 0.7501525940996948, "elapsed_seconds": 1.271743293 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370661159226509, "validation_loss": 0.577175882655752, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 1.583671719 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222784667825188, "validation_loss": 0.4819751270426132, "validation_accuracy": 0.8319430315361139, "elapsed_seconds": 1.900098152 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.4409767470448983, "validation_loss": 0.41512573871947256, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 2.21133266 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.3831498184941247, "validation_loss": 0.36713236366623175, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 2.525144359 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.34060897462005674, "validation_loss": 0.3310880763484648, "validation_accuracy": 0.8787385554425229, "elapsed_seconds": 2.841396184 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.30780011722625567, "validation_loss": 0.30337433190457336, "validation_accuracy": 0.8858596134282808, "elapsed_seconds": 3.155168061 } ], "validation_loss": 0.30337433190457336, "coordinator_cpu_threads": 1, "rate": 0.05, "coordinator_cpu_provider": "openblas", "test_loss": 0.2970538742908505, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332140263232, "fit_seconds": 4.025347966, "training_loss": 0.2923475264676603, "tensor_owned_device_peak_bytes": 6524456, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 6303328, "dtype": "Float32", "command_seconds": 4.631889841, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float32", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 1.0757496440000003, "initialization_seconds": 0, "failed": false, "nodes": [ { "response_read_wait_seconds": 0, "requests": 300, "coordinator_exchange_wall_seconds": 0, "inner_threads": 1, "backend": "cuda_cublas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 6303328, "request_write_seconds": 0, "request_payload_bytes": 0, "process_peak_rss_bytes": 331165696, "samples": 458760, "upload_preparation_seconds": 0.6446049009999999, "worker_total_seconds": 1.0742639050000002, "zero_sample_steps": 0, "response_payload_bytes": 0, "compute_seconds": 0.2802974630000002, "owned_device_tensor_peak_bytes": 6524456, "device": "cuda", "readback_seconds": 0.05001958200000001, "configuration_id": "atlas-03/C03", "platform": "atlas-03", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu32", "reference_trial": 1, "dtype": "Float32", "predeclared_tolerances": { "parameter_max_abs": 0.001, "loss_max_abs": 0.0005, "accuracy_max_abs": 0.005 }, "parameter_max_abs_error": 4.561617970466614e-06, "loss_abs_errors": { "initial_loss": 3.461216424582858e-09, "training_loss": 9.84575809770405e-09, "validation_loss": 9.405770595716234e-08, "test_loss": 5.494177884868279e-08 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "atlas03-gpu32", "repetition": 2, "started_utc": "2026-10-12T21:04:44.582221+00:00", "finished_utc": "2026-10-12T21:04:51.181922+00:00", "outer_command_wall_seconds": 6.598680917013553, "warmup_excluded_from_statistics": false, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 331112448, "test_accuracy": 0.8955239064089522, "parameter_sha256": "3492d0395ffb53158e162126537ba6af8c4f47d5180ead405ea211d6f07652de", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8858596134282808, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.3770934749329367, "validation_loss": 1.2565156643897568, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 0.316679754 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.1680702879389575, "validation_loss": 1.072663564672305, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 0.633138812 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784492964897984, "validation_loss": 0.8820205004872529, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 0.948193767 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923334564116187, "validation_loss": 0.7103084927900398, "validation_accuracy": 0.7501525940996948, "elapsed_seconds": 1.268483754 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370661159226509, "validation_loss": 0.577175882655752, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 1.584324315 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222784667825188, "validation_loss": 0.4819751270426132, "validation_accuracy": 0.8319430315361139, "elapsed_seconds": 1.901190611 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.4409767470448983, "validation_loss": 0.41512573871947256, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 2.217234942 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.3831498184941247, "validation_loss": 0.36713236366623175, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 2.536832664 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.34060897462005674, "validation_loss": 0.3310880763484648, "validation_accuracy": 0.8787385554425229, "elapsed_seconds": 2.85430091 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.30780011722625567, "validation_loss": 0.30337433190457336, "validation_accuracy": 0.8858596134282808, "elapsed_seconds": 3.168158778 } ], "validation_loss": 0.30337433190457336, "coordinator_cpu_threads": 1, "rate": 0.05, "coordinator_cpu_provider": "openblas", "test_loss": 0.2970538742908505, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332140263232, "fit_seconds": 4.045879262, "training_loss": 0.2923475264676603, "tensor_owned_device_peak_bytes": 6524456, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 6303328, "dtype": "Float32", "command_seconds": 4.670013525, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float32", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 1.0883052319999986, "initialization_seconds": 0, "failed": false, "nodes": [ { "response_read_wait_seconds": 0, "requests": 300, "coordinator_exchange_wall_seconds": 0, "inner_threads": 1, "backend": "cuda_cublas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 6303328, "request_write_seconds": 0, "request_payload_bytes": 0, "process_peak_rss_bytes": 331112448, "samples": 458760, "upload_preparation_seconds": 0.6526943540000001, "worker_total_seconds": 1.086779123, "zero_sample_steps": 0, "response_payload_bytes": 0, "compute_seconds": 0.28371388599999997, "owned_device_tensor_peak_bytes": 6524456, "device": "cuda", "readback_seconds": 0.051010612000000004, "configuration_id": "atlas-03/C03", "platform": "atlas-03", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu32", "reference_trial": 1, "dtype": "Float32", "predeclared_tolerances": { "parameter_max_abs": 0.001, "loss_max_abs": 0.0005, "accuracy_max_abs": 0.005 }, "parameter_max_abs_error": 4.561617970466614e-06, "loss_abs_errors": { "initial_loss": 3.461216424582858e-09, "training_loss": 9.84575809770405e-09, "validation_loss": 9.405770595716234e-08, "test_loss": 5.494177884868279e-08 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "atlas03-gpu32", "repetition": 3, "started_utc": "2026-10-12T21:04:51.700725+00:00", "finished_utc": "2026-10-12T21:04:57.814991+00:00", "outer_command_wall_seconds": 6.113490791001823, "warmup_excluded_from_statistics": false, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 331001856, "test_accuracy": 0.8955239064089522, "parameter_sha256": "3492d0395ffb53158e162126537ba6af8c4f47d5180ead405ea211d6f07652de", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8858596134282808, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.3770934749329367, "validation_loss": 1.2565156643897568, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 0.313996708 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.1680702879389575, "validation_loss": 1.072663564672305, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 0.626875807 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784492964897984, "validation_loss": 0.8820205004872529, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 0.945155166 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923334564116187, "validation_loss": 0.7103084927900398, "validation_accuracy": 0.7501525940996948, "elapsed_seconds": 1.263566426 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370661159226509, "validation_loss": 0.577175882655752, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 1.579234961 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222784667825188, "validation_loss": 0.4819751270426132, "validation_accuracy": 0.8319430315361139, "elapsed_seconds": 1.896999887 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.4409767470448983, "validation_loss": 0.41512573871947256, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 2.212340967 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.3831498184941247, "validation_loss": 0.36713236366623175, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 2.532778174 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.34060897462005674, "validation_loss": 0.3310880763484648, "validation_accuracy": 0.8787385554425229, "elapsed_seconds": 2.846151504 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.30780011722625567, "validation_loss": 0.30337433190457336, "validation_accuracy": 0.8858596134282808, "elapsed_seconds": 3.163349836 } ], "validation_loss": 0.30337433190457336, "coordinator_cpu_threads": 1, "rate": 0.05, "coordinator_cpu_provider": "openblas", "test_loss": 0.2970538742908505, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332140263232, "fit_seconds": 4.041986648, "training_loss": 0.2923475264676603, "tensor_owned_device_peak_bytes": 6524456, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 6303328, "dtype": "Float32", "command_seconds": 4.649221081, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float32", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 1.0837230520000005, "initialization_seconds": 0, "failed": false, "nodes": [ { "response_read_wait_seconds": 0, "requests": 300, "coordinator_exchange_wall_seconds": 0, "inner_threads": 1, "backend": "cuda_cublas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 6303328, "request_write_seconds": 0, "request_payload_bytes": 0, "process_peak_rss_bytes": 331001856, "samples": 458760, "upload_preparation_seconds": 0.6535847469999994, "worker_total_seconds": 1.0821805899999999, "zero_sample_steps": 0, "response_payload_bytes": 0, "compute_seconds": 0.2801538039999999, "owned_device_tensor_peak_bytes": 6524456, "device": "cuda", "readback_seconds": 0.049093760000000014, "configuration_id": "atlas-03/C03", "platform": "atlas-03", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu32", "reference_trial": 1, "dtype": "Float32", "predeclared_tolerances": { "parameter_max_abs": 0.001, "loss_max_abs": 0.0005, "accuracy_max_abs": 0.005 }, "parameter_max_abs_error": 4.561617970466614e-06, "loss_abs_errors": { "initial_loss": 3.461216424582858e-09, "training_loss": 9.84575809770405e-09, "validation_loss": 9.405770595716234e-08, "test_loss": 5.494177884868279e-08 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "atlas03-gpu64", "repetition": 1, "started_utc": "2026-10-12T21:05:04.365515+00:00", "finished_utc": "2026-10-12T21:05:09.806364+00:00", "outer_command_wall_seconds": 5.440311792001012, "warmup_excluded_from_statistics": false, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 324931584, "test_accuracy": 0.8955239064089522, "parameter_sha256": "558baffdc49fc87ee67d04a9cb17c77daf1671a544a5eafcd137576192b5a10e", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8857578840284842, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.3770934749530683, "validation_loss": 1.2565156669823923, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 0.277215312 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.1680702765782571, "validation_loss": 1.0726634152606465, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 0.548080624 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784489967242784, "validation_loss": 0.8820201711707012, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 0.817333068 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923336749857173, "validation_loss": 0.710308352958468, "validation_accuracy": 0.7502543234994914, "elapsed_seconds": 1.088850252 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370655949431792, "validation_loss": 0.5771756065192747, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 1.359913745 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222779060476093, "validation_loss": 0.4819741321230154, "validation_accuracy": 0.8320447609359105, "elapsed_seconds": 1.634895538 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.44097582388285234, "validation_loss": 0.41512437864999546, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 1.910043561 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.38314860589065225, "validation_loss": 0.36713147013723113, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 2.176309003 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.3406074100312188, "validation_loss": 0.3310860565229031, "validation_accuracy": 0.8786368260427263, "elapsed_seconds": 2.442637808 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.3077992643371988, "validation_loss": 0.3033731649951008, "validation_accuracy": 0.8857578840284842, "elapsed_seconds": 2.710004961 } ], "validation_loss": 0.3033731649951008, "coordinator_cpu_threads": 1, "rate": 0.05, "coordinator_cpu_provider": "openblas", "test_loss": 0.29705317871020637, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332128378295, "fit_seconds": 3.423176061, "training_loss": 0.29234691523654444, "tensor_owned_device_peak_bytes": 13048912, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 4202560, "dtype": "Float64", "command_seconds": 4.007730675, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float64", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 1.0366025189999992, "initialization_seconds": 0, "failed": false, "nodes": [ { "response_read_wait_seconds": 0, "requests": 300, "coordinator_exchange_wall_seconds": 0, "inner_threads": 1, "backend": "cuda_cublas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 4202560, "request_write_seconds": 0, "request_payload_bytes": 0, "process_peak_rss_bytes": 324931584, "samples": 458760, "upload_preparation_seconds": 0.41219602099999997, "worker_total_seconds": 1.0351131310000008, "zero_sample_steps": 0, "response_payload_bytes": 0, "compute_seconds": 0.4147275310000002, "owned_device_tensor_peak_bytes": 13048912, "device": "cuda", "readback_seconds": 0.05594505700000002, "configuration_id": "atlas-03/C03", "platform": "atlas-03", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu64", "reference_trial": 1, "dtype": "Float64", "predeclared_tolerances": { "parameter_max_abs": 1e-08, "loss_max_abs": 1e-08, "accuracy_max_abs": 1e-12 }, "parameter_max_abs_error": 3.3306690738754696e-16, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 0.0, "validation_loss": 5.551115123125783e-17, "test_loss": 5.551115123125783e-17 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "atlas03-gpu64", "repetition": 2, "started_utc": "2026-10-12T21:05:10.313192+00:00", "finished_utc": "2026-10-12T21:05:15.673453+00:00", "outer_command_wall_seconds": 5.359644708005362, "warmup_excluded_from_statistics": false, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 325730304, "test_accuracy": 0.8955239064089522, "parameter_sha256": "558baffdc49fc87ee67d04a9cb17c77daf1671a544a5eafcd137576192b5a10e", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8857578840284842, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.3770934749530683, "validation_loss": 1.2565156669823923, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 0.2752373 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.1680702765782571, "validation_loss": 1.0726634152606465, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 0.54327797 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784489967242784, "validation_loss": 0.8820201711707012, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 0.80785259 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923336749857173, "validation_loss": 0.710308352958468, "validation_accuracy": 0.7502543234994914, "elapsed_seconds": 1.079800249 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370655949431792, "validation_loss": 0.5771756065192747, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 1.347155112 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222779060476093, "validation_loss": 0.4819741321230154, "validation_accuracy": 0.8320447609359105, "elapsed_seconds": 1.612343813 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.44097582388285234, "validation_loss": 0.41512437864999546, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 1.878778839 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.38314860589065225, "validation_loss": 0.36713147013723113, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 2.140409316 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.3406074100312188, "validation_loss": 0.3310860565229031, "validation_accuracy": 0.8786368260427263, "elapsed_seconds": 2.407347621 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.3077992643371988, "validation_loss": 0.3033731649951008, "validation_accuracy": 0.8857578840284842, "elapsed_seconds": 2.670017877 } ], "validation_loss": 0.3033731649951008, "coordinator_cpu_threads": 1, "rate": 0.05, "coordinator_cpu_provider": "openblas", "test_loss": 0.29705317871020637, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332128378295, "fit_seconds": 3.380031545, "training_loss": 0.29234691523654444, "tensor_owned_device_peak_bytes": 13048912, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 4202560, "dtype": "Float64", "command_seconds": 3.955525727, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float64", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 1.0290266080000001, "initialization_seconds": 0, "failed": false, "nodes": [ { "response_read_wait_seconds": 0, "requests": 300, "coordinator_exchange_wall_seconds": 0, "inner_threads": 1, "backend": "cuda_cublas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 4202560, "request_write_seconds": 0, "request_payload_bytes": 0, "process_peak_rss_bytes": 324943872, "samples": 458760, "upload_preparation_seconds": 0.40694497400000007, "worker_total_seconds": 1.0275088589999999, "zero_sample_steps": 0, "response_payload_bytes": 0, "compute_seconds": 0.4143261709999999, "owned_device_tensor_peak_bytes": 13048912, "device": "cuda", "readback_seconds": 0.055599096000000035, "configuration_id": "atlas-03/C03", "platform": "atlas-03", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu64", "reference_trial": 1, "dtype": "Float64", "predeclared_tolerances": { "parameter_max_abs": 1e-08, "loss_max_abs": 1e-08, "accuracy_max_abs": 1e-12 }, "parameter_max_abs_error": 3.3306690738754696e-16, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 0.0, "validation_loss": 5.551115123125783e-17, "test_loss": 5.551115123125783e-17 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "atlas03-gpu64", "repetition": 3, "started_utc": "2026-10-12T21:05:16.236386+00:00", "finished_utc": "2026-10-12T21:05:21.563309+00:00", "outer_command_wall_seconds": 5.325457874991116, "warmup_excluded_from_statistics": false, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 324956160, "test_accuracy": 0.8955239064089522, "parameter_sha256": "558baffdc49fc87ee67d04a9cb17c77daf1671a544a5eafcd137576192b5a10e", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8857578840284842, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.3770934749530683, "validation_loss": 1.2565156669823923, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 0.277930806 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.1680702765782571, "validation_loss": 1.0726634152606465, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 0.542974256 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784489967242784, "validation_loss": 0.8820201711707012, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 0.803121148 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923336749857173, "validation_loss": 0.710308352958468, "validation_accuracy": 0.7502543234994914, "elapsed_seconds": 1.067164527 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370655949431792, "validation_loss": 0.5771756065192747, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 1.332007506 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222779060476093, "validation_loss": 0.4819741321230154, "validation_accuracy": 0.8320447609359105, "elapsed_seconds": 1.592197098 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.44097582388285234, "validation_loss": 0.41512437864999546, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 1.854502794 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.38314860589065225, "validation_loss": 0.36713147013723113, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 2.114901177 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.3406074100312188, "validation_loss": 0.3310860565229031, "validation_accuracy": 0.8786368260427263, "elapsed_seconds": 2.378416175 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.3077992643371988, "validation_loss": 0.3033731649951008, "validation_accuracy": 0.8857578840284842, "elapsed_seconds": 2.639413697 } ], "validation_loss": 0.3033731649951008, "coordinator_cpu_threads": 1, "rate": 0.05, "coordinator_cpu_provider": "openblas", "test_loss": 0.29705317871020637, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332128378295, "fit_seconds": 3.351845387, "training_loss": 0.29234691523654444, "tensor_owned_device_peak_bytes": 13048912, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 4202560, "dtype": "Float64", "command_seconds": 3.920242845, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float64", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 1.010918741, "initialization_seconds": 0, "failed": false, "nodes": [ { "response_read_wait_seconds": 0, "requests": 300, "coordinator_exchange_wall_seconds": 0, "inner_threads": 1, "backend": "cuda_cublas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 4202560, "request_write_seconds": 0, "request_payload_bytes": 0, "process_peak_rss_bytes": 324956160, "samples": 458760, "upload_preparation_seconds": 0.4014066510000002, "worker_total_seconds": 1.0095150980000012, "zero_sample_steps": 0, "response_payload_bytes": 0, "compute_seconds": 0.40619233300000024, "owned_device_tensor_peak_bytes": 13048912, "device": "cuda", "readback_seconds": 0.05416801799999993, "configuration_id": "atlas-03/C03", "platform": "atlas-03", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu64", "reference_trial": 1, "dtype": "Float64", "predeclared_tolerances": { "parameter_max_abs": 1e-08, "loss_max_abs": 1e-08, "accuracy_max_abs": 1e-12 }, "parameter_max_abs_error": 3.3306690738754696e-16, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 0.0, "validation_loss": 5.551115123125783e-17, "test_loss": 5.551115123125783e-17 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "01-02-gpu32", "repetition": 1, "started_utc": "2026-10-12T21:06:27.340991+00:00", "finished_utc": "2026-10-12T21:07:32.228718+00:00", "outer_command_wall_seconds": 64.88779833298759, "warmup_excluded_from_statistics": false, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 74383360, "test_accuracy": 0.8955239064089522, "parameter_sha256": "d4981c6fe0e53117073c0d8563cac476fa35e24af2658d8ecd892e7d5539bb07", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8858596134282808, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.3770934769285932, "validation_loss": 1.2565156643897568, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 9.877288583 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.168070280486428, "validation_loss": 1.072663564672305, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 15.828659 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784492893724769, "validation_loss": 0.8820205004872529, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 21.858273916 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923334494268212, "validation_loss": 0.7103084890670233, "validation_accuracy": 0.7501525940996948, "elapsed_seconds": 26.912038583 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370660988270472, "validation_loss": 0.5771758733421473, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 32.502619833 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222784888711614, "validation_loss": 0.4819751959608627, "validation_accuracy": 0.8319430315361139, "elapsed_seconds": 37.548834791 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.4409767913780941, "validation_loss": 0.41512575641895966, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 43.1463895 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.3831498941926472, "validation_loss": 0.3671321308139867, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 48.735022291 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.340608996330616, "validation_loss": 0.3310877792408362, "validation_accuracy": 0.8787385554425229, "elapsed_seconds": 56.621605166 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.3078003465981188, "validation_loss": 0.30337470066511013, "validation_accuracy": 0.8858596134282808, "elapsed_seconds": 64.23133575 } ], "validation_loss": 0.30337470066511013, "coordinator_cpu_threads": null, "rate": 0.05, "coordinator_cpu_provider": "apple_accelerate", "test_loss": 0.29705493784872467, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332174875397, "fit_seconds": 64.414978083, "training_loss": 0.29234811506937414, "tensor_owned_device_peak_bytes": 0, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 6303328, "dtype": "Float32", "command_seconds": 64.804851625, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float32", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 60.82774791199999, "initialization_seconds": 2.176992291, "failed": false, "nodes": [ { "response_read_wait_seconds": 5.540511544000001, "requests": 300, "coordinator_exchange_wall_seconds": 6.608140246, "inner_threads": null, "backend": "metal_mps", "target": "macos-arm64", "owned_host_tensor_peak_bytes": 983040, "request_write_seconds": 1.0675435869999996, "request_payload_bytes": 82667760, "process_peak_rss_bytes": 25493504, "samples": 229380, "upload_preparation_seconds": 0.267110162, "worker_total_seconds": 3.6027743710000024, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 3.2784082180000005, "owned_device_tensor_peak_bytes": 3329576, "device": "mps", "readback_seconds": 0.021237366000000004, "configuration_id": "atlas-01/C03", "platform": "atlas-01", "relative_batch_weight": 1 }, { "response_read_wait_seconds": 59.03837270800001, "requests": 300, "coordinator_exchange_wall_seconds": 60.31867120499993, "inner_threads": 1, "backend": "cuda_cublas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 983040, "request_write_seconds": 1.280212417, "request_payload_bytes": 82667760, "process_peak_rss_bytes": 252731392, "samples": 229380, "upload_preparation_seconds": 1.6343725100000008, "worker_total_seconds": 2.8092793789999995, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 0.7812318650000001, "owned_device_tensor_peak_bytes": 3329576, "device": "cuda", "readback_seconds": 0.14299656899999993, "configuration_id": "atlas-02/C03", "platform": "atlas-02", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu32", "reference_trial": 1, "dtype": "Float32", "predeclared_tolerances": { "parameter_max_abs": 0.001, "loss_max_abs": 0.0005, "accuracy_max_abs": 0.005 }, "parameter_max_abs_error": 3.8526952266693115e-05, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 5.787559557157707e-07, "validation_loss": 2.747028308158761e-07, "test_loss": 1.008616095310444e-06 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "01-02-gpu32", "repetition": 2, "started_utc": "2026-10-12T21:07:32.762292+00:00", "finished_utc": "2026-10-12T21:08:42.739788+00:00", "outer_command_wall_seconds": 69.97771516699868, "warmup_excluded_from_statistics": false, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 77430784, "test_accuracy": 0.8955239064089522, "parameter_sha256": "d4981c6fe0e53117073c0d8563cac476fa35e24af2658d8ecd892e7d5539bb07", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8858596134282808, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.3770934769285932, "validation_loss": 1.2565156643897568, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 9.993338625 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.168070280486428, "validation_loss": 1.072663564672305, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 17.8551795 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784492893724769, "validation_loss": 0.8820205004872529, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 25.024195167 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923334494268212, "validation_loss": 0.7103084890670233, "validation_accuracy": 0.7501525940996948, "elapsed_seconds": 31.687634458 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370660988270472, "validation_loss": 0.5771758733421473, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 38.637243167 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222784888711614, "validation_loss": 0.4819751959608627, "validation_accuracy": 0.8319430315361139, "elapsed_seconds": 44.8990655 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.4409767913780941, "validation_loss": 0.41512575641895966, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 50.239202667 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.3831498941926472, "validation_loss": 0.3671321308139867, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 57.36532875 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.340608996330616, "validation_loss": 0.3310877792408362, "validation_accuracy": 0.8787385554425229, "elapsed_seconds": 63.556858125 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.3078003465981188, "validation_loss": 0.30337470066511013, "validation_accuracy": 0.8858596134282808, "elapsed_seconds": 69.350484875 } ], "validation_loss": 0.30337470066511013, "coordinator_cpu_threads": null, "rate": 0.05, "coordinator_cpu_provider": "apple_accelerate", "test_loss": 0.29705493784872467, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332174875397, "fit_seconds": 69.533544166, "training_loss": 0.29234811506937414, "tensor_owned_device_peak_bytes": 0, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 6303328, "dtype": "Float32", "command_seconds": 69.92076175, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float32", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 65.64244092199993, "initialization_seconds": 2.329750791, "failed": false, "nodes": [ { "response_read_wait_seconds": 5.688616882999999, "requests": 300, "coordinator_exchange_wall_seconds": 6.897081250000001, "inner_threads": null, "backend": "metal_mps", "target": "macos-arm64", "owned_host_tensor_peak_bytes": 983040, "request_write_seconds": 1.2083657930000002, "request_payload_bytes": 82667760, "process_peak_rss_bytes": 23445504, "samples": 229380, "upload_preparation_seconds": 0.29392679999999993, "worker_total_seconds": 3.542504004000001, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 3.1871155310000003, "owned_device_tensor_peak_bytes": 3329576, "device": "mps", "readback_seconds": 0.022639126, "configuration_id": "atlas-01/C03", "platform": "atlas-01", "relative_batch_weight": 1 }, { "response_read_wait_seconds": 63.661730917999975, "requests": 300, "coordinator_exchange_wall_seconds": 65.10478625400002, "inner_threads": 1, "backend": "cuda_cublas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 983040, "request_write_seconds": 1.4429274139999988, "request_payload_bytes": 82667760, "process_peak_rss_bytes": 252841984, "samples": 229380, "upload_preparation_seconds": 1.695894748, "worker_total_seconds": 2.920432863, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 0.8144459129999998, "owned_device_tensor_peak_bytes": 3329576, "device": "cuda", "readback_seconds": 0.15217154900000004, "configuration_id": "atlas-02/C03", "platform": "atlas-02", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu32", "reference_trial": 1, "dtype": "Float32", "predeclared_tolerances": { "parameter_max_abs": 0.001, "loss_max_abs": 0.0005, "accuracy_max_abs": 0.005 }, "parameter_max_abs_error": 3.8526952266693115e-05, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 5.787559557157707e-07, "validation_loss": 2.747028308158761e-07, "test_loss": 1.008616095310444e-06 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "01-02-gpu32", "repetition": 3, "started_utc": "2026-10-12T21:08:43.268472+00:00", "finished_utc": "2026-10-12T21:09:50.135835+00:00", "outer_command_wall_seconds": 66.8675972080091, "warmup_excluded_from_statistics": false, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 77496320, "test_accuracy": 0.8955239064089522, "parameter_sha256": "d4981c6fe0e53117073c0d8563cac476fa35e24af2658d8ecd892e7d5539bb07", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8858596134282808, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.3770934769285932, "validation_loss": 1.2565156643897568, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 8.716643959 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.168070280486428, "validation_loss": 1.072663564672305, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 15.880695125 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784492893724769, "validation_loss": 0.8820205004872529, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 20.918853667 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923334494268212, "validation_loss": 0.7103084890670233, "validation_accuracy": 0.7501525940996948, "elapsed_seconds": 26.710836334 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370660988270472, "validation_loss": 0.5771758733421473, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 33.66853525 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222784888711614, "validation_loss": 0.4819751959608627, "validation_accuracy": 0.8319430315361139, "elapsed_seconds": 41.702471125 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.4409767913780941, "validation_loss": 0.41512575641895966, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 48.350602959 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.3831498941926472, "validation_loss": 0.3671321308139867, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 55.091696084 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.340608996330616, "validation_loss": 0.3310877792408362, "validation_accuracy": 0.8787385554425229, "elapsed_seconds": 61.081406375 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.3078003465981188, "validation_loss": 0.30337470066511013, "validation_accuracy": 0.8858596134282808, "elapsed_seconds": 66.231799125 } ], "validation_loss": 0.30337470066511013, "coordinator_cpu_threads": null, "rate": 0.05, "coordinator_cpu_provider": "apple_accelerate", "test_loss": 0.29705493784872467, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332174875397, "fit_seconds": 66.406795041, "training_loss": 0.29234811506937414, "tensor_owned_device_peak_bytes": 0, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 6303328, "dtype": "Float32", "command_seconds": 66.80565725, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float32", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 62.363844166000014, "initialization_seconds": 2.536115375, "failed": false, "nodes": [ { "response_read_wait_seconds": 5.730888381, "requests": 300, "coordinator_exchange_wall_seconds": 6.926093207000001, "inner_threads": null, "backend": "metal_mps", "target": "macos-arm64", "owned_host_tensor_peak_bytes": 983040, "request_write_seconds": 1.195105451000001, "request_payload_bytes": 82667760, "process_peak_rss_bytes": 24199168, "samples": 229380, "upload_preparation_seconds": 0.291751674, "worker_total_seconds": 3.5903741669999976, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 3.235207667999998, "owned_device_tensor_peak_bytes": 3329576, "device": "mps", "readback_seconds": 0.02372707299999998, "configuration_id": "atlas-01/C03", "platform": "atlas-01", "relative_batch_weight": 1 }, { "response_read_wait_seconds": 60.40910454899999, "requests": 300, "coordinator_exchange_wall_seconds": 61.83582804299997, "inner_threads": 1, "backend": "cuda_cublas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 983040, "request_write_seconds": 1.4266434240000008, "request_payload_bytes": 82667760, "process_peak_rss_bytes": 252841984, "samples": 229380, "upload_preparation_seconds": 1.6837423830000011, "worker_total_seconds": 2.8938203080000013, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 0.802364632, "owned_device_tensor_peak_bytes": 3329576, "device": "cuda", "readback_seconds": 0.14952447300000007, "configuration_id": "atlas-02/C03", "platform": "atlas-02", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu32", "reference_trial": 1, "dtype": "Float32", "predeclared_tolerances": { "parameter_max_abs": 0.001, "loss_max_abs": 0.0005, "accuracy_max_abs": 0.005 }, "parameter_max_abs_error": 3.8526952266693115e-05, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 5.787559557157707e-07, "validation_loss": 2.747028308158761e-07, "test_loss": 1.008616095310444e-06 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "01-03-gpu32", "repetition": 1, "started_utc": "2026-10-12T21:11:00.846721+00:00", "finished_utc": "2026-10-12T21:11:58.492091+00:00", "outer_command_wall_seconds": 57.645449082992855, "warmup_excluded_from_statistics": false, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 75169792, "test_accuracy": 0.8955239064089522, "parameter_sha256": "76a2caf28ba047bb311582e24dcb029de184d31a707f97d640024c9b33247040", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8858596134282808, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.377093485176267, "validation_loss": 1.2565156643897568, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 8.118216958 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.1680702864733972, "validation_loss": 1.072663564672305, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 13.913021958 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784492919683893, "validation_loss": 0.8820206904823636, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 19.919476375 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923334923880031, "validation_loss": 0.7103080904859735, "validation_accuracy": 0.7502543234994914, "elapsed_seconds": 24.89123575 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370656905827833, "validation_loss": 0.5771757672907498, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 30.355092 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222784275359083, "validation_loss": 0.48197514480273573, "validation_accuracy": 0.8320447609359105, "elapsed_seconds": 35.778515208 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.4409769844500581, "validation_loss": 0.41512621847925935, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 40.691235667 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.3831507439174298, "validation_loss": 0.3671325146727499, "validation_accuracy": 0.8679552390640896, "elapsed_seconds": 47.269333333 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.34060974856836146, "validation_loss": 0.3310890749597113, "validation_accuracy": 0.8787385554425229, "elapsed_seconds": 52.085296958 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.30780227909903857, "validation_loss": 0.30337603640338046, "validation_accuracy": 0.8858596134282808, "elapsed_seconds": 57.026137292 } ], "validation_loss": 0.30337603640338046, "coordinator_cpu_threads": null, "rate": 0.05, "coordinator_cpu_provider": "apple_accelerate", "test_loss": 0.29705585060017764, "optimizer": "sgd", "training_accuracy": 