A question. A few lines.
A place to begin.
Explore real Aner programs, from a first variable to a neural classifier. Download the source and follow every step.
An Iris classifier in a few lines
Split first, fit preprocessing on training rows, and evaluate a small dense classifier on held out data.
Explore the program Neural networksClassifying Wine measurements
Use the same concise pipeline with 13 features and the attributed UCI Wine recognition dataset.
Explore the program DataFrom a typed CSV to a tensor
Validate synthetic records, preserve missingness, summarize bounded batches, and explicitly collect complete rows.
Explore the program Machine learningKNN on Iris
Standardize training data, fit a nearest neighbor classifier, and inspect held out predictions.
Explore the program Machine learningClustering Iris
Fit K means and evaluate clusters without assuming that cluster IDs equal class labels.
Explore the program VisualizationSee a network learn XOR
Capture actual training steps for an offline topology, loss, and gradient report.
Explore the program LanguageA basic integer linked list
Build Node and LinkedList classes, append or prepend integers, and traverse nullable links after checking for null.
Explore the program LanguageA binary search tree
Use recursive insertion and lookup with typed nullable nodes.
Explore the program Neural networksA bounded Tensor matrix stress test
Repeat matched CPU and available accelerator matrix products with explicit synchronization, device metadata, and numerical agreement checks.
Explore the program Language modelsA cloud language model with an opaque key
Connect to the native OpenAI Responses adapter using a credential label configured through hidden terminal input.
Explore the program Model APIsA configured language model API
Expose one fixed local language model client through a bounded completion route without accepting request supplied provider configuration.
Explore the program Neural networksA controlled distributed Dense experiment
Reproduce the seeded 64 to 128 to 64 to 4 fixture with protected partitions, a fixed global batch and native sample weighted training.
Explore the program Neural networksA custom neural model class
Organize XOR parameters with a class, define forward and loss explicitly, differentiate with native Tensor operations, and update every parameter leaf.
Explore the program LanguageA cyclic graph
Explore object identity and cyclic references in Aner’s managed object heap.
Explore the program LanguageA linked list
Build nullable links and use methods to traverse and mutate a reference structure.
Explore the program LanguageA linked list of mixed values
Keep integer, String, Float64, and Bool payloads in one linked structure using Any fields and parameters.
Explore the program Language modelsA local language model in a few lines
Use the native Ollama adapter to create one explicit local inference request and read its typed response.
Explore the program Model APIsA model inference service class
Preload a fitted classifier once and retain an immutable native API application in a service class.
Explore the program Model APIsA native neural model API
Train a small classifier once, then serve validated feature rows on a native loopback listener.
Explore the program Language modelsA research assistant class
Organize instructions and a typed LLM client in a class while retaining explicit messages and generation controls.
Explore the program Neural networksA research class with native replica training
Organize an explicit Dense architecture and immutable training configuration in a class, then select replica placement independently.
Explore the program Language modelsA served model through vLLM
Connect to a served model identifier through the native compatible chat adapter with explicit endpoint and timeout.
Explore the program LanguageBroadcasting and matrix products
Combine compatible trailing axes without expanded inputs, distinguish elementwise multiplication from matrix products, and copy checked Float64 results into a Tensor.
Explore the program Neural networksClassifier precision policies
Train a seeded Iris classifier under one explicit execution profile, then inspect effective parameters, optimizer state, evaluation, and duration.
Explore the program Neural networksCompare native classifier devices
Fit the same seeded Iris classifier on CPU and available MPS or CUDA, report placement and precision, and inspect loss, duration, and held out accuracy.
Explore the program LanguageConcise generic classes
Use private fields, implicit receivers, and inferred result types without repetitive method syntax.
Explore the program Neural networksContinue Adam from a checkpoint
Save and restore fitted normalization, parameters, Adam moments and update count, then compare resumed and uninterrupted training.
Explore the program Neural networksDevelop a neural network
Configure custom Dense weights, training fitted normalization, cross entropy, and SGD; inspect captured training in development and run the same program in production.
