Open-source llm-training software
812 open-source libraries, tools and applications tagged llm-training, ranked by how closely llm-training is their primary purpose, then by adoption and maintenance health. Each entry links to a page with sources, use cases and when to choose or avoid it.
812 products, primary matches first, then adoption-weighted; health v2 shown.
| Product | Health v2 | Stars | Maturity |
|---|---|---|---|
| RVC-Project/Retrieval-based-Voice-Conversion-WebUI A WebUI-based framework for training and running retrieval-based voice conversion (RVC) models, letting users clone a voice timbre from as … | 77 | 38553 | active |
| tinygrad/tinygrad tinygrad is a small, hackable deep learning framework providing a PyTorch-like tensor library with autograd, plus a visible IR and compiler… | 88 | 33740 | active |
| hpcaitech/Open-Sora Open-Sora is an open-source initiative for efficiently producing high-quality videos from text prompts using deep learning models. It provi… | 47 | 29872 | active |
| MLX MLX is an array computation framework for machine learning on Apple silicon, developed by Apple ML research. It offers NumPy-like Python AP… | 95 | 28839 | active |
| MLflow MLflow is an open source AI engineering platform for agents, LLMs, and traditional ML models, providing experiment tracking, model registry… | 100 | 28441 | stable |
| facebookresearch/audiocraft A PyTorch library from Meta for audio processing and generation with deep learning, featuring the EnCodec neural audio codec and generative… | 58 | 23699 | active |
| microsoft/unilm Microsoft's collection of large-scale self-supervised pre-trained models spanning tasks, 100+ languages, and modalities (text, image, layou… | 66 | 22219 | active |
| ONNX Runtime ONNX Runtime is a cross-platform, high-performance machine-learning accelerator for running inference and training on ONNX models. It suppo… | 100 | 22115 | stable |
| karpathy/llama2.c A minimal, dependency-free implementation of the Llama 2 LLM architecture that trains in PyTorch and runs inference in a single ~700-line p… | 29 | 20192 | active |
| ConardLi/easy-dataset Easy Dataset is a self-hosted web application for building high-quality structured datasets for LLM fine-tuning, RAG, and model evaluation.… | 66 | 14989 | active |
| RWKV RWKV is a novel language model architecture that combines RNN efficiency (linear time, constant memory, no KV-cache) with transformer-level… | 66 | 14744 | active |
| Eclipse Deeplearning4j Eclipse Deeplearning4J is an open-source deep learning framework and ecosystem for the JVM, including the ND4J linear algebra library, the … | 76 | 14269 | active |
| mlfoundations/open_clip OpenCLIP is an open-source PyTorch implementation of CLIP and related multimodal contrastive models, with many pretrained image/text checkp… | 80 | 14157 | active |
| Physical-Intelligence/openpi Open-source repository from Physical Intelligence containing vision-language-action (VLA) models for robotics, including π₀, π₀-FAST, and π… | 64 | 13991 | active |
| modelscope/DiffSynth-Studio DiffSynth-Studio is an open-source diffusion model engine from the ModelScope community that integrates mainstream image, video, and audio … | 77 | 13229 | active |
| zai-org/CogVideo CogVideo/CogVideoX is an open-source family of text-to-video and image-to-video generation models from Zhipu AI (THUDM), with inference and… | 42 | 13055 | active |
| PKU-YuanGroup/Open-Sora-Plan Open-Sora Plan is an open-source effort to reproduce OpenAI's Sora text-to-video model, providing training and inference code for video gen… | 45 | 12209 | active |
| google/sentencepiece SentencePiece is a fast, lightweight unsupervised text tokenizer and detokenizer for neural network-based text generation systems, implemen… | 83 | 12132 | stable |
| NVIDIA/cosmos NVIDIA Cosmos is an open platform of omnimodal world foundation models, datasets, and tools for building Physical AI systems such as robots… | 69 | 11892 | active |
| SpeechBrain SpeechBrain is an open-source PyTorch-based speech toolkit for building conversational AI systems. It provides training recipes, pretrained… | 85 | 11845 | active |
| LTX-Video Official Python package from Lightricks providing inference pipelines and LoRA training for LTX-2/LTX-2.5, an open-weights DiT-based founda… | 45 | 11138 | active |
| karpathy/minbpe A minimal, clean Python implementation of the byte-level Byte Pair Encoding (BPE) algorithm used for tokenization in modern LLMs like GPT-4… | 26 | 10830 | stable |
| open-mmlab/Amphion Amphion is an open-source Python toolkit for audio, music, and speech generation, supporting tasks like text-to-speech, voice conversion, s… | 48 | 10304 | active |