0.8984000348766239, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332174875397, "fit_seconds": 57.208868792, "training_loss": 0.2923502071331597, "tensor_owned_device_peak_bytes": 0, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 6303328, "dtype": "Float32", "command_seconds": 57.587268958, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float32", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 54.109132078999984, "initialization_seconds": 1.586212709, "failed": false, "nodes": [ { "response_read_wait_seconds": 5.850273453999998, "requests": 300, "coordinator_exchange_wall_seconds": 7.047991165999994, "inner_threads": null, "backend": "metal_mps", "target": "macos-arm64", "owned_host_tensor_peak_bytes": 983040, "request_write_seconds": 1.1976290399999998, "request_payload_bytes": 82667760, "process_peak_rss_bytes": 24231936, "samples": 229380, "upload_preparation_seconds": 0.292591912, "worker_total_seconds": 3.697802285, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 3.3420961660000015, "owned_device_tensor_peak_bytes": 3329576, "device": "mps", "readback_seconds": 0.023648492000000004, "configuration_id": "atlas-01/C03", "platform": "atlas-01", "relative_batch_weight": 1 }, { "response_read_wait_seconds": 52.18226125400003, "requests": 300, "coordinator_exchange_wall_seconds": 53.59055008700001, "inner_threads": 1, "backend": "cuda_cublas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 983040, "request_write_seconds": 1.4080964589999994, "request_payload_bytes": 82667760, "process_peak_rss_bytes": 287035392, "samples": 229380, "upload_preparation_seconds": 1.0400027480000016, "worker_total_seconds": 1.5876060520000006, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 0.352285454, "owned_device_tensor_peak_bytes": 3329576, "device": "cuda", "readback_seconds": 0.06323915100000001, "configuration_id": "atlas-03/C03", "platform": "atlas-03", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu32", "reference_trial": 1, "dtype": "Float32", "predeclared_tolerances": { "parameter_max_abs": 0.001, "loss_max_abs": 0.0005, "accuracy_max_abs": 0.005 }, "parameter_max_abs_error": 5.525164306163788e-05, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 2.6708197412572687e-06, "validation_loss": 1.610441101140836e-06, "test_loss": 1.9213675482854242e-06 }, "accuracy_abs_errors": { "training_accuracy": 4.359577992840702e-05, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "01-03-gpu32", "repetition": 2, "started_utc": "2026-10-12T21:11:59.035948+00:00", "finished_utc": "2026-10-12T21:12:58.079455+00:00", "outer_command_wall_seconds": 59.04362216699519, "warmup_excluded_from_statistics": false, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 77758464, "test_accuracy": 0.8955239064089522, "parameter_sha256": "76a2caf28ba047bb311582e24dcb029de184d31a707f97d640024c9b33247040", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8858596134282808, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.377093485176267, "validation_loss": 1.2565156643897568, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 7.067693584 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.1680702864733972, "validation_loss": 1.072663564672305, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 12.561611417 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784492919683893, "validation_loss": 0.8820206904823636, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 17.387276917 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923334923880031, "validation_loss": 0.7103080904859735, "validation_accuracy": 0.7502543234994914, "elapsed_seconds": 22.261127292 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370656905827833, "validation_loss": 0.5771757672907498, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 28.33167875 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222784275359083, "validation_loss": 0.48197514480273573, "validation_accuracy": 0.8320447609359105, "elapsed_seconds": 34.480300209 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.4409769844500581, "validation_loss": 0.41512621847925935, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 40.907773375 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.3831507439174298, "validation_loss": 0.3671325146727499, "validation_accuracy": 0.8679552390640896, "elapsed_seconds": 45.8729535 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.34060974856836146, "validation_loss": 0.3310890749597113, "validation_accuracy": 0.8787385554425229, "elapsed_seconds": 51.27530875 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.30780227909903857, "validation_loss": 0.30337603640338046, "validation_accuracy": 0.8858596134282808, "elapsed_seconds": 58.416164667 } ], "validation_loss": 0.30337603640338046, "coordinator_cpu_threads": null, "rate": 0.05, "coordinator_cpu_provider": "apple_accelerate", "test_loss": 0.29705585060017764, "optimizer": "sgd", "training_accuracy": 0.8984000348766239, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332174875397, "fit_seconds": 58.597174292, "training_loss": 0.2923502071331597, "tensor_owned_device_peak_bytes": 0, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 6303328, "dtype": "Float32", "command_seconds": 58.984112334, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float32", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 55.77163029300002, "initialization_seconds": 1.315616167, "failed": false, "nodes": [ { "response_read_wait_seconds": 5.816193624999996, "requests": 300, "coordinator_exchange_wall_seconds": 7.023647721999998, "inner_threads": null, "backend": "metal_mps", "target": "macos-arm64", "owned_host_tensor_peak_bytes": 983040, "request_write_seconds": 1.2073680439999999, "request_payload_bytes": 82667760, "process_peak_rss_bytes": 24788992, "samples": 229380, "upload_preparation_seconds": 0.28917641099999997, "worker_total_seconds": 3.6478931899999996, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 3.296334863, "owned_device_tensor_peak_bytes": 3329576, "device": "mps", "readback_seconds": 0.023418092999999997, "configuration_id": "atlas-01/C03", "platform": "atlas-01", "relative_batch_weight": 1 }, { "response_read_wait_seconds": 53.80685994000001, "requests": 300, "coordinator_exchange_wall_seconds": 55.249875174000046, "inner_threads": 1, "backend": "cuda_cublas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 983040, "request_write_seconds": 1.442951583, "request_payload_bytes": 82667760, "process_peak_rss_bytes": 286707712, "samples": 229380, "upload_preparation_seconds": 1.0665977910000004, "worker_total_seconds": 1.6248352789999998, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 0.36445149700000007, "owned_device_tensor_peak_bytes": 3329576, "device": "cuda", "readback_seconds": 0.06346413999999996, "configuration_id": "atlas-03/C03", "platform": "atlas-03", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu32", "reference_trial": 1, "dtype": "Float32", "predeclared_tolerances": { "parameter_max_abs": 0.001, "loss_max_abs": 0.0005, "accuracy_max_abs": 0.005 }, "parameter_max_abs_error": 5.525164306163788e-05, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 2.6708197412572687e-06, "validation_loss": 1.610441101140836e-06, "test_loss": 1.9213675482854242e-06 }, "accuracy_abs_errors": { "training_accuracy": 4.359577992840702e-05, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "01-03-gpu32", "repetition": 3, "started_utc": "2026-10-12T21:12:58.642794+00:00", "finished_utc": "2026-10-12T21:14:03.500986+00:00", "outer_command_wall_seconds": 64.85838504199637, "warmup_excluded_from_statistics": false, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 78020608, "test_accuracy": 0.8955239064089522, "parameter_sha256": "76a2caf28ba047bb311582e24dcb029de184d31a707f97d640024c9b33247040", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8858596134282808, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.377093485176267, "validation_loss": 1.2565156643897568, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 7.65897825 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.1680702864733972, "validation_loss": 1.072663564672305, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 12.901491375 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784492919683893, "validation_loss": 0.8820206904823636, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 18.345523334 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923334923880031, "validation_loss": 0.7103080904859735, "validation_accuracy": 0.7502543234994914, "elapsed_seconds": 25.213640959 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370656905827833, "validation_loss": 0.5771757672907498, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 31.800692042 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222784275359083, "validation_loss": 0.48197514480273573, "validation_accuracy": 0.8320447609359105, "elapsed_seconds": 37.568827834 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.4409769844500581, "validation_loss": 0.41512621847925935, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 44.301601 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.3831507439174298, "validation_loss": 0.3671325146727499, "validation_accuracy": 0.8679552390640896, "elapsed_seconds": 50.669520167 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.34060974856836146, "validation_loss": 0.3310890749597113, "validation_accuracy": 0.8787385554425229, "elapsed_seconds": 57.938484834 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.30780227909903857, "validation_loss": 0.30337603640338046, "validation_accuracy": 0.8858596134282808, "elapsed_seconds": 64.250439125 } ], "validation_loss": 0.30337603640338046, "coordinator_cpu_threads": null, "rate": 0.05, "coordinator_cpu_provider": "apple_accelerate", "test_loss": 0.29705585060017764, "optimizer": "sgd", "training_accuracy": 0.8984000348766239, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332174875397, "fit_seconds": 64.432534459, "training_loss": 0.2923502071331597, "tensor_owned_device_peak_bytes": 0, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 6303328, "dtype": "Float32", "command_seconds": 64.820838792, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float32", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 61.30786578899997, "initialization_seconds": 1.5760365, "failed": false, "nodes": [ { "response_read_wait_seconds": 5.858686630999997, "requests": 300, "coordinator_exchange_wall_seconds": 7.047510581000003, "inner_threads": null, "backend": "metal_mps", "target": "macos-arm64", "owned_host_tensor_peak_bytes": 983040, "request_write_seconds": 1.188724290000001, "request_payload_bytes": 82667760, "process_peak_rss_bytes": 25034752, "samples": 229380, "upload_preparation_seconds": 0.2974659540000005, "worker_total_seconds": 3.705357743999997, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 3.346444957999999, "owned_device_tensor_peak_bytes": 3329576, "device": "mps", "readback_seconds": 0.022949124999999994, "configuration_id": "atlas-01/C03", "platform": "atlas-01", "relative_batch_weight": 1 }, { "response_read_wait_seconds": 59.355597614000004, "requests": 300, "coordinator_exchange_wall_seconds": 60.785204422000014, "inner_threads": 1, "backend": "cuda_cublas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 983040, "request_write_seconds": 1.4294870309999994, "request_payload_bytes": 82667760, "process_peak_rss_bytes": 286941184, "samples": 229380, "upload_preparation_seconds": 1.0515673750000007, "worker_total_seconds": 1.6082218649999993, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 0.359499999, "owned_device_tensor_peak_bytes": 3329576, "device": "cuda", "readback_seconds": 0.06348462099999996, "configuration_id": "atlas-03/C03", "platform": "atlas-03", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu32", "reference_trial": 1, "dtype": "Float32", "predeclared_tolerances": { "parameter_max_abs": 0.001, "loss_max_abs": 0.0005, "accuracy_max_abs": 0.005 }, "parameter_max_abs_error": 5.525164306163788e-05, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 2.6708197412572687e-06, "validation_loss": 1.610441101140836e-06, "test_loss": 1.9213675482854242e-06 }, "accuracy_abs_errors": { "training_accuracy": 4.359577992840702e-05, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "02-03-gpu32", "repetition": 1, "started_utc": "2026-10-12T21:15:40.212872+00:00", "finished_utc": "2026-10-12T21:17:02.736789+00:00", "outer_command_wall_seconds": 82.52440370799741, "warmup_excluded_from_statistics": false, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 77627392, "test_accuracy": 0.8955239064089522, "parameter_sha256": "cbf657f3b09c0dec409f57f27268269a2c3fad35ede16055b7f4412d0779c3d0", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8857578840284842, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.3770934871719234, "validation_loss": 1.2565156643897568, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 12.441479541 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.1680703039042089, "validation_loss": 1.072663564672305, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 19.927762958 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784492979553586, "validation_loss": 0.8820206997959684, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 26.413152875 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.792333496246792, "validation_loss": 0.7103080904859735, "validation_accuracy": 0.7502543234994914, "elapsed_seconds": 32.559044291 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370656875892986, "validation_loss": 0.5771757579771452, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 41.697115916 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222784380131045, "validation_loss": 0.4819751541102769, "validation_accuracy": 0.8320447609359105, "elapsed_seconds": 49.241847916 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.4409769955275104, "validation_loss": 0.41512625572761463, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 56.640403083 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.3831508259147429, "validation_loss": 0.3671326767270466, "validation_accuracy": 0.8679552390640896, "elapsed_seconds": 65.018776291 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.34060971804885004, "validation_loss": 0.33108892592990874, "validation_accuracy": 0.8787385554425229, "elapsed_seconds": 73.460169125 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.3078019323654272, "validation_loss": 0.3033758939464604, "validation_accuracy": 0.8857578840284842, "elapsed_seconds": 81.815366666 } ], "validation_loss": 0.3033758939464604, "coordinator_cpu_threads": null, "rate": 0.05, "coordinator_cpu_provider": "apple_accelerate", "test_loss": 0.2970558105929094, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332174875397, "fit_seconds": 82.012750791, "training_loss": 0.2923499547917588, "tensor_owned_device_peak_bytes": 0, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 6303328, "dtype": "Float32", "command_seconds": 82.4643615, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float32", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 76.98600395499996, "initialization_seconds": 3.481013958, "failed": false, "nodes": [ { "response_read_wait_seconds": 67.15449578600001, "requests": 300, "coordinator_exchange_wall_seconds": 68.58777321100001, "inner_threads": 1, "backend": "cuda_cublas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 983040, "request_write_seconds": 1.4331425359999996, "request_payload_bytes": 82667760, "process_peak_rss_bytes": 252645376, "samples": 229380, "upload_preparation_seconds": 1.6567668230000003, "worker_total_seconds": 2.8483384549999995, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 0.7925328389999999, "owned_device_tensor_peak_bytes": 3329576, "device": "cuda", "readback_seconds": 0.14461619200000006, "configuration_id": "atlas-02/C03", "platform": "atlas-02", "relative_batch_weight": 1 }, { "response_read_wait_seconds": 66.43971983499999, "requests": 300, "coordinator_exchange_wall_seconds": 67.86396450300003, "inner_threads": 1, "backend": "cuda_cublas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 983040, "request_write_seconds": 1.424187879999999, "request_payload_bytes": 82667760, "process_peak_rss_bytes": 286822400, "samples": 229380, "upload_preparation_seconds": 1.0825509750000002, "worker_total_seconds": 1.6464566009999997, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 0.3691265440000001, "owned_device_tensor_peak_bytes": 3329576, "device": "cuda", "readback_seconds": 0.06162792100000003, "configuration_id": "atlas-03/C03", "platform": "atlas-03", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu32", "reference_trial": 1, "dtype": "Float32", "predeclared_tolerances": { "parameter_max_abs": 0.001, "loss_max_abs": 0.0005, "accuracy_max_abs": 0.005 }, "parameter_max_abs_error": 5.7095661759376526e-05, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 2.418478340371344e-06, "validation_loss": 1.4679841811027217e-06, "test_loss": 1.8813602800649143e-06 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.00010172939979657958, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "02-03-gpu32", "repetition": 2, "started_utc": "2026-10-12T21:17:03.346744+00:00", "finished_utc": "2026-10-12T21:18:35.510878+00:00", "outer_command_wall_seconds": 92.16468412500399, "warmup_excluded_from_statistics": false, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 77725696, "test_accuracy": 0.8955239064089522, "parameter_sha256": "cbf657f3b09c0dec409f57f27268269a2c3fad35ede16055b7f4412d0779c3d0", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8857578840284842, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.3770934871719234, "validation_loss": 1.2565156643897568, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 14.678294167 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.1680703039042089, "validation_loss": 1.072663564672305, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 23.080242083 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784492979553586, "validation_loss": 0.8820206997959684, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 30.819576875 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.792333496246792, "validation_loss": 0.7103080904859735, "validation_accuracy": 0.7502543234994914, "elapsed_seconds": 38.743766833 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370656875892986, "validation_loss": 0.5771757579771452, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 46.249155542 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222784380131045, "validation_loss": 0.4819751541102769, "validation_accuracy": 0.8320447609359105, "elapsed_seconds": 56.214431208 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.4409769955275104, "validation_loss": 0.41512625572761463, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 66.577043125 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.3831508259147429, "validation_loss": 0.3671326767270466, "validation_accuracy": 0.8679552390640896, "elapsed_seconds": 74.956468375 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.34060971804885004, "validation_loss": 0.33108892592990874, "validation_accuracy": 0.8787385554425229, "elapsed_seconds": 82.935187083 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.3078019323654272, "validation_loss": 0.3033758939464604, "validation_accuracy": 0.8857578840284842, "elapsed_seconds": 91.496647083 } ], "validation_loss": 0.3033758939464604, "coordinator_cpu_threads": null, "rate": 0.05, "coordinator_cpu_provider": "apple_accelerate", "test_loss": 0.2970558105929094, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332174875397, "fit_seconds": 91.678021833, "training_loss": 0.2923499547917588, "tensor_owned_device_peak_bytes": 0, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 6303328, "dtype": "Float32", "command_seconds": 92.129275166, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float32", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 86.50576891800002, "initialization_seconds": 3.696369333, "failed": false, "nodes": [ { "response_read_wait_seconds": 76.93777121300002, "requests": 300, "coordinator_exchange_wall_seconds": 78.28922679199995, "inner_threads": 1, "backend": "cuda_cublas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 983040, "request_write_seconds": 1.3513404990000002, "request_payload_bytes": 82667760, "process_peak_rss_bytes": 252858368, "samples": 229380, "upload_preparation_seconds": 1.7106720770000006, "worker_total_seconds": 2.9448951880000003, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 0.8241611529999998, "owned_device_tensor_peak_bytes": 3329576, "device": "cuda", "readback_seconds": 0.15175968399999998, "configuration_id": "atlas-02/C03", "platform": "atlas-02", "relative_batch_weight": 1 }, { "response_read_wait_seconds": 72.59485933, "requests": 300, "coordinator_exchange_wall_seconds": 73.94015379599995, "inner_threads": 1, "backend": "cuda_cublas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 983040, "request_write_seconds": 1.3452226289999993, "request_payload_bytes": 82667760, "process_peak_rss_bytes": 286904320, "samples": 229380, "upload_preparation_seconds": 1.092290365000001, "worker_total_seconds": 1.646046648999999, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 0.36056593099999973, "owned_device_tensor_peak_bytes": 3329576, "device": "cuda", "readback_seconds": 0.06239568800000004, "configuration_id": "atlas-03/C03", "platform": "atlas-03", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu32", "reference_trial": 1, "dtype": "Float32", "predeclared_tolerances": { "parameter_max_abs": 0.001, "loss_max_abs": 0.0005, "accuracy_max_abs": 0.005 }, "parameter_max_abs_error": 5.7095661759376526e-05, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 2.418478340371344e-06, "validation_loss": 1.4679841811027217e-06, "test_loss": 1.8813602800649143e-06 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.00010172939979657958, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "02-03-gpu32", "repetition": 3, "started_utc": "2026-10-12T21:18:36.132001+00:00", "finished_utc": "2026-10-12T21:20:07.242095+00:00", "outer_command_wall_seconds": 91.10916891599481, "warmup_excluded_from_statistics": false, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 75268096, "test_accuracy": 0.8955239064089522, "parameter_sha256": "cbf657f3b09c0dec409f57f27268269a2c3fad35ede16055b7f4412d0779c3d0", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8857578840284842, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.3770934871719234, "validation_loss": 1.2565156643897568, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 12.235564209 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.1680703039042089, "validation_loss": 1.072663564672305, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 21.771244667 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784492979553586, "validation_loss": 0.8820206997959684, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 30.701174334 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.792333496246792, "validation_loss": 0.7103080904859735, "validation_accuracy": 0.7502543234994914, "elapsed_seconds": 39.455617792 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370656875892986, "validation_loss": 0.5771757579771452, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 47.908013209 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222784380131045, "validation_loss": 0.4819751541102769, "validation_accuracy": 0.8320447609359105, "elapsed_seconds": 55.943725084 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.4409769955275104, "validation_loss": 0.41512625572761463, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 63.683780209 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.3831508259147429, "validation_loss": 0.3671326767270466, "validation_accuracy": 0.8679552390640896, "elapsed_seconds": 72.109687334 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.34060971804885004, "validation_loss": 0.33108892592990874, "validation_accuracy": 0.8787385554425229, "elapsed_seconds": 81.881025292 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.3078019323654272, "validation_loss": 0.3033758939464604, "validation_accuracy": 0.8857578840284842, "elapsed_seconds": 90.419067834 } ], "validation_loss": 0.3033758939464604, "coordinator_cpu_threads": null, "rate": 0.05, "coordinator_cpu_provider": "apple_accelerate", "test_loss": 0.2970558105929094, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332174875397, "fit_seconds": 90.603120209, "training_loss": 0.2923499547917588, "tensor_owned_device_peak_bytes": 0, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 6303328, "dtype": "Float32", "command_seconds": 91.037551042, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float32", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 85.57617219800002, "initialization_seconds": 3.537018333, "failed": false, "nodes": [ { "response_read_wait_seconds": 79.19499566500002, "requests": 300, "coordinator_exchange_wall_seconds": 80.50158036999997, "inner_threads": 1, "backend": "cuda_cublas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 983040, "request_write_seconds": 1.3065204980000003, "request_payload_bytes": 82667760, "process_peak_rss_bytes": 252796928, "samples": 229380, "upload_preparation_seconds": 1.6798352950000002, "worker_total_seconds": 2.9094893740000005, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 0.8207718829999999, "owned_device_tensor_peak_bytes": 3329576, "device": "cuda", "readback_seconds": 0.15176089000000012, "configuration_id": "atlas-02/C03", "platform": "atlas-02", "relative_batch_weight": 1 }, { "response_read_wait_seconds": 73.17139229200005, "requests": 300, "coordinator_exchange_wall_seconds": 74.47368540199999, "inner_threads": 1, "backend": "cuda_cublas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 983040, "request_write_seconds": 1.3022185680000002, "request_payload_bytes": 82667760, "process_peak_rss_bytes": 286883840, "samples": 229380, "upload_preparation_seconds": 1.0510841480000002, "worker_total_seconds": 1.6180502700000008, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 0.3666001520000001, "owned_device_tensor_peak_bytes": 3329576, "device": "cuda", "readback_seconds": 0.06479273999999999, "configuration_id": "atlas-03/C03", "platform": "atlas-03", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu32", "reference_trial": 1, "dtype": "Float32", "predeclared_tolerances": { "parameter_max_abs": 0.001, "loss_max_abs": 0.0005, "accuracy_max_abs": 0.005 }, "parameter_max_abs_error": 5.7095661759376526e-05, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 2.418478340371344e-06, "validation_loss": 1.4679841811027217e-06, "test_loss": 1.8813602800649143e-06 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.00010172939979657958, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "01-02-03-gpu32", "repetition": 1, "started_utc": "2026-10-12T21:21:34.113387+00:00", "finished_utc": "2026-10-12T21:23:00.909544+00:00", "outer_command_wall_seconds": 86.79663883299509, "warmup_excluded_from_statistics": false, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 76365824, "test_accuracy": 0.8955239064089522, "parameter_sha256": "267925aebd1e6e485966fd3926f1b0e321368eb6395472f921d629c627c888a5", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8857578840284842, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.3770934779472095, "validation_loss": 1.2565156643897568, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 13.421200458 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.1680702881156562, "validation_loss": 1.072663564672305, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 20.058269291 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.978449299727543, "validation_loss": 0.8820205004872529, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 29.448020708 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923334504246493, "validation_loss": 0.7103084927900398, "validation_accuracy": 0.7501525940996948, "elapsed_seconds": 37.2770965 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370661070981077, "validation_loss": 0.577175882655752, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 45.465128791 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222785054353695, "validation_loss": 0.481975248113411, "validation_accuracy": 0.8319430315361139, "elapsed_seconds": 53.736764333 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.44097682620333817, "validation_loss": 0.41512579553488677, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 63.534095208 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.3831498520357044, "validation_loss": 0.36713215224861734, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 71.117132166 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.34060882918919294, "validation_loss": 0.3310873703760145, "validation_accuracy": 0.8787385554425229, "elapsed_seconds": 79.004380708 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.3077999734350523, "validation_loss": 0.30337328036464484, "validation_accuracy": 0.8857578840284842, "elapsed_seconds": 86.112170833 } ], "validation_loss": 0.30337328036464484, "coordinator_cpu_threads": null, "rate": 0.05, "coordinator_cpu_provider": "apple_accelerate", "test_loss": 0.29705295057650866, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332174875397, "fit_seconds": 86.292820584, "training_loss": 0.2923465431937054, "tensor_owned_device_peak_bytes": 0, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 6303328, "dtype": "Float32", "command_seconds": 86.745891042, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float32", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 81.28054237900005, "initialization_seconds": 3.57189175, "failed": false, "nodes": [ { "response_read_wait_seconds": 5.224934279000005, "requests": 300, "coordinator_exchange_wall_seconds": 6.0719885110000025, "inner_threads": null, "backend": "metal_mps", "target": "macos-arm64", "owned_host_tensor_peak_bytes": 655360, "request_write_seconds": 0.8469903709999993, "request_payload_bytes": 61870640, "process_peak_rss_bytes": 21774336, "samples": 152920, "upload_preparation_seconds": 0.2199126280000001, "worker_total_seconds": 3.4963926999999972, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 3.215899031000002, "owned_device_tensor_peak_bytes": 2264616, "device": "mps", "readback_seconds": 0.023044838999999998, "configuration_id": "atlas-01/C03", "platform": "atlas-01", "relative_batch_weight": 1 }, { "response_read_wait_seconds": 74.27892257999996, "requests": 300, "coordinator_exchange_wall_seconds": 75.22893874099996, "inner_threads": 1, "backend": "cuda_cublas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 655360, "request_write_seconds": 0.9499169850000007, "request_payload_bytes": 61870640, "process_peak_rss_bytes": 252145664, "samples": 152920, "upload_preparation_seconds": 1.3072958470000007, "worker_total_seconds": 2.478999484000003, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 0.7740768829999997, "owned_device_tensor_peak_bytes": 2264616, "device": "cuda", "readback_seconds": 0.14073625700000006, "configuration_id": "atlas-02/C03", "platform": "atlas-02", "relative_batch_weight": 1 }, { "response_read_wait_seconds": 68.02293487299998, "requests": 300, "coordinator_exchange_wall_seconds": 68.97260412699998, "inner_threads": 1, "backend": "cuda_cublas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 655360, "request_write_seconds": 0.9496139069999998, "request_payload_bytes": 61870640, "process_peak_rss_bytes": 286322688, "samples": 152920, "upload_preparation_seconds": 0.9706038950000002, "worker_total_seconds": 1.6777562749999992, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 0.46720008700000026, "owned_device_tensor_peak_bytes": 2264616, "device": "cuda", "readback_seconds": 0.07297718599999994, "configuration_id": "atlas-03/C03", "platform": "atlas-03", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu32", "reference_trial": 1, "dtype": "Float32", "predeclared_tolerances": { "parameter_max_abs": 0.001, "loss_max_abs": 0.0005, "accuracy_max_abs": 0.005 }, "parameter_max_abs_error": 4.633516073226929e-05, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 9.931197130219616e-07, "validation_loss": 1.1455976344776353e-06, "test_loss": 9.786561206981403e-07 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.00010172939979657958, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "01-02-03-gpu32", "repetition": 2, "started_utc": "2026-10-12T21:23:01.558721+00:00", "finished_utc": "2026-10-12T21:24:27.652051+00:00", "outer_command_wall_seconds": 86.09376583401172, "warmup_excluded_from_statistics": false, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 77922304, "test_accuracy": 0.8955239064089522, "parameter_sha256": "267925aebd1e6e485966fd3926f1b0e321368eb6395472f921d629c627c888a5", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8857578840284842, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.3770934779472095, "validation_loss": 1.2565156643897568, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 13.088093666 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.1680702881156562, "validation_loss": 1.072663564672305, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 20.483112958 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.978449299727543, "validation_loss": 0.8820205004872529, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 29.110661666 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923334504246493, "validation_loss": 0.7103084927900398, "validation_accuracy": 0.7501525940996948, "elapsed_seconds": 38.658434541 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370661070981077, "validation_loss": 0.577175882655752, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 46.866637416 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222785054353695, "validation_loss": 0.481975248113411, "validation_accuracy": 0.8319430315361139, "elapsed_seconds": 53.7820955 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.44097682620333817, "validation_loss": 0.41512579553488677, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 60.06701025 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.3831498520357044, "validation_loss": 0.36713215224861734, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 69.574346791 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.34060882918919294, "validation_loss": 0.3310873703760145, "validation_accuracy": 0.8787385554425229, "elapsed_seconds": 77.942009916 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.3077999734350523, "validation_loss": 0.30337328036464484, "validation_accuracy": 0.8857578840284842, "elapsed_seconds": 85.411605625 } ], "validation_loss": 0.30337328036464484, "coordinator_cpu_threads": null, "rate": 0.05, "coordinator_cpu_provider": "apple_accelerate", "test_loss": 0.29705295057650866, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332174875397, "fit_seconds": 85.590917916, "training_loss": 0.2923465431937054, "tensor_owned_device_peak_bytes": 0, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 6303328, "dtype": "Float32", "command_seconds": 86.053667791, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float32", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 80.45096650299998, "initialization_seconds": 3.627995167, "failed": false, "nodes": [ { "response_read_wait_seconds": 5.417965785999997, "requests": 300, "coordinator_exchange_wall_seconds": 6.368811862000002, "inner_threads": null, "backend": "metal_mps", "target": "macos-arm64", "owned_host_tensor_peak_bytes": 655360, "request_write_seconds": 0.9507761470000002, "request_payload_bytes": 61870640, "process_peak_rss_bytes": 22003712, "samples": 152920, "upload_preparation_seconds": 0.2503211539999999, "worker_total_seconds": 3.480475675999999, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 3.1651296360000023, "owned_device_tensor_peak_bytes": 2264616, "device": "mps", "readback_seconds": 0.024552329999999994, "configuration_id": "atlas-01/C03", "platform": "atlas-01", "relative_batch_weight": 1 }, { "response_read_wait_seconds": 75.50457230200001, "requests": 300, "coordinator_exchange_wall_seconds": 76.56417875299995, "inner_threads": 1, "backend": "cuda_cublas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 655360, "request_write_seconds": 1.0594542839999999, "request_payload_bytes": 61870640, "process_peak_rss_bytes": 252153856, "samples": 152920, "upload_preparation_seconds": 1.307947765000001, "worker_total_seconds": 2.474586277, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 0.7703047389999997, "owned_device_tensor_peak_bytes": 2264616, "device": "cuda", "readback_seconds": 0.1406611379999999, "configuration_id": "atlas-02/C03", "platform": "atlas-02", "relative_batch_weight": 1 }, { "response_read_wait_seconds": 67.35432496700007, "requests": 300, "coordinator_exchange_wall_seconds": 68.41272091100004, "inner_threads": 1, "backend": "cuda_cublas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 655360, "request_write_seconds": 1.0583404930000002, "request_payload_bytes": 61870640, "process_peak_rss_bytes": 286232576, "samples": 152920, "upload_preparation_seconds": 0.9694459050000002, "worker_total_seconds": 1.6511017689999994, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 0.4391299450000002, "owned_device_tensor_peak_bytes": 2264616, "device": "cuda", "readback_seconds": 0.07221494900000006, "configuration_id": "atlas-03/C03", "platform": "atlas-03", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu32", "reference_trial": 1, "dtype": "Float32", "predeclared_tolerances": { "parameter_max_abs": 0.001, "loss_max_abs": 0.0005, "accuracy_max_abs": 0.005 }, "parameter_max_abs_error": 4.633516073226929e-05, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 9.931197130219616e-07, "validation_loss": 1.1455976344776353e-06, "test_loss": 9.786561206981403e-07 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.00010172939979657958, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "01-02-03-gpu32", "repetition": 3, "started_utc": "2026-10-12T21:24:28.317151+00:00", "finished_utc": "2026-10-12T21:25:54.393411+00:00", "outer_command_wall_seconds": 86.07671895899693, "warmup_excluded_from_statistics": false, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 77938688, "test_accuracy": 0.8955239064089522, "parameter_sha256": "267925aebd1e6e485966fd3926f1b0e321368eb6395472f921d629c627c888a5", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8857578840284842, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.3770934779472095, "validation_loss": 1.2565156643897568, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 14.2945245 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.1680702881156562, "validation_loss": 1.072663564672305, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 22.088109416 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.978449299727543, "validation_loss": 0.8820205004872529, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 30.470619166 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923334504246493, "validation_loss": 0.7103084927900398, "validation_accuracy": 0.7501525940996948, "elapsed_seconds": 37.831608375 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370661070981077, "validation_loss": 0.577175882655752, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 46.762338583 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222785054353695, "validation_loss": 0.481975248113411, "validation_accuracy": 0.8319430315361139, "elapsed_seconds": 53.68550825 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.44097682620333817, "validation_loss": 0.41512579553488677, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 