Explore the program Model APIsExplicit HTTPS and API authentication
Configure a permitted hostname, trusted TLS files, an opaque inbound bearer reference and request bounds for an external listener.
Explore the program Language modelsExplicit language model messages
Supply ordered typed messages, output budget, sampling temperature, finish reason and reported token usage.
Explore the program DataExplore the teaching catalogue
Inspect offline datasets, dimensions, citations, licenses, and source versions.
Explore the program LanguageFixed and dynamic typed arrays
Declare an element type, compare fixed length with doubling capacity, and see find switch from linear to binary search after sorting.
Explore the program LanguageFunctions and control flow
Use recursion, conditions, loops, and local scope.
Explore the program LanguageGeneric linked structures
Reuse Node<T> and LinkedList<T> for integers, floating point values, and Employee objects.
Explore the program LanguageHello, Aner
Start with scalar arithmetic in a complete Aner program.
Explore the program LanguageInspect a mixed linked list
Append, prepend, and remove mixed payloads while named development captures show Node references and the resulting list.
Explore the program Neural networksInspect Tensor precision and devices
Convert precision explicitly, select an available native device, and compare matrix results and resident gradients with CPU Float32.
Explore the program Neural networksIris with mean squared error
Fit a classifier using squared softmax probability errors, training fitted preprocessing, and a held out evaluation.
Explore the program Neural networksIris: the full training loop
Inspect explicit parameters, softmax cross entropy, SGD, and held out evaluation.
Explore the program Neural networksLearning XOR
Train a small dense network on synthetic XOR data with native reverse mode differentiation.
Explore the program LanguageList, Dictionary, and Set basics
Store mixed values, look up ordered Dictionary entries, remove Set duplicates, and stably sort numbers or Strings without imports.
Explore the program LanguageMatrices, reshape, and shared storage
Create explicit shapes and element coordinates, share fixed reshape storage, copy dynamic data safely, and inspect detached transpose and axis permutations.
Explore the program Neural networksMNIST with a concise training configuration
Train an audited 54,000 row Dense classifier with held out validation, lazy mini batches, Adam, and explicit precision.
Explore the program Neural networksMNIST with explicit forward and optimization functions
Expose Dense initialization and activation topology in a builder function, then compose immutable training and held out validation contracts.
Explore the program Neural networksMNIST with model classes
Organize Dense parameters and objectives in classes while retaining audited sources and explicit training and evaluation loops.
Explore the program Neural networksNative neural training with one explicit context
Fit a protected Iris partition with the same model and configuration on a strict CPU, CUDA or Metal/MPS context.
Explore the program LanguageNested collections and shared values
Combine Lists, Dictionaries, and user objects, inspect shallow copies, and traverse Dictionary snapshots in insertion order.
Explore the program Neural networksNeural research with explicit devices
Choose He and Xavier initialization, add Dense layers, train with Adam and measured progress, inspect CPU placement, and evaluate held out Iris data.
Explore the program LanguageNumerical arrays with native storage
Run checked vector arithmetic, dot products, cumulative sums, and population statistics, then copy Float64 values into a Tensor.
Explore the program Neural networksPortable Tensor precision policies
Run the same seeded source with preserved Float64, required compatibility adaptation, or intentional Float32 selection; inspect requested and actual execution before and after training.
Explore the program Model APIsPrepare a model for inference serving
Train with a protected split and save one new classifier checkpoint before starting a separate API process.
Explore the program Scientific computingReproducible native CPU parallelism
Compare exact serial and automatic CPU sample accounting under the same scientific simulation, then inspect measured execution and the result.
Explore the program Neural networksSave and resume an Iris classifier
Save a versioned fitted checkpoint, restore normalization and weights, then continue training from the immutable fitted snapshot.
Explore the program LanguageTypes and arithmetic
Explore typed bindings, strings, booleans, comparisons, and numerical expressions.
Explore the program