Table of Contents

AiDotNet Documentation

The most comprehensive AI/ML framework for .NET with 4,300+ implementations across 60+ feature categories.

Get Started | View on GitHub | Interactive Playground

Feature Highlights

Category Count Key Features
Neural Networks 100+ CNN, RNN, Transformer, GAN, VAE, GNN
Classical ML 106+ Classification, Regression, Clustering
Computer Vision 50+ YOLO v8-11, DETR, Mask R-CNN, OCR
Audio Processing 90+ Whisper, TTS, Music Generation
Reinforcement Learning 80+ DQN, PPO, SAC, Multi-Agent
RAG & Embeddings 50+ Vector stores, Retrievers, Rerankers
LoRA Fine-tuning 37+ QLoRA, DoRA, AdaLoRA
Distributed Training 10+ DDP, FSDP, ZeRO

Quick Start

Installation

dotnet add package AiDotNet

Hello World

using AiDotNet;

var result = await new AiModelBuilder<double, double[], double>()
    .ConfigureModel(new NeuralNetwork<double>(inputSize: 4, hiddenSize: 16, outputSize: 3))
    .ConfigureOptimizer(new AdamOptimizer<double>())
    .ConfigurePreprocessing()
    .BuildAsync(features, labels);

// Use result.Predict() directly - this is the facade pattern
var prediction = result.Predict(newSample);

What do you want to build?

Task Documentation
Classify data Classification Tutorial
Detect objects Computer Vision Tutorial
Process text & RAG NLP & RAG Tutorial
Fine-tune LLMs LLM Fine-tuning Tutorial
Train RL agents Reinforcement Learning Tutorial
Scale training Distributed Training Tutorial
Deploy models Deployment Tutorial

Why AiDotNet?

AiDotNet is the most feature-complete AI/ML framework for .NET, designed to match and exceed the capabilities of Python frameworks while providing native .NET performance and developer experience.

Comprehensive Framework Comparison

Feature AiDotNet TorchSharp TensorFlow.NET ML.NET Accord.NET
Neural Network Architectures 100+ 50+ 30+ ~10 ~15
Classical ML Algorithms 106+ None None ~30 ~50
Computer Vision Models 50+ Via PyTorch Via TF Limited Basic
Audio Processing 90+ Limited Limited None Basic
Reinforcement Learning 80+ agents Manual Limited None None
LoRA/Fine-tuning 37+ adapters Manual None None None
HuggingFace Integration Native Partial Partial None None
Distributed Training DDP/FSDP/ZeRO DDP only MirroredStrategy None None

Performance Advantages

Benchmark AiDotNet TorchSharp TensorFlow.NET
SIMD Optimizations Native Via LibTorch Via TF Runtime
Memory Support Native No No
Span Operations Full Limited Limited
AOT Compilation Supported Limited No
Startup Time Fast Slow (Python runtime) Slow (TF runtime)

Key Performance Features:

  • SIMD-accelerated tensor operations - Native AVX2/AVX-512 support
  • BLAS integration - Optional Intel MKL/OpenBLAS for matrix operations
  • GPU acceleration - CUDA and OpenCL support without Python dependencies
  • Memory efficient - Uses Memory/Span for zero-copy operations
  • No Python runtime - Pure .NET execution, no interop overhead

Why Not TorchSharp?

TorchSharp wraps PyTorch's C++ runtime (LibTorch), which means:

  • Large runtime dependency (~700MB+ LibTorch binaries)
  • Slower startup - Must load PyTorch runtime
  • Limited .NET integration - Array copying between .NET and LibTorch
  • No classical ML - Only deep learning, no traditional algorithms
  • Manual everything - No AutoML, no hyperparameter optimization

AiDotNet provides:

  • Pure .NET implementation - No external runtime dependencies
  • Instant startup - No runtime initialization overhead
  • Native .NET types - Memory, Span, IAsyncEnumerable
  • 106+ classical ML algorithms - Full traditional ML support
  • Built-in AutoML - Automatic model selection and tuning

Why Not TensorFlow.NET?

TensorFlow.NET wraps TensorFlow's C runtime, which means:

  • Complex setup - Requires TensorFlow native libraries
  • Version compatibility issues - TF version must match wrapper version
  • Limited Keras support - Incomplete high-level API
  • Resource heavy - TensorFlow runtime consumes significant memory

AiDotNet provides:

  • Simple NuGet install - Just dotnet add package AiDotNet
  • Always compatible - No version matching required
  • High-level API - AiModelBuilder for easy model creation
  • Lightweight - Only load what you use

Why Not ML.NET?

ML.NET is Microsoft's official ML library, but:

  • Limited neural networks - Only basic architectures (~10)
  • No computer vision - Must use ONNX models
  • No audio processing - No speech/audio support
  • No reinforcement learning - No RL agents
  • No HuggingFace - No transformer model support

AiDotNet provides:

  • 100+ neural network architectures - CNN, RNN, Transformer, GAN, VAE, GNN
  • 50+ computer vision models - YOLO v8-11, DETR, Mask R-CNN, SAM
  • 90+ audio models - Whisper, TTS, music generation
  • 80+ RL agents - DQN, PPO, SAC, multi-agent systems
  • Native HuggingFace - Load and fine-tune transformer models

Feature Depth Comparison

Neural Networks

Architecture Type AiDotNet Others
Convolutional (CNN) 15+ variants Basic
Recurrent (RNN/LSTM/GRU) 10+ variants Basic
Transformer 20+ variants Manual
GAN 15+ variants Manual
VAE 10+ variants Manual
Graph Neural Networks 10+ variants None/.NET
Diffusion Models 20+ variants None/.NET
NeRF/3D 5+ variants None/.NET

Training Capabilities

Capability AiDotNet TorchSharp ML.NET
Mixed Precision (FP16/BF16) Yes Yes No
Gradient Checkpointing Yes Yes No
Multi-GPU Training DDP/FSDP/ZeRO DDP No
AutoML Built-in No AutoML.NET
Hyperparameter Optimization Built-in No Limited
Meta-Learning 15+ methods No No
Self-Supervised Learning 10+ methods Manual No

Summary: When to Use AiDotNet

Choose AiDotNet when you need:

  • A single framework that does everything (classical ML + deep learning + RL)
  • Native .NET performance without Python/C++ runtime dependencies
  • State-of-the-art models (YOLO v11, Whisper, Stable Diffusion)
  • HuggingFace model integration
  • Distributed training (DDP, FSDP, ZeRO)
  • LoRA fine-tuning for LLMs
  • Production deployment with AiDotNet.Serving

Consider alternatives when:

  • You need PyTorch ecosystem compatibility → TorchSharp
  • You have existing TensorFlow models → TensorFlow.NET
  • You only need basic ML with Microsoft support → ML.NET

Getting Help

About

AiDotNet is developed and maintained by Ooples Finance with contributions from the community.

Licensed under Apache License 2.0.

Edit this page