59.77911775 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.3831498520357044, "validation_loss": 0.36713215224861734, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 68.868621833 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.34060882918919294, "validation_loss": 0.3310873703760145, "validation_accuracy": 0.8787385554425229, "elapsed_seconds": 77.172159583 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.3077999734350523, "validation_loss": 0.30337328036464484, "validation_accuracy": 0.8857578840284842, "elapsed_seconds": 85.3885745 } ], "validation_loss": 0.30337328036464484, "coordinator_cpu_threads": null, "rate": 0.05, "coordinator_cpu_provider": "apple_accelerate", "test_loss": 0.29705295057650866, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332174875397, "fit_seconds": 85.56780825, "training_loss": 0.2923465431937054, "tensor_owned_device_peak_bytes": 0, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 6303328, "dtype": "Float32", "command_seconds": 86.037888583, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float32", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 79.88829062200001, "initialization_seconds": 4.154696167, "failed": false, "nodes": [ { "response_read_wait_seconds": 5.396259586000003, "requests": 300, "coordinator_exchange_wall_seconds": 6.3356772550000064, "inner_threads": null, "backend": "metal_mps", "target": "macos-arm64", "owned_host_tensor_peak_bytes": 655360, "request_write_seconds": 0.9393523770000002, "request_payload_bytes": 61870640, "process_peak_rss_bytes": 21807104, "samples": 152920, "upload_preparation_seconds": 0.24906942500000023, "worker_total_seconds": 3.4586895430000006, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 3.146254414000001, "owned_device_tensor_peak_bytes": 2264616, "device": "mps", "readback_seconds": 0.023416451999999997, "configuration_id": "atlas-01/C03", "platform": "atlas-01", "relative_batch_weight": 1 }, { "response_read_wait_seconds": 73.03388654199996, "requests": 300, "coordinator_exchange_wall_seconds": 74.09127016700003, "inner_threads": 1, "backend": "cuda_cublas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 655360, "request_write_seconds": 1.057320459, "request_payload_bytes": 61870640, "process_peak_rss_bytes": 252051456, "samples": 152920, "upload_preparation_seconds": 1.3187351210000007, "worker_total_seconds": 2.511008313, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 0.7919749250000002, "owned_device_tensor_peak_bytes": 2264616, "device": "cuda", "readback_seconds": 0.14422723300000018, "configuration_id": "atlas-02/C03", "platform": "atlas-02", "relative_batch_weight": 1 }, { "response_read_wait_seconds": 71.42854133200002, "requests": 300, "coordinator_exchange_wall_seconds": 72.48310226199992, "inner_threads": 1, "backend": "cuda_cublas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 655360, "request_write_seconds": 1.0544526330000004, "request_payload_bytes": 61870640, "process_peak_rss_bytes": 286314496, "samples": 152920, "upload_preparation_seconds": 0.9608044390000001, "worker_total_seconds": 1.643087887, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 0.442767139, "owned_device_tensor_peak_bytes": 2264616, "device": "cuda", "readback_seconds": 0.07239806899999998, "configuration_id": "atlas-03/C03", "platform": "atlas-03", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu32", "reference_trial": 1, "dtype": "Float32", "predeclared_tolerances": { "parameter_max_abs": 0.001, "loss_max_abs": 0.0005, "accuracy_max_abs": 0.005 }, "parameter_max_abs_error": 4.633516073226929e-05, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 9.931197130219616e-07, "validation_loss": 1.1455976344776353e-06, "test_loss": 9.786561206981403e-07 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.00010172939979657958, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "02-03-gpu64", "repetition": 1, "started_utc": "2026-10-12T21:27:50.883871+00:00", "finished_utc": "2026-10-12T21:29:47.882374+00:00", "outer_command_wall_seconds": 116.99921195799834, "warmup_excluded_from_statistics": false, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 76529664, "test_accuracy": 0.8955239064089522, "parameter_sha256": "2e41fec29bc85fdf18b8447ee70bad448e44b1ed687e126d50c34e7270df079d", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8857578840284842, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.377093474953068, "validation_loss": 1.2565156669823923, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 15.988621125 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.1680702765782571, "validation_loss": 1.0726634152606465, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 26.763810208 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784489967242784, "validation_loss": 0.8820201711707012, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 37.077493875 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923336749857174, "validation_loss": 0.7103083529584678, "validation_accuracy": 0.7502543234994914, "elapsed_seconds": 49.44496725 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370655949431793, "validation_loss": 0.5771756065192747, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 61.2291545 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222779060476093, "validation_loss": 0.4819741321230155, "validation_accuracy": 0.8320447609359105, "elapsed_seconds": 71.429050292 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.44097582388285234, "validation_loss": 0.41512437864999546, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 81.033892 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.38314860589065225, "validation_loss": 0.36713147013723113, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 93.402497875 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.3406074100312188, "validation_loss": 0.3310860565229032, "validation_accuracy": 0.8786368260427263, "elapsed_seconds": 104.583983833 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.3077992643371988, "validation_loss": 0.3033731649951008, "validation_accuracy": 0.8857578840284842, "elapsed_seconds": 116.261746792 } ], "validation_loss": 0.3033731649951008, "coordinator_cpu_threads": null, "rate": 0.05, "coordinator_cpu_provider": "apple_accelerate", "test_loss": 0.29705317871020637, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332128378295, "fit_seconds": 116.467892458, "training_loss": 0.2923469152365444, "tensor_owned_device_peak_bytes": 0, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 4202560, "dtype": "Float64", "command_seconds": 116.93126075, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float64", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 111.007873166, "initialization_seconds": 4.062272708, "failed": false, "nodes": [ { "response_read_wait_seconds": 104.58419616099997, "requests": 300, "coordinator_exchange_wall_seconds": 107.19696121899996, "inner_threads": 1, "backend": "cuda_cublas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 393216, "request_write_seconds": 2.612709755, "request_payload_bytes": 165262320, "process_peak_rss_bytes": 254693376, "samples": 229380, "upload_preparation_seconds": 1.1209860590000003, "worker_total_seconds": 3.444995925000001, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 1.773451806000001, "owned_device_tensor_peak_bytes": 6659152, "device": "cuda", "readback_seconds": 0.16438954099999997, "configuration_id": "atlas-02/C03", "platform": "atlas-02", "relative_batch_weight": 1 }, { "response_read_wait_seconds": 96.94922095000004, "requests": 300, "coordinator_exchange_wall_seconds": 99.56707587099999, "inner_threads": 1, "backend": "cuda_cublas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 393216, "request_write_seconds": 2.6178147880000004, "request_payload_bytes": 165262320, "process_peak_rss_bytes": 282214400, "samples": 229380, "upload_preparation_seconds": 0.45382250100000004, "worker_total_seconds": 1.1185561859999993, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 0.4439623549999999, "owned_device_tensor_peak_bytes": 6659152, "device": "cuda", "readback_seconds": 0.06328919300000004, "configuration_id": "atlas-03/C03", "platform": "atlas-03", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu64", "reference_trial": 1, "dtype": "Float64", "predeclared_tolerances": { "parameter_max_abs": 1e-08, "loss_max_abs": 1e-08, "accuracy_max_abs": 1e-12 }, "parameter_max_abs_error": 3.3306690738754696e-16, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 5.551115123125783e-17, "validation_loss": 5.551115123125783e-17, "test_loss": 5.551115123125783e-17 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "02-03-gpu64", "repetition": 2, "started_utc": "2026-10-12T21:29:48.553814+00:00", "finished_utc": "2026-10-12T21:31:54.803212+00:00", "outer_command_wall_seconds": 126.23090900000534, "warmup_excluded_from_statistics": false, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 76169216, "test_accuracy": 0.8955239064089522, "parameter_sha256": "2e41fec29bc85fdf18b8447ee70bad448e44b1ed687e126d50c34e7270df079d", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8857578840284842, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.377093474953068, "validation_loss": 1.2565156669823923, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 15.949470166 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.1680702765782571, "validation_loss": 1.0726634152606465, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 27.538246833 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784489967242784, "validation_loss": 0.8820201711707012, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 39.132941583 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923336749857174, "validation_loss": 0.7103083529584678, "validation_accuracy": 0.7502543234994914, "elapsed_seconds": 50.726370166 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370655949431793, "validation_loss": 0.5771756065192747, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 63.6551475 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222779060476093, "validation_loss": 0.4819741321230155, "validation_accuracy": 0.8320447609359105, "elapsed_seconds": 77.326266208 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.44097582388285234, "validation_loss": 0.41512437864999546, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 89.419806291 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.38314860589065225, "validation_loss": 0.36713147013723113, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 100.660684583 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.3406074100312188, "validation_loss": 0.3310860565229032, "validation_accuracy": 0.8786368260427263, "elapsed_seconds": 112.950963916 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.3077992643371988, "validation_loss": 0.3033731649951008, "validation_accuracy": 0.8857578840284842, "elapsed_seconds": 125.491803791 } ], "validation_loss": 0.3033731649951008, "coordinator_cpu_threads": null, "rate": 0.05, "coordinator_cpu_provider": "apple_accelerate", "test_loss": 0.29705317871020637, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332128378295, "fit_seconds": 125.694160875, "training_loss": 0.2923469152365444, "tensor_owned_device_peak_bytes": 0, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 4202560, "dtype": "Float64", "command_seconds": 126.164727625, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float64", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 120.20119451000001, "initialization_seconds": 4.104409833, "failed": false, "nodes": [ { "response_read_wait_seconds": 108.28300708299992, "requests": 300, "coordinator_exchange_wall_seconds": 110.89321180400003, "inner_threads": 1, "backend": "cuda_cublas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 393216, "request_write_seconds": 2.610150838999998, "request_payload_bytes": 165262320, "process_peak_rss_bytes": 254693376, "samples": 229380, "upload_preparation_seconds": 1.132229757, "worker_total_seconds": 3.4631106159999994, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 1.7728895749999996, "owned_device_tensor_peak_bytes": 6659152, "device": "cuda", "readback_seconds": 0.167989555, "configuration_id": "atlas-02/C03", "platform": "atlas-02", "relative_batch_weight": 1 }, { "response_read_wait_seconds": 107.93112454199999, "requests": 300, "coordinator_exchange_wall_seconds": 110.54037168700005, "inner_threads": 1, "backend": "cuda_cublas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 393216, "request_write_seconds": 2.609176034999999, "request_payload_bytes": 165262320, "process_peak_rss_bytes": 281931776, "samples": 229380, "upload_preparation_seconds": 0.45233597800000014, "worker_total_seconds": 1.1019800560000002, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 0.43328174500000005, "owned_device_tensor_peak_bytes": 6659152, "device": "cuda", "readback_seconds": 0.06230874499999999, "configuration_id": "atlas-03/C03", "platform": "atlas-03", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu64", "reference_trial": 1, "dtype": "Float64", "predeclared_tolerances": { "parameter_max_abs": 1e-08, "loss_max_abs": 1e-08, "accuracy_max_abs": 1e-12 }, "parameter_max_abs_error": 3.3306690738754696e-16, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 5.551115123125783e-17, "validation_loss": 5.551115123125783e-17, "test_loss": 5.551115123125783e-17 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "02-03-gpu64", "repetition": 3, "started_utc": "2026-10-12T21:31:55.489718+00:00", "finished_utc": "2026-10-12T21:34:00.144617+00:00", "outer_command_wall_seconds": 124.655269416995, "warmup_excluded_from_statistics": false, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 75055104, "test_accuracy": 0.8955239064089522, "parameter_sha256": "2e41fec29bc85fdf18b8447ee70bad448e44b1ed687e126d50c34e7270df079d", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8857578840284842, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.377093474953068, "validation_loss": 1.2565156669823923, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 17.837028291 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.1680702765782571, "validation_loss": 1.0726634152606465, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 29.710270625 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784489967242784, "validation_loss": 0.8820201711707012, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 41.597968916 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923336749857174, "validation_loss": 0.7103083529584678, "validation_accuracy": 0.7502543234994914, "elapsed_seconds": 54.631133166 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370655949431793, "validation_loss": 0.5771756065192747, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 66.563338291 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222779060476093, "validation_loss": 0.4819741321230155, "validation_accuracy": 0.8320447609359105, "elapsed_seconds": 77.3345915 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.44097582388285234, "validation_loss": 0.41512437864999546, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 87.610939958 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.38314860589065225, "validation_loss": 0.36713147013723113, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 99.592040708 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.3406074100312188, "validation_loss": 0.3310860565229032, "validation_accuracy": 0.8786368260427263, "elapsed_seconds": 110.900311166 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.3077992643371988, "validation_loss": 0.3033731649951008, "validation_accuracy": 0.8857578840284842, "elapsed_seconds": 123.920459166 } ], "validation_loss": 0.3033731649951008, "coordinator_cpu_threads": null, "rate": 0.05, "coordinator_cpu_provider": "apple_accelerate", "test_loss": 0.29705317871020637, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332128378295, "fit_seconds": 124.125102542, "training_loss": 0.2923469152365444, "tensor_owned_device_peak_bytes": 0, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 4202560, "dtype": "Float64", "command_seconds": 124.584389833, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float64", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 119.00228512599999, "initialization_seconds": 3.715385375, "failed": false, "nodes": [ { "response_read_wait_seconds": 111.21368550599998, "requests": 300, "coordinator_exchange_wall_seconds": 113.86994150799987, "inner_threads": 1, "backend": "cuda_cublas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 393216, "request_write_seconds": 2.656146796, "request_payload_bytes": 165262320, "process_peak_rss_bytes": 254758912, "samples": 229380, "upload_preparation_seconds": 1.1247009410000004, "worker_total_seconds": 3.4443716670000013, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 1.7582898530000002, "owned_device_tensor_peak_bytes": 6659152, "device": "cuda", "readback_seconds": 0.17331558899999996, "configuration_id": "atlas-02/C03", "platform": "atlas-02", "relative_batch_weight": 1 }, { "response_read_wait_seconds": 94.17421533999999, "requests": 300, "coordinator_exchange_wall_seconds": 96.842004244, "inner_threads": 1, "backend": "cuda_cublas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 393216, "request_write_seconds": 2.6677069549999985, "request_payload_bytes": 165262320, "process_peak_rss_bytes": 282042368, "samples": 229380, "upload_preparation_seconds": 0.45668940900000027, "worker_total_seconds": 1.1111053540000004, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 0.434931702, "owned_device_tensor_peak_bytes": 6659152, "device": "cuda", "readback_seconds": 0.062440942000000006, "configuration_id": "atlas-03/C03", "platform": "atlas-03", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu64", "reference_trial": 1, "dtype": "Float64", "predeclared_tolerances": { "parameter_max_abs": 1e-08, "loss_max_abs": 1e-08, "accuracy_max_abs": 1e-12 }, "parameter_max_abs_error": 3.3306690738754696e-16, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 5.551115123125783e-17, "validation_loss": 5.551115123125783e-17, "test_loss": 5.551115123125783e-17 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "01-02-03-cpu32", "repetition": 1, "started_utc": "2026-10-12T21:35:22.826943+00:00", "finished_utc": "2026-10-12T21:36:51.990229+00:00", "outer_command_wall_seconds": 89.16344270898844, "warmup_excluded_from_statistics": false, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 76414976, "test_accuracy": 0.8955239064089522, "parameter_sha256": "0515a7177dd3e8dd33592beb68b3d4ee4068c8d014af92de819605ffeffb41e6", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8858596134282808, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.377093476616772, "validation_loss": 1.2565156643897568, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 11.003902708 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.1680702856314797, "validation_loss": 1.072663564672305, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 18.803377708 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784492924101361, "validation_loss": 0.8820205098008576, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 27.825012333 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923334511782175, "validation_loss": 0.7103085021036445, "validation_accuracy": 0.7501525940996948, "elapsed_seconds": 38.406651 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.63706611166629, "validation_loss": 0.577175882655752, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 47.76885925 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222784787122826, "validation_loss": 0.4819751335609238, "validation_accuracy": 0.8319430315361139, "elapsed_seconds": 56.188603833 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.4409767381969684, "validation_loss": 0.41512574337627495, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 63.984500833 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.38314981840577533, "validation_loss": 0.36713236366623175, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 72.247348125 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.3406089726244003, "validation_loss": 0.33108807820997305, "validation_accuracy": 0.8787385554425229, "elapsed_seconds": 80.429953625 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.3078001165168622, "validation_loss": 0.30337432724777097, "validation_accuracy": 0.8858596134282808, "elapsed_seconds": 88.4635365 } ], "validation_loss": 0.30337432724777097, "coordinator_cpu_threads": null, "rate": 0.05, "coordinator_cpu_provider": "apple_accelerate", "test_loss": 0.2970538696340481, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332174875397, "fit_seconds": 88.645885167, "training_loss": 0.2923475254698321, "tensor_owned_device_peak_bytes": 0, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 6303328, "dtype": "Float32", "command_seconds": 89.109699709, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float32", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 83.408396453, "initialization_seconds": 3.706035625, "failed": false, "nodes": [ { "response_read_wait_seconds": 3.0818655430000006, "requests": 300, "coordinator_exchange_wall_seconds": 3.9981469220000023, "inner_threads": null, "backend": "apple_accelerate", "target": "macos-arm64", "owned_host_tensor_peak_bytes": 6834532, "request_write_seconds": 0.9162247140000003, "request_payload_bytes": 61870640, "process_peak_rss_bytes": 25133056, "samples": 152920, "upload_preparation_seconds": 0.15840661500000008, "worker_total_seconds": 1.2889534940000007, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 1.0918180099999997, "owned_device_tensor_peak_bytes": 0, "device": "cpu", "readback_seconds": 0.018997038999999993, "configuration_id": "atlas-01/C03", "platform": "atlas-01", "relative_batch_weight": 1 }, { "response_read_wait_seconds": 77.784782799, "requests": 300, "coordinator_exchange_wall_seconds": 78.81341042199998, "inner_threads": 1, "backend": "openblas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 6834532, "request_write_seconds": 1.028517587, "request_payload_bytes": 61870640, "process_peak_rss_bytes": 124428288, "samples": 152920, "upload_preparation_seconds": 1.348304754, "worker_total_seconds": 13.917500425000002, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 12.001523366999999, "owned_device_tensor_peak_bytes": 0, "device": "cpu", "readback_seconds": 0.14739254699999993, "configuration_id": "atlas-02/C03", "platform": "atlas-02", "relative_batch_weight": 1 }, { "response_read_wait_seconds": 73.57592612399998, "requests": 300, "coordinator_exchange_wall_seconds": 74.60627787600002, "inner_threads": 1, "backend": "openblas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 6834532, "request_write_seconds": 1.0302939589999998, "request_payload_bytes": 61870640, "process_peak_rss_bytes": 124391424, "samples": 152920, "upload_preparation_seconds": 0.9753726200000005, "worker_total_seconds": 5.588757929000004, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 4.426885362000004, "owned_device_tensor_peak_bytes": 0, "device": "cpu", "readback_seconds": 0.03694794100000001, "configuration_id": "atlas-03/C03", "platform": "atlas-03", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu32", "reference_trial": 1, "dtype": "Float32", "predeclared_tolerances": { "parameter_max_abs": 0.001, "loss_max_abs": 0.0005, "accuracy_max_abs": 0.005 }, "parameter_max_abs_error": 4.561617970466614e-06, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 1.0843586306652497e-08, "validation_loss": 9.871450834531004e-08, "test_loss": 5.9598581236830483e-08 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "01-02-03-cpu32", "repetition": 2, "started_utc": "2026-10-12T21:36:52.737796+00:00", "finished_utc": "2026-10-12T21:38:25.479464+00:00", "outer_command_wall_seconds": 92.74185400000715, "warmup_excluded_from_statistics": false, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 77889536, "test_accuracy": 0.8955239064089522, "parameter_sha256": "0515a7177dd3e8dd33592beb68b3d4ee4068c8d014af92de819605ffeffb41e6", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8858596134282808, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.377093476616772, "validation_loss": 1.2565156643897568, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 15.009750042 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.1680702856314797, "validation_loss": 1.072663564672305, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 24.134078833 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784492924101361, "validation_loss": 0.8820205098008576, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 33.54404475 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923334511782175, "validation_loss": 0.7103085021036445, "validation_accuracy": 0.7501525940996948, "elapsed_seconds": 41.998970708 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.63706611166629, "validation_loss": 0.577175882655752, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 49.715118875 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222784787122826, "validation_loss": 0.4819751335609238, "validation_accuracy": 0.8319430315361139, "elapsed_seconds": 57.070782958 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.4409767381969684, "validation_loss": 0.41512574337627495, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 66.806895542 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.38314981840577533, "validation_loss": 0.36713236366623175, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 75.607657292 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.3406089726244003, "validation_loss": 0.33108807820997305, "validation_accuracy": 0.8787385554425229, "elapsed_seconds": 84.037268125 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.3078001165168622, "validation_loss": 0.30337432724777097, "validation_accuracy": 0.8858596134282808, "elapsed_seconds": 92.046533292 } ], "validation_loss": 0.30337432724777097, "coordinator_cpu_threads": null, "rate": 0.05, "coordinator_cpu_provider": "apple_accelerate", "test_loss": 0.2970538696340481, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332174875397, "fit_seconds": 92.231870458, "training_loss": 0.2923475254698321, "tensor_owned_device_peak_bytes": 0, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 6303328, "dtype": "Float32", "command_seconds": 92.696210667, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float32", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 87.13350698900007, "initialization_seconds": 3.610389666, "failed": false, "nodes": [ { "response_read_wait_seconds": 2.944582882999999, "requests": 300, "coordinator_exchange_wall_seconds": 3.811123540000002, "inner_threads": null, "backend": "apple_accelerate", "target": "macos-arm64", "owned_host_tensor_peak_bytes": 6834532, "request_write_seconds": 0.8664736270000004, "request_payload_bytes": 61870640, "process_peak_rss_bytes": 24117248, "samples": 152920, "upload_preparation_seconds": 0.15058891999999988, "worker_total_seconds": 1.2409844569999993, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 1.0529558259999998, "owned_device_tensor_peak_bytes": 0, "device": "cpu", "readback_seconds": 0.018368426, "configuration_id": "atlas-01/C03", "platform": "atlas-01", "relative_batch_weight": 1 }, { "response_read_wait_seconds": 79.15330272000001, "requests": 300, "coordinator_exchange_wall_seconds": 80.12637662399999, "inner_threads": 1, "backend": "openblas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 6834532, "request_write_seconds": 0.9729863700000007, "request_payload_bytes": 61870640, "process_peak_rss_bytes": 124116992, "samples": 152920, "upload_preparation_seconds": 1.3649168740000002, "worker_total_seconds": 13.892850112000001, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 11.949936006, "owned_device_tensor_peak_bytes": 0, "device": "cpu", "readback_seconds": 0.146685316, "configuration_id": "atlas-02/C03", "platform": "atlas-02", "relative_batch_weight": 1 }, { "response_read_wait_seconds": 74.19305296100005, "requests": 300, "coordinator_exchange_wall_seconds": 75.160350051, "inner_threads": 1, "backend": "openblas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 6834532, "request_write_seconds": 0.9672401329999998, "request_payload_bytes": 61870640, "process_peak_rss_bytes": 124420096, "samples": 152920, "upload_preparation_seconds": 0.9685700180000002, "worker_total_seconds": 5.551128372999997, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 4.399906118000004, "owned_device_tensor_peak_bytes": 0, "device": "cpu", "readback_seconds": 0.036184635999999985, "configuration_id": "atlas-03/C03", "platform": "atlas-03", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu32", "reference_trial": 1, "dtype": "Float32", "predeclared_tolerances": { "parameter_max_abs": 0.001, "loss_max_abs": 0.0005, "accuracy_max_abs": 0.005 }, "parameter_max_abs_error": 4.561617970466614e-06, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 1.0843586306652497e-08, "validation_loss": 9.871450834531004e-08, "test_loss": 5.9598581236830483e-08 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "01-02-03-cpu32", "repetition": 3, "started_utc": "2026-10-12T21:38:26.225883+00:00", "finished_utc": "2026-10-12T21:39:54.443759+00:00", "outer_command_wall_seconds": 88.21805833300459, "warmup_excluded_from_statistics": false, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 77873152, "test_accuracy": 0.8955239064089522, "parameter_sha256": "0515a7177dd3e8dd33592beb68b3d4ee4068c8d014af92de819605ffeffb41e6", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8858596134282808, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.377093476616772, "validation_loss": 1.2565156643897568, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 11.686840417 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.1680702856314797, "validation_loss": 1.072663564672305, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 20.859513625 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784492924101361, "validation_loss": 0.8820205098008576, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 29.420555542 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923334511782175, "validation_loss": 0.7103085021036445, "validation_accuracy": 0.7501525940996948, "elapsed_seconds": 38.216397875 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.63706611166629, "validation_loss": 0.577175882655752, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 46.650260417 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222784787122826, "validation_loss": 0.4819751335609238, "validation_accuracy": 0.8319430315361139, "elapsed_seconds": 55.248262 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.4409767381969684, "validation_loss": 0.41512574337627495, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 63.574626459 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.38314981840577533, "validation_loss": 0.36713236366623175, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 71.896393209 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.3406089726244003, "validation_loss": 0.33108807820997305, "validation_accuracy": 0.8787385554425229, "elapsed_seconds": 79.942112917 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.3078001165168622, "validation_loss": 0.30337432724777097, "validation_accuracy": 0.8858596134282808, "elapsed_seconds": 87.459792625 } ], "validation_loss": 0.30337432724777097, "coordinator_cpu_threads": null, "rate": 0.05, "coordinator_cpu_provider": "apple_accelerate", "test_loss": 0.2970538696340481, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332174875397, "fit_seconds": 87.638751083, "training_loss": 0.2923475254698321, "tensor_owned_device_peak_bytes": 0, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 6303328, "dtype": "Float32", "command_seconds": 88.129241583, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float32", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 82.48221805299991, "initialization_seconds": 3.64844775, "failed": false, "nodes": [ { "response_read_wait_seconds": 3.093941788, "requests": 300, "coordinator_exchange_wall_seconds": 4.0202211650000015, "inner_threads": null, "backend": "apple_accelerate", "target": "macos-arm64", "owned_host_tensor_peak_bytes": 6834532, "request_write_seconds": 0.9262225830000002, "request_payload_bytes": 61870640, "process_peak_rss_bytes": 24690688, "samples": 152920, "upload_preparation_seconds": 0.15966467000000004, "worker_total_seconds": 1.2952419939999993, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 1.0963840000000002, "owned_device_tensor_peak_bytes": 0, "device": "cpu", "readback_seconds": 0.019318249000000006, "configuration_id": "atlas-01/C03", "platform": "atlas-01", "relative_batch_weight": 1 }, { "response_read_wait_seconds": 77.513491078, "requests": 300, "coordinator_exchange_wall_seconds": 78.54798208700002, "inner_threads": 1, "backend": "openblas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 6834532, "request_write_seconds": 1.0344487010000003, "request_payload_bytes": 61870640, "process_peak_rss_bytes": 124194816, "samples": 152920, "upload_preparation_seconds": 1.3208315059999995, "worker_total_seconds": 13.59844539500001, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 11.702780269000003, "owned_device_tensor_peak_bytes": 0, "device": "cpu", "readback_seconds": 0.14600822400000005, "configuration_id": "atlas-02/C03", "platform": "atlas-02", "relative_batch_weight": 1 }, { "response_read_wait_seconds": 65.71419562500003, "requests": 300, "coordinator_exchange_wall_seconds": 66.752902796, "inner_threads": 1, "backend": "openblas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 6834532, "request_write_seconds": 1.0386550049999992, "request_payload_bytes": 61870640, "process_peak_rss_bytes": 124260352, "samples": 152920, "upload_preparation_seconds": 0.9642789750000003, "worker_total_seconds": 5.463456344000003, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 4.324123430000001, "owned_device_tensor_peak_bytes": 0, "device": "cpu", "readback_seconds": 0.03586423799999997, "configuration_id": "atlas-03/C03", "platform": "atlas-03", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu32", "reference_trial": 1, "dtype": "Float32", "predeclared_tolerances": { "parameter_max_abs": 0.001, "loss_max_abs": 0.0005, "accuracy_max_abs": 0.005 }, "parameter_max_abs_error": 4.561617970466614e-06, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 1.0843586306652497e-08, "validation_loss": 9.871450834531004e-08, "test_loss": 5.9598581236830483e-08 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "01-02-03-cpu64", "repetition": 1, "started_utc": "2026-10-12T21:41:53.751355+00:00", "finished_utc": "2026-10-12T21:43:42.388119+00:00", "outer_command_wall_seconds": 108.63698216599005, "warmup_excluded_from_statistics": false, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 70926336, "test_accuracy": 0.8955239064089522, "parameter_sha256": "41b2bc8c6a1cd76505a0568ed6c5f96c91a1999298441dbd8cc81399f9ada917", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8857578840284842, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.377093474953068, "validation_loss": 1.2565156669823925, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 14.906687458 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.1680702765782573, "validation_loss": 1.0726634152606465, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 26.721767708 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784489967242783, "validation_loss": 0.8820201711707012, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 37.208723416 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923336749857173, "validation_loss": 0.710308352958468, "validation_accuracy": 0.7502543234994914, "elapsed_seconds": 48.335735083 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370655949431792, "validation_loss": 0.5771756065192747, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 58.265553125 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222779060476092, "validation_loss": 0.4819741321230154, "validation_accuracy": 0.8320447609359105, "elapsed_seconds": 67.9275475 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.44097582388285234, "validation_loss": 0.4151243786499954, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 77.70821025 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.38314860589065225, "validation_loss": 0.3671314701372311, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 88.048695916 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.3406074100312187, "validation_loss": 0.3310860565229032, "validation_accuracy": 0.8786368260427263, "elapsed_seconds": 98.471638666 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.3077992643371988, "validation_loss": 0.3033731649951008, "validation_accuracy": 0.8857578840284842, "elapsed_seconds": 107.902759666 } ], "validation_loss": 0.3033731649951008, "coordinator_cpu_threads": null, "rate": 0.05, "coordinator_cpu_provider": "apple_accelerate", "test_loss": 0.2970531787102063, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332128378295, "fit_seconds": 108.103035375, "training_loss": 0.29234691523654444, "tensor_owned_device_peak_bytes": 0, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 4202560, "dtype": "Float64", "command_seconds": 108.564510833, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float64", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 103.10029991900007, "initialization_seconds": 3.571098291, "failed": false, "nodes": [ { "response_read_wait_seconds": 3.839887451999999, "requests": 300, "coordinator_exchange_wall_seconds": 5.567488794999997, "inner_threads": null, "backend": "apple_accelerate", "target": "macos-arm64", "owned_host_tensor_peak_bytes": 5398608, "request_write_seconds": 1.7275411719999993, "request_payload_bytes": 123668080, "process_peak_rss_bytes": 20987904, "samples": 152920, "upload_preparation_seconds": 0.10331621500000006, "worker_total_seconds": 1.4972432159999998, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 1.3599068440000004, "owned_device_tensor_peak_bytes": 0, "device": "cpu", "readback_seconds": 0.015248027000000003, "configuration_id": "atlas-01/C03", "platform": "atlas-01", "relative_batch_weight": 1 }, { "response_read_wait_seconds": 93.57786412000006, "requests": 300, "coordinator_exchange_wall_seconds": 95.71556680100005, "inner_threads": 1, "backend": "openblas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 5398608, "request_write_seconds": 2.1376533849999997, "request_payload_bytes": 123668080, "process_peak_rss_bytes": 123088896, "samples": 152920, "upload_preparation_seconds": 0.8924193459999997, "worker_total_seconds": 13.227574789999995, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 11.839041184000004, "owned_device_tensor_peak_bytes": 0, "device": "cpu", "readback_seconds": 0.16050355099999997, "configuration_id": "atlas-02/C03", "platform": "atlas-02", "relative_batch_weight": 1 }, { "response_read_wait_seconds": 90.68079633100007, "requests": 300, "coordinator_exchange_wall_seconds": 92.81445225600002, "inner_threads": 1, "backend": "openblas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 5398608, "request_write_seconds": 2.1335961649999993, "request_payload_bytes": 123668080, "process_peak_rss_bytes": 122892288, "samples": 152920, "upload_preparation_seconds": 0.45744333199999965, "worker_total_seconds": 4.483889233999999, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 3.8670035659999997, "owned_device_tensor_peak_bytes": 0, "device": "cpu", "readback_seconds": 0.038350755000000035, "configuration_id": "atlas-03/C03", "platform": "atlas-03", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu64", "reference_trial": 