Comprehensive collection of repositories for open source LLMs, fine-tuning, inference, RAG, agents, evaluation, and production deployment.
| Repository | Description | Stars |
|---|---|---|
| meta-llama/llama | Meta's LLaMA — foundational open LLM used by most OSS models |  |
| meta-llama/llama3 | LLaMA 3 — Meta's most capable open model family |  |
| mistralai/mistral-src | Mistral 7B — outperforms LLaMA 2 13B on most benchmarks |  |
| google/gemma | Google Gemma — lightweight open models from Google DeepMind |  |
| microsoft/phi | Microsoft Phi — small but powerful models |  |
| EleutherAI/gpt-neox | Large-scale autoregressive LM training framework |  |
| EleutherAI/gpt-j | GPT-J 6B — open source GPT-3-style model |  |
| THUDM/ChatGLM-6B | Open bilingual Chinese-English dialogue language model |  |
| THUDM/ChatGLM3 | ChatGLM3 — third generation bilingual LLM |  |
| BlinkDL/RWKV-LM | RWKV — RNN with transformer-level performance, runs on CPU |  |
| databrickslabs/dolly | Dolly — Databricks open instruction-following LLM |  |
| tiiuae/falcon-llm | Falcon LLM — TII open model, top of open LLM leaderboard |  |
| openlm-research/open_llama | Open reproduction of LLaMA using RedPajama dataset |  |
| mosaicml/llm-foundry | LLM training code for MosaicML's MPT models |  |
| deepseek-ai/DeepSeek-LLM | DeepSeek LLM — strong open model from China |  |
| deepseek-ai/DeepSeek-Coder | DeepSeek Coder — best open code model |  |
| Qwen/Qwen | Alibaba's Qwen — strong multilingual open LLM |  |
| 01-ai/Yi | Yi models — bilingual 6B/34B open LLMs |  |
| Repository | Description | Stars |
|---|---|---|
| ollama/ollama | Run LLaMA, Mistral, Gemma, Phi locally — one command |  |
| ggerganov/llama.cpp | LLM inference in C/C++ — CPU, Metal, CUDA, Vulkan |  |
| vllm-project/vllm | PagedAttention — 24x higher throughput than HuggingFace |  |
| lm-sys/FastChat | Training, serving, and evaluating LLM chatbots |  |
| nomic-ai/gpt4all | Run LLMs on your laptop — cross-platform desktop app |  |
| Mozilla-Ocho/llamafile | Run LLMs as a single executable file — no install needed |  |
| ggerganov/whisper.cpp | Port of OpenAI's Whisper in C++ — CPU-friendly |  |
| mlc-ai/mlc-llm | Run LLMs natively on any device — iOS, Android, browser |  |
| LostRuins/koboldcpp | Easy-to-use llama.cpp wrapper with KoboldAI UI |  |
| oobabooga/text-generation-webui | Gradio web UI for running LLMs locally |  |
| huggingface/text-generation-inference | HuggingFace production LLM serving toolkit |  |
| skypilot-org/skypilot | Run LLMs on any cloud — lowest cost, best GPU availability |  |
| Repository | Description | Stars |
|---|---|---|
| huggingface/peft | LoRA, QLoRA, prefix tuning, P-tuning — all in one library |  |
| huggingface/trl | RLHF, PPO, DPO, SFT — train LLMs with RL |  |
| hiyouga/LLaMA-Factory | Unified fine-tuning for 100+ LLMs — LoRA, full, QLoRA |  |
| artidoro/qlora | QLoRA — fine-tune 65B model on single GPU |  |
| tloen/alpaca-lora | Instruct-tune LLaMA on consumer hardware using LoRA |  |
| microsoft/DeepSpeed | ZeRO optimization — train massive models on limited hardware |  |
| Lightning-AI/lit-gpt | Fine-tune and pretrain LLMs — Lightning clean codebase |  |
| OpenAccess-AI-Collective/axolotl | Streamline fine-tuning of AI models — supports many formats |  |
| jondurbin/torchtune | PyTorch native fine-tuning library for LLMs |  |
| unslothai/unsloth | Fine-tune LLaMA, Mistral 2x faster — 70% less memory |  |
| huggingface/alignment-handbook | Robust recipes for alignment — SFT, DPO, RLHF |  |
| Repository | Description | Stars |
|---|---|---|
| run-llama/llama_index | Complete RAG framework — loaders, chunking, retrieval, synthesis |  |
| langchain-ai/langchain | RAG chains, document loaders, retrievers, vector stores |  |
| chroma-core/chroma | AI-native open-source vector database — embeddings storage |  |
| qdrant/qdrant | High-performance vector search engine in Rust |  |
| weaviate/weaviate | ML-first vector database — hybrid search, GraphQL |  |
| milvus-io/milvus | Cloud-native vector DB for billion-scale similarity search |  |
| pgvector/pgvector | Vector similarity search inside PostgreSQL |  |
| BerriAI/litellm | Call 100+ LLM APIs using the OpenAI format — unified interface |  |
| deepset-ai/haystack | End-to-end NLP framework for RAG and QA pipelines |  |
| embedchain/embedchain | Mem0 — memory layer for LLM applications |  |
| whyhow-ai/knowledge-graph-studio | GraphRAG — knowledge graph based RAG by Microsoft |  |
| Repository | Description | Stars |
|---|---|---|
| langchain-ai/langgraph | Build stateful multi-actor agent workflows as graphs |  |
| microsoft/autogen | Multi-agent conversations — agents collaborate to solve tasks |  |
