Open-Source Training and RLHF Frameworks

FATE is a self-hosted federated learning framework for enterprises and institutions that want to train models together while protecting data privacy. It supports collaboration across participating organizations and uses homomorphic encryption and multi-party computation (MPC) for secure computation.

Tianshou is a Python reinforcement learning library built on PyTorch and Gymnasium. It's for researchers developing learning algorithms and developers training agents in custom environments. Its two API levels let you choose between a high-level interface for applying existing algorithms and modular interfaces for changing how an algorithm learns.

APPFL is an MIT-licensed Python framework for researchers and data scientists who want to train machine learning models across distributed clients while keeping training data decentralized. It supports user-defined models and lets teams test privacy-preserving federated learning methods in simulations or real distributed experiments.

TorchMD-NET is an open-source Python library for researchers who train neural network potentials for molecular dynamics. It exposes its models as PyTorch modules and connects them to GPU-accelerated simulation software, including ACEMD, OpenMM and TorchMD. It's available under the MIT license.

Sentence Transformers is a Python library for developers building search and document retrieval on their own hardware. It helps match content by meaning and rank results against a query.

Transformer Engine is a library for developers building and running Transformer models on NVIDIA GPUs. It reduces compute and memory demands in training and inference through low-precision arithmetic and optimized model components.

Fish Diffusion is a Python framework for developers and researchers training diffusion models for text-to-speech, singing voice synthesis and singing voice conversion. It works with your own datasets and provides code you can adapt for different voice generation tasks.

ms-swift is a Python framework for developers and researchers training and deploying language and multimodal models on their own hardware. It covers model adaptation through evaluation and serving, with support for Qwen3, DeepSeek-R1, Llama4, Mistral, Qwen3-VL and InternVL3.5.

Flash Linear Attention is a Python library for researchers and developers building language models on their own hardware. It supplies reusable components for exploring linear attention, sparse attention and state space architectures.

GLiNER is a Python framework for extracting named entities from text using small models that run locally on CPUs and consumer hardware. It's for developers building document processing, personal data detection or search systems who need to choose their own entity types without training a model for each task.

LitGPT is a Python toolkit for developers and researchers who want to train and deploy language models on their own hardware. Its directly editable model implementations suit custom applications and research experiments.

NeMo RL is NVIDIA's Python library for researchers and engineering teams adapting language and multimodal models on their own hardware. It supports experiments on a GPU workstation and distributed training across multiple nodes.

MLX Tune is a Python library for adapting AI models on an Apple Silicon Mac. Formerly unsloth-mlx, it is an independent project and is not affiliated with Unsloth. Its Unsloth-compatible API lets you prototype training locally with minimal changes to existing scripts, then move to NVIDIA hardware with Unsloth. Training runs through Apple's MLX framework and doesn't require a cloud GPU.

MaxText is an LLM training library for researchers and engineers building models from scratch or adapting existing ones. It provides a reference implementation teams can modify for their own model designs, with training that scales from a single host to large clusters.

MusicGen is Meta AI's model for generating music from written descriptions. It's part of AudioCraft, a PyTorch library for developers and audio researchers building music generation tools in their own computing environment.

NeuralForecast is an open-source Python library for data scientists and developers who want to train and compare neural time series forecasting models in their own Python environment. It brings recurrent networks and forecasting-specific architectures under one interface, so comparing models doesn't require working with separate research implementations. The license is Apache 2.0.

torchtitan is a training platform for researchers and developers experimenting with generative AI models on their own GPU servers or clusters. Its replaceable components and extension points support work on model architectures and training infrastructure.

SkyRL is a Python library for researchers and developers who want to train LLMs with reinforcement learning on their own hardware. Its Tinker API support lets you run existing Tinker training scripts against a server you control without changing the scripts. It's open source under Apache 2.0.

Stable Baselines3 is a Python library for training reinforcement learning agents on your own hardware using PyTorch. It builds on Stable Baselines and gives researchers and developers tested algorithm implementations for comparing approaches or building their own projects. It's free and open source under the MIT license.

Stable Audio Tools is a Python toolkit for developers and audio researchers who want to generate audio or train models on their own recordings. It supports work with pretrained models as well as training from scratch.

CleanRL is a Python deep reinforcement learning library for researchers and developers who want to study or modify an algorithm without tracing it through a large modular codebase. Each algorithm variant lives in one standalone file. Experiments can run locally, while Docker and AWS integration support cloud execution through AWS Batch.

AutoGluon is a Python AutoML library for developers and data scientists who want predictive models without selecting every algorithm themselves. It chooses combinations of classic machine learning and foundation models for the task.

MiniMind is a toolkit for learners and developers who want to understand language models by training small ones on their own hardware. Its core training algorithms use native PyTorch code that you can inspect and change.

Ultralytics YOLO is a Python computer vision library for developers building applications that interpret images and video. Models can run on hardware you control for tasks such as visual inspection, inventory monitoring and robotics perception.

Agent Lightning is a Python framework for developers and researchers training AI agents through reinforcement learning. It keeps the agent's existing tools, context and task environment involved in training, so feedback comes from the workflow the agent actually uses.

DeepPoseKit is an open-source Python toolkit for researchers measuring animal movement and behavior in images or video. You define the body points you want to track and train a model on your own annotations. It runs through TensorFlow and Keras on your own hardware, with Windows support documented, and uses the Apache 2.0 license.

Lightning Pose is an open-source package for researchers who estimate animal poses from video for behavior analysis. It supports 2D and 3D pose estimation with single-camera and multiple-camera datasets, including footage where body parts become hidden from view.

Megatron-LM is a Python framework for researchers training large language models on NVIDIA GPU infrastructure. It provides reference training scripts for experimenting with models and studying distributed training.

PaddleX is a low-code toolkit for developers building AI applications with PaddlePaddle models. It brings pretrained pipelines and model training into one tool for applications that process documents, images or time-series data.

MLForecast is an open-source Python library for data scientists and developers who need to forecast many time series with machine learning. It combines time series feature engineering with scikit-learn-compatible models, so teams can use regressors such as LightGBM and LinearRegression rather than adopt a dedicated forecasting model. Its license is Apache 2.0.

Avalanche is a Python library for researchers and practitioners studying continual learning, where models learn across a sequence of data or tasks. Built on PyTorch, it brings benchmark datasets, training strategies and evaluation into one framework, reducing the custom code needed to compare experiments. It's open source under the MIT license and can train models on your own CPU or CUDA GPU.

XGBoost is a machine learning library for training and running gradient-boosted tree models on your own hardware. Data scientists use it for classification, regression and ranking, with parallel computation for larger workloads. The code is licensed under Apache 2.0.

MosaicML Composer is a Python library for training neural networks on your own GPUs or multi-node clusters. Built on PyTorch, it's aimed at researchers and engineers who want distributed training with room to customize their experiments. It's open source under Apache 2.0, and prebuilt Docker images are available.