1, "dtype": "Float64", "predeclared_tolerances": { "parameter_max_abs": 1e-08, "loss_max_abs": 1e-08, "accuracy_max_abs": 1e-12 }, "parameter_max_abs_error": 2.7755575615628914e-16, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 0.0, "validation_loss": 5.551115123125783e-17, "test_loss": 1.1102230246251565e-16 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "01-02-03-cpu64", "repetition": 2, "started_utc": "2026-10-12T21:43:43.145563+00:00", "finished_utc": "2026-10-12T21:45:37.585234+00:00", "outer_command_wall_seconds": 114.44000404200051, "warmup_excluded_from_statistics": false, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 71483392, "test_accuracy": 0.8955239064089522, "parameter_sha256": "41b2bc8c6a1cd76505a0568ed6c5f96c91a1999298441dbd8cc81399f9ada917", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8857578840284842, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.377093474953068, "validation_loss": 1.2565156669823925, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 15.577016417 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.1680702765782573, "validation_loss": 1.0726634152606465, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 26.0532695 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784489967242783, "validation_loss": 0.8820201711707012, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 35.792331084 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923336749857173, "validation_loss": 0.710308352958468, "validation_accuracy": 0.7502543234994914, "elapsed_seconds": 45.687528167 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370655949431792, "validation_loss": 0.5771756065192747, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 57.607157709 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222779060476092, "validation_loss": 0.4819741321230154, "validation_accuracy": 0.8320447609359105, "elapsed_seconds": 69.0522235 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.44097582388285234, "validation_loss": 0.4151243786499954, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 80.180738834 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.38314860589065225, "validation_loss": 0.3671314701372311, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 91.542893209 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.3406074100312187, "validation_loss": 0.3310860565229032, "validation_accuracy": 0.8786368260427263, "elapsed_seconds": 103.055112625 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.3077992643371988, "validation_loss": 0.3033731649951008, "validation_accuracy": 0.8857578840284842, "elapsed_seconds": 113.698078 } ], "validation_loss": 0.3033731649951008, "coordinator_cpu_threads": null, "rate": 0.05, "coordinator_cpu_provider": "apple_accelerate", "test_loss": 0.2970531787102063, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332128378295, "fit_seconds": 113.89337375, "training_loss": 0.29234691523654444, "tensor_owned_device_peak_bytes": 0, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 4202560, "dtype": "Float64", "command_seconds": 114.357245166, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float64", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 108.85374649399996, "initialization_seconds": 3.568028291, "failed": false, "nodes": [ { "response_read_wait_seconds": 3.913244859999998, "requests": 300, "coordinator_exchange_wall_seconds": 5.695344581000001, "inner_threads": null, "backend": "apple_accelerate", "target": "macos-arm64", "owned_host_tensor_peak_bytes": 5398608, "request_write_seconds": 1.782044620999998, "request_payload_bytes": 123668080, "process_peak_rss_bytes": 20840448, "samples": 152920, "upload_preparation_seconds": 0.10499419699999991, "worker_total_seconds": 1.526615741000001, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 1.3873297540000007, "owned_device_tensor_peak_bytes": 0, "device": "cpu", "readback_seconds": 0.015355873999999993, "configuration_id": "atlas-01/C03", "platform": "atlas-01", "relative_batch_weight": 1 }, { "response_read_wait_seconds": 104.37799353499994, "requests": 300, "coordinator_exchange_wall_seconds": 106.56334433700003, "inner_threads": 1, "backend": "openblas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 5398608, "request_write_seconds": 2.1852947489999988, "request_payload_bytes": 123668080, "process_peak_rss_bytes": 122904576, "samples": 152920, "upload_preparation_seconds": 0.9124430939999998, "worker_total_seconds": 13.369700921999998, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 11.957221955999998, "owned_device_tensor_peak_bytes": 0, "device": "cpu", "readback_seconds": 0.16115268799999996, "configuration_id": "atlas-02/C03", "platform": "atlas-02", "relative_batch_weight": 1 }, { "response_read_wait_seconds": 92.89171421000005, "requests": 300, "coordinator_exchange_wall_seconds": 95.08160861699996, "inner_threads": 1, "backend": "openblas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 5398608, "request_write_seconds": 2.1897552950000008, "request_payload_bytes": 123668080, "process_peak_rss_bytes": 123039744, "samples": 152920, "upload_preparation_seconds": 0.4670489530000005, "worker_total_seconds": 4.534127982999999, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 3.906021795999999, "owned_device_tensor_peak_bytes": 0, "device": "cpu", "readback_seconds": 0.039353534999999995, "configuration_id": "atlas-03/C03", "platform": "atlas-03", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu64", "reference_trial": 1, "dtype": "Float64", "predeclared_tolerances": { "parameter_max_abs": 1e-08, "loss_max_abs": 1e-08, "accuracy_max_abs": 1e-12 }, "parameter_max_abs_error": 2.7755575615628914e-16, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 0.0, "validation_loss": 5.551115123125783e-17, "test_loss": 1.1102230246251565e-16 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "01-02-03-cpu64", "repetition": 3, "started_utc": "2026-10-12T21:45:38.382843+00:00", "finished_utc": "2026-10-12T21:47:35.452462+00:00", "outer_command_wall_seconds": 117.02706866699737, "warmup_excluded_from_statistics": false, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 73039872, "test_accuracy": 0.8955239064089522, "parameter_sha256": "41b2bc8c6a1cd76505a0568ed6c5f96c91a1999298441dbd8cc81399f9ada917", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8857578840284842, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.377093474953068, "validation_loss": 1.2565156669823925, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 16.746186916 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.1680702765782573, "validation_loss": 1.0726634152606465, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 26.973442708 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784489967242783, "validation_loss": 0.8820201711707012, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 37.163750375 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923336749857173, "validation_loss": 0.710308352958468, "validation_accuracy": 0.7502543234994914, "elapsed_seconds": 47.817894416 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370655949431792, "validation_loss": 0.5771756065192747, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 58.694929333 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222779060476092, "validation_loss": 0.4819741321230154, "validation_accuracy": 0.8320447609359105, "elapsed_seconds": 69.075709541 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.44097582388285234, "validation_loss": 0.4151243786499954, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 81.801101791 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.38314860589065225, "validation_loss": 0.3671314701372311, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 93.339549125 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.3406074100312187, "validation_loss": 0.3310860565229032, "validation_accuracy": 0.8786368260427263, "elapsed_seconds": 105.281066333 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.3077992643371988, "validation_loss": 0.3033731649951008, "validation_accuracy": 0.8857578840284842, "elapsed_seconds": 116.265963291 } ], "validation_loss": 0.3033731649951008, "coordinator_cpu_threads": null, "rate": 0.05, "coordinator_cpu_provider": "apple_accelerate", "test_loss": 0.2970531787102063, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332128378295, "fit_seconds": 116.4694985, "training_loss": 0.29234691523654444, "tensor_owned_device_peak_bytes": 0, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 4202560, "dtype": "Float64", "command_seconds": 116.938115166, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float64", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 111.66095254399998, "initialization_seconds": 3.390874083, "failed": false, "nodes": [ { "response_read_wait_seconds": 3.7663322510000006, "requests": 300, "coordinator_exchange_wall_seconds": 5.434352178, "inner_threads": null, "backend": "apple_accelerate", "target": "macos-arm64", "owned_host_tensor_peak_bytes": 5398608, "request_write_seconds": 1.6679660429999998, "request_payload_bytes": 123668080, "process_peak_rss_bytes": 25673728, "samples": 152920, "upload_preparation_seconds": 0.10130259600000005, "worker_total_seconds": 1.477964198000001, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 1.3427640759999984, "owned_device_tensor_peak_bytes": 0, "device": "cpu", "readback_seconds": 0.015336034999999996, "configuration_id": "atlas-01/C03", "platform": "atlas-01", "relative_batch_weight": 1 }, { "response_read_wait_seconds": 104.35145782700003, "requests": 300, "coordinator_exchange_wall_seconds": 106.41689174800003, "inner_threads": 1, "backend": "openblas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 5398608, "request_write_seconds": 2.0653792869999994, "request_payload_bytes": 123668080, "process_peak_rss_bytes": 122826752, "samples": 152920, "upload_preparation_seconds": 0.8999936510000001, "worker_total_seconds": 13.327921916999996, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 11.927826143999997, "owned_device_tensor_peak_bytes": 0, "device": "cpu", "readback_seconds": 0.16321454900000001, "configuration_id": "atlas-02/C03", "platform": "atlas-02", "relative_batch_weight": 1 }, { "response_read_wait_seconds": 94.24916463099996, "requests": 300, "coordinator_exchange_wall_seconds": 96.32528441800005, "inner_threads": 1, "backend": "openblas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 5398608, "request_write_seconds": 2.0760746739999996, "request_payload_bytes": 123668080, "process_peak_rss_bytes": 122687488, "samples": 152920, "upload_preparation_seconds": 0.4658746279999999, "worker_total_seconds": 4.476447578000001, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 3.8578807020000028, "owned_device_tensor_peak_bytes": 0, "device": "cpu", "readback_seconds": 0.03728029299999998, "configuration_id": "atlas-03/C03", "platform": "atlas-03", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu64", "reference_trial": 1, "dtype": "Float64", "predeclared_tolerances": { "parameter_max_abs": 1e-08, "loss_max_abs": 1e-08, "accuracy_max_abs": 1e-12 }, "parameter_max_abs_error": 2.7755575615628914e-16, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 0.0, "validation_loss": 5.551115123125783e-17, "test_loss": 1.1102230246251565e-16 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "02-03-gpu32-w1-7", "repetition": 1, "started_utc": "2026-10-12T21:49:13.642377+00:00", "finished_utc": "2026-10-12T21:50:46.891718+00:00", "outer_command_wall_seconds": 93.24883054199745, "warmup_excluded_from_statistics": false, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 75546624, "test_accuracy": 0.8955239064089522, "parameter_sha256": "7b775ba5290afda4e7d8f423f17c59214a43cd340d4e1f62d0bcd21e508a56d0", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8857578840284842, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.3770934739558967, "validation_loss": 1.2565156643897568, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 13.994734541 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.168070296236003, "validation_loss": 1.072663564672305, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 21.339542916 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784492987595977, "validation_loss": 0.8820205004872529, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 29.25987175 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923334554839502, "validation_loss": 0.7103084927900398, "validation_accuracy": 0.7501525940996948, "elapsed_seconds": 36.507014958 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370661065030486, "validation_loss": 0.577175882655752, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 45.250734541 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222785073355307, "validation_loss": 0.481975248113411, "validation_accuracy": 0.8319430315361139, "elapsed_seconds": 54.494141833 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.4409768236892788, "validation_loss": 0.41512579553488677, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 63.448872583 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.3831498536207961, "validation_loss": 0.36713215224861734, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 72.720057541 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.3406088334786849, "validation_loss": 0.331087377828111, "validation_accuracy": 0.8787385554425229, "elapsed_seconds": 81.843666791 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.3077999742386418, "validation_loss": 0.30337328036464484, "validation_accuracy": 0.8857578840284842, "elapsed_seconds": 92.549535 } ], "validation_loss": 0.30337328036464484, "coordinator_cpu_threads": null, "rate": 0.05, "coordinator_cpu_provider": "apple_accelerate", "test_loss": 0.29705295057650866, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332174875397, "fit_seconds": 92.727323583, "training_loss": 0.29234655017850286, "tensor_owned_device_peak_bytes": 0, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 6303328, "dtype": "Float32", "command_seconds": 93.18005375, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float32", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 87.24128903799996, "initialization_seconds": 3.949542334, "failed": false, "nodes": [ { "response_read_wait_seconds": 50.69311366600001, "requests": 300, "coordinator_exchange_wall_seconds": 51.246885000999995, "inner_threads": 1, "backend": "cuda_cublas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 245760, "request_write_seconds": 0.5537185480000001, "request_payload_bytes": 35875600, "process_peak_rss_bytes": 251711488, "samples": 57350, "upload_preparation_seconds": 0.8638853949999997, "worker_total_seconds": 1.8202823890000008, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 0.648182167, "owned_device_tensor_peak_bytes": 933416, "device": "cuda", "readback_seconds": 0.13533233299999997, "configuration_id": "atlas-02/C03", "platform": "atlas-02", "relative_batch_weight": 1 }, { "response_read_wait_seconds": 82.62908840499999, "requests": 300, "coordinator_exchange_wall_seconds": 84.79015824300004, "inner_threads": 1, "backend": "cuda_cublas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 1720320, "request_write_seconds": 2.1609882789999997, "request_payload_bytes": 129459920, "process_peak_rss_bytes": 288571392, "samples": 401410, "upload_preparation_seconds": 1.1265795959999994, "worker_total_seconds": 1.6740156149999985, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 0.35420495699999993, "owned_device_tensor_peak_bytes": 5725736, "device": "cuda", "readback_seconds": 0.062500059, "configuration_id": "atlas-03/C03", "platform": "atlas-03", "relative_batch_weight": 7 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu32", "reference_trial": 1, "dtype": "Float32", "predeclared_tolerances": { "parameter_max_abs": 0.001, "loss_max_abs": 0.0005, "accuracy_max_abs": 0.005 }, "parameter_max_abs_error": 4.634261131286621e-05, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 9.861349155593224e-07, "validation_loss": 1.1455976344776353e-06, "test_loss": 9.786561206981403e-07 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.00010172939979657958, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "02-03-gpu32-w1-7", "repetition": 2, "started_utc": "2026-10-12T21:50:47.688131+00:00", "finished_utc": "2026-10-12T21:52:19.820556+00:00", "outer_command_wall_seconds": 92.13198762499087, "warmup_excluded_from_statistics": false, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 77479936, "test_accuracy": 0.8955239064089522, "parameter_sha256": "7b775ba5290afda4e7d8f423f17c59214a43cd340d4e1f62d0bcd21e508a56d0", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8857578840284842, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.3770934739558967, "validation_loss": 1.2565156643897568, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 16.972418875 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.168070296236003, "validation_loss": 1.072663564672305, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 25.601841292 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784492987595977, "validation_loss": 0.8820205004872529, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 32.036054167 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923334554839502, "validation_loss": 0.7103084927900398, "validation_accuracy": 0.7501525940996948, "elapsed_seconds": 40.809456667 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370661065030486, "validation_loss": 0.577175882655752, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 50.11987275 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222785073355307, "validation_loss": 0.481975248113411, "validation_accuracy": 0.8319430315361139, "elapsed_seconds": 58.308101417 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.4409768236892788, "validation_loss": 0.41512579553488677, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 66.680593833 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.3831498536207961, "validation_loss": 0.36713215224861734, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 74.744550208 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.3406088334786849, "validation_loss": 0.331087377828111, "validation_accuracy": 0.8787385554425229, "elapsed_seconds": 83.049846458 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.3077999742386418, "validation_loss": 0.30337328036464484, "validation_accuracy": 0.8857578840284842, "elapsed_seconds": 91.464022 } ], "validation_loss": 0.30337328036464484, "coordinator_cpu_threads": null, "rate": 0.05, "coordinator_cpu_provider": "apple_accelerate", "test_loss": 0.29705295057650866, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332174875397, "fit_seconds": 91.641994208, "training_loss": 0.29234655017850286, "tensor_owned_device_peak_bytes": 0, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 6303328, "dtype": "Float32", "command_seconds": 92.093433917, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float32", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 86.35572246200006, "initialization_seconds": 3.730089625, "failed": false, "nodes": [ { "response_read_wait_seconds": 47.13218325199995, "requests": 300, "coordinator_exchange_wall_seconds": 47.69015449500003, "inner_threads": 1, "backend": "cuda_cublas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 245760, "request_write_seconds": 0.5579191199999997, "request_payload_bytes": 35875600, "process_peak_rss_bytes": 251682816, "samples": 57350, "upload_preparation_seconds": 0.8400615010000003, "worker_total_seconds": 1.7712620580000007, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 0.6278459390000003, "owned_device_tensor_peak_bytes": 933416, "device": "cuda", "readback_seconds": 0.13438590700000008, "configuration_id": "atlas-02/C03", "platform": "atlas-02", "relative_batch_weight": 1 }, { "response_read_wait_seconds": 82.66911189099999, "requests": 300, "coordinator_exchange_wall_seconds": 84.83596416000006, "inner_threads": 1, "backend": "cuda_cublas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 1720320, "request_write_seconds": 2.1667847920000005, "request_payload_bytes": 129459920, "process_peak_rss_bytes": 288174080, "samples": 401410, "upload_preparation_seconds": 1.119587932999999, "worker_total_seconds": 1.6696630329999997, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 0.35501172800000025, "owned_device_tensor_peak_bytes": 5725736, "device": "cuda", "readback_seconds": 0.06263750900000004, "configuration_id": "atlas-03/C03", "platform": "atlas-03", "relative_batch_weight": 7 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu32", "reference_trial": 1, "dtype": "Float32", "predeclared_tolerances": { "parameter_max_abs": 0.001, "loss_max_abs": 0.0005, "accuracy_max_abs": 0.005 }, "parameter_max_abs_error": 4.634261131286621e-05, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 9.861349155593224e-07, "validation_loss": 1.1455976344776353e-06, "test_loss": 9.786561206981403e-07 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.00010172939979657958, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "02-03-gpu32-w1-7", "repetition": 3, "started_utc": "2026-10-12T21:52:20.662153+00:00", "finished_utc": "2026-10-12T21:53:51.164937+00:00", "outer_command_wall_seconds": 90.50230108300457, "warmup_excluded_from_statistics": false, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 77643776, "test_accuracy": 0.8955239064089522, "parameter_sha256": "7b775ba5290afda4e7d8f423f17c59214a43cd340d4e1f62d0bcd21e508a56d0", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8857578840284842, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.3770934739558967, "validation_loss": 1.2565156643897568, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 12.19051825 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.168070296236003, "validation_loss": 1.072663564672305, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 21.419508792 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784492987595977, "validation_loss": 0.8820205004872529, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 30.7901285 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923334554839502, "validation_loss": 0.7103084927900398, "validation_accuracy": 0.7501525940996948, "elapsed_seconds": 38.806903667 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370661065030486, "validation_loss": 0.577175882655752, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 46.47613125 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222785073355307, "validation_loss": 0.481975248113411, "validation_accuracy": 0.8319430315361139, "elapsed_seconds": 54.681346833 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.4409768236892788, "validation_loss": 0.41512579553488677, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 62.857529417 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.3831498536207961, "validation_loss": 0.36713215224861734, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 71.902250208 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.3406088334786849, "validation_loss": 0.331087377828111, "validation_accuracy": 0.8787385554425229, "elapsed_seconds": 80.579985625 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.3077999742386418, "validation_loss": 0.30337328036464484, "validation_accuracy": 0.8857578840284842, "elapsed_seconds": 89.8204215 } ], "validation_loss": 0.30337328036464484, "coordinator_cpu_threads": null, "rate": 0.05, "coordinator_cpu_provider": "apple_accelerate", "test_loss": 0.29705295057650866, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332174875397, "fit_seconds": 90.005007041, "training_loss": 0.29234655017850286, "tensor_owned_device_peak_bytes": 0, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 6303328, "dtype": "Float32", "command_seconds": 90.461175125, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float32", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 84.335202698, "initialization_seconds": 4.212591333, "failed": false, "nodes": [ { "response_read_wait_seconds": 49.793614289000026, "requests": 300, "coordinator_exchange_wall_seconds": 50.300012469000016, "inner_threads": 1, "backend": "cuda_cublas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 245760, "request_write_seconds": 0.5063517589999998, "request_payload_bytes": 35875600, "process_peak_rss_bytes": 251740160, "samples": 57350, "upload_preparation_seconds": 0.8483147890000005, "worker_total_seconds": 1.7833610479999993, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 0.6286405510000002, "owned_device_tensor_peak_bytes": 933416, "device": "cuda", "readback_seconds": 0.13205223499999996, "configuration_id": "atlas-02/C03", "platform": "atlas-02", "relative_batch_weight": 1 }, { "response_read_wait_seconds": 80.11868316900004, "requests": 300, "coordinator_exchange_wall_seconds": 82.07265038100005, "inner_threads": 1, "backend": "cuda_cublas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 1720320, "request_write_seconds": 1.9538403350000004, "request_payload_bytes": 129459920, "process_peak_rss_bytes": 288477184, "samples": 401410, "upload_preparation_seconds": 1.1299737530000002, "worker_total_seconds": 1.6912474249999987, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 0.3653267259999998, "owned_device_tensor_peak_bytes": 5725736, "device": "cuda", "readback_seconds": 0.06264641600000001, "configuration_id": "atlas-03/C03", "platform": "atlas-03", "relative_batch_weight": 7 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu32", "reference_trial": 1, "dtype": "Float32", "predeclared_tolerances": { "parameter_max_abs": 0.001, "loss_max_abs": 0.0005, "accuracy_max_abs": 0.005 }, "parameter_max_abs_error": 4.634261131286621e-05, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 9.861349155593224e-07, "validation_loss": 1.1455976344776353e-06, "test_loss": 9.786561206981403e-07 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.00010172939979657958, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } } ], "excluded_warmups": [ { "case": "atlas01-cpu32", "repetition": 0, "started_utc": "2026-10-12T20:57:45.410645+00:00", "finished_utc": "2026-10-12T20:57:48.460228+00:00", "outer_command_wall_seconds": 3.049420666997321, "warmup_excluded_from_statistics": true, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 112132096, "test_accuracy": 0.8955239064089522, "parameter_sha256": "a8aeb21461c306c7ff0e8ef9fdd8aacd26058a8de7a405d931627793679b2c75", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8858596134282808, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.3770934789242497, "validation_loss": 1.2565156643897568, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 0.251483917 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.168070280486428, "validation_loss": 1.072663564672305, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 0.509068292 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784492905028291, "validation_loss": 0.8820205004872529, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 0.763373167 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923334447027282, "validation_loss": 0.7103084927900398, "validation_accuracy": 0.7501525940996948, "elapsed_seconds": 1.013658709 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370661036836642, "validation_loss": 0.577175882655752, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 1.265881667 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222784697760033, "validation_loss": 0.48197512238581086, "validation_accuracy": 0.8319430315361139, "elapsed_seconds": 1.516110875 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.44097673220999917, "validation_loss": 0.4151257340626702, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 1.768818334 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.3831498027939214, "validation_loss": 0.36713235249111875, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 2.013266167 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.3406089502044476, "validation_loss": 0.3310881936901826, "validation_accuracy": 0.8787385554425229, "elapsed_seconds": 2.255975167 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.3078001272045378, "validation_loss": 0.3033744259622793, "validation_accuracy": 0.8858596134282808, "elapsed_seconds": 2.498773375 } ], "validation_loss": 0.3033744259622793, "coordinator_cpu_threads": null, "rate": 0.05, "coordinator_cpu_provider": "apple_accelerate", "test_loss": 0.29705392923262935, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332174875397, "fit_seconds": 2.687472375, "training_loss": 0.2923475363134184, "tensor_owned_device_peak_bytes": 0, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 21086868, "dtype": "Float32", "command_seconds": 2.987563958, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float32", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 1.9314693690000002, "initialization_seconds": 0, "failed": false, "nodes": [ { "response_read_wait_seconds": 0, "requests": 300, "coordinator_exchange_wall_seconds": 0, "inner_threads": null, "backend": "apple_accelerate", "target": "macos-arm64", "owned_host_tensor_peak_bytes": 21086868, "request_write_seconds": 0, "request_payload_bytes": 0, "process_peak_rss_bytes": 108806144, "samples": 458760, "upload_preparation_seconds": 0.16226138000000018, "worker_total_seconds": 1.930616410999999, "zero_sample_steps": 0, "response_payload_bytes": 0, "compute_seconds": 1.744908203, "owned_device_tensor_peak_bytes": 0, "device": "cpu", "readback_seconds": 0.012085206, "configuration_id": "atlas-01/C03", "platform": "atlas-01", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu32", "reference_trial": 1, "dtype": "Float32", "predeclared_tolerances": { "parameter_max_abs": 0.001, "loss_max_abs": 0.0005, "accuracy_max_abs": 0.005 }, "parameter_max_abs_error": 0.0, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 0.0, "validation_loss": 0.0, "test_loss": 0.0 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "atlas01-cpu64", "repetition": 0, "started_utc": "2026-10-12T20:57:57.978867+00:00", "finished_utc": "2026-10-12T20:58:01.420879+00:00", "outer_command_wall_seconds": 3.441918792013894, "warmup_excluded_from_statistics": true, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 115146752, "test_accuracy": 0.8955239064089522, "parameter_sha256": "97b5b013f99579d2b33679af40a39c73e0cd9e28faee9d9df1ac1ad46da08f4c", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8857578840284842, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.3770934749530683, "validation_loss": 1.2565156669823923, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 0.28732575 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.1680702765782571, "validation_loss": 1.0726634152606465, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 0.600480084 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784489967242787, "validation_loss": 0.8820201711707012, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 0.887859292 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923336749857169, "validation_loss": 0.710308352958468, "validation_accuracy": 0.7502543234994914, "elapsed_seconds": 1.178308625 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.637065594943179, "validation_loss": 0.5771756065192747, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 1.46814575 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222779060476093, "validation_loss": 0.4819741321230154, "validation_accuracy": 0.8320447609359105, "elapsed_seconds": 1.780412875 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.4409758238828523, "validation_loss": 0.4151243786499954, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 2.061415417 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.38314860589065225, "validation_loss": 0.36713147013723113, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 2.342698209 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.3406074100312188, "validation_loss": 0.3310860565229031, "validation_accuracy": 0.8786368260427263, "elapsed_seconds": 2.624226209 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.30779926433719895, "validation_loss": 0.30337316499510086, "validation_accuracy": 0.8857578840284842, "elapsed_seconds": 2.901779625 } ], "validation_loss": 0.30337316499510086, "coordinator_cpu_threads": null, "rate": 0.05, "coordinator_cpu_provider": "apple_accelerate", "test_loss": 0.2970531787102064, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332128378295, "fit_seconds": 3.094263541, "training_loss": 0.29234691523654444, "tensor_owned_device_peak_bytes": 0, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 16495216, "dtype": "Float64", "command_seconds": 3.393792584, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float64", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 2.367014832, "initialization_seconds": 0, "failed": false, "nodes": [ { "response_read_wait_seconds": 0, "requests": 300, "coordinator_exchange_wall_seconds": 0, "inner_threads": null, "backend": "apple_accelerate", "target": "macos-arm64", "owned_host_tensor_peak_bytes": 16495216, "request_write_seconds": 0, "request_payload_bytes": 0, "process_peak_rss_bytes": 115113984, "samples": 458760, "upload_preparation_seconds": 0.10921516899999996, "worker_total_seconds": 2.365851219999999, "zero_sample_steps": 0, "response_payload_bytes": 0, "compute_seconds": 2.235735917999999, "owned_device_tensor_peak_bytes": 0, "device": "cpu", "readback_seconds": 0.009738838, "configuration_id": "atlas-01/C03", "platform": "atlas-01", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu64", "reference_trial": 1, "dtype": "Float64", "predeclared_tolerances": { "parameter_max_abs": 1e-08, "loss_max_abs": 1e-08, "accuracy_max_abs": 1e-12 }, "parameter_max_abs_error": 0.0, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 0.0, "validation_loss": 0.0, "test_loss": 0.0 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "atlas02-cpu32", "repetition": 0, "started_utc": "2026-10-12T20:58:12.237515+00:00", "finished_utc": "2026-10-12T20:58:35.725968+00:00", "outer_command_wall_seconds": 23.47286449999956, "warmup_excluded_from_statistics": true, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 177848320, "test_accuracy": 0.8955239064089522, "parameter_sha256": "44166434f82221f11d83b4836f9b584b873bc9705b804eb25b3e2186648823a0", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8857578840284842, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.3770934829155626, "validation_loss": 1.2565156643897568, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 1.779836802 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.168070276495115, "validation_loss": 1.072663564672305, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 3.566808775 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784492964897984, "validation_loss": 0.8820205098008576, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 5.759231762 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923334504246493, "validation_loss": 0.7103084927900398, "validation_accuracy": 0.7501525940996948, "elapsed_seconds": 7.536405708 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370661056793206, "validation_loss": 0.577175882655752, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 9.339211357 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222785087614635, "validation_loss": 0.481975248113411, "validation_accuracy": 0.8319430315361139, "elapsed_seconds": 11.103047412 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.4409768225446347, "validation_loss": 0.41512579553488677, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 13.497924601 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.3831498494190038, "validation_loss": 0.36713215224861734, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 15.588201516 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.3406088343784194, "validation_loss": 0.33108737317130865, "validation_accuracy": 0.8787385554425229, "elapsed_seconds": 17.25415147 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.30779997538653403, "validation_loss": 0.3033732850214472, "validation_accuracy": 0.8857578840284842, "elapsed_seconds": 18.926843629 } ], "validation_loss": 0.3033732850214472, "coordinator_cpu_threads": 1, "rate": 0.05, "coordinator_cpu_provider": "openblas", "test_loss": 0.29705295057650866, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.599133218017636, "fit_seconds": 20.221955579, "training_loss": 0.29234654818284644, "tensor_owned_device_peak_bytes": 0, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 21086868, "dtype": "Float32", "command_seconds": 21.262666738, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float32", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 15.352909224, "initialization_seconds": 0, "failed": false, "nodes": [ { "response_read_wait_seconds": 0, "requests": 300, "coordinator_exchange_wall_seconds": 0, "inner_threads": 1, "backend": "openblas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 21086868, "request_write_seconds": 0, "request_payload_bytes": 0, "process_peak_rss_bytes": 177848320, "samples": 458760, "upload_preparation_seconds": 1.158970562, "worker_total_seconds": 15.348747676000004, "zero_sample_steps": 0, "response_payload_bytes": 0, "compute_seconds": 13.718721355000012, "owned_device_tensor_peak_bytes": 0, "device": "cpu", "readback_seconds": 0.07190662700000008, "configuration_id": "atlas-02/C03", "platform": "atlas-02", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu32", "reference_trial": 1, "dtype": "Float32", "predeclared_tolerances": { "parameter_max_abs": 0.001, "loss_max_abs": 0.0005, "accuracy_max_abs": 0.005 }, "parameter_max_abs_error": 4.6350061893463135e-05, "loss_abs_errors": { "initial_loss": 5.300964112109341e-10, "training_loss": 9.881305719772193e-07, "validation_loss": 1.1409408320894876e-06, "test_loss": 9.786561206981403e-07 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.00010172939979657958, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "atlas02-cpu64", "repetition": 0, "started_utc": "2026-10-12T20:59:42.300833+00:00", "finished_utc": "2026-10-12T21:00:01.956174+00:00", "outer_command_wall_seconds": 19.655189582990715, "warmup_excluded_from_statistics": true, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 172433408, "test_accuracy": 0.8955239064089522, "parameter_sha256": "012115e11d05b6a7def740254a06068ea009e1acd63dc6799684cada7ccf2b63", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8857578840284842, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.3770934749530683, "validation_loss": 1.2565156669823923, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 1.546670492 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.1680702765782573, "validation_loss": 1.0726634152606465, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 3.066817163 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784489967242787, "validation_loss": 0.8820201711707012, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 4.610343636 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923336749857169, "validation_loss": 0.7103083529584681, "validation_accuracy": 0.7502543234994914, "elapsed_seconds": 6.478356091 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.637065594943179, "validation_loss": 0.5771756065192747, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 8.014073002 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222779060476093, "validation_loss": 0.48197413212301554, "validation_accuracy": 0.8320447609359105, "elapsed_seconds": 9.541281413 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.4409758238828523, "validation_loss": 0.41512437864999546, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 11.064449876 