Stars
Fully local web research and report writing assistant
Awesome ModelContextProtocol resources - A curated list of MCP resources
Build resilient language agents as graphs.
[ICML 2024] LLMCompiler: An LLM Compiler for Parallel Function Calling
EasyR1: An Efficient, Scalable, Multi-Modality RL Training Framework based on veRL
**LightAgent** is an extremely lightweight active Agentic Framework with memory, tools , and a Tree of Thought (`ToT`). It supports swarm-like multi-agent collaboration, automated tool generation, …
verl: Volcano Engine Reinforcement Learning for LLMs
Use PEFT or Full-parameter to CPT/SFT/DPO/GRPO 500+ LLMs (Qwen3, Qwen3-MoE, Llama4, InternLM3, DeepSeek-R1, ...) and 200+ MLLMs (Qwen2.5-VL, Qwen2.5-Omni, Qwen2-Audio, Ovis2, InternVL3, Llava, GLM4…
分享一些好用的 Dify DSL 工作流程，自用、学习两相宜。 Sharing some Dify workflows.
PIKE-RAG: sPecIalized KnowledgE and Rationale Augmented Generation
RAGFlow is an open-source RAG (Retrieval-Augmented Generation) engine based on deep document understanding.
LLMs can generate feedback on their work, use it to improve the output, and repeat this process iteratively.
Efficient Triton Kernels for LLM Training
RAGChecker: A Fine-grained Framework For Diagnosing RAG
Training and serving XLM-RoBERTa for named entity recognition on custom dataset with PyTorch.
Reference implementation for DPO (Direct Preference Optimization)
Implementation of "Decoding-time Realignment of Language Models", ICML 2024.
A generative speech model for daily dialogue.
Official inference library for Mistral models
InternLM-XComposer2.5-OmniLive: A Comprehensive Multimodal System for Long-term Streaming Video and Audio Interactions
ChatPilot: Chat Agent Web UI，实现Chat对话前端，支持Google搜索、文件网址对话（RAG）、代码解释器功能，复现了Kimi Chat(文件，拖进来；网址，发出来)。
OneDiff: An out-of-the-box acceleration library for diffusion models.
Medusa: Simple Framework for Accelerating LLM Generation with Multiple Decoding Heads
Unified Efficient Fine-Tuning of 100+ LLMs & VLMs (ACL 2024)