| Metaflow Metaflow is a human-centric Python framework from Netflix for building, managing, and deploying real-life AI/ML and data science systems. I… | 92 | 10280 | stable |
| InternVL InternVL is a family of open-source multimodal large language models (vision-language models) that combine vision transformers with LLMs to… | 34 | 10163 | active |
| Spring AI Spring AI is an application framework for AI engineering that provides Spring-friendly, portable abstractions for integrating AI models int… | 94 | 9542 | active |
| OneFlow OneFlow is an open-source deep learning framework written in C++ with a PyTorch-like Python API, focused on scalable and efficient distribu… | 44 | 9437 | active |
| NVlabs/Sana SANA is an efficiency-oriented PyTorch codebase for high-resolution text-to-image and text-to-video generation built on Linear Diffusion Tr… | 70 | 9129 | active |
| deepseek-ai/DeepGEMM DeepGEMM is a high-performance CUDA BLAS kernel library for NVIDIA tensor cores, providing FP8, FP4, and BF16 GEMMs plus fused MoE and othe… | 82 | 8937 | active |
| sebastianstarke/AI4Animation AI4Animation is a deep learning framework for data-driven character animation and control, built around Unity with a Python remake (AI4Anim… | 63 | 8873 | active |
| datajuicer/data-juicer Data-Juicer is a Python library and data processing system for cleaning, deduplicating, synthesizing, and analyzing data for foundation mod… | 94 | 7132 | active |
| albertan017/LLM4Decompile LLM4Decompile is an open-source series of large language models (1.3B to 33B) trained to decompile binary code back into readable, executab… | 53 | 7076 | active |
| SenseNova SenseNova-U is a series of open-weight unified multimodal models (e.g., SenseNova-U1.5-8B-MoT) built on the NEO-unify architecture that com… | 58 | 6884 | active |
| open-edge-platform/anomalib Anomalib is a Python deep learning library for anomaly detection, offering state-of-the-art unsupervised algorithms for detecting and local… | 96 | 6196 | active |
| SamuelSchmidgall/AgentLaboratory Agent Laboratory is an end-to-end autonomous research workflow framework that uses specialized LLM-driven agents to assist human researcher… | 34 | 5869 | active |
| AidLux (AidLearning) AidLux (originally AidLearning) is an AIoT development platform that runs a native Ubuntu Linux environment with GUI, deep learning tooling… | 63 | 5812 | active |
| MiniGPT-4 Official code for MiniGPT-4 and MiniGPT-v2, vision-language models that align a frozen visual encoder with a frozen LLM (Vicuna) via a sing… | 30 | 25605 | maintenance |
| NVIDIA/Model-Optimizer NVIDIA Model Optimizer (ModelOpt) is a Python library of state-of-the-art model optimization techniques including quantization, pruning, di… | 92 | 5386 | active |
| Kiln-AI/Kiln Kiln is a free desktop app and MIT-licensed Python library for the full AI development loop: evals, prompt optimization, RAG, agents, fine-… | 91 | 5207 | active |
| EasyPhoto EasyPhoto is a Stable Diffusion WebUI plugin for generating AI portraits by training a personal 'digital doppelganger' from 5-20 user photo… | 29 | 5147 | active |
| microsoft/muzic Muzic is a Microsoft Research project providing deep learning models for music understanding and generation, including MusicBERT, SongMASS,… | 62 | 4960 | active |
| openxla/xla XLA (Accelerated Linear Algebra) is an open-source machine learning compiler that optimizes models from frameworks like PyTorch, TensorFlow… | 77 | 4605 | active |
| Kwai-Kolors/Kolors Kolors is a large-scale latent diffusion model for photorealistic text-to-image synthesis, trained with bilingual (Chinese and English) tex… | 24 | 4598 | active |
| fixie-ai/ultravox Ultravox is a fast multimodal LLM that understands human speech directly without a separate ASR stage, by projecting audio into the LLM's e… | 40 | 4577 | active |
| lucas-maes/le-wm LeWorldModel (LeWM) is the official PyTorch codebase for a JEPA-based world model that trains stably end-to-end from raw pixels using only … | 49 | 4547 | active |
| zai-org/GLM-4.5 GLM-4.5 (and successors GLM-4.6/4.7) is a family of open-weight Mixture-of-Experts foundation models from Z.ai focused on agentic tasks, re… | 44 | 4424 | active |
| ThilinaRajapakse/simpletransformers Simple Transformers is a Python library built on Hugging Face Transformers that lets users train, fine-tune, and evaluate Transformer model… | 57 | 4256 | active |
| ArcInstitute/evo2 Evo 2 is a DNA foundation language model (1B-40B parameters) that models genomes at single-nucleotide resolution with up to 1 million base … | 56 | 4247 | active |
| SwanHubX/SwanLab SwanLab is an open-source AI training tracking and visualization platform with a Python SDK, CLI, and cloud or self-hosted dashboard. It in… | 88 | 4237 | active |