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.38314860589065225, "validation_loss": 0.367131470137231, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 12.60228479 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.3406074100312188, "validation_loss": 0.3310860565229031, "validation_accuracy": 0.8786368260427263, "elapsed_seconds": 14.128848353 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.30779926433719895, "validation_loss": 0.30337316499510086, "validation_accuracy": 0.8857578840284842, "elapsed_seconds": 15.652775259 } ], "validation_loss": 0.30337316499510086, "coordinator_cpu_threads": 1, "rate": 0.05, "coordinator_cpu_provider": "openblas", "test_loss": 0.29705317871020637, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332128378295, "fit_seconds": 16.800824795, "training_loss": 0.29234691523654444, "tensor_owned_device_peak_bytes": 0, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 16495216, "dtype": "Float64", "command_seconds": 17.692334678, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float64", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 13.010435647000008, "initialization_seconds": 0, "failed": false, "nodes": [ { "response_read_wait_seconds": 0, "requests": 300, "coordinator_exchange_wall_seconds": 0, "inner_threads": 1, "backend": "openblas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 16495216, "request_write_seconds": 0, "request_payload_bytes": 0, "process_peak_rss_bytes": 172433408, "samples": 458760, "upload_preparation_seconds": 0.5601694949999999, "worker_total_seconds": 13.007778716000004, "zero_sample_steps": 0, "response_payload_bytes": 0, "compute_seconds": 12.130655559000003, "owned_device_tensor_peak_bytes": 0, "device": "cpu", "readback_seconds": 0.04750998199999997, "configuration_id": "atlas-02/C03", "platform": "atlas-02", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu64", "reference_trial": 1, "dtype": "Float64", "predeclared_tolerances": { "parameter_max_abs": 1e-08, "loss_max_abs": 1e-08, "accuracy_max_abs": 1e-12 }, "parameter_max_abs_error": 3.3306690738754696e-16, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 0.0, "validation_loss": 0.0, "test_loss": 5.551115123125783e-17 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "atlas03-cpu32", "repetition": 0, "started_utc": "2026-10-12T21:01:03.980685+00:00", "finished_utc": "2026-10-12T21:01:19.386534+00:00", "outer_command_wall_seconds": 15.405253333010478, "warmup_excluded_from_statistics": true, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 177950720, "test_accuracy": 0.8955239064089522, "parameter_sha256": "8dcc7cf64ee5fc2adf55c88ca37831aad3c93a301687a1cf8e1174efd2007af8", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8857578840284842, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.3770934789242497, "validation_loss": 1.2565156643897568, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 1.178166394 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.168070280486428, "validation_loss": 1.072663564672305, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 2.367461833 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784492885071727, "validation_loss": 0.8820205004872529, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 3.558453498 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923334484289929, "validation_loss": 0.7103084927900398, "validation_accuracy": 0.7501525940996948, "elapsed_seconds": 4.748853475 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370661056793206, "validation_loss": 0.577175882655752, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 5.916787308 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.522278505767979, "validation_loss": 0.48197524997491925, "validation_accuracy": 0.8319430315361139, "elapsed_seconds": 7.103647229 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.44097682454029113, "validation_loss": 0.41512579553488677, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 8.33728205 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.3831498552734491, "validation_loss": 0.3671321569054197, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 9.551066284 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.3406088323827629, "validation_loss": 0.3310873796896192, "validation_accuracy": 0.8787385554425229, "elapsed_seconds": 10.71622538 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.30779997638436224, "validation_loss": 0.3033732850214472, "validation_accuracy": 0.8857578840284842, "elapsed_seconds": 11.907946349 } ], "validation_loss": 0.3033732850214472, "coordinator_cpu_threads": 1, "rate": 0.05, "coordinator_cpu_provider": "openblas", "test_loss": 0.29705295057650866, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332140263232, "fit_seconds": 12.802943787, "training_loss": 0.2923465511763311, "tensor_owned_device_peak_bytes": 0, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 21086868, "dtype": "Float32", "command_seconds": 13.327374162, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float32", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 9.748438409999999, "initialization_seconds": 0, "failed": false, "nodes": [ { "response_read_wait_seconds": 0, "requests": 300, "coordinator_exchange_wall_seconds": 0, "inner_threads": 1, "backend": "openblas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 21086868, "request_write_seconds": 0, "request_payload_bytes": 0, "process_peak_rss_bytes": 177950720, "samples": 458760, "upload_preparation_seconds": 0.704570182, "worker_total_seconds": 9.746718342000001, "zero_sample_steps": 0, "response_payload_bytes": 0, "compute_seconds": 8.678560248999995, "owned_device_tensor_peak_bytes": 0, "device": "cpu", "readback_seconds": 0.046774402, "configuration_id": "atlas-03/C03", "platform": "atlas-03", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu32", "reference_trial": 1, "dtype": "Float32", "predeclared_tolerances": { "parameter_max_abs": 0.001, "loss_max_abs": 0.0005, "accuracy_max_abs": 0.005 }, "parameter_max_abs_error": 4.635751247406006e-05, "loss_abs_errors": { "initial_loss": 3.461216424582858e-09, "training_loss": 9.851370872948628e-07, "validation_loss": 1.1409408320894876e-06, "test_loss": 9.786561206981403e-07 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.00010172939979657958, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "atlas03-cpu64", "repetition": 0, "started_utc": "2026-10-12T21:02:05.030324+00:00", "finished_utc": "2026-10-12T21:02:19.264999+00:00", "outer_command_wall_seconds": 14.234295957998256, "warmup_excluded_from_statistics": true, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 172437504, "test_accuracy": 0.8955239064089522, "parameter_sha256": "0b9cb6964e88ea01eaaf5fc901715b34b184b5c0942cb2e7439c70e1603c1b4c", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8857578840284842, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.3770934749530683, "validation_loss": 1.2565156669823923, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 1.07689208 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.1680702765782573, "validation_loss": 1.0726634152606465, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 2.171737283 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784489967242784, "validation_loss": 0.8820201711707012, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 3.271321396 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923336749857169, "validation_loss": 0.7103083529584678, "validation_accuracy": 0.7502543234994914, "elapsed_seconds": 4.3297166 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370655949431789, "validation_loss": 0.5771756065192747, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 5.404688821 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222779060476093, "validation_loss": 0.4819741321230154, "validation_accuracy": 0.8320447609359105, "elapsed_seconds": 6.474495124 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.4409758238828523, "validation_loss": 0.41512437864999546, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 7.54563011 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.38314860589065225, "validation_loss": 0.36713147013723113, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 8.619295179 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.3406074100312188, "validation_loss": 0.3310860565229031, "validation_accuracy": 0.8786368260427263, "elapsed_seconds": 9.681282209 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.307799264337199, "validation_loss": 0.3033731649951008, "validation_accuracy": 0.8857578840284842, "elapsed_seconds": 10.756736846 } ], "validation_loss": 0.3033731649951008, "coordinator_cpu_threads": 1, "rate": 0.05, "coordinator_cpu_provider": "openblas", "test_loss": 0.29705317871020637, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332128378295, "fit_seconds": 11.578024317, "training_loss": 0.29234691523654444, "tensor_owned_device_peak_bytes": 0, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 16495216, "dtype": "Float64", "command_seconds": 12.085365206, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float64", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 8.946393093000003, "initialization_seconds": 0, "failed": false, "nodes": [ { "response_read_wait_seconds": 0, "requests": 300, "coordinator_exchange_wall_seconds": 0, "inner_threads": 1, "backend": "openblas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 16495216, "request_write_seconds": 0, "request_payload_bytes": 0, "process_peak_rss_bytes": 172437504, "samples": 458760, "upload_preparation_seconds": 0.35804231799999997, "worker_total_seconds": 8.944913559, "zero_sample_steps": 0, "response_payload_bytes": 0, "compute_seconds": 8.310706494999998, "owned_device_tensor_peak_bytes": 0, "device": "cpu", "readback_seconds": 0.033648264999999976, "configuration_id": "atlas-03/C03", "platform": "atlas-03", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu64", "reference_trial": 1, "dtype": "Float64", "predeclared_tolerances": { "parameter_max_abs": 1e-08, "loss_max_abs": 1e-08, "accuracy_max_abs": 1e-12 }, "parameter_max_abs_error": 2.220446049250313e-16, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 0.0, "validation_loss": 5.551115123125783e-17, "test_loss": 5.551115123125783e-17 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "atlas01-gpu32", "repetition": 0, "started_utc": "2026-10-12T21:03:01.333474+00:00", "finished_utc": "2026-10-12T21:03:04.749401+00:00", "outer_command_wall_seconds": 3.4155962499935413, "warmup_excluded_from_statistics": true, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 84574208, "test_accuracy": 0.8955239064089522, "parameter_sha256": "e6e47d12ca151780c09565d005880563cd9bf9b214e322e63a6dc6d1df72288a", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8857578840284842, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.3770934714717202, "validation_loss": 1.2565156457625473, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 0.304338541 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.1680702799563316, "validation_loss": 1.072663564672305, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 0.559521416 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784492745375777, "validation_loss": 0.8820205004872529, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 0.818017833 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923334404463672, "validation_loss": 0.7103084890670233, "validation_accuracy": 0.7501525940996948, "elapsed_seconds": 1.154683625 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370660917097257, "validation_loss": 0.577175882655752, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 1.462587375 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222784927962121, "validation_loss": 0.4819752387998063, "validation_accuracy": 0.8319430315361139, "elapsed_seconds": 1.729325416 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.4409768185533219, "validation_loss": 0.41512578622128204, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 1.985930666 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.3831498435645586, "validation_loss": 0.3671321541161891, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 2.260957708 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.34060883298301897, "validation_loss": 0.33108755570825943, "validation_accuracy": 0.8787385554425229, "elapsed_seconds": 2.524291916 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.3078000803066117, "validation_loss": 0.3033739387867773, "validation_accuracy": 0.8857578840284842, "elapsed_seconds": 2.779439708 } ], "validation_loss": 0.3033739387867773, "coordinator_cpu_threads": null, "rate": 0.05, "coordinator_cpu_provider": "apple_accelerate", "test_loss": 0.2970536025106482, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332174875397, "fit_seconds": 2.961131334, "training_loss": 0.29234693766326264, "tensor_owned_device_peak_bytes": 6524456, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 6303328, "dtype": "Float32", "command_seconds": 3.337731334, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float32", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 2.1605423409999984, "initialization_seconds": 0, "failed": false, "nodes": [ { "response_read_wait_seconds": 0, "requests": 300, "coordinator_exchange_wall_seconds": 0, "inner_threads": null, "backend": "metal_mps", "target": "macos-arm64", "owned_host_tensor_peak_bytes": 6303328, "request_write_seconds": 0, "request_payload_bytes": 0, "process_peak_rss_bytes": 84574208, "samples": 458760, "upload_preparation_seconds": 0.24579895699999982, "worker_total_seconds": 2.159393827999999, "zero_sample_steps": 0, "response_payload_bytes": 0, "compute_seconds": 1.8673816680000002, "owned_device_tensor_peak_bytes": 6524456, "device": "mps", "readback_seconds": 0.01851034300000002, "configuration_id": "atlas-01/C03", "platform": "atlas-01", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu32", "reference_trial": 1, "dtype": "Float32", "predeclared_tolerances": { "parameter_max_abs": 0.001, "loss_max_abs": 0.0005, "accuracy_max_abs": 0.005 }, "parameter_max_abs_error": 4.6253204345703125e-05, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 5.986501557808666e-07, "validation_loss": 4.87175501995285e-07, "test_loss": 3.2672198113647966e-07 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.00010172939979657958, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "atlas02-gpu32", "repetition": 0, "started_utc": "2026-10-12T21:03:15.653189+00:00", "finished_utc": "2026-10-12T21:03:25.241666+00:00", "outer_command_wall_seconds": 9.587779541994678, "warmup_excluded_from_statistics": true, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 296083456, "test_accuracy": 0.8955239064089522, "parameter_sha256": "ed951a02e2b2e52170c840ab714347f81580d9334a40d7bbb4d5bc5024e36067", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8857578840284842, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.377093479454346, "validation_loss": 1.2565156643897568, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 0.464715175 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.1680702799563316, "validation_loss": 1.072663564672305, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 0.925197047 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784492688156565, "validation_loss": 0.8820204688234222, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 1.377412261 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923333773649147, "validation_loss": 0.7103079675415404, "validation_accuracy": 0.7500508646998982, "elapsed_seconds": 1.844678312 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370662267595382, "validation_loss": 0.5771763893740587, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 2.306257475 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222790034815743, "validation_loss": 0.4819753710275264, "validation_accuracy": 0.8318413021363174, "elapsed_seconds": 2.772053544 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.4409766082142524, "validation_loss": 0.4151265396652241, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 3.250647328 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.38315067391354407, "validation_loss": 0.3671332475067034, "validation_accuracy": 0.868056968463886, "elapsed_seconds": 3.728233644 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.3406097975944681, "validation_loss": 0.3310883492989312, "validation_accuracy": 0.8787385554425229, "elapsed_seconds": 4.204007505 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.3078013175200518, "validation_loss": 0.30337606328913785, "validation_accuracy": 0.8857578840284842, "elapsed_seconds": 4.670535611 } ], "validation_loss": 0.30337606328913785, "coordinator_cpu_threads": 1, "rate": 0.05, "coordinator_cpu_provider": "openblas", "test_loss": 0.29705589171101093, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.599133218017636, "fit_seconds": 5.90137563, "training_loss": 0.2923495870296973, "tensor_owned_device_peak_bytes": 6524456, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 6303328, "dtype": "Float32", "command_seconds": 7.239048507, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float32", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 1.647846244000001, "initialization_seconds": 0, "failed": false, "nodes": [ { "response_read_wait_seconds": 0, "requests": 300, "coordinator_exchange_wall_seconds": 0, "inner_threads": 1, "backend": "cuda_cublas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 6303328, "request_write_seconds": 0, "request_payload_bytes": 0, "process_peak_rss_bytes": 296083456, "samples": 458760, "upload_preparation_seconds": 0.9633048629999998, "worker_total_seconds": 1.6454725870000007, "zero_sample_steps": 0, "response_payload_bytes": 0, "compute_seconds": 0.48091154599999997, "owned_device_tensor_peak_bytes": 6524456, "device": "cuda", "readback_seconds": 0.06619939800000006, "configuration_id": "atlas-02/C03", "platform": "atlas-02", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu32", "reference_trial": 1, "dtype": "Float32", "predeclared_tolerances": { "parameter_max_abs": 0.001, "loss_max_abs": 0.0005, "accuracy_max_abs": 0.005 }, "parameter_max_abs_error": 5.9798359870910645e-05, "loss_abs_errors": { "initial_loss": 5.300964112109341e-10, "training_loss": 2.050716278900744e-06, "validation_loss": 1.6373268585390122e-06, "test_loss": 1.9624783815785918e-06 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.00010172939979657958, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "atlas02-gpu64", "repetition": 0, "started_utc": "2026-10-12T21:03:52.804805+00:00", "finished_utc": "2026-10-12T21:04:01.297614+00:00", "outer_command_wall_seconds": 8.492361542012077, "warmup_excluded_from_statistics": true, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 300441600, "test_accuracy": 0.8955239064089522, "parameter_sha256": "a238a137db9084d9bbc52207f21c90e2e4fd813de1533ee1fecc17d7f66c0b55", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8857578840284842, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.3770934749530683, "validation_loss": 1.2565156669823923, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 0.472543847 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.1680702765782571, "validation_loss": 1.0726634152606465, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 0.970271407 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784489967242784, "validation_loss": 0.8820201711707012, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 1.448970971 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923336749857173, "validation_loss": 0.7103083529584678, "validation_accuracy": 0.7502543234994914, "elapsed_seconds": 1.911379363 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370655949431792, "validation_loss": 0.5771756065192747, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 2.383683269 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222779060476093, "validation_loss": 0.4819741321230154, "validation_accuracy": 0.8320447609359105, "elapsed_seconds": 2.846828839 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.4409758238828523, "validation_loss": 0.4151243786499954, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 3.345352195 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.38314860589065225, "validation_loss": 0.36713147013723113, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 3.814536788 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.3406074100312188, "validation_loss": 0.3310860565229031, "validation_accuracy": 0.8786368260427263, "elapsed_seconds": 4.29606622 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.3077992643371988, "validation_loss": 0.3033731649951009, "validation_accuracy": 0.8857578840284842, "elapsed_seconds": 4.789742985 } ], "validation_loss": 0.3033731649951009, "coordinator_cpu_threads": 1, "rate": 0.05, "coordinator_cpu_provider": "openblas", "test_loss": 0.29705317871020637, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332128378295, "fit_seconds": 5.839324199, "training_loss": 0.29234691523654444, "tensor_owned_device_peak_bytes": 13048912, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 4202560, "dtype": "Float64", "command_seconds": 6.873237744, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float64", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 2.2663030490000016, "initialization_seconds": 0, "failed": false, "nodes": [ { "response_read_wait_seconds": 0, "requests": 300, "coordinator_exchange_wall_seconds": 0, "inner_threads": 1, "backend": "cuda_cublas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 4202560, "request_write_seconds": 0, "request_payload_bytes": 0, "process_peak_rss_bytes": 300441600, "samples": 458760, "upload_preparation_seconds": 0.6663317319999998, "worker_total_seconds": 2.263715747, "zero_sample_steps": 0, "response_payload_bytes": 0, "compute_seconds": 1.249973174, "owned_device_tensor_peak_bytes": 13048912, "device": "cuda", "readback_seconds": 0.07780340699999999, "configuration_id": "atlas-02/C03", "platform": "atlas-02", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu64", "reference_trial": 1, "dtype": "Float64", "predeclared_tolerances": { "parameter_max_abs": 1e-08, "loss_max_abs": 1e-08, "accuracy_max_abs": 1e-12 }, "parameter_max_abs_error": 3.3306690738754696e-16, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 0.0, "validation_loss": 5.551115123125783e-17, "test_loss": 5.551115123125783e-17 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "atlas03-gpu32", "repetition": 0, "started_utc": "2026-10-12T21:04:31.016470+00:00", "finished_utc": "2026-10-12T21:04:37.631225+00:00", "outer_command_wall_seconds": 6.6138601250131615, "warmup_excluded_from_statistics": true, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 331055104, "test_accuracy": 0.8955239064089522, "parameter_sha256": "3492d0395ffb53158e162126537ba6af8c4f47d5180ead405ea211d6f07652de", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8858596134282808, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.3770934749329367, "validation_loss": 1.2565156643897568, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 0.317130394 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.1680702879389575, "validation_loss": 1.072663564672305, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 0.631834429 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784492964897984, "validation_loss": 0.8820205004872529, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 0.940836973 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923334564116187, "validation_loss": 0.7103084927900398, "validation_accuracy": 0.7501525940996948, "elapsed_seconds": 1.254240877 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370661159226509, "validation_loss": 0.577175882655752, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 1.567642016 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222784667825188, "validation_loss": 0.4819751270426132, "validation_accuracy": 0.8319430315361139, "elapsed_seconds": 1.88276756 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.4409767470448983, "validation_loss": 0.41512573871947256, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 2.196966135 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.3831498184941247, "validation_loss": 0.36713236366623175, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 2.512298861 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.34060897462005674, "validation_loss": 0.3310880763484648, "validation_accuracy": 0.8787385554425229, "elapsed_seconds": 2.829481781 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.30780011722625567, "validation_loss": 0.30337433190457336, "validation_accuracy": 0.8858596134282808, "elapsed_seconds": 3.140557342 } ], "validation_loss": 0.30337433190457336, "coordinator_cpu_threads": 1, "rate": 0.05, "coordinator_cpu_provider": "openblas", "test_loss": 0.2970538742908505, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332140263232, "fit_seconds": 4.010166623, "training_loss": 0.2923475264676603, "tensor_owned_device_peak_bytes": 6524456, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 6303328, "dtype": "Float32", "command_seconds": 4.622983736, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float32", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 1.0723118530000004, "initialization_seconds": 0, "failed": false, "nodes": [ { "response_read_wait_seconds": 0, "requests": 300, "coordinator_exchange_wall_seconds": 0, "inner_threads": 1, "backend": "cuda_cublas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 6303328, "request_write_seconds": 0, "request_payload_bytes": 0, "process_peak_rss_bytes": 331055104, "samples": 458760, "upload_preparation_seconds": 0.6449716120000006, "worker_total_seconds": 1.0708317399999994, "zero_sample_steps": 0, "response_payload_bytes": 0, "compute_seconds": 0.27833039400000004, "owned_device_tensor_peak_bytes": 6524456, "device": "cuda", "readback_seconds": 0.04935571199999998, "configuration_id": "atlas-03/C03", "platform": "atlas-03", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu32", "reference_trial": 1, "dtype": "Float32", "predeclared_tolerances": { "parameter_max_abs": 0.001, "loss_max_abs": 0.0005, "accuracy_max_abs": 0.005 }, "parameter_max_abs_error": 4.561617970466614e-06, "loss_abs_errors": { "initial_loss": 3.461216424582858e-09, "training_loss": 9.84575809770405e-09, "validation_loss": 9.405770595716234e-08, "test_loss": 5.494177884868279e-08 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "atlas03-gpu64", "repetition": 0, "started_utc": "2026-10-12T21:04:58.304106+00:00", "finished_utc": "2026-10-12T21:05:03.882042+00:00", "outer_command_wall_seconds": 5.577275874995394, "warmup_excluded_from_statistics": true, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 325672960, "test_accuracy": 0.8955239064089522, "parameter_sha256": "558baffdc49fc87ee67d04a9cb17c77daf1671a544a5eafcd137576192b5a10e", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8857578840284842, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.3770934749530683, "validation_loss": 1.2565156669823923, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 0.286575973 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.1680702765782571, "validation_loss": 1.0726634152606465, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 0.563498858 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784489967242784, "validation_loss": 0.8820201711707012, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 0.841784853 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923336749857173, "validation_loss": 0.710308352958468, "validation_accuracy": 0.7502543234994914, "elapsed_seconds": 1.113011794 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370655949431792, "validation_loss": 0.5771756065192747, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 1.388583175 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222779060476093, "validation_loss": 0.4819741321230154, "validation_accuracy": 0.8320447609359105, "elapsed_seconds": 1.662748677 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.44097582388285234, "validation_loss": 0.41512437864999546, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 1.937911793 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.38314860589065225, "validation_loss": 0.36713147013723113, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 2.21139743 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.3406074100312188, "validation_loss": 0.3310860565229031, "validation_accuracy": 0.8786368260427263, "elapsed_seconds": 2.487264023 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.3077992643371988, "validation_loss": 0.3033731649951008, "validation_accuracy": 0.8857578840284842, "elapsed_seconds": 2.753584447 } ], "validation_loss": 0.3033731649951008, "coordinator_cpu_threads": 1, "rate": 0.05, "coordinator_cpu_provider": "openblas", "test_loss": 0.29705317871020637, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332128378295, "fit_seconds": 3.475173471, "training_loss": 0.29234691523654444, "tensor_owned_device_peak_bytes": 13048912, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 4202560, "dtype": "Float64", "command_seconds": 4.05719003, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float64", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 1.062177131, "initialization_seconds": 0, "failed": false, "nodes": [ { "response_read_wait_seconds": 0, "requests": 300, "coordinator_exchange_wall_seconds": 0, "inner_threads": 1, "backend": "cuda_cublas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 4202560, "request_write_seconds": 0, "request_payload_bytes": 0, "process_peak_rss_bytes": 325083136, "samples": 458760, "upload_preparation_seconds": 0.42055923900000003, "worker_total_seconds": 1.0605821979999992, "zero_sample_steps": 0, "response_payload_bytes": 0, "compute_seconds": 0.4274700339999997, "owned_device_tensor_peak_bytes": 13048912, "device": "cuda", "readback_seconds": 0.056410975999999995, "configuration_id": "atlas-03/C03", "platform": "atlas-03", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu64", "reference_trial": 1, "dtype": "Float64", "predeclared_tolerances": { "parameter_max_abs": 1e-08, "loss_max_abs": 1e-08, "accuracy_max_abs": 1e-12 }, "parameter_max_abs_error": 3.3306690738754696e-16, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 0.0, "validation_loss": 5.551115123125783e-17, "test_loss": 5.551115123125783e-17 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "01-02-gpu32", "repetition": 0, "started_utc": "2026-10-12T21:05:22.081331+00:00", "finished_utc": "2026-10-12T21:06:26.820749+00:00", "outer_command_wall_seconds": 64.73958308300644, "warmup_excluded_from_statistics": true, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 78266368, "test_accuracy": 0.8955239064089522, "parameter_sha256": "d4981c6fe0e53117073c0d8563cac476fa35e24af2658d8ecd892e7d5539bb07", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8858596134282808, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.3770934769285932, "validation_loss": 1.2565156643897568, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 8.746244291 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.168070280486428, "validation_loss": 1.072663564672305, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 16.292603625 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784492893724769, "validation_loss": 0.8820205004872529, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 23.333769166 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923334494268212, "validation_loss": 0.7103084890670233, "validation_accuracy": 0.7501525940996948, "elapsed_seconds": 30.351758458 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370660988270472, "validation_loss": 0.5771758733421473, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 36.678942458 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222784888711614, "validation_loss": 0.4819751959608627, "validation_accuracy": 0.8319430315361139, "elapsed_seconds": 42.359690875 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.4409767913780941, "validation_loss": 0.41512575641895966, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 47.959577083 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.3831498941926472, "validation_loss": 0.3671321308139867, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 53.514541833 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.340608996330616, "validation_loss": 0.3310877792408362, "validation_accuracy": 0.8787385554425229, "elapsed_seconds": 58.999707333 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.3078003465981188, "validation_loss": 0.30337470066511013, "validation_accuracy": 0.8858596134282808, "elapsed_seconds": 64.058257416 } ], "validation_loss": 0.30337470066511013, "coordinator_cpu_threads": null, "rate": 0.05, "coordinator_cpu_provider": "apple_accelerate", "test_loss": 0.29705493784872467, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332174875397, "fit_seconds": 64.2693505, "training_loss": 0.29234811506937414, "tensor_owned_device_peak_bytes": 0, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 6303328, "dtype": "Float32", "command_seconds": 64.660003708, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float32", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 60.38416687900002, "initialization_seconds": 2.469252625, "failed": false, "nodes": [ { "response_read_wait_seconds": 5.877354080000004, "requests": 300, "coordinator_exchange_wall_seconds": 6.926406029999998, "inner_threads": null, "backend": "metal_mps", "target": "macos-arm64", "owned_host_tensor_peak_bytes": 983040, "request_write_seconds": 1.0489763320000003, "request_payload_bytes": 82667760, "process_peak_rss_bytes": 24707072, "samples": 229380, "upload_preparation_seconds": 0.26727945300000006, "worker_total_seconds": 3.9411224920000025, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 3.611328246999999, "owned_device_tensor_peak_bytes": 3329576, "device": "mps", "readback_seconds": 0.025695535999999995, "configuration_id": "atlas-01/C03", "platform": "atlas-01", "relative_batch_weight": 1 }, { "response_read_wait_seconds": 58.65864878699998, "requests": 300, "coordinator_exchange_wall_seconds": 59.92227559199998, "inner_threads": 1, "backend": "cuda_cublas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 983040, "request_write_seconds": 1.2635202579999996, "request_payload_bytes": 82667760, "process_peak_rss_bytes": 252731392, "samples": 229380, "upload_preparation_seconds": 1.67697006, "worker_total_seconds": 2.9071535979999985, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 0.8204131400000004, "owned_device_tensor_peak_bytes": 3329576, "device": "cuda", "readback_seconds": 0.15230885300000005, "configuration_id": "atlas-02/C03", "platform": "atlas-02", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu32", "reference_trial": 1, "dtype": "Float32", "predeclared_tolerances": { "parameter_max_abs": 0.001, "loss_max_abs": 0.0005, "accuracy_max_abs": 0.005 }, "parameter_max_abs_error": 3.8526952266693115e-05, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 5.787559557157707e-07, "validation_loss": 2.747028308158761e-07, "test_loss": 1.008616095310444e-06 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "01-03-gpu32", "repetition": 0, "started_utc": "2026-10-12T21:09:50.674974+00:00", "finished_utc": "2026-10-12T21:11:00.289716+00:00", "outer_command_wall_seconds": 69.61488270800328, "warmup_excluded_from_statistics": true, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 77971456, "test_accuracy": 0.8955239064089522, "parameter_sha256": "76a2caf28ba047bb311582e24dcb029de184d31a707f97d640024c9b33247040", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8858596134282808, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.377093485176267, "validation_loss": 1.2565156643897568, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 9.490744 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.1680702864733972, "validation_loss": 1.072663564672305, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 15.205177875 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784492919683893, "validation_loss": 0.8820206904823636, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 21.625301791 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923334923880031, "validation_loss": 0.7103080904859735, "validation_accuracy": 0.7502543234994914, "elapsed_seconds": 28.793399916 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370656905827833, "validation_loss": 0.5771757672907498, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 35.147466166 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222784275359083, "validation_loss": 0.48197514480273573, "validation_accuracy": 0.8320447609359105, "elapsed_seconds": 41.978165583 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.4409769844500581, "validation_loss": 0.41512621847925935, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 48.910739666 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.3831507439174298, "validation_loss": 0.3671325146727499, "validation_accuracy": 0.8679552390640896, "elapsed_seconds": 55.230234458 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.34060974856836146, "validation_loss": 0.3310890749597113, "validation_accuracy": 0.8787385554425229, "elapsed_seconds": 62.200927708 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.30780227909903857, "validation_loss": 0.30337603640338046, "validation_accuracy": 0.8858596134282808, "elapsed_seconds": 68.993155958 } ], "validation_loss": 0.30337603640338046, "coordinator_cpu_threads": null, "rate": 0.05, "coordinator_cpu_provider": "apple_accelerate", "test_loss": 0.29705585060017764, "optimizer": "sgd", "training_accuracy": 0.8984000348766239, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332174875397, "fit_seconds": 69.176242084, "training_loss": 0.2923502071331597, "tensor_owned_device_peak_bytes": 0, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 6303328, "dtype": "Float32", "command_seconds": 69.564732625, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float32", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 65.86550407899999, "initialization_seconds": 1.868351, "failed": false, "nodes": [ { "response_read_wait_seconds": 5.7124138659999995, "requests": 300, "coordinator_exchange_wall_seconds": 6.819589626000005, "inner_threads": null, "backend": "metal_mps", "target": "macos-arm64", "owned_host_tensor_peak_bytes": 983040, "request_write_seconds": 1.1071086220000008, "request_payload_bytes": 82667760, "process_peak_rss_bytes": 23003136, "samples": 229380, "upload_preparation_seconds": 0.27509722200000003, "worker_total_seconds": 3.7087138409999985, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 3.371315613, "owned_device_tensor_peak_bytes": 3329576, "device": "mps", "readback_seconds": 0.023875389, "configuration_id": "atlas-01/C03", "platform": "atlas-01", "relative_batch_weight": 1 }, { "response_read_wait_seconds": 64.06248253499999, "requests": 300, "coordinator_exchange_wall_seconds": 65.38049750400006, "inner_threads": 1, "backend": "cuda_cublas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 983040, "request_write_seconds": 1.3179383750000004, "request_payload_bytes": 82667760, "process_peak_rss_bytes": 286916608, "samples": 229380, "upload_preparation_seconds": 1.0532853790000003, "worker_total_seconds": 1.6054541470000003, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 0.3555906759999999, "owned_device_tensor_peak_bytes": 3329576, "device": "cuda", "readback_seconds": 0.06211730699999996, "configuration_id": "atlas-03/C03", "platform": "atlas-03", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu32", "reference_trial": 