| crewAIInc/crewAI | Role-based multi-agent framework — autonomous crews |  |
| Significant-Gravitas/AutoGPT | Autonomous AI agents — goal-driven with long-term memory |  |
| geekan/MetaGPT | Multi-agent framework — assigns roles like PM, engineer, QA |  |
| pydantic/pydantic-ai | Type-safe agent framework by Pydantic team |  |
| agno-agi/agno | Full-stack framework for building multi-modal AI agents |  |
| openai/swarm | Educational multi-agent orchestration framework by OpenAI |  |
| browser-use/browser-use | Make websites accessible to AI agents — browser automation |  |
| microsoft/TaskWeaver | Code-first agent framework — data analytics tasks |  |
| Repository | Description | Stars |
|---|---|---|
| dair-ai/Prompt-Engineering-Guide | Comprehensive guide — zero-shot, few-shot, CoT, ReAct, RAG |  |
| openai/openai-cookbook | OpenAI official examples and best practices |  |
| anthropics/anthropic-cookbook | Claude API examples — RAG, tools, vision, agents |  |
| brexhq/prompt-engineering | Brex's internal prompt engineering guide |  |
| microsoft/promptflow | Build, evaluate, and deploy LLM flows — Microsoft |  |
| hegelai/prompttools | Open source tools for testing and experimenting with prompts |  |
| ErikBjare/gptme | Personal AI assistant in your terminal — tool use, code execution |  |
| Repository | Description | Stars |
|---|---|---|
| open-webui/open-webui | Self-hosted ChatGPT-like UI — works with Ollama and OpenAI |  |
| lobehub/lobe-chat | Open source ChatGPT alternative — plugins, vision, TTS |  |
| danny-avila/LibreChat | Enhanced ChatGPT clone — multi-model, self-hostable |  |
| mckaywrigley/chatbot-ui | Open source ChatGPT UI — simple and clean |  |
| oobabooga/text-generation-webui | Gradio web UI for LLMs — GPTQ, GGUF, ExLlama support | |
| Yidadaa/ChatGPT-Next-Web | Well-designed cross-platform ChatGPT UI |  |
| lencx/ChatGPT | ChatGPT desktop app — macOS, Windows, Linux |  |
| Repository | Description | Stars |
|---|---|---|
| openai/evals | OpenAI framework for evaluating LLMs and prompts |  |
| EleutherAI/lm-evaluation-harness | 200+ NLP benchmarks — MMLU, HellaSwag, ARC, TruthfulQA |  |
| google/BIG-bench | Beyond Imitation Game — 200+ challenging tasks for LLMs |  |
| allenai/OLMO | Open Language Model — fully open weights, data, and training code |  |
| lm-sys/FastChat | Chatbot Arena — human preference evaluation of LLMs | |
| confident-ai/deepeval | LLM evaluation framework — like pytest for LLMs |  |
| brainlid/langchain | UpTrain — evaluate and improve LLM applications |  |
| explodinggradients/ragas | Evaluate RAG pipelines — faithfulness, relevance, recall |  |
| Repository | Description | Stars |
|---|---|---|
| anthropics/constitutional-ai | Anthropic's human feedback dataset for RLHF research |  |
| openai/safety-gym | Environments for developing safe RL algorithms |  |
| llm-attacks/llm-attacks | Universal adversarial attacks on aligned LLMs |  |
| guardrails-ai/guardrails | Add guardrails to LLM outputs — validation and correction |  |
| NVIDIA/NeMo-Guardrails | Programmable guardrails for LLM-based apps — NVIDIA |  |
| Repository | Description | Stars |
|---|---|---|
| qwopqwop200/GPTQ-for-LLaMa | GPTQ quantization for LLaMA — 4-bit inference |  |
| TimDettmers/bitsandbytes | 8-bit and 4-bit quantization — run large models on small GPUs |  |
| IST-DASLab/gptq | GPTQ — accurate post-training quantization for GPT models |  |
| casper-hansen/AutoAWQ | AWQ quantization — 4-bit with minimal accuracy loss |  |
| Repository | Description | Stars |
|---|---|---|
| bigcode-project/starcoder2 | StarCoder 2 — best open code model, 15B parameters |  |
| deepseek-ai/DeepSeek-Coder | DeepSeek Coder — top open code LLM | |
| Codium-ai/pr-agent | AI-powered PR reviews and code analysis |  |
| continuedev/continue | Open source AI code assistant — VS Code and JetBrains |  |
| cline/cline | Autonomous coding agent in VS Code — uses any LLM |  |
| aider-chat/aider | AI pair programming in terminal — edit code with LLMs |  |
| OpenDevin/OpenDevin | OpenHands — open source Devin AI software engineer |  |
| mentat-ai/mentat | AI coding assistant that understands your codebase |  |
| Repository | Description | Stars |
|---|---|---|
| haotian-liu/LLaVA | LLaVA — large language and vision assistant |  |
| ggerganov/llama.cpp | Supports LLaVA multimodal inference locally | |
| microsoft/LLaVA-Med | LLaVA for medical visual question answering |  |
| THUDM/CogVLM | Visual language model — strong image understanding |  |
| openai/CLIP | CLIP — connects text and images, zero-shot classification |  |
| salesforce/BLIP | Bootstrapping language-image pretraining for vision-language |  |