| LLaDA Official PyTorch implementation of LLaDA, a family of large language diffusion models (8B base/instruct, MoE, and iLLaDA variants) with pre… | 59 | 3986 | active |
| AnswerDotAI/RAGatouille RAGatouille is a Python library that makes it easy to use and train state-of-the-art late-interaction retrieval models like ColBERT in any … | 24 | 3964 | active |
| LazyAGI/LazyLLM LazyLLM is a low-code Python framework for building multi-agent LLM applications, covering prototype assembly, data feedback, and iterative… | 92 | 3889 | active |
| Stability-AI/stable-audio-tools Stability AI's training and inference toolkit for conditional audio generation models, including Stable Audio Open. It supports training cu… | 73 | 3868 | active |
| NVlabs/VILA VILA is a family of open-source vision language models (VLMs) optimized for efficient video and multi-image understanding, spanning edge, d… | 54 | 3865 | active |
| HazyResearch/ThunderKittens ThunderKittens is a C++/CUDA framework of tile-based primitives for writing fast deep learning GPU kernels. It embeds natively into CUDA so… | 68 | 3754 | active |
| starVLA/starVLA StarVLA is an open-source, Lego-like modular codebase for developing Vision-Language-Action (VLA) models for generalist robots. It unifies … | 67 | 3745 | active |
| Polyaxon Polyaxon is an open-source AI engineering control plane that teams self-host on their own Kubernetes clusters to schedule, track, and gover… | 75 | 3735 | active |
| NVlabs/GR00T-WholeBodyControl NVIDIA's unified platform for developing, training, and deploying whole-body controllers for humanoid robots, including the decoupled WBC m… | 60 | 3654 | active |
| ZhaoJ9014/face.evoLVe A high-performance face recognition library built on PaddlePaddle and PyTorch, providing comprehensive tools for face-related analytics and… | 34 | 3594 | active |
| unitreerobotics/unitree_rl_gym A reinforcement learning gym environment for training locomotion policies on Unitree robots (Go2, H1, H1_2, G1) using Isaac Gym and Mujoco.… | 39 | 3583 | active |
| zhaoyingjun/chatbot A Chinese chatbot project that trains conversational models (Seq2Seq and GPT-style, including small SLM variants) on custom corpora with lo… | 49 | 3582 | active |
| huggingface/datatrove DataTrove is a Python library from Hugging Face for processing, filtering, and deduplicating large-scale text data through customizable pip… | 87 | 3390 | active |
| magenta/ddsp DDSP is a Python library of differentiable digital signal processing components (synthesizers, filters, waveshapers) that can be embedded i… | 68 | 3380 | active |
| Captcha Solution A deep learning training tool for image CAPTCHA recognition built on TensorFlow, using CNN/ResNet/DenseNet backbones with GRU/LSTM recurren… | 52 | 3215 | active |
| Eladlev/AutoPrompt AutoPrompt is a Python framework for optimizing LLM prompts using Intent-based Prompt Calibration. It iteratively generates challenging edg… | 39 | 3017 | active |
| intel/neural-compressor Intel Neural Compressor is an open-source Python library providing state-of-the-art low-bit quantization (INT8/FP8/MXFP8/INT4/MXFP4/NVFP4),… | 95 | 2722 | active |
| prophesier/diff-svc Diff-SVC is a deep learning project that performs singing voice conversion using diffusion models, transforming input singing audio into a … | 58 | 2719 | active |
| TMElyralab/MusePose MusePose is a diffusion-based, pose-guided image-to-video generation framework for creating virtual human videos, where a character in a re… | 25 | 2706 | active |
| InfiniteYou InfiniteYou (InfU) is a research framework from ByteDance for identity-preserved text-to-image generation built on Diffusion Transformers l… | 34 | 2685 | active |
| openlake-project/openlake OpenLake is a high-performance distributed storage engine written in Rust (built on io_uring, RDMA, and GPUDirect) designed to feed GPUs du… | 78 | 2662 | active |
| lucidrains/audiolm-pytorch A PyTorch implementation of AudioLM, Google Research's language modeling approach to audio generation, including a MIT-licensed SoundStream… | 34 | 2625 | active |
| data-infra/cube-studio CubeStudio is an open-source, cloud-native, all-in-one AI platform covering the full machine learning lifecycle (MLOps/MaaS/LLMOps), includ… | 75 | 2546 | active |
| aigc-apps/EasyAnimate EasyAnimate is an end-to-end Python pipeline for high-resolution, long video and image generation based on transformer diffusion (DiT) mode… | 17 | 2266 | active |
| Alpha-VLLM/Lumina-T2X Lumina-T2X is a unified framework for text-to-any-modality generation built on flow-based large diffusion transformers. It supports generat… | 25 | 2250 | active |