1, "dtype": "Float32", "predeclared_tolerances": { "parameter_max_abs": 0.001, "loss_max_abs": 0.0005, "accuracy_max_abs": 0.005 }, "parameter_max_abs_error": 5.525164306163788e-05, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 2.6708197412572687e-06, "validation_loss": 1.610441101140836e-06, "test_loss": 1.9213675482854242e-06 }, "accuracy_abs_errors": { "training_accuracy": 4.359577992840702e-05, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "02-03-gpu32", "repetition": 0, "started_utc": "2026-10-12T21:14:04.084129+00:00", "finished_utc": "2026-10-12T21:15:39.640508+00:00", "outer_command_wall_seconds": 95.56755437499669, "warmup_excluded_from_statistics": true, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 74743808, "test_accuracy": 0.8955239064089522, "parameter_sha256": "cbf657f3b09c0dec409f57f27268269a2c3fad35ede16055b7f4412d0779c3d0", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8857578840284842, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.3770934871719234, "validation_loss": 1.2565156643897568, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 14.099307625 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.1680703039042089, "validation_loss": 1.072663564672305, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 23.173418417 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784492979553586, "validation_loss": 0.8820206997959684, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 32.125925917 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.792333496246792, "validation_loss": 0.7103080904859735, "validation_accuracy": 0.7502543234994914, "elapsed_seconds": 41.35906825 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370656875892986, "validation_loss": 0.5771757579771452, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 51.365916625 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222784380131045, "validation_loss": 0.4819751541102769, "validation_accuracy": 0.8320447609359105, "elapsed_seconds": 61.09195325 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.4409769955275104, "validation_loss": 0.41512625572761463, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 70.004898375 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.3831508259147429, "validation_loss": 0.3671326767270466, "validation_accuracy": 0.8679552390640896, "elapsed_seconds": 77.819035209 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.34060971804885004, "validation_loss": 0.33108892592990874, "validation_accuracy": 0.8787385554425229, "elapsed_seconds": 85.973215917 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.3078019323654272, "validation_loss": 0.3033758939464604, "validation_accuracy": 0.8857578840284842, "elapsed_seconds": 94.858135292 } ], "validation_loss": 0.3033758939464604, "coordinator_cpu_threads": null, "rate": 0.05, "coordinator_cpu_provider": "apple_accelerate", "test_loss": 0.2970558105929094, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332174875397, "fit_seconds": 95.03769775, "training_loss": 0.2923499547917588, "tensor_owned_device_peak_bytes": 0, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 6303328, "dtype": "Float32", "command_seconds": 95.500020792, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float32", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 89.552289162, "initialization_seconds": 3.967912291, "failed": false, "nodes": [ { "response_read_wait_seconds": 81.76176778300005, "requests": 300, "coordinator_exchange_wall_seconds": 83.15705933399998, "inner_threads": 1, "backend": "cuda_cublas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 983040, "request_write_seconds": 1.3951505409999996, "request_payload_bytes": 82667760, "process_peak_rss_bytes": 252661760, "samples": 229380, "upload_preparation_seconds": 1.6805691330000014, "worker_total_seconds": 2.9115065489999985, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 0.8238766070000008, "owned_device_tensor_peak_bytes": 3329576, "device": "cuda", "readback_seconds": 0.14680787600000003, "configuration_id": "atlas-02/C03", "platform": "atlas-02", "relative_batch_weight": 1 }, { "response_read_wait_seconds": 77.97323866399998, "requests": 300, "coordinator_exchange_wall_seconds": 79.37411287499997, "inner_threads": 1, "backend": "cuda_cublas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 983040, "request_write_seconds": 1.400813331, "request_payload_bytes": 82667760, "process_peak_rss_bytes": 286781440, "samples": 229380, "upload_preparation_seconds": 1.0792778760000001, "worker_total_seconds": 1.6620221430000004, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 0.38421525500000003, "owned_device_tensor_peak_bytes": 3329576, "device": "cuda", "readback_seconds": 0.06359029499999999, "configuration_id": "atlas-03/C03", "platform": "atlas-03", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu32", "reference_trial": 1, "dtype": "Float32", "predeclared_tolerances": { "parameter_max_abs": 0.001, "loss_max_abs": 0.0005, "accuracy_max_abs": 0.005 }, "parameter_max_abs_error": 5.7095661759376526e-05, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 2.418478340371344e-06, "validation_loss": 1.4679841811027217e-06, "test_loss": 1.8813602800649143e-06 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.00010172939979657958, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "01-02-03-gpu32", "repetition": 0, "started_utc": "2026-10-12T21:20:07.885621+00:00", "finished_utc": "2026-10-12T21:21:33.485203+00:00", "outer_command_wall_seconds": 85.60007712499646, "warmup_excluded_from_statistics": true, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 78118912, "test_accuracy": 0.8955239064089522, "parameter_sha256": "267925aebd1e6e485966fd3926f1b0e321368eb6395472f921d629c627c888a5", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8857578840284842, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.3770934779472095, "validation_loss": 1.2565156643897568, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 12.734590292 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.1680702881156562, "validation_loss": 1.072663564672305, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 21.334082 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.978449299727543, "validation_loss": 0.8820205004872529, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 29.046520875 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923334504246493, "validation_loss": 0.7103084927900398, "validation_accuracy": 0.7501525940996948, "elapsed_seconds": 36.737624542 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370661070981077, "validation_loss": 0.577175882655752, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 46.822999708 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222785054353695, "validation_loss": 0.481975248113411, "validation_accuracy": 0.8319430315361139, "elapsed_seconds": 55.141711917 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.44097682620333817, "validation_loss": 0.41512579553488677, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 62.263293542 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.3831498520357044, "validation_loss": 0.36713215224861734, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 70.529649875 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.34060882918919294, "validation_loss": 0.3310873703760145, "validation_accuracy": 0.8787385554425229, "elapsed_seconds": 78.874027292 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.3077999734350523, "validation_loss": 0.30337328036464484, "validation_accuracy": 0.8857578840284842, "elapsed_seconds": 84.854626292 } ], "validation_loss": 0.30337328036464484, "coordinator_cpu_threads": null, "rate": 0.05, "coordinator_cpu_provider": "apple_accelerate", "test_loss": 0.29705295057650866, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332174875397, "fit_seconds": 85.038873417, "training_loss": 0.2923465431937054, "tensor_owned_device_peak_bytes": 0, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 6303328, "dtype": "Float32", "command_seconds": 85.513868917, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float32", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 79.48423029400001, "initialization_seconds": 4.100220041, "failed": false, "nodes": [ { "response_read_wait_seconds": 5.306526963000006, "requests": 300, "coordinator_exchange_wall_seconds": 6.201332535000001, "inner_threads": null, "backend": "metal_mps", "target": "macos-arm64", "owned_host_tensor_peak_bytes": 655360, "request_write_seconds": 0.8947482399999999, "request_payload_bytes": 61870640, "process_peak_rss_bytes": 22167552, "samples": 152920, "upload_preparation_seconds": 0.23415579900000008, "worker_total_seconds": 3.4744991629999995, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 3.177466288, "owned_device_tensor_peak_bytes": 2264616, "device": "mps", "readback_seconds": 0.02359312399999999, "configuration_id": "atlas-01/C03", "platform": "atlas-01", "relative_batch_weight": 1 }, { "response_read_wait_seconds": 73.09806750899997, "requests": 300, "coordinator_exchange_wall_seconds": 74.09301870399999, "inner_threads": 1, "backend": "cuda_cublas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 655360, "request_write_seconds": 0.9948952450000004, "request_payload_bytes": 61870640, "process_peak_rss_bytes": 251990016, "samples": 152920, "upload_preparation_seconds": 1.3329841280000008, "worker_total_seconds": 2.526531604, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 0.7928417719999993, "owned_device_tensor_peak_bytes": 2264616, "device": "cuda", "readback_seconds": 0.14141310799999998, "configuration_id": "atlas-02/C03", "platform": "atlas-02", "relative_batch_weight": 1 }, { "response_read_wait_seconds": 61.40420241599995, "requests": 300, "coordinator_exchange_wall_seconds": 62.40696596500003, "inner_threads": 1, "backend": "cuda_cublas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 655360, "request_write_seconds": 1.0026990840000003, "request_payload_bytes": 61870640, "process_peak_rss_bytes": 286343168, "samples": 152920, "upload_preparation_seconds": 0.9471422430000005, "worker_total_seconds": 1.6062523249999998, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 0.42203035700000024, "owned_device_tensor_peak_bytes": 2264616, "device": "cuda", "readback_seconds": 0.07208261100000003, "configuration_id": "atlas-03/C03", "platform": "atlas-03", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu32", "reference_trial": 1, "dtype": "Float32", "predeclared_tolerances": { "parameter_max_abs": 0.001, "loss_max_abs": 0.0005, "accuracy_max_abs": 0.005 }, "parameter_max_abs_error": 4.633516073226929e-05, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 9.931197130219616e-07, "validation_loss": 1.1455976344776353e-06, "test_loss": 9.786561206981403e-07 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.00010172939979657958, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "02-03-gpu64", "repetition": 0, "started_utc": "2026-10-12T21:25:55.065989+00:00", "finished_utc": "2026-10-12T21:27:50.219287+00:00", "outer_command_wall_seconds": 115.15401437500259, "warmup_excluded_from_statistics": true, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 76021760, "test_accuracy": 0.8955239064089522, "parameter_sha256": "2e41fec29bc85fdf18b8447ee70bad448e44b1ed687e126d50c34e7270df079d", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8857578840284842, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.377093474953068, "validation_loss": 1.2565156669823923, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 16.12656475 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.1680702765782571, "validation_loss": 1.0726634152606465, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 26.834443833 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784489967242784, "validation_loss": 0.8820201711707012, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 39.286660125 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923336749857174, "validation_loss": 0.7103083529584678, "validation_accuracy": 0.7502543234994914, "elapsed_seconds": 50.762432083 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370655949431793, "validation_loss": 0.5771756065192747, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 61.861711375 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222779060476093, "validation_loss": 0.4819741321230155, "validation_accuracy": 0.8320447609359105, "elapsed_seconds": 74.065822791 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.44097582388285234, "validation_loss": 0.41512437864999546, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 84.69395 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.38314860589065225, "validation_loss": 0.36713147013723113, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 94.797147916 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.3406074100312188, "validation_loss": 0.3310860565229032, "validation_accuracy": 0.8786368260427263, "elapsed_seconds": 104.190977708 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.3077992643371988, "validation_loss": 0.3033731649951008, "validation_accuracy": 0.8857578840284842, "elapsed_seconds": 114.424461458 } ], "validation_loss": 0.3033731649951008, "coordinator_cpu_threads": null, "rate": 0.05, "coordinator_cpu_provider": "apple_accelerate", "test_loss": 0.29705317871020637, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332128378295, "fit_seconds": 114.626010708, "training_loss": 0.2923469152365444, "tensor_owned_device_peak_bytes": 0, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 4202560, "dtype": "Float64", "command_seconds": 115.079214208, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float64", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 108.75344520400007, "initialization_seconds": 4.480273959, "failed": false, "nodes": [ { "response_read_wait_seconds": 96.97324445399998, "requests": 300, "coordinator_exchange_wall_seconds": 99.59451492300005, "inner_threads": 1, "backend": "cuda_cublas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 393216, "request_write_seconds": 2.621154334000001, "request_payload_bytes": 165262320, "process_peak_rss_bytes": 254672896, "samples": 229380, "upload_preparation_seconds": 1.097188956, "worker_total_seconds": 3.421205883, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 1.7691627810000001, "owned_device_tensor_peak_bytes": 6659152, "device": "cuda", "readback_seconds": 0.16161648099999995, "configuration_id": "atlas-02/C03", "platform": "atlas-02", "relative_batch_weight": 1 }, { "response_read_wait_seconds": 97.71336742200003, "requests": 300, "coordinator_exchange_wall_seconds": 100.33254196500003, "inner_threads": 1, "backend": "cuda_cublas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 393216, "request_write_seconds": 2.6191075460000017, "request_payload_bytes": 165262320, "process_peak_rss_bytes": 282333184, "samples": 229380, "upload_preparation_seconds": 0.44741871000000016, "worker_total_seconds": 1.1036737510000008, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 0.4342578450000002, "owned_device_tensor_peak_bytes": 6659152, "device": "cuda", "readback_seconds": 0.06386925900000001, "configuration_id": "atlas-03/C03", "platform": "atlas-03", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu64", "reference_trial": 1, "dtype": "Float64", "predeclared_tolerances": { "parameter_max_abs": 1e-08, "loss_max_abs": 1e-08, "accuracy_max_abs": 1e-12 }, "parameter_max_abs_error": 3.3306690738754696e-16, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 5.551115123125783e-17, "validation_loss": 5.551115123125783e-17, "test_loss": 5.551115123125783e-17 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "01-02-03-cpu32", "repetition": 0, "started_utc": "2026-10-12T21:34:00.845146+00:00", "finished_utc": "2026-10-12T21:35:22.123635+00:00", "outer_command_wall_seconds": 81.27868870800012, "warmup_excluded_from_statistics": true, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 77676544, "test_accuracy": 0.8955239064089522, "parameter_sha256": "0515a7177dd3e8dd33592beb68b3d4ee4068c8d014af92de819605ffeffb41e6", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8858596134282808, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.377093476616772, "validation_loss": 1.2565156643897568, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 12.13179925 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.1680702856314797, "validation_loss": 1.072663564672305, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 19.472795292 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784492924101361, "validation_loss": 0.8820205098008576, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 26.815865583 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923334511782175, "validation_loss": 0.7103085021036445, "validation_accuracy": 0.7501525940996948, "elapsed_seconds": 33.838339542 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.63706611166629, "validation_loss": 0.577175882655752, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 40.432530958 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222784787122826, "validation_loss": 0.4819751335609238, "validation_accuracy": 0.8319430315361139, "elapsed_seconds": 47.241279708 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.4409767381969684, "validation_loss": 0.41512574337627495, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 55.682459458 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.38314981840577533, "validation_loss": 0.36713236366623175, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 64.821309375 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.3406089726244003, "validation_loss": 0.33108807820997305, "validation_accuracy": 0.8787385554425229, "elapsed_seconds": 73.418497542 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.3078001165168622, "validation_loss": 0.30337432724777097, "validation_accuracy": 0.8858596134282808, "elapsed_seconds": 80.560801125 } ], "validation_loss": 0.30337432724777097, "coordinator_cpu_threads": null, "rate": 0.05, "coordinator_cpu_provider": "apple_accelerate", "test_loss": 0.2970538696340481, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332174875397, "fit_seconds": 80.743491583, "training_loss": 0.2923475254698321, "tensor_owned_device_peak_bytes": 0, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 6303328, "dtype": "Float32", "command_seconds": 81.215138791, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float32", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 75.86044349699999, "initialization_seconds": 3.422008625, "failed": false, "nodes": [ { "response_read_wait_seconds": 2.8950631949999983, "requests": 300, "coordinator_exchange_wall_seconds": 3.748421306999999, "inner_threads": null, "backend": "apple_accelerate", "target": "macos-arm64", "owned_host_tensor_peak_bytes": 6834532, "request_write_seconds": 0.8533067680000003, "request_payload_bytes": 61870640, "process_peak_rss_bytes": 23756800, "samples": 152920, "upload_preparation_seconds": 0.1485217880000001, "worker_total_seconds": 1.2233465459999993, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 1.0380956760000002, "owned_device_tensor_peak_bytes": 0, "device": "cpu", "readback_seconds": 0.018250369, "configuration_id": "atlas-01/C03", "platform": "atlas-01", "relative_batch_weight": 1 }, { "response_read_wait_seconds": 69.60150283399997, "requests": 300, "coordinator_exchange_wall_seconds": 70.53996233500004, "inner_threads": 1, "backend": "openblas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 6834532, "request_write_seconds": 0.9383336269999999, "request_payload_bytes": 61870640, "process_peak_rss_bytes": 124370944, "samples": 152920, "upload_preparation_seconds": 1.3378761519999995, "worker_total_seconds": 13.76024367299999, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 11.857667987999992, "owned_device_tensor_peak_bytes": 0, "device": "cpu", "readback_seconds": 0.14817662999999998, "configuration_id": "atlas-02/C03", "platform": "atlas-02", "relative_batch_weight": 1 }, { "response_read_wait_seconds": 66.98049921400005, "requests": 300, "coordinator_exchange_wall_seconds": 67.92628737099999, "inner_threads": 1, "backend": "openblas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 6834532, "request_write_seconds": 0.9457282869999999, "request_payload_bytes": 61870640, "process_peak_rss_bytes": 124448768, "samples": 152920, "upload_preparation_seconds": 0.9647414199999996, "worker_total_seconds": 5.617922715999999, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 4.464594031999999, "owned_device_tensor_peak_bytes": 0, "device": "cpu", "readback_seconds": 0.03709636999999999, "configuration_id": "atlas-03/C03", "platform": "atlas-03", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu32", "reference_trial": 1, "dtype": "Float32", "predeclared_tolerances": { "parameter_max_abs": 0.001, "loss_max_abs": 0.0005, "accuracy_max_abs": 0.005 }, "parameter_max_abs_error": 4.561617970466614e-06, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 1.0843586306652497e-08, "validation_loss": 9.871450834531004e-08, "test_loss": 5.9598581236830483e-08 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "01-02-03-cpu64", "repetition": 0, "started_utc": "2026-10-12T21:39:55.196012+00:00", "finished_utc": "2026-10-12T21:41:52.981474+00:00", "outer_command_wall_seconds": 117.78573633299675, "warmup_excluded_from_statistics": true, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 72925184, "test_accuracy": 0.8955239064089522, "parameter_sha256": "41b2bc8c6a1cd76505a0568ed6c5f96c91a1999298441dbd8cc81399f9ada917", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8857578840284842, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.377093474953068, "validation_loss": 1.2565156669823925, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 17.070421125 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.1680702765782573, "validation_loss": 1.0726634152606465, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 27.760898334 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784489967242783, "validation_loss": 0.8820201711707012, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 38.764758 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923336749857173, "validation_loss": 0.710308352958468, "validation_accuracy": 0.7502543234994914, "elapsed_seconds": 49.399004959 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370655949431792, "validation_loss": 0.5771756065192747, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 60.103717084 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222779060476092, "validation_loss": 0.4819741321230154, "validation_accuracy": 0.8320447609359105, "elapsed_seconds": 71.906559625 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.44097582388285234, "validation_loss": 0.4151243786499954, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 83.823334834 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.38314860589065225, "validation_loss": 0.3671314701372311, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 95.195944334 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.3406074100312187, "validation_loss": 0.3310860565229032, "validation_accuracy": 0.8786368260427263, "elapsed_seconds": 106.255588959 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.3077992643371988, "validation_loss": 0.3033731649951008, "validation_accuracy": 0.8857578840284842, "elapsed_seconds": 117.062359167 } ], "validation_loss": 0.3033731649951008, "coordinator_cpu_threads": null, "rate": 0.05, "coordinator_cpu_provider": "apple_accelerate", "test_loss": 0.2970531787102063, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332128378295, "fit_seconds": 117.262093792, "training_loss": 0.29234691523654444, "tensor_owned_device_peak_bytes": 0, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 4202560, "dtype": "Float64", "command_seconds": 117.729577417, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float64", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 112.07988546600002, "initialization_seconds": 3.767031917, "failed": false, "nodes": [ { "response_read_wait_seconds": 3.835400244999999, "requests": 300, "coordinator_exchange_wall_seconds": 5.525068798999997, "inner_threads": null, "backend": "apple_accelerate", "target": "macos-arm64", "owned_host_tensor_peak_bytes": 5398608, "request_write_seconds": 1.6896159579999992, "request_payload_bytes": 123668080, "process_peak_rss_bytes": 20430848, "samples": 152920, "upload_preparation_seconds": 0.10181038099999992, "worker_total_seconds": 1.493291511000001, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 1.357170673, "owned_device_tensor_peak_bytes": 0, "device": "cpu", "readback_seconds": 0.015514420000000001, "configuration_id": "atlas-01/C03", "platform": "atlas-01", "relative_batch_weight": 1 }, { "response_read_wait_seconds": 103.45107758500008, "requests": 300, "coordinator_exchange_wall_seconds": 105.54949871499997, "inner_threads": 1, "backend": "openblas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 5398608, "request_write_seconds": 2.0982967080000017, "request_payload_bytes": 123668080, "process_peak_rss_bytes": 122888192, "samples": 152920, "upload_preparation_seconds": 0.898226373, "worker_total_seconds": 13.308884870000004, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 11.913351262999987, "owned_device_tensor_peak_bytes": 0, "device": "cpu", "readback_seconds": 0.1604217169999999, "configuration_id": "atlas-02/C03", "platform": "atlas-02", "relative_batch_weight": 1 }, { "response_read_wait_seconds": 92.84670632799995, "requests": 300, "coordinator_exchange_wall_seconds": 94.94475455000003, "inner_threads": 1, "backend": "openblas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 5398608, "request_write_seconds": 2.097993712000002, "request_payload_bytes": 123668080, "process_peak_rss_bytes": 122707968, "samples": 152920, "upload_preparation_seconds": 0.47593388800000025, "worker_total_seconds": 4.495309421999996, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 3.8611696270000007, "owned_device_tensor_peak_bytes": 0, "device": "cpu", "readback_seconds": 0.03835445099999999, "configuration_id": "atlas-03/C03", "platform": "atlas-03", "relative_batch_weight": 1 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu64", "reference_trial": 1, "dtype": "Float64", "predeclared_tolerances": { "parameter_max_abs": 1e-08, "loss_max_abs": 1e-08, "accuracy_max_abs": 1e-12 }, "parameter_max_abs_error": 2.7755575615628914e-16, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 0.0, "validation_loss": 5.551115123125783e-17, "test_loss": 1.1102230246251565e-16 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } }, { "case": "02-03-gpu32-w1-7", "repetition": 0, "started_utc": "2026-10-12T21:47:36.249309+00:00", "finished_utc": "2026-10-12T21:49:12.845327+00:00", "outer_command_wall_seconds": 96.59534758300288, "warmup_excluded_from_statistics": true, "state": "PASS", "native": { "epochs": 10, "process_peak_rss_bytes": 77725696, "test_accuracy": 0.8955239064089522, "parameter_sha256": "7b775ba5290afda4e7d8f423f17c59214a43cd340d4e1f62d0bcd21e508a56d0", "test_selection_sha256": "0032029550bedde5a54824100b56ba6de1ec4c6d3fc3203fce4fa18311272aa0", "evaluation_device": "cpu", "validation_rows": 9830, "canonical_device": "cpu", "training_selection_sha256": "b1339234a2470c33fdf8e0a4dae2262ffca2e018696b7e462373b9e1529c0159", "global_batch": 1536, "dataset_source_sha256": "72bde3eff8b278c07be2746ea50d317557c7534244eee0452f1f7907e0f36562", "validation_accuracy": 0.8857578840284842, "test_rows": 9830, "history": [ { "epoch": 1, "updates": 30, "loss_before_updates": 1.3770934739558967, "validation_loss": 1.2565156643897568, "validation_accuracy": 0.42675483214649035, "elapsed_seconds": 15.895174041 }, { "epoch": 2, "updates": 60, "loss_before_updates": 1.168070296236003, "validation_loss": 1.072663564672305, "validation_accuracy": 0.576703967446592, "elapsed_seconds": 26.051440291 }, { "epoch": 3, "updates": 90, "loss_before_updates": 0.9784492987595977, "validation_loss": 0.8820205004872529, "validation_accuracy": 0.6831129196337742, "elapsed_seconds": 33.832212791 }, { "epoch": 4, "updates": 120, "loss_before_updates": 0.7923334554839502, "validation_loss": 0.7103084927900398, "validation_accuracy": 0.7501525940996948, "elapsed_seconds": 42.564675583 }, { "epoch": 5, "updates": 150, "loss_before_updates": 0.6370661065030486, "validation_loss": 0.577175882655752, "validation_accuracy": 0.7958290946083418, "elapsed_seconds": 51.586851791 }, { "epoch": 6, "updates": 180, "loss_before_updates": 0.5222785073355307, "validation_loss": 0.481975248113411, "validation_accuracy": 0.8319430315361139, "elapsed_seconds": 59.279952625 }, { "epoch": 7, "updates": 210, "loss_before_updates": 0.4409768236892788, "validation_loss": 0.41512579553488677, "validation_accuracy": 0.8533062054933875, "elapsed_seconds": 68.89562525 }, { "epoch": 8, "updates": 240, "loss_before_updates": 0.3831498536207961, "validation_loss": 0.36713215224861734, "validation_accuracy": 0.8681586978636826, "elapsed_seconds": 78.803372833 }, { "epoch": 9, "updates": 270, "loss_before_updates": 0.3406088334786849, "validation_loss": 0.331087377828111, "validation_accuracy": 0.8787385554425229, "elapsed_seconds": 87.370968416 }, { "epoch": 10, "updates": 300, "loss_before_updates": 0.3077999742386418, "validation_loss": 0.30337328036464484, "validation_accuracy": 0.8857578840284842, "elapsed_seconds": 95.923373041 } ], "validation_loss": 0.30337328036464484, "coordinator_cpu_threads": null, "rate": 0.05, "coordinator_cpu_provider": "apple_accelerate", "test_loss": 0.29705295057650866, "optimizer": "sgd", "training_accuracy": 0.8983564390966955, "model": [ 64, 128, 64, 4 ], "gradient_accumulation": "Float64", "initial_loss": 1.5991332174875397, "fit_seconds": 96.099011083, "training_loss": 0.29234655017850286, "tensor_owned_device_peak_bytes": 0, "source_sha256": "43aaceb4725f52a5136643b824d790c26600f92378c6cf753287c0b069d5211a", "parameters": 16836, "tensor_owned_host_peak_bytes": 6303328, "dtype": "Float32", "command_seconds": 96.564945334, "validation_selection_sha256": "132f697d6ead0297570749a407872ddeb1987e2ffd8f53f3313d205b7c21499b", "training_rows": 45876, "seed": 42, "replica_execution": { "dtype": "Float32", "host_accumulation": "Float64", "transport": "portable-host-buffers", "elapsed_seconds": 90.68487058299996, "initialization_seconds": 3.915627166, "failed": false, "nodes": [ { "response_read_wait_seconds": 54.90928757299996, "requests": 300, "coordinator_exchange_wall_seconds": 55.42423544900004, "inner_threads": 1, "backend": "cuda_cublas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 245760, "request_write_seconds": 0.5148984849999997, "request_payload_bytes": 35875600, "process_peak_rss_bytes": 251752448, "samples": 57350, "upload_preparation_seconds": 0.8554224539999999, "worker_total_seconds": 1.8019077080000003, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 0.6352282370000008, "owned_device_tensor_peak_bytes": 933416, "device": "cuda", "readback_seconds": 0.13762357799999997, "configuration_id": "atlas-02/C03", "platform": "atlas-02", "relative_batch_weight": 1 }, { "response_read_wait_seconds": 84.32314129500007, "requests": 300, "coordinator_exchange_wall_seconds": 86.347758223, "inner_threads": 1, "backend": "cuda_cublas", "target": "linux-x86_64", "owned_host_tensor_peak_bytes": 1720320, "request_write_seconds": 2.0245749999999996, "request_payload_bytes": 129459920, "process_peak_rss_bytes": 288497664, "samples": 401410, "upload_preparation_seconds": 1.1468986990000005, "worker_total_seconds": 1.7134027259999998, "zero_sample_steps": 0, "response_payload_bytes": 40441800, "compute_seconds": 0.37050897099999985, "owned_device_tensor_peak_bytes": 5725736, "device": "cuda", "readback_seconds": 0.06267061499999999, "configuration_id": "atlas-03/C03", "platform": "atlas-03", "relative_batch_weight": 7 } ] } }, "numerical_comparison": { "state": "PASS", "reference_case": "atlas01-cpu32", "reference_trial": 1, "dtype": "Float32", "predeclared_tolerances": { "parameter_max_abs": 0.001, "loss_max_abs": 0.0005, "accuracy_max_abs": 0.005 }, "parameter_max_abs_error": 4.634261131286621e-05, "loss_abs_errors": { "initial_loss": 0.0, "training_loss": 9.861349155593224e-07, "validation_loss": 1.1455976344776353e-06, "test_loss": 9.786561206981403e-07 }, "accuracy_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.00010172939979657958, "test_accuracy": 0.0 }, "failed_fields": [] }, "external_process_gpu_memory": { "status": "UNAVAILABLE_NOT_REQUESTED", "peak_bytes": null, "requested_interval_ms": null } } ], "summary": [ { "case": "atlas01-cpu32", "label": "Atlas-01 CPU \u00b7 Float32", "dtype": "Float32", "kind": "standalone", "replicas": [ { "configuration_id": "atlas-01/C03", "platform": "atlas-01", "device": "cpu", "relative_batch_weight": 1 } ], "coordinator_configuration_id": "atlas-01/C03", "numerical_reference_case": "atlas01-cpu32", "repetitions": 3, "outer_command_wall_seconds": { "median": 3.0628960830072174, "min": 3.0537536250049016, "max": 3.1175427089910954 }, "fit_seconds": { "median": 2.721057125, "min": 2.670434875, "max": 2.743213375 }, "command_seconds": { "median": 3.0174825, "min": 2.961245625, "max": 3.033698916 }, "quality": { "initial_loss": { "median": 1.5991332174875397, "min": 1.5991332174875397, "max": 1.5991332174875397 }, "training_loss": { "median": 0.2923475363134184, "min": 0.2923475363134184, "max": 0.2923475363134184 }, "validation_loss": { "median": 0.3033744259622793, "min": 0.3033744259622793, "max": 0.3033744259622793 }, "test_loss": { "median": 0.29705392923262935, "min": 0.29705392923262935, "max": 0.29705392923262935 }, "training_accuracy": { "median": 0.8983564390966955, "min": 0.8983564390966955, "max": 0.8983564390966955 }, "validation_accuracy": { "median": 0.8858596134282808, "min": 0.8858596134282808, "max": 0.8858596134282808 }, "test_accuracy": { "median": 0.8955239064089522, "min": 0.8955239064089522, "max": 0.8955239064089522 } }, "parameter_max_abs_error": 0.0, "loss_max_abs_errors": { "initial_loss": 0.0, "training_loss": 0.0, "validation_loss": 0.0, "test_loss": 0.0 }, "accuracy_max_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "coordinator_memory": { "process_peak_rss_bytes": { "median": 121389056, "min": 104071168, "max": 127565824 }, "tensor_owned_host_peak_bytes": { "median": 21086868, "min": 21086868, "max": 21086868 }, "tensor_owned_device_peak_bytes": { "median": 0, "min": 0, "max": 0 } }, "replica_metrics": [ { "configuration_id": "atlas-01/C03", "platform": "atlas-01", "device": "cpu", "relative_batch_weight": 1, "response_read_wait_seconds": { "median": 0, "min": 0, "max": 0 }, "requests": { "median": 300, "min": 300, "max": 300 }, "coordinator_exchange_wall_seconds": { "median": 0, "min": 0, "max": 0 }, "inner_threads": null, "owned_host_tensor_peak_bytes": { "median": 21086868, "min": 21086868, "max": 21086868 }, "request_write_seconds": { "median": 0, "min": 0, "max": 0 }, "request_payload_bytes": { "median": 0, "min": 0, "max": 0 }, "process_peak_rss_bytes": { "median": 121356288, "min": 104038400, "max": 127533056 }, "samples": { "median": 458760, "min": 458760, "max": 458760 }, "upload_preparation_seconds": { "median": 0.16428429600000008, "min": 0.16247137999999994, "max": 0.16654621 }, "worker_total_seconds": { "median": 1.959291791, "min": 1.9309346270000005, "max": 1.986230381 }, "zero_sample_steps": { "median": 0, "min": 0, "max": 0 }, "response_payload_bytes": { "median": 0, "min": 0, "max": 0 }, "compute_seconds": { "median": 1.7714715769999996, "min": 1.7451295810000005, "max": 1.7956941709999998 }, "owned_device_tensor_peak_bytes": { "median": 0, "min": 0, "max": 0 }, "readback_seconds": { "median": 0.012126252999999997, "min": 0.012026081000000004, "max": 0.012358166999999995 }, "backend": "apple_accelerate", "target": "macos-arm64" } ], "fastest_selected_baseline": null }, { "case": "atlas01-cpu64", "label": "Atlas-01 CPU \u00b7 Float64", "dtype": "Float64", "kind": "standalone", "replicas": [ { "configuration_id": "atlas-01/C03", "platform": "atlas-01", "device": "cpu", "relative_batch_weight": 1 } ], "coordinator_configuration_id": "atlas-01/C03", "numerical_reference_case": "atlas01-cpu64", "repetitions": 3, "outer_command_wall_seconds": { "median": 3.4516792920039734, "min": 3.430949834000785, "max": 3.477942041005008 }, "fit_seconds": { "median": 3.086051291, "min": 3.077315833, "max": 3.097414333 }, "command_seconds": { "median": 3.387386792, "min": 3.37957175, "max": 3.39695175 }, "quality": { "initial_loss": { "median": 1.5991332128378295, "min": 1.5991332128378295, "max": 1.5991332128378295 }, "training_loss": { "median": 0.29234691523654444, "min": 0.29234691523654444, "max": 0.29234691523654444 }, "validation_loss": { "median": 0.30337316499510086, "min": 0.30337316499510086, "max": 0.30337316499510086 }, "test_loss": { "median": 0.2970531787102064, "min": 0.2970531787102064, "max": 0.2970531787102064 }, "training_accuracy": { "median": 0.8983564390966955, "min": 0.8983564390966955, "max": 0.8983564390966955 }, "validation_accuracy": { "median": 0.8857578840284842, "min": 0.8857578840284842, "max": 0.8857578840284842 }, "test_accuracy": { "median": 0.8955239064089522, "min": 0.8955239064089522, "max": 0.8955239064089522 } }, "parameter_max_abs_error": 0.0, "loss_max_abs_errors": { "initial_loss": 0.0, "training_loss": 0.0, "validation_loss": 0.0, "test_loss": 0.0 }, "accuracy_max_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "coordinator_memory": { "process_peak_rss_bytes": { "median": 94257152, "min": 94224384, "max": 112525312 }, "tensor_owned_host_peak_bytes": { "median": 16495216, "min": 16495216, "max": 16495216 }, "tensor_owned_device_peak_bytes": { "median": 0, "min": 0, "max": 0 } }, "replica_metrics": [ { "configuration_id": "atlas-01/C03", "platform": "atlas-01", "device": "cpu", "relative_batch_weight": 1, "response_read_wait_seconds": { "median": 0, "min": 0, "max": 0 }, "requests": { "median": 300, "min": 300, "max": 300 }, "coordinator_exchange_wall_seconds": { "median": 0, "min": 0, "max": 0 }, "inner_threads": null, "owned_host_tensor_peak_bytes": { "median": 16495216, "min": 16495216, "max": 16495216 }, "request_write_seconds": { "median": 0, "min": 0, "max": 0 }, "request_payload_bytes": { "median": 0, "min": 0, "max": 0 }, "process_peak_rss_bytes": { "median": 94224384, "min": 94191616, "max": 112492544 }, "samples": { "median": 458760, "min": 458760, "max": 458760 }, "upload_preparation_seconds": { "median": 0.107875924, "min": 0.10784828699999997, "max": 0.10974542300000004 }, "worker_total_seconds": { "median": 2.3510604120000003, "min": 2.3420942970000005, "max": 2.3609744629999994 }, "zero_sample_steps": { "median": 0, "min": 0, "max": 0 }, "response_payload_bytes": { "median": 0, "min": 0, "max": 0 }, "compute_seconds": { "median": 2.2227794629999997, "min": 2.213592166000002, "max": 2.2300271619999994 }, "owned_device_tensor_peak_bytes": { "median": 0, "min": 0, "max": 0 }, "readback_seconds": { "median": 0.009549411, "min": 0.009374091000000001, "max": 0.009962339000000002 }, "backend": "apple_accelerate", "target": "macos-arm64" } ], "fastest_selected_baseline": null }, { "case": "atlas02-cpu32", "label": "Atlas-02 CPU \u00b7 Float32", "dtype": "Float32", "kind": "standalone", "replicas": [ { "configuration_id": "atlas-02/C03", "platform": "atlas-02", "device": "cpu", "relative_batch_weight": 1 } ], "coordinator_configuration_id": "atlas-02/C03", "numerical_reference_case": "atlas01-cpu32", "repetitions": 3, "outer_command_wall_seconds": { "median": 22.104401999997208, "min": 20.88762570900144, "max": 22.852109415995074 }, "fit_seconds": { "median": 19.335128503, "min": 17.864245818, "max": 20.186910193 }, "command_seconds": { "median": 20.064694395, "min": 18.971624858, "max": 21.367686295 }, "quality": { "initial_loss": { "median": 1.599133218017636, "min": 1.599133218017636, "max": 1.599133218017636 }, "training_loss": { "median": 0.29234654818284644, "min": 0.29234654818284644, "max": 0.29234654818284644 }, "validation_loss": { "median": 0.3033732850214472, "min": 0.3033732850214472, "max": 0.3033732850214472 }, "test_loss": { "median": 0.29705295057650866, "min": 0.29705295057650866, "max": 0.29705295057650866 }, "training_accuracy": { "median": 0.8983564390966955, "min": 0.8983564390966955, "max": 0.8983564390966955 }, "validation_accuracy": { "median": 0.8857578840284842, "min": 0.8857578840284842, "max": 0.8857578840284842 }, "test_accuracy": { "median": 0.8955239064089522, "min": 0.8955239064089522, "max": 0.8955239064089522 } }, "parameter_max_abs_error": 4.6350061893463135e-05, "loss_max_abs_errors": { "initial_loss": 5.300964112109341e-10, "training_loss": 9.881305719772193e-07, "validation_loss": 1.1409408320894876e-06, "test_loss": 9.786561206981403e-07 }, "accuracy_max_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.00010172939979657958, "test_accuracy": 0.0 }, "coordinator_memory": { "process_peak_rss_bytes": { "median": 177963008, "min": 177876992, "max": 177995776 }, "tensor_owned_host_peak_bytes": { "median": 21086868, "min": 21086868, "max": 21086868 }, "tensor_owned_device_peak_bytes": { "median": 0, "min": 0, "max": 0 } }, "replica_metrics": [ { "configuration_id": "atlas-02/C03", "platform": "atlas-02", "device": "cpu", "relative_batch_weight": 1, "response_read_wait_seconds": { "median": 0, "min": 0, "max": 0 }, "requests": { "median": 300, "min": 300, "max": 300 }, "coordinator_exchange_wall_seconds": { "median": 0, "min": 0, "max": 0 }, "inner_threads": { "median": 1, "min": 1, "max": 1 }, "owned_host_tensor_peak_bytes": { "median": 21086868, "min": 21086868, "max": 21086868 }, "request_write_seconds": { "median": 0, "min": 0, "max": 0 }, "request_payload_bytes": { "median": 0, "min": 0, "max": 0 }, "process_peak_rss_bytes": { "median": 177963008, "min": 177876992, "max": 177995776 }, "samples": { "median": 458760, "min": 458760, "max": 458760 }, "upload_preparation_seconds": { "median": 1.0965687200000003, "min": 1.0238003210000002, "max": 1.15576639 }, "worker_total_seconds": { "median": 14.506156379000004, "min": 13.444564214000014, "max": 15.424223340000001 }, "zero_sample_steps": { "median": 0, "min": 0, "max": 0 }, "response_payload_bytes": { "median": 0, "min": 0, "max": 0 }, "compute_seconds": { "median": 12.980122186000001, "min": 12.017443807999992, "max": 13.792235244999995 }, "owned_device_tensor_peak_bytes": { "median": 0, "min": 0, "max": 0 }, "readback_seconds": { "median": 0.064397052, "min": 0.060066732000000025, "max": 0.07318645099999999 }, "backend": "openblas", "target": "linux-x86_64" } ], "fastest_selected_baseline": null }, { "case": "atlas02-cpu64", "label": "Atlas-02 CPU \u00b7 Float64", "dtype": "Float64", "kind": "standalone", "replicas": [ { "configuration_id": "atlas-02/C03", "platform": "atlas-02", "device": "cpu", "relative_batch_weight": 1 } ], "coordinator_configuration_id": "atlas-02/C03", "numerical_reference_case": "atlas01-cpu64", "repetitions": 3, "outer_command_wall_seconds": { "median": 20.65155120800773, "min": 19.701527957993676, "max": 20.810743416994228 }, "fit_seconds": { "median": 18.017975232, "min": 16.659829462, "max": 18.071835351 }, "command_seconds": { "median": 18.795863525, "min": 17.671156699, "max": 19.072891984 }, "quality": { "initial_loss": { "median": 1.5991332128378295, "min": 1.5991332128378295, "max": 1.5991332128378295 }, "training_loss": { "median": 0.29234691523654444, "min": 0.29234691523654444, "max": 0.29234691523654444 }, "validation_loss": { "median": 0.30337316499510086, "min": 0.30337316499510086, "max": 0.30337316499510086 }, "test_loss": { "median": 0.29705317871020637, "min": 0.29705317871020637, "max": 0.29705317871020637 }, "training_accuracy": { "median": 0.8983564390966955, "min": 0.8983564390966955, "max": 0.8983564390966955 }, "validation_accuracy": { "median": 0.8857578840284842, "min": 0.8857578840284842, "max": 0.8857578840284842 }, "test_accuracy": { "median": 0.8955239064089522, "min": 0.8955239064089522, "max": 0.8955239064089522 } }, "parameter_max_abs_error": 3.3306690738754696e-16, "loss_max_abs_errors": { "initial_loss": 0.0, "training_loss": 0.0, "validation_loss": 0.0, "test_loss": 5.551115123125783e-17 }, "accuracy_max_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "coordinator_memory": { "process_peak_rss_bytes": { "median": 172453888, "min": 172425216, "max": 172531712 }, "tensor_owned_host_peak_bytes": { "median": 16495216, "min": 16495216, "max": 16495216 }, "tensor_owned_device_peak_bytes": { "median": 0, "min": 0, "max": 0 } }, "replica_metrics": [ { "configuration_id": "atlas-02/C03", "platform": "atlas-02", "device": "cpu", "relative_batch_weight": 1, "response_read_wait_seconds": { "median": 0, "min": 0, "max": 0 }, "requests": { "median": 300, "min": 300, "max": 300 }, "coordinator_exchange_wall_seconds": { "median": 0, "min": 0, "max": 0 }, "inner_threads": { "median": 1, "min": 1, "max": 1 }, "owned_host_tensor_peak_bytes": { "median": 16495216, "min": 16495216, "max": 16495216 }, "request_write_seconds": { "median": 0, "min": 0, "max": 0 }, "request_payload_bytes": { "median": 0, "min": 0, "max": 0 }, "process_peak_rss_bytes": { "median": 172453888, "min": 172425216, "max": 172531712 }, "samples": { "median": 458760, "min": 458760, "max": 458760 }, "upload_preparation_seconds": { "median": 0.5774402320000003, "min": 0.5563793830000007, "max": 0.6052092589999998 }, "worker_total_seconds": { "median": 13.412996769999998, "min": 12.879265018999998, "max": 14.06670137800001 }, "zero_sample_steps": { "median": 0, "min": 0, "max": 0 }, "response_payload_bytes": { "median": 0, "min": 0, "max": 0 }, "compute_seconds": { "median": 12.507452861000004, "min": 12.006613135999997, "max": 13.120414492000009 }, "owned_device_tensor_peak_bytes": { "median": 0, "min": 0, "max": 0 }, "readback_seconds": { "median": 0.050057721999999985, "min": 0.048369314, "max": 0.05226979300000003 }, "backend": "openblas", "target": "linux-x86_64" } ], "fastest_selected_baseline": null }, { "case": "atlas03-cpu32", "label": "Atlas-03 CPU \u00b7 Float32", "dtype": "Float32", "kind": "standalone", "replicas": [ { "configuration_id": "atlas-03/C03", "platform": "atlas-03", "device": "cpu", "relative_batch_weight": 1 } ], "coordinator_configuration_id": "atlas-03/C03", "numerical_reference_case": "atlas01-cpu32", "repetitions": 3, "outer_command_wall_seconds": { "median": 14.828693917006603, "min": 14.590928958001314, "max": 15.160592833010014 }, "fit_seconds": { "median": 12.708668223, "min": 12.587794052, "max": 12.860272355 }, "command_seconds": { "median": 13.231820473, "min": 13.101270396, "max": 13.380385378 }, "quality": { "initial_loss": { "median": 1.5991332140263232, "min": 1.5991332140263232, "max": 1.5991332140263232 }, "training_loss": { "median": 0.2923465511763311, "min": 0.2923465511763311, "max": 0.2923465511763311 }, "validation_loss": { "median": 0.3033732850214472, "min": 0.3033732850214472, "max": 0.3033732850214472 }, "test_loss": { "median": 0.29705295057650866, "min": 0.29705295057650866, "max": 0.29705295057650866 }, "training_accuracy": { "median": 0.8983564390966955, "min": 0.8983564390966955, "max": 0.8983564390966955 }, "validation_accuracy": { "median": 0.8857578840284842, "min": 0.8857578840284842, "max": 0.8857578840284842 }, "test_accuracy": { "median": 0.8955239064089522, "min": 0.8955239064089522, "max": 0.8955239064089522 } }, "parameter_max_abs_error": 4.635751247406006e-05, "loss_max_abs_errors": { "initial_loss": 3.461216424582858e-09, "training_loss": 9.851370872948628e-07, "validation_loss": 1.1409408320894876e-06, "test_loss": 9.786561206981403e-07 }, "accuracy_max_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.00010172939979657958, "test_accuracy": 0.0 }, "coordinator_memory": { "process_peak_rss_bytes": { "median": 177893376, "min": 177852416, "max": 177975296 }, "tensor_owned_host_peak_bytes": { "median": 21086868, "min": 21086868, "max": 21086868 }, "tensor_owned_device_peak_bytes": { "median": 0, "min": 0, "max": 0 } }, "replica_metrics": [ { "configuration_id": "atlas-03/C03", "platform": "atlas-03", "device": "cpu", "relative_batch_weight": 1, "response_read_wait_seconds": { "median": 0, "min": 0, "max": 0 }, "requests": { "median": 300, "min": 300, "max": 300 }, "coordinator_exchange_wall_seconds": { "median": 0, "min": 0, "max": 0 }, "inner_threads": { "median": 1, "min": 1, "max": 1 }, "owned_host_tensor_peak_bytes": { "median": 21086868, "min": 21086868, "max": 21086868 }, "request_write_seconds": { "median": 0, "min": 0, "max": 0 }, "request_payload_bytes": { "median": 0, "min": 0, "max": 0 }, "process_peak_rss_bytes": { "median": 177893376, "min": 177852416, "max": 177975296 }, "samples": { "median": 458760, "min": 458760, "max": 458760 }, "upload_preparation_seconds": { "median": 0.7007955530000008, "min": 0.6872004380000003, "max": 0.7012716800000004 }, "worker_total_seconds": { "median": 9.686538782000001, "min": 9.584013947999999, "max": 9.768997478999998 }, "zero_sample_steps": { "median": 0, "min": 0, "max": 0 }, "response_payload_bytes": { "median": 0, "min": 0, "max": 0 }, "compute_seconds": { "median": 8.630125725999997, "min": 8.543419429, "max": 8.709238542999996 }, "owned_device_tensor_peak_bytes": { "median": 0, "min": 0, "max": 0 }, "readback_seconds": { "median": 0.045808335000000054, "min": 0.045342648, "max": 0.04580889400000001 }, "backend": "openblas", "target": "linux-x86_64" } ], "fastest_selected_baseline": null }, { "case": "atlas03-cpu64", "label": "Atlas-03 CPU \u00b7 Float64", "dtype": "Float64", "kind": "standalone", "replicas": [ { "configuration_id": "atlas-03/C03", "platform": "atlas-03", "device": "cpu", "relative_batch_weight": 1 } ], "coordinator_configuration_id": "atlas-03/C03", "numerical_reference_case": "atlas01-cpu64", "repetitions": 3, "outer_command_wall_seconds": { "median": 13.518542040998, "min": 13.406571915998938, "max": 13.942002208001213 }, "fit_seconds": { "median": 11.45661501, "min": 11.436723592, "max": 11.479060799 }, "command_seconds": { "median": 11.956218479, "min": 11.936004619, "max": 11.984865174 }, "quality": { "initial_loss": { "median": 1.5991332128378295, "min": 1.5991332128378295, "max": 1.5991332128378295 }, "training_loss": { "median": 0.29234691523654444, "min": 0.29234691523654444, "max": 0.29234691523654444 }, "validation_loss": { "median": 0.3033731649951008, "min": 0.3033731649951008, "max": 0.3033731649951008 }, "test_loss": { "median": 0.29705317871020637, "min": 0.29705317871020637, "max": 0.29705317871020637 }, "training_accuracy": { "median": 0.8983564390966955, "min": 0.8983564390966955, "max": 0.8983564390966955 }, "validation_accuracy": { "median": 0.8857578840284842, "min": 0.8857578840284842, "max": 0.8857578840284842 }, "test_accuracy": { "median": 0.8955239064089522, "min": 0.8955239064089522, "max": 0.8955239064089522 } }, "parameter_max_abs_error": 2.220446049250313e-16, "loss_max_abs_errors": { "initial_loss": 0.0, "training_loss": 0.0, "validation_loss": 5.551115123125783e-17, "test_loss": 5.551115123125783e-17 }, "accuracy_max_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "coordinator_memory": { "process_peak_rss_bytes": { "median": 172462080, "min": 172457984, "max": 172683264 }, "tensor_owned_host_peak_bytes": { "median": 16495216, "min": 16495216, "max": 16495216 }, "tensor_owned_device_peak_bytes": { "median": 0, "min": 0, "max": 0 } }, "replica_metrics": [ { "configuration_id": "atlas-03/C03", "platform": "atlas-03", "device": "cpu", "relative_batch_weight": 1, "response_read_wait_seconds": { "median": 0, "min": 0, "max": 0 }, "requests": { "median": 300, "min": 300, "max": 300 }, "coordinator_exchange_wall_seconds": { "median": 0, "min": 0, "max": 0 }, "inner_threads": { "median": 1, "min": 1, "max": 1 }, "owned_host_tensor_peak_bytes": { "median": 16495216, "min": 16495216, "max": 16495216 }, "request_write_seconds": { "median": 0, "min": 0, "max": 0 }, "request_payload_bytes": { "median": 0, "min": 0, "max": 0 }, "process_peak_rss_bytes": { "median": 172462080, "min": 172457984, "max": 172683264 }, "samples": { "median": 458760, "min": 458760, "max": 458760 }, "upload_preparation_seconds": { "median": 0.3564620449999999, "min": 0.35629507399999993, "max": 0.3585980489999996 }, "worker_total_seconds": { "median": 8.894932573000004, "min": 8.886005835000004, "max": 8.897904400999998 }, "zero_sample_steps": { "median": 0, "min": 0, "max": 0 }, "response_payload_bytes": { "median": 0, "min": 0, "max": 0 }, "compute_seconds": { "median": 8.261263486, "min": 8.256238315999996, "max": 8.265567833999997 }, "owned_device_tensor_peak_bytes": { "median": 0, "min": 0, "max": 0 }, "readback_seconds": { "median": 0.032498655, "min": 0.03217572499999999, "max": 0.034002631000000005 }, "backend": "openblas", "target": "linux-x86_64" } ], "fastest_selected_baseline": null }, { "case": "atlas01-gpu32", "label": "Atlas-01 MPS \u00b7 Float32", "dtype": "Float32", "kind": "standalone", "replicas": [ { "configuration_id": "atlas-01/C03", "platform": "atlas-01", "device": "mps", "relative_batch_weight": 1 } ], "coordinator_configuration_id": "atlas-01/C03", "numerical_reference_case": "atlas01-cpu32", "repetitions": 3, "outer_command_wall_seconds": { "median": 3.2152430829883087, "min": 3.0367226250091335, "max": 3.255333624998457 }, "fit_seconds": { "median": 2.804163125, "min": 2.632902875, "max": 2.816833625 }, "command_seconds": { "median": 3.161081833, "min": 2.984413792, "max": 3.189324208 }, "quality": { "initial_loss": { "median": 1.5991332174875397, "min": 1.5991332174875397, "max": 1.5991332174875397 }, "training_loss": { "median": 0.29234693766326264, "min": 0.29234693766326264, "max": 0.29234693766326264 }, "validation_loss": { "median": 0.3033739387867773, "min": 0.3033739387867773, "max": 0.3033739387867773 }, "test_loss": { "median": 0.2970536025106482, "min": 0.2970536025106482, "max": 0.2970536025106482 }, "training_accuracy": { "median": 0.8983564390966955, "min": 0.8983564390966955, "max": 0.8983564390966955 }, "validation_accuracy": { "median": 0.8857578840284842, "min": 0.8857578840284842, "max": 0.8857578840284842 }, "test_accuracy": { "median": 0.8955239064089522, "min": 0.8955239064089522, "max": 0.8955239064089522 } }, "parameter_max_abs_error": 4.6253204345703125e-05, "loss_max_abs_errors": { "initial_loss": 0.0, "training_loss": 5.986501557808666e-07, "validation_loss": 4.87175501995285e-07, "test_loss": 3.2672198113647966e-07 }, "accuracy_max_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.00010172939979657958, "test_accuracy": 0.0 }, "coordinator_memory": { "process_peak_rss_bytes": { "median": 85131264, "min": 84983808, "max": 87965696 }, "tensor_owned_host_peak_bytes": { "median": 6303328, "min": 6303328, "max": 6303328 }, "tensor_owned_device_peak_bytes": { "median": 6524456, "min": 6524456, "max": 6524456 } }, "replica_metrics": [ { "configuration_id": "atlas-01/C03", "platform": "atlas-01", "device": "mps", "relative_batch_weight": 1, "response_read_wait_seconds": { "median": 0, "min": 0, "max": 0 }, "requests": { "median": 300, "min": 300, "max": 300 }, "coordinator_exchange_wall_seconds": { "median": 0, "min": 0, "max": 0 }, "inner_threads": null, "owned_host_tensor_peak_bytes": { "median": 6303328, "min": 6303328, "max": 6303328 }, "request_write_seconds": { "median": 0, "min": 0, "max": 0 }, "request_payload_bytes": { "median": 0, "min": 0, "max": 0 }, "process_peak_rss_bytes": { "median": 85131264, "min": 84983808, "max": 87965696 }, "samples": { "median": 458760, "min": 458760, "max": 458760 }, "upload_preparation_seconds": { "median": 0.24405004799999996, "min": 0.23252137699999997, "max": 0.250555169 }, "worker_total_seconds": { "median": 1.9918613789999977, "min": 1.858832952, "max": 2.0093898709999993 }, "zero_sample_steps": { "median": 0, "min": 0, "max": 0 }, "response_payload_bytes": { "median": 0, "min": 0, "max": 0 }, "compute_seconds": { "median": 1.7054683319999997, "min": 1.5886115010000001, "max": 1.7175838759999995 }, "owned_device_tensor_peak_bytes": { "median": 6524456, "min": 6524456, "max": 6524456 }, "readback_seconds": { "median": 0.016662950999999995, "min": 0.015358756000000006, "max": 0.017307832999999998 }, "backend": "metal_mps", "target": "macos-arm64" } ], "fastest_selected_baseline": null }, { "case": "atlas02-gpu32", "label": "Atlas-02 CUDA \u00b7 Float32", "dtype": "Float32", "kind": "standalone", "replicas": [ { "configuration_id": "atlas-02/C03", "platform": "atlas-02", "device": "cuda", "relative_batch_weight": 1 } ], "coordinator_configuration_id": "atlas-02/C03", "numerical_reference_case": "atlas01-cpu32", "repetitions": 3, "outer_command_wall_seconds": { "median": 8.69213504200161, "min": 8.515092125002411, "max": 8.770283082994865 }, "fit_seconds": { "median": 5.944281978, "min": 5.904053987, "max": 6.029027588 }, "command_seconds": { "median": 7.097920784, "min": 7.032981553, "max": 7.155840428 }, "quality": { "initial_loss": { "median": 1.599133218017636, "min": 1.599133218017636, "max": 1.599133218017636 }, "training_loss": { "median": 0.2923495870296973, "min": 0.2923495870296973, "max": 0.2923495870296973 }, "validation_loss": { "median": 0.30337606328913785, "min": 0.30337606328913785, "max": 0.30337606328913785 }, "test_loss": { "median": 0.29705589171101093, "min": 0.29705589171101093, "max": 0.29705589171101093 }, "training_accuracy": { "median": 0.8983564390966955, "min": 0.8983564390966955, "max": 0.8983564390966955 }, "validation_accuracy": { "median": 0.8857578840284842, "min": 0.8857578840284842, "max": 0.8857578840284842 }, "test_accuracy": { "median": 0.8955239064089522, "min": 0.8955239064089522, "max": 0.8955239064089522 } }, "parameter_max_abs_error": 5.9798359870910645e-05, "loss_max_abs_errors": { "initial_loss": 5.300964112109341e-10, "training_loss": 2.050716278900744e-06, "validation_loss": 1.6373268585390122e-06, "test_loss": 1.9624783815785918e-06 }, "accuracy_max_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.00010172939979657958, "test_accuracy": 0.0 }, "coordinator_memory": { "process_peak_rss_bytes": { "median": 296333312, "min": 296271872, "max": 296456192 }, "tensor_owned_host_peak_bytes": { "median": 6303328, "min": 6303328, "max": 6303328 }, "tensor_owned_device_peak_bytes": { "median": 6524456, "min": 6524456, "max": 6524456 } }, "replica_metrics": [ { "configuration_id": "atlas-02/C03", "platform": "atlas-02", "device": "cuda", "relative_batch_weight": 1, "response_read_wait_seconds": { "median": 0, "min": 0, "max": 0 }, "requests": { "median": 300, "min": 300, "max": 300 }, "coordinator_exchange_wall_seconds": { "median": 0, "min": 0, "max": 0 }, "inner_threads": { "median": 1, "min": 1, "max": 1 }, "owned_host_tensor_peak_bytes": { "median": 6303328, "min": 6303328, "max": 6303328 }, "request_write_seconds": { "median": 0, "min": 0, "max": 0 }, "request_payload_bytes": { "median": 0, "min": 0, "max": 0 }, "process_peak_rss_bytes": { "median": 296333312, "min": 296271872, "max": 296456192 }, "samples": { "median": 458760, "min": 458760, "max": 458760 }, "upload_preparation_seconds": { "median": 0.9739203370000002, "min": 0.9662449480000002, "max": 0.9876225599999999 }, "worker_total_seconds": { "median": 1.6844404309999996, "min": 1.6604763869999994, "max": 1.703028964 }, "zero_sample_steps": { "median": 0, "min": 0, "max": 0 }, "response_payload_bytes": { "median": 0, "min": 0, "max": 0 }, "compute_seconds": { "median": 0.5040637470000001, "min": 0.48920735600000026, "max": 0.5048317030000001 }, "owned_device_tensor_peak_bytes": { "median": 6524456, "min": 6524456, "max": 6524456 }, "readback_seconds": { "median": 0.06745282600000004, "min": 0.06735880900000002, "max": 0.07029516799999998 }, "backend": "cuda_cublas", "target": "linux-x86_64" } ], "fastest_selected_baseline": null }, { "case": "atlas02-gpu64", "label": "Atlas-02 CUDA \u00b7 Float64", "dtype": "Float64", "kind": "standalone", "replicas": [ { "configuration_id": "atlas-02/C03", "platform": "atlas-02", "device": "cuda", "relative_batch_weight": 1 } ], "coordinator_configuration_id": "atlas-02/C03", "numerical_reference_case": "atlas01-cpu64", "repetitions": 3, "outer_command_wall_seconds": { "median": 9.508705583997653, "min": 8.796772208006587, "max": 9.726846584002487 }, "fit_seconds": { "median": 5.958052482, "min": 5.843198569, "max": 6.5129025 }, "command_seconds": { "median": 7.217335922, "min": 7.141771782, "max": 7.993388733 }, "quality": { "initial_loss": { "median": 1.5991332128378295, "min": 1.5991332128378295, "max": 1.5991332128378295 }, "training_loss": { "median": 0.29234691523654444, "min": 0.29234691523654444, "max": 0.29234691523654444 }, "validation_loss": { "median": 0.3033731649951009, "min": 0.3033731649951009, "max": 0.3033731649951009 }, "test_loss": { "median": 0.29705317871020637, "min": 0.29705317871020637, "max": 0.29705317871020637 }, "training_accuracy": { "median": 0.8983564390966955, "min": 0.8983564390966955, "max": 0.8983564390966955 }, "validation_accuracy": { "median": 0.8857578840284842, "min": 0.8857578840284842, "max": 0.8857578840284842 }, "test_accuracy": { "median": 0.8955239064089522, "min": 0.8955239064089522, "max": 0.8955239064089522 } }, "parameter_max_abs_error": 3.3306690738754696e-16, "loss_max_abs_errors": { "initial_loss": 0.0, "training_loss": 0.0, "validation_loss": 5.551115123125783e-17, "test_loss": 5.551115123125783e-17 }, "accuracy_max_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "coordinator_memory": { "process_peak_rss_bytes": { "median": 300347392, "min": 300224512, "max": 300429312 }, "tensor_owned_host_peak_bytes": { "median": 4202560, "min": 4202560, "max": 4202560 }, "tensor_owned_device_peak_bytes": { "median": 13048912, "min": 13048912, "max": 13048912 } }, "replica_metrics": [ { "configuration_id": "atlas-02/C03", "platform": "atlas-02", "device": "cuda", "relative_batch_weight": 1, "response_read_wait_seconds": { "median": 0, "min": 0, "max": 0 }, "requests": { "median": 300, "min": 300, "max": 300 }, "coordinator_exchange_wall_seconds": { "median": 0, "min": 0, "max": 0 }, "inner_threads": { "median": 1, "min": 1, "max": 1 }, "owned_host_tensor_peak_bytes": { "median": 4202560, "min": 4202560, "max": 4202560 }, "request_write_seconds": { "median": 0, "min": 0, "max": 0 }, "request_payload_bytes": { "median": 0, "min": 0, "max": 0 }, "process_peak_rss_bytes": { "median": 300347392, "min": 300224512, "max": 300429312 }, "samples": { "median": 458760, "min": 458760, "max": 458760 }, "upload_preparation_seconds": { "median": 0.6689899059999999, "min": 0.6621139690000001, "max": 0.8554335600000007 }, "worker_total_seconds": { "median": 2.260629614, "min": 2.258230271999999, "max": 2.575395199 }, "zero_sample_steps": { "median": 0, "min": 0, "max": 0 }, "response_payload_bytes": { "median": 0, "min": 0, "max": 0 }, "compute_seconds": { "median": 1.2547439029999992, "min": 1.2478117600000003, "max": 1.3179493970000007 }, "owned_device_tensor_peak_bytes": { "median": 13048912, "min": 13048912, "max": 13048912 }, "readback_seconds": { "median": 0.07557949400000004, "min": 0.075353733, "max": 0.11018652099999998 }, "backend": "cuda_cublas", "target": "linux-x86_64" } ], "fastest_selected_baseline": null }, { "case": "atlas03-gpu32", "label": "Atlas-03 CUDA \u00b7 Float32", "dtype": "Float32", "kind": "standalone", "replicas": [ { "configuration_id": "atlas-03/C03", "platform": "atlas-03", "device": "cuda", "relative_batch_weight": 1 } ], "coordinator_configuration_id": "atlas-03/C03", "numerical_reference_case": "atlas01-cpu32", "repetitions": 3, "outer_command_wall_seconds": { "median": 6.113490791001823, "min": 6.021835834006197, "max": 6.598680917013553 }, "fit_seconds": { "median": 4.041986648, "min": 4.025347966, "max": 4.045879262 }, "command_seconds": { "median": 4.649221081, "min": 4.631889841, "max": 4.670013525 }, "quality": { "initial_loss": { "median": 1.5991332140263232, "min": 1.5991332140263232, "max": 1.5991332140263232 }, "training_loss": { "median": 0.2923475264676603, "min": 0.2923475264676603, "max": 0.2923475264676603 }, "validation_loss": { "median": 0.30337433190457336, "min": 0.30337433190457336, "max": 0.30337433190457336 }, "test_loss": { "median": 0.2970538742908505, "min": 0.2970538742908505, "max": 0.2970538742908505 }, "training_accuracy": { "median": 0.8983564390966955, "min": 0.8983564390966955, "max": 0.8983564390966955 }, "validation_accuracy": { "median": 0.8858596134282808, "min": 0.8858596134282808, "max": 0.8858596134282808 }, "test_accuracy": { "median": 0.8955239064089522, "min": 0.8955239064089522, "max": 0.8955239064089522 } }, "parameter_max_abs_error": 4.561617970466614e-06, "loss_max_abs_errors": { "initial_loss": 3.461216424582858e-09, "training_loss": 9.84575809770405e-09, "validation_loss": 9.405770595716234e-08, "test_loss": 5.494177884868279e-08 }, "accuracy_max_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "coordinator_memory": { "process_peak_rss_bytes": { "median": 331112448, "min": 331001856, "max": 331165696 }, "tensor_owned_host_peak_bytes": { "median": 6303328, "min": 6303328, "max": 6303328 }, "tensor_owned_device_peak_bytes": { "median": 6524456, "min": 6524456, "max": 6524456 } }, "replica_metrics": [ { "configuration_id": "atlas-03/C03", "platform": "atlas-03", "device": "cuda", "relative_batch_weight": 1, "response_read_wait_seconds": { "median": 0, "min": 0, "max": 0 }, "requests": { "median": 300, "min": 300, "max": 300 }, "coordinator_exchange_wall_seconds": { "median": 0, "min": 0, "max": 0 }, "inner_threads": { "median": 1, "min": 1, "max": 1 }, "owned_host_tensor_peak_bytes": { "median": 6303328, "min": 6303328, "max": 6303328 }, "request_write_seconds": { "median": 0, "min": 0, "max": 0 }, "request_payload_bytes": { "median": 0, "min": 0, "max": 0 }, "process_peak_rss_bytes": { "median": 331112448, "min": 331001856, "max": 331165696 }, "samples": { "median": 458760, "min": 458760, "max": 458760 }, "upload_preparation_seconds": { "median": 0.6526943540000001, "min": 0.6446049009999999, "max": 0.6535847469999994 }, "worker_total_seconds": { "median": 1.0821805899999999, "min": 1.0742639050000002, "max": 1.086779123 }, "zero_sample_steps": { "median": 0, "min": 0, "max": 0 }, "response_payload_bytes": { "median": 0, "min": 0, "max": 0 }, "compute_seconds": { "median": 0.2802974630000002, "min": 0.2801538039999999, "max": 0.28371388599999997 }, "owned_device_tensor_peak_bytes": { "median": 6524456, "min": 6524456, "max": 6524456 }, "readback_seconds": { "median": 0.05001958200000001, "min": 0.049093760000000014, "max": 0.051010612000000004 }, "backend": "cuda_cublas", "target": "linux-x86_64" } ], "fastest_selected_baseline": null }, { "case": "atlas03-gpu64", "label": "Atlas-03 CUDA \u00b7 Float64", "dtype": "Float64", "kind": "standalone", "replicas": [ { "configuration_id": "atlas-03/C03", "platform": "atlas-03", "device": "cuda", "relative_batch_weight": 1 } ], "coordinator_configuration_id": "atlas-03/C03", "numerical_reference_case": "atlas01-cpu64", "repetitions": 3, "outer_command_wall_seconds": { "median": 5.359644708005362, "min": 5.325457874991116, "max": 5.440311792001012 }, "fit_seconds": { "median": 3.380031545, "min": 3.351845387, "max": 3.423176061 }, "command_seconds": { "median": 3.955525727, "min": 3.920242845, "max": 4.007730675 }, "quality": { "initial_loss": { "median": 1.5991332128378295, "min": 1.5991332128378295, "max": 1.5991332128378295 }, "training_loss": { "median": 0.29234691523654444, "min": 0.29234691523654444, "max": 0.29234691523654444 }, "validation_loss": { "median": 0.3033731649951008, "min": 0.3033731649951008, "max": 0.3033731649951008 }, "test_loss": { "median": 0.29705317871020637, "min": 0.29705317871020637, "max": 0.29705317871020637 }, "training_accuracy": { "median": 0.8983564390966955, "min": 0.8983564390966955, "max": 0.8983564390966955 }, "validation_accuracy": { "median": 0.8857578840284842, "min": 0.8857578840284842, "max": 0.8857578840284842 }, "test_accuracy": { "median": 0.8955239064089522, "min": 0.8955239064089522, "max": 0.8955239064089522 } }, "parameter_max_abs_error": 3.3306690738754696e-16, "loss_max_abs_errors": { "initial_loss": 0.0, "training_loss": 0.0, "validation_loss": 5.551115123125783e-17, "test_loss": 5.551115123125783e-17 }, "accuracy_max_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "coordinator_memory": { "process_peak_rss_bytes": { "median": 324956160, "min": 324931584, "max": 325730304 }, "tensor_owned_host_peak_bytes": { "median": 4202560, "min": 4202560, "max": 4202560 }, "tensor_owned_device_peak_bytes": { "median": 13048912, "min": 13048912, "max": 13048912 } }, "replica_metrics": [ { "configuration_id": "atlas-03/C03", "platform": "atlas-03", "device": "cuda", "relative_batch_weight": 1, "response_read_wait_seconds": { "median": 0, "min": 0, "max": 0 }, "requests": { "median": 300, "min": 300, "max": 300 }, "coordinator_exchange_wall_seconds": { "median": 0, "min": 0, "max": 0 }, "inner_threads": { "median": 1, "min": 1, "max": 1 }, "owned_host_tensor_peak_bytes": { "median": 4202560, "min": 4202560, "max": 4202560 }, "request_write_seconds": { "median": 0, "min": 0, "max": 0 }, "request_payload_bytes": { "median": 0, "min": 0, "max": 0 }, "process_peak_rss_bytes": { "median": 324943872, "min": 324931584, "max": 324956160 }, "samples": { "median": 458760, "min": 458760, "max": 458760 }, "upload_preparation_seconds": { "median": 0.40694497400000007, "min": 0.4014066510000002, "max": 0.41219602099999997 }, "worker_total_seconds": { "median": 1.0275088589999999, "min": 1.0095150980000012, "max": 1.0351131310000008 }, "zero_sample_steps": { "median": 0, "min": 0, "max": 0 }, "response_payload_bytes": { "median": 0, "min": 0, "max": 0 }, "compute_seconds": { "median": 0.4143261709999999, "min": 0.40619233300000024, "max": 0.4147275310000002 }, "owned_device_tensor_peak_bytes": { "median": 13048912, "min": 13048912, "max": 13048912 }, "readback_seconds": { "median": 0.055599096000000035, "min": 0.05416801799999993, "max": 0.05594505700000002 }, "backend": "cuda_cublas", "target": "linux-x86_64" } ], "fastest_selected_baseline": null }, { "case": "01-02-gpu32", "label": "Atlas-01 MPS + Atlas-02 CUDA \u00b7 Float32", "dtype": "Float32", "kind": "group", "replicas": [ { "configuration_id": "atlas-01/C03", "platform": "atlas-01", "device": "mps", "relative_batch_weight": 1 }, { "configuration_id": "atlas-02/C03", "platform": "atlas-02", "device": "cuda", "relative_batch_weight": 1 } ], "coordinator_configuration_id": "atlas-01/C03", "numerical_reference_case": "atlas01-cpu32", "repetitions": 3, "outer_command_wall_seconds": { "median": 66.8675972080091, "min": 64.88779833298759, "max": 69.97771516699868 }, "fit_seconds": { "median": 66.406795041, "min": 64.414978083, "max": 69.533544166 }, "command_seconds": { "median": 66.80565725, "min": 64.804851625, "max": 69.92076175 }, "quality": { "initial_loss": { "median": 1.5991332174875397, "min": 1.5991332174875397, "max": 1.5991332174875397 }, "training_loss": { "median": 0.29234811506937414, "min": 0.29234811506937414, "max": 0.29234811506937414 }, "validation_loss": { "median": 0.30337470066511013, "min": 0.30337470066511013, "max": 0.30337470066511013 }, "test_loss": { "median": 0.29705493784872467, "min": 0.29705493784872467, "max": 0.29705493784872467 }, "training_accuracy": { "median": 0.8983564390966955, "min": 0.8983564390966955, "max": 0.8983564390966955 }, "validation_accuracy": { "median": 0.8858596134282808, "min": 0.8858596134282808, "max": 0.8858596134282808 }, "test_accuracy": { "median": 0.8955239064089522, "min": 0.8955239064089522, "max": 0.8955239064089522 } }, "parameter_max_abs_error": 3.8526952266693115e-05, "loss_max_abs_errors": { "initial_loss": 0.0, "training_loss": 5.787559557157707e-07, "validation_loss": 2.747028308158761e-07, "test_loss": 1.008616095310444e-06 }, "accuracy_max_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "coordinator_memory": { "process_peak_rss_bytes": { "median": 77430784, "min": 74383360, "max": 77496320 }, "tensor_owned_host_peak_bytes": { "median": 6303328, "min": 6303328, "max": 6303328 }, "tensor_owned_device_peak_bytes": { "median": 0, "min": 0, "max": 0 } }, "replica_metrics": [ { "configuration_id": "atlas-01/C03", "platform": "atlas-01", "device": "mps", "relative_batch_weight": 1, "response_read_wait_seconds": { "median": 5.688616882999999, "min": 5.540511544000001, "max": 5.730888381 }, "requests": { "median": 300, "min": 300, "max": 300 }, "coordinator_exchange_wall_seconds": { "median": 6.897081250000001, "min": 6.608140246, "max": 6.926093207000001 }, "inner_threads": null, "owned_host_tensor_peak_bytes": { "median": 983040, "min": 983040, "max": 983040 }, "request_write_seconds": { "median": 1.195105451000001, "min": 1.0675435869999996, "max": 1.2083657930000002 }, "request_payload_bytes": { "median": 82667760, "min": 82667760, "max": 82667760 }, "process_peak_rss_bytes": { "median": 24199168, "min": 23445504, "max": 25493504 }, "samples": { "median": 229380, "min": 229380, "max": 229380 }, "upload_preparation_seconds": { "median": 0.291751674, "min": 0.267110162, "max": 0.29392679999999993 }, "worker_total_seconds": { "median": 3.5903741669999976, "min": 3.542504004000001, "max": 3.6027743710000024 }, "zero_sample_steps": { "median": 0, "min": 0, "max": 0 }, "response_payload_bytes": { "median": 40441800, "min": 40441800, "max": 40441800 }, "compute_seconds": { "median": 3.235207667999998, "min": 3.1871155310000003, "max": 3.2784082180000005 }, "owned_device_tensor_peak_bytes": { "median": 3329576, "min": 3329576, "max": 3329576 }, "readback_seconds": { "median": 0.022639126, "min": 0.021237366000000004, "max": 0.02372707299999998 }, "backend": "metal_mps", "target": "macos-arm64" }, { "configuration_id": "atlas-02/C03", "platform": "atlas-02", "device": "cuda", "relative_batch_weight": 1, "response_read_wait_seconds": { "median": 60.40910454899999, "min": 59.03837270800001, "max": 63.661730917999975 }, "requests": { "median": 300, "min": 300, "max": 300 }, "coordinator_exchange_wall_seconds": { "median": 61.83582804299997, "min": 60.31867120499993, "max": 65.10478625400002 }, "inner_threads": { "median": 1, "min": 1, "max": 1 }, "owned_host_tensor_peak_bytes": { "median": 983040, "min": 983040, "max": 983040 }, "request_write_seconds": { "median": 1.4266434240000008, "min": 1.280212417, "max": 1.4429274139999988 }, "request_payload_bytes": { "median": 82667760, "min": 82667760, "max": 82667760 }, "process_peak_rss_bytes": { "median": 252841984, "min": 252731392, "max": 252841984 }, "samples": { "median": 229380, "min": 229380, "max": 229380 }, "upload_preparation_seconds": { "median": 1.6837423830000011, "min": 1.6343725100000008, "max": 1.695894748 }, "worker_total_seconds": { "median": 2.8938203080000013, "min": 2.8092793789999995, "max": 2.920432863 }, "zero_sample_steps": { "median": 0, "min": 0, "max": 0 }, "response_payload_bytes": { "median": 40441800, "min": 40441800, "max": 40441800 }, "compute_seconds": { "median": 0.802364632, "min": 0.7812318650000001, "max": 0.8144459129999998 }, "owned_device_tensor_peak_bytes": { "median": 3329576, "min": 3329576, "max": 3329576 }, "readback_seconds": { "median": 0.14952447300000007, "min": 0.14299656899999993, "max": 0.15217154900000004 }, "backend": "cuda_cublas", "target": "linux-x86_64" } ], "fastest_selected_baseline": { "fit_seconds": { "case": "atlas01-gpu32", "seconds": 2.804163125, "speedup": 0.042227051061095346 }, "outer_command_wall_seconds": { "case": "atlas01-gpu32", "seconds": 3.2152430829883087, "speedup": 0.048083723914685565 } } }, { "case": "01-03-gpu32", "label": "Atlas-01 MPS + Atlas-03 CUDA \u00b7 Float32", "dtype": "Float32", "kind": "group", "replicas": [ { "configuration_id": "atlas-01/C03", "platform": "atlas-01", "device": "mps", "relative_batch_weight": 1 }, { "configuration_id": "atlas-03/C03", "platform": "atlas-03", "device": "cuda", "relative_batch_weight": 1 } ], "coordinator_configuration_id": "atlas-01/C03", "numerical_reference_case": "atlas01-cpu32", "repetitions": 3, "outer_command_wall_seconds": { "median": 59.04362216699519, "min": 57.645449082992855, "max": 64.85838504199637 }, "fit_seconds": { "median": 58.597174292, "min": 57.208868792, "max": 64.432534459 }, "command_seconds": { "median": 58.984112334, "min": 57.587268958, "max": 64.820838792 }, "quality": { "initial_loss": { "median": 1.5991332174875397, "min": 1.5991332174875397, "max": 1.5991332174875397 }, "training_loss": { "median": 0.2923502071331597, "min": 0.2923502071331597, "max": 0.2923502071331597 }, "validation_loss": { "median": 0.30337603640338046, "min": 0.30337603640338046, "max": 0.30337603640338046 }, "test_loss": { "median": 0.29705585060017764, "min": 0.29705585060017764, "max": 0.29705585060017764 }, "training_accuracy": { "median": 0.8984000348766239, "min": 0.8984000348766239, "max": 0.8984000348766239 }, "validation_accuracy": { "median": 0.8858596134282808, "min": 0.8858596134282808, "max": 0.8858596134282808 }, "test_accuracy": { "median": 0.8955239064089522, "min": 0.8955239064089522, "max": 0.8955239064089522 } }, "parameter_max_abs_error": 5.525164306163788e-05, "loss_max_abs_errors": { "initial_loss": 0.0, "training_loss": 2.6708197412572687e-06, "validation_loss": 1.610441101140836e-06, "test_loss": 1.9213675482854242e-06 }, "accuracy_max_abs_errors": { "training_accuracy": 4.359577992840702e-05, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "coordinator_memory": { "process_peak_rss_bytes": { "median": 77758464, "min": 75169792, "max": 78020608 }, "tensor_owned_host_peak_bytes": { "median": 6303328, "min": 6303328, "max": 6303328 }, "tensor_owned_device_peak_bytes": { "median": 0, "min": 0, "max": 0 } }, "replica_metrics": [ { "configuration_id": "atlas-01/C03", "platform": "atlas-01", "device": "mps", "relative_batch_weight": 1, "response_read_wait_seconds": { "median": 5.850273453999998, "min": 5.816193624999996, "max": 5.858686630999997 }, "requests": { "median": 300, "min": 300, "max": 300 }, "coordinator_exchange_wall_seconds": { "median": 7.047510581000003, "min": 7.023647721999998, "max": 7.047991165999994 }, "inner_threads": null, "owned_host_tensor_peak_bytes": { "median": 983040, "min": 983040, "max": 983040 }, "request_write_seconds": { "median": 1.1976290399999998, "min": 1.188724290000001, "max": 1.2073680439999999 }, "request_payload_bytes": { "median": 82667760, "min": 82667760, "max": 82667760 }, "process_peak_rss_bytes": { "median": 24788992, "min": 24231936, "max": 25034752 }, "samples": { "median": 229380, "min": 229380, "max": 229380 }, "upload_preparation_seconds": { "median": 0.292591912, "min": 0.28917641099999997, "max": 0.2974659540000005 }, "worker_total_seconds": { "median": 3.697802285, "min": 3.6478931899999996, "max": 3.705357743999997 }, "zero_sample_steps": { "median": 0, "min": 0, "max": 0 }, "response_payload_bytes": { "median": 40441800, "min": 40441800, "max": 40441800 }, "compute_seconds": { "median": 3.3420961660000015, "min": 3.296334863, "max": 3.346444957999999 }, "owned_device_tensor_peak_bytes": { "median": 3329576, "min": 3329576, "max": 3329576 }, "readback_seconds": { "median": 0.023418092999999997, "min": 0.022949124999999994, "max": 0.023648492000000004 }, "backend": "metal_mps", "target": "macos-arm64" }, { "configuration_id": "atlas-03/C03", "platform": "atlas-03", "device": "cuda", "relative_batch_weight": 1, "response_read_wait_seconds": { "median": 53.80685994000001, "min": 52.18226125400003, "max": 59.355597614000004 }, "requests": { "median": 300, "min": 300, "max": 300 }, "coordinator_exchange_wall_seconds": { "median": 55.249875174000046, "min": 53.59055008700001, "max": 60.785204422000014 }, "inner_threads": { "median": 1, "min": 1, "max": 1 }, "owned_host_tensor_peak_bytes": { "median": 983040, "min": 983040, "max": 983040 }, "request_write_seconds": { "median": 1.4294870309999994, "min": 1.4080964589999994, "max": 1.442951583 }, "request_payload_bytes": { "median": 82667760, "min": 82667760, "max": 82667760 }, "process_peak_rss_bytes": { "median": 286941184, "min": 286707712, "max": 287035392 }, "samples": { "median": 229380, "min": 229380, "max": 229380 }, "upload_preparation_seconds": { "median": 1.0515673750000007, "min": 1.0400027480000016, "max": 1.0665977910000004 }, "worker_total_seconds": { "median": 1.6082218649999993, "min": 1.5876060520000006, "max": 1.6248352789999998 }, "zero_sample_steps": { "median": 0, "min": 0, "max": 0 }, "response_payload_bytes": { "median": 40441800, "min": 40441800, "max": 40441800 }, "compute_seconds": { "median": 0.359499999, "min": 0.352285454, "max": 0.36445149700000007 }, "owned_device_tensor_peak_bytes": { "median": 3329576, "min": 3329576, "max": 3329576 }, "readback_seconds": { "median": 0.06346413999999996, "min": 0.06323915100000001, "max": 0.06348462099999996 }, "backend": "cuda_cublas", "target": "linux-x86_64" } ], "fastest_selected_baseline": { "fit_seconds": { "case": "atlas01-gpu32", "seconds": 2.804163125, "speedup": 0.047854920631946574 }, "outer_command_wall_seconds": { "case": "atlas01-gpu32", "seconds": 3.2152430829883087, "speedup": 0.05445538340948192 } } }, { "case": "02-03-gpu32", "label": "Atlas-02 CUDA + Atlas-03 CUDA \u00b7 Float32", "dtype": "Float32", "kind": "group", "replicas": [ { "configuration_id": "atlas-02/C03", "platform": "atlas-02", "device": "cuda", "relative_batch_weight": 1 }, { "configuration_id": "atlas-03/C03", "platform": "atlas-03", "device": "cuda", "relative_batch_weight": 1 } ], "coordinator_configuration_id": "atlas-01/C03", "numerical_reference_case": "atlas01-cpu32", "repetitions": 3, "outer_command_wall_seconds": { "median": 91.10916891599481, "min": 82.52440370799741, "max": 92.16468412500399 }, "fit_seconds": { "median": 90.603120209, "min": 82.012750791, "max": 91.678021833 }, "command_seconds": { "median": 91.037551042, "min": 82.4643615, "max": 92.129275166 }, "quality": { "initial_loss": { "median": 1.5991332174875397, "min": 1.5991332174875397, "max": 1.5991332174875397 }, "training_loss": { "median": 0.2923499547917588, "min": 0.2923499547917588, "max": 0.2923499547917588 }, "validation_loss": { "median": 0.3033758939464604, "min": 0.3033758939464604, "max": 0.3033758939464604 }, "test_loss": { "median": 0.2970558105929094, "min": 0.2970558105929094, "max": 0.2970558105929094 }, "training_accuracy": { "median": 0.8983564390966955, "min": 0.8983564390966955, "max": 0.8983564390966955 }, "validation_accuracy": { "median": 0.8857578840284842, "min": 0.8857578840284842, "max": 0.8857578840284842 }, "test_accuracy": { "median": 0.8955239064089522, "min": 0.8955239064089522, "max": 0.8955239064089522 } }, "parameter_max_abs_error": 5.7095661759376526e-05, "loss_max_abs_errors": { "initial_loss": 0.0, "training_loss": 2.418478340371344e-06, "validation_loss": 1.4679841811027217e-06, "test_loss": 1.8813602800649143e-06 }, "accuracy_max_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.00010172939979657958, "test_accuracy": 0.0 }, "coordinator_memory": { "process_peak_rss_bytes": { "median": 77627392, "min": 75268096, "max": 77725696 }, "tensor_owned_host_peak_bytes": { "median": 6303328, "min": 6303328, "max": 6303328 }, "tensor_owned_device_peak_bytes": { "median": 0, "min": 0, "max": 0 } }, "replica_metrics": [ { "configuration_id": "atlas-02/C03", "platform": "atlas-02", "device": "cuda", "relative_batch_weight": 1, "response_read_wait_seconds": { "median": 76.93777121300002, "min": 67.15449578600001, "max": 79.19499566500002 }, "requests": { "median": 300, "min": 300, "max": 300 }, "coordinator_exchange_wall_seconds": { "median": 78.28922679199995, "min": 68.58777321100001, "max": 80.50158036999997 }, "inner_threads": { "median": 1, "min": 1, "max": 1 }, "owned_host_tensor_peak_bytes": { "median": 983040, "min": 983040, "max": 983040 }, "request_write_seconds": { "median": 1.3513404990000002, "min": 1.3065204980000003, "max": 1.4331425359999996 }, "request_payload_bytes": { "median": 82667760, "min": 82667760, "max": 82667760 }, "process_peak_rss_bytes": { "median": 252796928, "min": 252645376, "max": 252858368 }, "samples": { "median": 229380, "min": 229380, "max": 229380 }, "upload_preparation_seconds": { "median": 1.6798352950000002, "min": 1.6567668230000003, "max": 1.7106720770000006 }, "worker_total_seconds": { "median": 2.9094893740000005, "min": 2.8483384549999995, "max": 2.9448951880000003 }, "zero_sample_steps": { "median": 0, "min": 0, "max": 0 }, "response_payload_bytes": { "median": 40441800, "min": 40441800, "max": 40441800 }, "compute_seconds": { "median": 0.8207718829999999, "min": 0.7925328389999999, "max": 0.8241611529999998 }, "owned_device_tensor_peak_bytes": { "median": 3329576, "min": 3329576, "max": 3329576 }, "readback_seconds": { "median": 0.15175968399999998, "min": 0.14461619200000006, "max": 0.15176089000000012 }, "backend": "cuda_cublas", "target": "linux-x86_64" }, { "configuration_id": "atlas-03/C03", "platform": "atlas-03", "device": "cuda", "relative_batch_weight": 1, "response_read_wait_seconds": { "median": 72.59485933, "min": 66.43971983499999, "max": 73.17139229200005 }, "requests": { "median": 300, "min": 300, "max": 300 }, "coordinator_exchange_wall_seconds": { "median": 73.94015379599995, "min": 67.86396450300003, "max": 74.47368540199999 }, "inner_threads": { "median": 1, "min": 1, "max": 1 }, "owned_host_tensor_peak_bytes": { "median": 983040, "min": 983040, "max": 983040 }, "request_write_seconds": { "median": 1.3452226289999993, "min": 1.3022185680000002, "max": 1.424187879999999 }, "request_payload_bytes": { "median": 82667760, "min": 82667760, "max": 82667760 }, "process_peak_rss_bytes": { "median": 286883840, "min": 286822400, "max": 286904320 }, "samples": { "median": 229380, "min": 229380, "max": 229380 }, "upload_preparation_seconds": { "median": 1.0825509750000002, "min": 1.0510841480000002, "max": 1.092290365000001 }, "worker_total_seconds": { "median": 1.646046648999999, "min": 1.6180502700000008, "max": 1.6464566009999997 }, "zero_sample_steps": { "median": 0, "min": 0, "max": 0 }, "response_payload_bytes": { "median": 40441800, "min": 40441800, "max": 40441800 }, "compute_seconds": { "median": 0.3666001520000001, "min": 0.36056593099999973, "max": 0.3691265440000001 }, "owned_device_tensor_peak_bytes": { "median": 3329576, "min": 3329576, "max": 3329576 }, "readback_seconds": { "median": 0.06239568800000004, "min": 0.06162792100000003, "max": 0.06479273999999999 }, "backend": "cuda_cublas", "target": "linux-x86_64" } ], "fastest_selected_baseline": { "fit_seconds": { "case": "atlas03-gpu32", "seconds": 4.041986648, "speedup": 0.04461200275085551 }, "outer_command_wall_seconds": { "case": "atlas03-gpu32", "seconds": 6.113490791001823, "speedup": 0.06710071954051772 } } }, { "case": "01-02-03-gpu32", "label": "Atlas-01 MPS + Atlas-02 CUDA + Atlas-03 CUDA \u00b7 Float32", "dtype": "Float32", "kind": "group", "replicas": [ { "configuration_id": "atlas-01/C03", "platform": "atlas-01", "device": "mps", "relative_batch_weight": 1 }, { "configuration_id": "atlas-02/C03", "platform": "atlas-02", "device": "cuda", "relative_batch_weight": 1 }, { "configuration_id": "atlas-03/C03", "platform": "atlas-03", "device": "cuda", "relative_batch_weight": 1 } ], "coordinator_configuration_id": "atlas-01/C03", "numerical_reference_case": "atlas01-cpu32", "repetitions": 3, "outer_command_wall_seconds": { "median": 86.09376583401172, "min": 86.07671895899693, "max": 86.79663883299509 }, "fit_seconds": { "median": 85.590917916, "min": 85.56780825, "max": 86.292820584 }, "command_seconds": { "median": 86.053667791, "min": 86.037888583, "max": 86.745891042 }, "quality": { "initial_loss": { "median": 1.5991332174875397, "min": 1.5991332174875397, "max": 1.5991332174875397 }, "training_loss": { "median": 0.2923465431937054, "min": 0.2923465431937054, "max": 0.2923465431937054 }, "validation_loss": { "median": 0.30337328036464484, "min": 0.30337328036464484, "max": 0.30337328036464484 }, "test_loss": { "median": 0.29705295057650866, "min": 0.29705295057650866, "max": 0.29705295057650866 }, "training_accuracy": { "median": 0.8983564390966955, "min": 0.8983564390966955, "max": 0.8983564390966955 }, "validation_accuracy": { "median": 0.8857578840284842, "min": 0.8857578840284842, "max": 0.8857578840284842 }, "test_accuracy": { "median": 0.8955239064089522, "min": 0.8955239064089522, "max": 0.8955239064089522 } }, "parameter_max_abs_error": 4.633516073226929e-05, "loss_max_abs_errors": { "initial_loss": 0.0, "training_loss": 9.931197130219616e-07, "validation_loss": 1.1455976344776353e-06, "test_loss": 9.786561206981403e-07 }, "accuracy_max_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.00010172939979657958, "test_accuracy": 0.0 }, "coordinator_memory": { "process_peak_rss_bytes": { "median": 77922304, "min": 76365824, "max": 77938688 }, "tensor_owned_host_peak_bytes": { "median": 6303328, "min": 6303328, "max": 6303328 }, "tensor_owned_device_peak_bytes": { "median": 0, "min": 0, "max": 0 } }, "replica_metrics": [ { "configuration_id": "atlas-01/C03", "platform": "atlas-01", "device": "mps", "relative_batch_weight": 1, "response_read_wait_seconds": { "median": 5.396259586000003, "min": 5.224934279000005, "max": 5.417965785999997 }, "requests": { "median": 300, "min": 300, "max": 300 }, "coordinator_exchange_wall_seconds": { "median": 6.3356772550000064, "min": 6.0719885110000025, "max": 6.368811862000002 }, "inner_threads": null, "owned_host_tensor_peak_bytes": { "median": 655360, "min": 655360, "max": 655360 }, "request_write_seconds": { "median": 0.9393523770000002, "min": 0.8469903709999993, "max": 0.9507761470000002 }, "request_payload_bytes": { "median": 61870640, "min": 61870640, "max": 61870640 }, "process_peak_rss_bytes": { "median": 21807104, "min": 21774336, "max": 22003712 }, "samples": { "median": 152920, "min": 152920, "max": 152920 }, "upload_preparation_seconds": { "median": 0.24906942500000023, "min": 0.2199126280000001, "max": 0.2503211539999999 }, "worker_total_seconds": { "median": 3.480475675999999, "min": 3.4586895430000006, "max": 3.4963926999999972 }, "zero_sample_steps": { "median": 0, "min": 0, "max": 0 }, "response_payload_bytes": { "median": 40441800, "min": 40441800, "max": 40441800 }, "compute_seconds": { "median": 3.1651296360000023, "min": 3.146254414000001, "max": 3.215899031000002 }, "owned_device_tensor_peak_bytes": { "median": 2264616, "min": 2264616, "max": 2264616 }, "readback_seconds": { "median": 0.023416451999999997, "min": 0.023044838999999998, "max": 0.024552329999999994 }, "backend": "metal_mps", "target": "macos-arm64" }, { "configuration_id": "atlas-02/C03", "platform": "atlas-02", "device": "cuda", "relative_batch_weight": 1, "response_read_wait_seconds": { "median": 74.27892257999996, "min": 73.03388654199996, "max": 75.50457230200001 }, "requests": { "median": 300, "min": 300, "max": 300 }, "coordinator_exchange_wall_seconds": { "median": 75.22893874099996, "min": 74.09127016700003, "max": 76.56417875299995 }, "inner_threads": { "median": 1, "min": 1, "max": 1 }, "owned_host_tensor_peak_bytes": { "median": 655360, "min": 655360, "max": 655360 }, "request_write_seconds": { "median": 1.057320459, "min": 0.9499169850000007, "max": 1.0594542839999999 }, "request_payload_bytes": { "median": 61870640, "min": 61870640, "max": 61870640 }, "process_peak_rss_bytes": { "median": 252145664, "min": 252051456, "max": 252153856 }, "samples": { "median": 152920, "min": 152920, "max": 152920 }, "upload_preparation_seconds": { "median": 1.307947765000001, "min": 1.3072958470000007, "max": 1.3187351210000007 }, "worker_total_seconds": { "median": 2.478999484000003, "min": 2.474586277, "max": 2.511008313 }, "zero_sample_steps": { "median": 0, "min": 0, "max": 0 }, "response_payload_bytes": { "median": 40441800, "min": 40441800, "max": 40441800 }, "compute_seconds": { "median": 0.7740768829999997, "min": 0.7703047389999997, "max": 0.7919749250000002 }, "owned_device_tensor_peak_bytes": { "median": 2264616, "min": 2264616, "max": 2264616 }, "readback_seconds": { "median": 0.14073625700000006, "min": 0.1406611379999999, "max": 0.14422723300000018 }, "backend": "cuda_cublas", "target": "linux-x86_64" }, { "configuration_id": "atlas-03/C03", "platform": "atlas-03", "device": "cuda", "relative_batch_weight": 1, "response_read_wait_seconds": { "median": 68.02293487299998, "min": 67.35432496700007, "max": 71.42854133200002 }, "requests": { "median": 300, "min": 300, "max": 300 }, "coordinator_exchange_wall_seconds": { "median": 68.97260412699998, "min": 68.41272091100004, "max": 72.48310226199992 }, "inner_threads": { "median": 1, "min": 1, "max": 1 }, "owned_host_tensor_peak_bytes": { "median": 655360, "min": 655360, "max": 655360 }, "request_write_seconds": { "median": 1.0544526330000004, "min": 0.9496139069999998, "max": 1.0583404930000002 }, "request_payload_bytes": { "median": 61870640, "min": 61870640, "max": 61870640 }, "process_peak_rss_bytes": { "median": 286314496, "min": 286232576, "max": 286322688 }, "samples": { "median": 152920, "min": 152920, "max": 152920 }, "upload_preparation_seconds": { "median": 0.9694459050000002, "min": 0.9608044390000001, "max": 0.9706038950000002 }, "worker_total_seconds": { "median": 1.6511017689999994, "min": 1.643087887, "max": 1.6777562749999992 }, "zero_sample_steps": { "median": 0, "min": 0, "max": 0 }, "response_payload_bytes": { "median": 40441800, "min": 40441800, "max": 40441800 }, "compute_seconds": { "median": 0.442767139, "min": 0.4391299450000002, "max": 0.46720008700000026 }, "owned_device_tensor_peak_bytes": { "median": 2264616, "min": 2264616, "max": 2264616 }, "readback_seconds": { "median": 0.07239806899999998, "min": 0.07221494900000006, "max": 0.07297718599999994 }, "backend": "cuda_cublas", "target": "linux-x86_64" } ], "fastest_selected_baseline": { "fit_seconds": { "case": "atlas01-gpu32", "seconds": 2.804163125, "speedup": 0.032762391072286914 }, "outer_command_wall_seconds": { "case": "atlas01-gpu32", "seconds": 3.2152430829883087, "speedup": 0.03734582930414821 } } }, { "case": "02-03-gpu64", "label": "Atlas-02 CUDA + Atlas-03 CUDA \u00b7 Float64", "dtype": "Float64", "kind": "group", "replicas": [ { "configuration_id": "atlas-02/C03", "platform": "atlas-02", "device": "cuda", "relative_batch_weight": 1 }, { "configuration_id": "atlas-03/C03", "platform": "atlas-03", "device": "cuda", "relative_batch_weight": 1 } ], "coordinator_configuration_id": "atlas-01/C03", "numerical_reference_case": "atlas01-cpu64", "repetitions": 3, "outer_command_wall_seconds": { "median": 124.655269416995, "min": 116.99921195799834, "max": 126.23090900000534 }, "fit_seconds": { "median": 124.125102542, "min": 116.467892458, "max": 125.694160875 }, "command_seconds": { "median": 124.584389833, "min": 116.93126075, "max": 126.164727625 }, "quality": { "initial_loss": { "median": 1.5991332128378295, "min": 1.5991332128378295, "max": 1.5991332128378295 }, "training_loss": { "median": 0.2923469152365444, "min": 0.2923469152365444, "max": 0.2923469152365444 }, "validation_loss": { "median": 0.3033731649951008, "min": 0.3033731649951008, "max": 0.3033731649951008 }, "test_loss": { "median": 0.29705317871020637, "min": 0.29705317871020637, "max": 0.29705317871020637 }, "training_accuracy": { "median": 0.8983564390966955, "min": 0.8983564390966955, "max": 0.8983564390966955 }, "validation_accuracy": { "median": 0.8857578840284842, "min": 0.8857578840284842, "max": 0.8857578840284842 }, "test_accuracy": { "median": 0.8955239064089522, "min": 0.8955239064089522, "max": 0.8955239064089522 } }, "parameter_max_abs_error": 3.3306690738754696e-16, "loss_max_abs_errors": { "initial_loss": 0.0, "training_loss": 5.551115123125783e-17, "validation_loss": 5.551115123125783e-17, "test_loss": 5.551115123125783e-17 }, "accuracy_max_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "coordinator_memory": { "process_peak_rss_bytes": { "median": 76169216, "min": 75055104, "max": 76529664 }, "tensor_owned_host_peak_bytes": { "median": 4202560, "min": 4202560, "max": 4202560 }, "tensor_owned_device_peak_bytes": { "median": 0, "min": 0, "max": 0 } }, "replica_metrics": [ { "configuration_id": "atlas-02/C03", "platform": "atlas-02", "device": "cuda", "relative_batch_weight": 1, "response_read_wait_seconds": { "median": 108.28300708299992, "min": 104.58419616099997, "max": 111.21368550599998 }, "requests": { "median": 300, "min": 300, "max": 300 }, "coordinator_exchange_wall_seconds": { "median": 110.89321180400003, "min": 107.19696121899996, "max": 113.86994150799987 }, "inner_threads": { "median": 1, "min": 1, "max": 1 }, "owned_host_tensor_peak_bytes": { "median": 393216, "min": 393216, "max": 393216 }, "request_write_seconds": { "median": 2.612709755, "min": 2.610150838999998, "max": 2.656146796 }, "request_payload_bytes": { "median": 165262320, "min": 165262320, "max": 165262320 }, "process_peak_rss_bytes": { "median": 254693376, "min": 254693376, "max": 254758912 }, "samples": { "median": 229380, "min": 229380, "max": 229380 }, "upload_preparation_seconds": { "median": 1.1247009410000004, "min": 1.1209860590000003, "max": 1.132229757 }, "worker_total_seconds": { "median": 3.444995925000001, "min": 3.4443716670000013, "max": 3.4631106159999994 }, "zero_sample_steps": { "median": 0, "min": 0, "max": 0 }, "response_payload_bytes": { "median": 40441800, "min": 40441800, "max": 40441800 }, "compute_seconds": { "median": 1.7728895749999996, "min": 1.7582898530000002, "max": 1.773451806000001 }, "owned_device_tensor_peak_bytes": { "median": 6659152, "min": 6659152, "max": 6659152 }, "readback_seconds": { "median": 0.167989555, "min": 0.16438954099999997, "max": 0.17331558899999996 }, "backend": "cuda_cublas", "target": "linux-x86_64" }, { "configuration_id": "atlas-03/C03", "platform": "atlas-03", "device": "cuda", "relative_batch_weight": 1, "response_read_wait_seconds": { "median": 96.94922095000004, "min": 94.17421533999999, "max": 107.93112454199999 }, "requests": { "median": 300, "min": 300, "max": 300 }, "coordinator_exchange_wall_seconds": { "median": 99.56707587099999, "min": 96.842004244, "max": 110.54037168700005 }, "inner_threads": { "median": 1, "min": 1, "max": 1 }, "owned_host_tensor_peak_bytes": { "median": 393216, "min": 393216, "max": 393216 }, "request_write_seconds": { "median": 2.6178147880000004, "min": 2.609176034999999, "max": 2.6677069549999985 }, "request_payload_bytes": { "median": 165262320, "min": 165262320, "max": 165262320 }, "process_peak_rss_bytes": { "median": 282042368, "min": 281931776, "max": 282214400 }, "samples": { "median": 229380, "min": 229380, "max": 229380 }, "upload_preparation_seconds": { "median": 0.45382250100000004, "min": 0.45233597800000014, "max": 0.45668940900000027 }, "worker_total_seconds": { "median": 1.1111053540000004, "min": 1.1019800560000002, "max": 1.1185561859999993 }, "zero_sample_steps": { "median": 0, "min": 0, "max": 0 }, "response_payload_bytes": { "median": 40441800, "min": 40441800, "max": 40441800 }, "compute_seconds": { "median": 0.434931702, "min": 0.43328174500000005, "max": 0.4439623549999999 }, "owned_device_tensor_peak_bytes": { "median": 6659152, "min": 6659152, "max": 6659152 }, "readback_seconds": { "median": 0.062440942000000006, "min": 0.06230874499999999, "max": 0.06328919300000004 }, "backend": "cuda_cublas", "target": "linux-x86_64" } ], "fastest_selected_baseline": { "fit_seconds": { "case": "atlas03-gpu64", "seconds": 3.380031545, "speedup": 0.027230845943158877 }, "outer_command_wall_seconds": { "case": "atlas03-gpu64", "seconds": 5.359644708005362, "speedup": 0.042995733217473193 } } }, { "case": "01-02-03-cpu32", "label": "Atlas-01 CPU + Atlas-02 CPU + Atlas-03 CPU \u00b7 Float32", "dtype": "Float32", "kind": "group", "replicas": [ { "configuration_id": "atlas-01/C03", "platform": "atlas-01", "device": "cpu", "relative_batch_weight": 1 }, { "configuration_id": "atlas-02/C03", "platform": "atlas-02", "device": "cpu", "relative_batch_weight": 1 }, { "configuration_id": "atlas-03/C03", "platform": "atlas-03", "device": "cpu", "relative_batch_weight": 1 } ], "coordinator_configuration_id": "atlas-01/C03", "numerical_reference_case": "atlas01-cpu32", "repetitions": 3, "outer_command_wall_seconds": { "median": 89.16344270898844, "min": 88.21805833300459, "max": 92.74185400000715 }, "fit_seconds": { "median": 88.645885167, "min": 87.638751083, "max": 92.231870458 }, "command_seconds": { "median": 89.109699709, "min": 88.129241583, "max": 92.696210667 }, "quality": { "initial_loss": { "median": 1.5991332174875397, "min": 1.5991332174875397, "max": 1.5991332174875397 }, "training_loss": { "median": 0.2923475254698321, "min": 0.2923475254698321, "max": 0.2923475254698321 }, "validation_loss": { "median": 0.30337432724777097, "min": 0.30337432724777097, "max": 0.30337432724777097 }, "test_loss": { "median": 0.2970538696340481, "min": 0.2970538696340481, "max": 0.2970538696340481 }, "training_accuracy": { "median": 0.8983564390966955, "min": 0.8983564390966955, "max": 0.8983564390966955 }, "validation_accuracy": { "median": 0.8858596134282808, "min": 0.8858596134282808, "max": 0.8858596134282808 }, "test_accuracy": { "median": 0.8955239064089522, "min": 0.8955239064089522, "max": 0.8955239064089522 } }, "parameter_max_abs_error": 4.561617970466614e-06, "loss_max_abs_errors": { "initial_loss": 0.0, "training_loss": 1.0843586306652497e-08, "validation_loss": 9.871450834531004e-08, "test_loss": 5.9598581236830483e-08 }, "accuracy_max_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "coordinator_memory": { "process_peak_rss_bytes": { "median": 77873152, "min": 76414976, "max": 77889536 }, "tensor_owned_host_peak_bytes": { "median": 6303328, "min": 6303328, "max": 6303328 }, "tensor_owned_device_peak_bytes": { "median": 0, "min": 0, "max": 0 } }, "replica_metrics": [ { "configuration_id": "atlas-01/C03", "platform": "atlas-01", "device": "cpu", "relative_batch_weight": 1, "response_read_wait_seconds": { "median": 3.0818655430000006, "min": 2.944582882999999, "max": 3.093941788 }, "requests": { "median": 300, "min": 300, "max": 300 }, "coordinator_exchange_wall_seconds": { "median": 3.9981469220000023, "min": 3.811123540000002, "max": 4.0202211650000015 }, "inner_threads": null, "owned_host_tensor_peak_bytes": { "median": 6834532, "min": 6834532, "max": 6834532 }, "request_write_seconds": { "median": 0.9162247140000003, "min": 0.8664736270000004, "max": 0.9262225830000002 }, "request_payload_bytes": { "median": 61870640, "min": 61870640, "max": 61870640 }, "process_peak_rss_bytes": { "median": 24690688, "min": 24117248, "max": 25133056 }, "samples": { "median": 152920, "min": 152920, "max": 152920 }, "upload_preparation_seconds": { "median": 0.15840661500000008, "min": 0.15058891999999988, "max": 0.15966467000000004 }, "worker_total_seconds": { "median": 1.2889534940000007, "min": 1.2409844569999993, "max": 1.2952419939999993 }, "zero_sample_steps": { "median": 0, "min": 0, "max": 0 }, "response_payload_bytes": { "median": 40441800, "min": 40441800, "max": 40441800 }, "compute_seconds": { "median": 1.0918180099999997, "min": 1.0529558259999998, "max": 1.0963840000000002 }, "owned_device_tensor_peak_bytes": { "median": 0, "min": 0, "max": 0 }, "readback_seconds": { "median": 0.018997038999999993, "min": 0.018368426, "max": 0.019318249000000006 }, "backend": "apple_accelerate", "target": "macos-arm64" }, { "configuration_id": "atlas-02/C03", "platform": "atlas-02", "device": "cpu", "relative_batch_weight": 1, "response_read_wait_seconds": { "median": 77.784782799, "min": 77.513491078, "max": 79.15330272000001 }, "requests": { "median": 300, "min": 300, "max": 300 }, "coordinator_exchange_wall_seconds": { "median": 78.81341042199998, "min": 78.54798208700002, "max": 80.12637662399999 }, "inner_threads": { "median": 1, "min": 1, "max": 1 }, "owned_host_tensor_peak_bytes": { "median": 6834532, "min": 6834532, "max": 6834532 }, "request_write_seconds": { "median": 1.028517587, "min": 0.9729863700000007, "max": 1.0344487010000003 }, "request_payload_bytes": { "median": 61870640, "min": 61870640, "max": 61870640 }, "process_peak_rss_bytes": { "median": 124194816, "min": 124116992, "max": 124428288 }, "samples": { "median": 152920, "min": 152920, "max": 152920 }, "upload_preparation_seconds": { "median": 1.348304754, "min": 1.3208315059999995, "max": 1.3649168740000002 }, "worker_total_seconds": { "median": 13.892850112000001, "min": 13.59844539500001, "max": 13.917500425000002 }, "zero_sample_steps": { "median": 0, "min": 0, "max": 0 }, "response_payload_bytes": { "median": 40441800, "min": 40441800, "max": 40441800 }, "compute_seconds": { "median": 11.949936006, "min": 11.702780269000003, "max": 12.001523366999999 }, "owned_device_tensor_peak_bytes": { "median": 0, "min": 0, "max": 0 }, "readback_seconds": { "median": 0.146685316, "min": 0.14600822400000005, "max": 0.14739254699999993 }, "backend": "openblas", "target": "linux-x86_64" }, { "configuration_id": "atlas-03/C03", "platform": "atlas-03", "device": "cpu", "relative_batch_weight": 1, "response_read_wait_seconds": { "median": 73.57592612399998, "min": 65.71419562500003, "max": 74.19305296100005 }, "requests": { "median": 300, "min": 300, "max": 300 }, "coordinator_exchange_wall_seconds": { "median": 74.60627787600002, "min": 66.752902796, "max": 75.160350051 }, "inner_threads": { "median": 1, "min": 1, "max": 1 }, "owned_host_tensor_peak_bytes": { "median": 6834532, "min": 6834532, "max": 6834532 }, "request_write_seconds": { "median": 1.0302939589999998, "min": 0.9672401329999998, "max": 1.0386550049999992 }, "request_payload_bytes": { "median": 61870640, "min": 61870640, "max": 61870640 }, "process_peak_rss_bytes": { "median": 124391424, "min": 124260352, "max": 124420096 }, "samples": { "median": 152920, "min": 152920, "max": 152920 }, "upload_preparation_seconds": { "median": 0.9685700180000002, "min": 0.9642789750000003, "max": 0.9753726200000005 }, "worker_total_seconds": { "median": 5.551128372999997, "min": 5.463456344000003, "max": 5.588757929000004 }, "zero_sample_steps": { "median": 0, "min": 0, "max": 0 }, "response_payload_bytes": { "median": 40441800, "min": 40441800, "max": 40441800 }, "compute_seconds": { "median": 4.399906118000004, "min": 4.324123430000001, "max": 4.426885362000004 }, "owned_device_tensor_peak_bytes": { "median": 0, "min": 0, "max": 0 }, "readback_seconds": { "median": 0.036184635999999985, "min": 0.03586423799999997, "max": 0.03694794100000001 }, "backend": "openblas", "target": "linux-x86_64" } ], "fastest_selected_baseline": { "fit_seconds": { "case": "atlas01-cpu32", "seconds": 2.721057125, "speedup": 0.03069580860830483 }, "outer_command_wall_seconds": { "case": "atlas01-cpu32", "seconds": 3.0628960830072174, "speedup": 0.03435147847536456 } } }, { "case": "01-02-03-cpu64", "label": "Atlas-01 CPU + Atlas-02 CPU + Atlas-03 CPU \u00b7 Float64", "dtype": "Float64", "kind": "group", "replicas": [ { "configuration_id": "atlas-01/C03", "platform": "atlas-01", "device": "cpu", "relative_batch_weight": 1 }, { "configuration_id": "atlas-02/C03", "platform": "atlas-02", "device": "cpu", "relative_batch_weight": 1 }, { "configuration_id": "atlas-03/C03", "platform": "atlas-03", "device": "cpu", "relative_batch_weight": 1 } ], "coordinator_configuration_id": "atlas-01/C03", "numerical_reference_case": "atlas01-cpu64", "repetitions": 3, "outer_command_wall_seconds": { "median": 114.44000404200051, "min": 108.63698216599005, "max": 117.02706866699737 }, "fit_seconds": { "median": 113.89337375, "min": 108.103035375, "max": 116.4694985 }, "command_seconds": { "median": 114.357245166, "min": 108.564510833, "max": 116.938115166 }, "quality": { "initial_loss": { "median": 1.5991332128378295, "min": 1.5991332128378295, "max": 1.5991332128378295 }, "training_loss": { "median": 0.29234691523654444, "min": 0.29234691523654444, "max": 0.29234691523654444 }, "validation_loss": { "median": 0.3033731649951008, "min": 0.3033731649951008, "max": 0.3033731649951008 }, "test_loss": { "median": 0.2970531787102063, "min": 0.2970531787102063, "max": 0.2970531787102063 }, "training_accuracy": { "median": 0.8983564390966955, "min": 0.8983564390966955, "max": 0.8983564390966955 }, "validation_accuracy": { "median": 0.8857578840284842, "min": 0.8857578840284842, "max": 0.8857578840284842 }, "test_accuracy": { "median": 0.8955239064089522, "min": 0.8955239064089522, "max": 0.8955239064089522 } }, "parameter_max_abs_error": 2.7755575615628914e-16, "loss_max_abs_errors": { "initial_loss": 0.0, "training_loss": 0.0, "validation_loss": 5.551115123125783e-17, "test_loss": 1.1102230246251565e-16 }, "accuracy_max_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.0, "test_accuracy": 0.0 }, "coordinator_memory": { "process_peak_rss_bytes": { "median": 71483392, "min": 70926336, "max": 73039872 }, "tensor_owned_host_peak_bytes": { "median": 4202560, "min": 4202560, "max": 4202560 }, "tensor_owned_device_peak_bytes": { "median": 0, "min": 0, "max": 0 } }, "replica_metrics": [ { "configuration_id": "atlas-01/C03", "platform": "atlas-01", "device": "cpu", "relative_batch_weight": 1, "response_read_wait_seconds": { "median": 3.839887451999999, "min": 3.7663322510000006, "max": 3.913244859999998 }, "requests": { "median": 300, "min": 300, "max": 300 }, "coordinator_exchange_wall_seconds": { "median": 5.567488794999997, "min": 5.434352178, "max": 5.695344581000001 }, "inner_threads": null, "owned_host_tensor_peak_bytes": { "median": 5398608, "min": 5398608, "max": 5398608 }, "request_write_seconds": { "median": 1.7275411719999993, "min": 1.6679660429999998, "max": 1.782044620999998 }, "request_payload_bytes": { "median": 123668080, "min": 123668080, "max": 123668080 }, "process_peak_rss_bytes": { "median": 20987904, "min": 20840448, "max": 25673728 }, "samples": { "median": 152920, "min": 152920, "max": 152920 }, "upload_preparation_seconds": { "median": 0.10331621500000006, "min": 0.10130259600000005, "max": 0.10499419699999991 }, "worker_total_seconds": { "median": 1.4972432159999998, "min": 1.477964198000001, "max": 1.526615741000001 }, "zero_sample_steps": { "median": 0, "min": 0, "max": 0 }, "response_payload_bytes": { "median": 40441800, "min": 40441800, "max": 40441800 }, "compute_seconds": { "median": 1.3599068440000004, "min": 1.3427640759999984, "max": 1.3873297540000007 }, "owned_device_tensor_peak_bytes": { "median": 0, "min": 0, "max": 0 }, "readback_seconds": { "median": 0.015336034999999996, "min": 0.015248027000000003, "max": 0.015355873999999993 }, "backend": "apple_accelerate", "target": "macos-arm64" }, { "configuration_id": "atlas-02/C03", "platform": "atlas-02", "device": "cpu", "relative_batch_weight": 1, "response_read_wait_seconds": { "median": 104.35145782700003, "min": 93.57786412000006, "max": 104.37799353499994 }, "requests": { "median": 300, "min": 300, "max": 300 }, "coordinator_exchange_wall_seconds": { "median": 106.41689174800003, "min": 95.71556680100005, "max": 106.56334433700003 }, "inner_threads": { "median": 1, "min": 1, "max": 1 }, "owned_host_tensor_peak_bytes": { "median": 5398608, "min": 5398608, "max": 5398608 }, "request_write_seconds": { "median": 2.1376533849999997, "min": 2.0653792869999994, "max": 2.1852947489999988 }, "request_payload_bytes": { "median": 123668080, "min": 123668080, "max": 123668080 }, "process_peak_rss_bytes": { "median": 122904576, "min": 122826752, "max": 123088896 }, "samples": { "median": 152920, "min": 152920, "max": 152920 }, "upload_preparation_seconds": { "median": 0.8999936510000001, "min": 0.8924193459999997, "max": 0.9124430939999998 }, "worker_total_seconds": { "median": 13.327921916999996, "min": 13.227574789999995, "max": 13.369700921999998 }, "zero_sample_steps": { "median": 0, "min": 0, "max": 0 }, "response_payload_bytes": { "median": 40441800, "min": 40441800, "max": 40441800 }, "compute_seconds": { "median": 11.927826143999997, "min": 11.839041184000004, "max": 11.957221955999998 }, "owned_device_tensor_peak_bytes": { "median": 0, "min": 0, "max": 0 }, "readback_seconds": { "median": 0.16115268799999996, "min": 0.16050355099999997, "max": 0.16321454900000001 }, "backend": "openblas", "target": "linux-x86_64" }, { "configuration_id": "atlas-03/C03", "platform": "atlas-03", "device": "cpu", "relative_batch_weight": 1, "response_read_wait_seconds": { "median": 92.89171421000005, "min": 90.68079633100007, "max": 94.24916463099996 }, "requests": { "median": 300, "min": 300, "max": 300 }, "coordinator_exchange_wall_seconds": { "median": 95.08160861699996, "min": 92.81445225600002, "max": 96.32528441800005 }, "inner_threads": { "median": 1, "min": 1, "max": 1 }, "owned_host_tensor_peak_bytes": { "median": 5398608, "min": 5398608, "max": 5398608 }, "request_write_seconds": { "median": 2.1335961649999993, "min": 2.0760746739999996, "max": 2.1897552950000008 }, "request_payload_bytes": { "median": 123668080, "min": 123668080, "max": 123668080 }, "process_peak_rss_bytes": { "median": 122892288, "min": 122687488, "max": 123039744 }, "samples": { "median": 152920, "min": 152920, "max": 152920 }, "upload_preparation_seconds": { "median": 0.4658746279999999, "min": 0.45744333199999965, "max": 0.4670489530000005 }, "worker_total_seconds": { "median": 4.483889233999999, "min": 4.476447578000001, "max": 4.534127982999999 }, "zero_sample_steps": { "median": 0, "min": 0, "max": 0 }, "response_payload_bytes": { "median": 40441800, "min": 40441800, "max": 40441800 }, "compute_seconds": { "median": 3.8670035659999997, "min": 3.8578807020000028, "max": 3.906021795999999 }, "owned_device_tensor_peak_bytes": { "median": 0, "min": 0, "max": 0 }, "readback_seconds": { "median": 0.038350755000000035, "min": 0.03728029299999998, "max": 0.039353534999999995 }, "backend": "openblas", "target": "linux-x86_64" } ], "fastest_selected_baseline": { "fit_seconds": { "case": "atlas01-cpu64", "seconds": 3.086051291, "speedup": 0.02709596870643232 }, "outer_command_wall_seconds": { "case": "atlas01-cpu64", "seconds": 3.4516792920039734, "speedup": 0.0301614747473897 } } }, { "case": "02-03-gpu32-w1-7", "label": "Atlas-02 CUDA + Atlas-03 CUDA \u00b7 Float32 \u00b7 allocation 1:7", "dtype": "Float32", "kind": "group", "replicas": [ { "configuration_id": "atlas-02/C03", "platform": "atlas-02", "device": "cuda", "relative_batch_weight": 1 }, { "configuration_id": "atlas-03/C03", "platform": "atlas-03", "device": "cuda", "relative_batch_weight": 7 } ], "coordinator_configuration_id": "atlas-01/C03", "numerical_reference_case": "atlas01-cpu32", "repetitions": 3, "outer_command_wall_seconds": { "median": 92.13198762499087, "min": 90.50230108300457, "max": 93.24883054199745 }, "fit_seconds": { "median": 91.641994208, "min": 90.005007041, "max": 92.727323583 }, "command_seconds": { "median": 92.093433917, "min": 90.461175125, "max": 93.18005375 }, "quality": { "initial_loss": { "median": 1.5991332174875397, "min": 1.5991332174875397, "max": 1.5991332174875397 }, "training_loss": { "median": 0.29234655017850286, "min": 0.29234655017850286, "max": 0.29234655017850286 }, "validation_loss": { "median": 0.30337328036464484, "min": 0.30337328036464484, "max": 0.30337328036464484 }, "test_loss": { "median": 0.29705295057650866, "min": 0.29705295057650866, "max": 0.29705295057650866 }, "training_accuracy": { "median": 0.8983564390966955, "min": 0.8983564390966955, "max": 0.8983564390966955 }, "validation_accuracy": { "median": 0.8857578840284842, "min": 0.8857578840284842, "max": 0.8857578840284842 }, "test_accuracy": { "median": 0.8955239064089522, "min": 0.8955239064089522, "max": 0.8955239064089522 } }, "parameter_max_abs_error": 4.634261131286621e-05, "loss_max_abs_errors": { "initial_loss": 0.0, "training_loss": 9.861349155593224e-07, "validation_loss": 1.1455976344776353e-06, "test_loss": 9.786561206981403e-07 }, "accuracy_max_abs_errors": { "training_accuracy": 0.0, "validation_accuracy": 0.00010172939979657958, "test_accuracy": 0.0 }, "coordinator_memory": { "process_peak_rss_bytes": { "median": 77479936, "min": 75546624, "max": 77643776 }, "tensor_owned_host_peak_bytes": { "median": 6303328, "min": 6303328, "max": 6303328 }, "tensor_owned_device_peak_bytes": { "median": 0, "min": 0, "max": 0 } }, "replica_metrics": [ { "configuration_id": "atlas-02/C03", "platform": "atlas-02", "device": "cuda", "relative_batch_weight": 1, "response_read_wait_seconds": { "median": 49.793614289000026, "min": 47.13218325199995, "max": 50.69311366600001 }, "requests": { "median": 300, "min": 300, "max": 300 }, "coordinator_exchange_wall_seconds": { "median": 50.300012469000016, "min": 47.69015449500003, "max": 51.246885000999995 }, "inner_threads": { "median": 1, "min": 1, "max": 1 }, "owned_host_tensor_peak_bytes": { "median": 245760, "min": 245760, "max": 245760 }, "request_write_seconds": { "median": 0.5537185480000001, "min": 0.5063517589999998, "max": 0.5579191199999997 }, "request_payload_bytes": { "median": 35875600, "min": 35875600, "max": 35875600 }, "process_peak_rss_bytes": { "median": 251711488, "min": 251682816, "max": 251740160 }, "samples": { "median": 57350, "min": 57350, "max": 57350 }, "upload_preparation_seconds": { "median": 0.8483147890000005, "min": 0.8400615010000003, "max": 0.8638853949999997 }, "worker_total_seconds": { "median": 1.7833610479999993, "min": 1.7712620580000007, "max": 1.8202823890000008 }, "zero_sample_steps": { "median": 0, "min": 0, "max": 0 }, "response_payload_bytes": { "median": 40441800, "min": 40441800, "max": 40441800 }, "compute_seconds": { "median": 0.6286405510000002, "min": 0.6278459390000003, "max": 0.648182167 }, "owned_device_tensor_peak_bytes": { "median": 933416, "min": 933416, "max": 933416 }, "readback_seconds": { "median": 0.13438590700000008, "min": 0.13205223499999996, "max": 0.13533233299999997 }, "backend": "cuda_cublas", "target": "linux-x86_64" }, { "configuration_id": "atlas-03/C03", "platform": "atlas-03", "device": "cuda", "relative_batch_weight": 7, "response_read_wait_seconds": { "median": 82.62908840499999, "min": 80.11868316900004, "max": 82.66911189099999 }, "requests": { "median": 300, "min": 300, "max": 300 }, "coordinator_exchange_wall_seconds": { "median": 84.79015824300004, "min": 82.07265038100005, "max": 84.83596416000006 }, "inner_threads": { "median": 1, "min": 1, "max": 1 }, "owned_host_tensor_peak_bytes": { "median": 1720320, "min": 1720320, "max": 1720320 }, "request_write_seconds": { "median": 2.1609882789999997, "min": 1.9538403350000004, "max": 2.1667847920000005 }, "request_payload_bytes": { "median": 129459920, "min": 129459920, "max": 129459920 }, "process_peak_rss_bytes": { "median": 288477184, "min": 288174080, "max": 288571392 }, "samples": { "median": 401410, "min": 401410, "max": 401410 }, "upload_preparation_seconds": { "median": 1.1265795959999994, "min": 1.119587932999999, "max": 1.1299737530000002 }, "worker_total_seconds": { "median": 1.6740156149999985, "min": 1.6696630329999997, "max": 1.6912474249999987 }, "zero_sample_steps": { "median": 0, "min": 0, "max": 0 }, "response_payload_bytes": { "median": 40441800, "min": 40441800, "max": 40441800 }, "compute_seconds": { "median": 0.35501172800000025, "min": 0.35420495699999993, "max": 0.3653267259999998 }, "owned_device_tensor_peak_bytes": { "median": 5725736, "min": 5725736, "max": 5725736 }, "readback_seconds": { "median": 0.06263750900000004, "min": 0.062500059, "max": 0.06264641600000001 }, "backend": "cuda_cublas", "target": "linux-x86_64" } ], "fastest_selected_baseline": { "fit_seconds": { "case": "atlas03-gpu32", "seconds": 4.041986648, "speedup": 0.04410627117984683 }, "outer_command_wall_seconds": { "case": "atlas03-gpu32", "seconds": 6.113490791001823, "speedup": 0.06635578965131904 } } } ] }