| microsoft/LLaVA-Med LLaVA-Med is a large language-and-vision assistant fine-tuned for the biomedicine domain, built on the LLaVA multimodal architecture. It su… | 38 | 2232 | active |
| apple/ml-ferret Apple's Ferret, an end-to-end multimodal large language model (MLLM) that accepts any-form referring and grounds anything in its responses,… | 70 | 8655 | maintenance |
| lifeiteng/vall-e An unofficial PyTorch implementation of VALL-E, a zero-shot text-to-speech model that treats TTS as a conditional language modeling task ov… | 37 | 2221 | active |
| NX-AI/xlstm Official PyTorch implementation of xLSTM, an extended Long Short-Term Memory recurrent architecture with exponential gating and matrix memo… | 76 | 2205 | active |
| Avalanche Avalanche is an end-to-end continual learning library built on PyTorch, developed by ContinualAI. It provides modules for benchmarks, train… | 24 | 2104 | active |
| ali-vilab/In-Context-LoRA Official repository for In-Context LoRA (IC-LoRA), a framework for adapting Diffusion Transformers to diverse visual generation tasks via L… | 22 | 2081 | active |
| tensorflow/privacy TensorFlow Privacy is a Python library providing TensorFlow optimizers for training machine learning models with differential privacy. It i… | 64 | 2036 | active |
| bytetriper/RAE Official PyTorch implementation of 'Diffusion Transformers with Representation Autoencoders' (RAE), a two-stage image generation pipeline u… | 45 | 2032 | active |
| NUS-HPC-AI-Lab/VideoSys VideoSys is an open-source Python library providing easy and efficient infrastructure for video generation, supporting training, inference,… | 31 | 2025 | active |
| meta-pytorch/opacus Opacus is a PyTorch library for training neural networks with differential privacy via DP-SGD, requiring minimal code changes through its P… | 73 | 1960 | active |
| tensorlayer/TensorLayer TensorLayer is a TensorFlow-based deep learning and reinforcement learning library offering customizable neural layers for researchers and … | 23 | 7383 | maintenance |
| PixArt-alpha/PixArt-sigma PixArt-Σ is a PyTorch implementation of a diffusion transformer model for high-resolution (up to 4K) text-to-image generation, trained with… | 26 | 1939 | active |
| NVIDIA-NeMo/Curator NVIDIA NeMo Curator is a scalable, GPU-accelerated toolkit for preprocessing and curating large text, image, video, and audio datasets for … | 88 | 1814 | active |
| RAVE RAVE is the official PyTorch implementation of a realtime audio variational autoencoder for fast, high-quality neural audio synthesis. It s… | 51 | 1811 | active |
| Stability-AI/StableCascade Official codebase for Stable Cascade, a text-to-image generation model built on the Würstchen architecture with a highly compressed latent … | 26 | 6528 | maintenance |
| bespokelabsai/curator Bespoke Curator is a Python library for bulk LLM inference and scalable synthetic data curation for post-training and structured data extra… | 89 | 1744 | active |
| LLPhant/LLPhant LLPhant is a comprehensive PHP generative AI framework inspired by LangChain and LlamaIndex, providing tools for working with LLMs, embeddi… | 96 | 1722 | active |
| feizc/FluxMusic FluxMusic is the official PyTorch implementation of a rectified flow Transformer model for text-to-music generation, from the paper 'Flux t… | 24 | 1709 | active |
| amazon-far/holosoma Holosoma is a Python framework for training and deploying reinforcement learning policies on humanoid robots, supporting locomotion and who… | 85 | 1702 | active |
| PaddlePaddle/PaddleVideo PaddleVideo is a video understanding toolkit built on PaddlePaddle, offering state-of-the-art models for action recognition, temporal actio… | 23 | 1702 | active |
| MLRun MLRun is an open-source MLOps and AI orchestration framework for building, training, deploying, and monitoring machine learning and generat… | 92 | 1697 | active |
| scvi-tools scvi-tools is a Python library for deep probabilistic modeling and analysis of single-cell and spatial omics data, built on PyTorch, PyTorc… | 92 | 1697 | stable |
| microsoft/Mage Mage is a family of lightweight 4B-parameter multimodal models from Microsoft, including Mage-VL for image and video understanding and Mage… | 55 | 1649 | active |
| JIA-Lab-research/ControlNeXt ControlNeXt is the official implementation of a controllable generation method for images and videos, built on Stable Diffusion XL, Stable … | 24 | 1646 | active |
| MoonshotAI/Kimi-Linear Kimi Linear is a hybrid linear attention architecture (Kimi Delta Attention, based on Gated DeltaNet) released by Moonshot AI with 48B-para… | 37 | 1607 | active |
| SonyResearch/micro_diffusion Official implementation of Sony Research's micro-budget approach to training large-scale text-to-image diffusion transformer models from sc… | 24 | 1598 | active |