grpo

A framework for agentic tool use training with reinforcement learning

从 MiniMind 源码读起,再延伸到现代大模型技术体系的中文学习笔记。主线逐行精读预训练 / SFT / DPO / PPO / GRPO 与训练机制;附录 17 篇进阶卷覆盖量化、投机解码、RLHF 全景、模型代际史等 MiniMind 没涉及、但进阶绕不开的主题。

Your AI intranet: network the computers you already own for inference and training.

NEWTON: Agentic Planning for Physically Grounded Video Generation

DataClaw: Agentic Tailoring Multimodal Data from Raw Streams — coming soon (code, weights, dataset & DataClaw-val upon acceptance).

FlowSteer: agents designing agentic workflows via reinforced progressive canvas editing.

An implementation of GRPO for Unsloth's VLMs training

A travel agent based on Qwen2.5, fine-tuned by SFT + DPO/PPO/GRPO using traveling question-answer dataset, a mindmap can be output using the response. A RAG system is build upon the tuned qwen2, using Prompt-Template + Tool-Use + Chroma embedding database + LangChain

R1-Track: Direct Application of MLLMs to Visual Object Tracking via Reinforcement Learning.

🏆Winning Project | ModelGate is a contract-aware AI control plane that ingests customer contracts, extracts SLA/privacy/routing constraints, and generates an OpenAI-compatible endpoint that automatically routes every request to the optimal model. Simple queries go to cheap models. Complex queries go to premium ones.

[CVPR 2026 Highlight] ReAG: Reasoning-Augmented Generation for Knowledge-based Visual Question Answering

A fine-tuning SDK — any open model, any harness, any method. 12 training methods behind one argument; pure-stdlib core.

[NeurIPS 2025] Official repository of the paper "Unlocking Aha Moments via Reinforcement Learning: Advancing Collaborative Visual Comprehension and Generation"

A lightweight post-training framework for LLMs and VLMs. 51 algorithms, 38 verified models. Scales with DeepSpeed, vLLM, and Ray.

Multi-node distributed LLM training framework

LLM finetuning for Sudoku solving

Adversarial LLM arena using GRPO + SFT to train a model that generates questions hard LLMs can't solve

Advanced LLM fine-tuning techniques: SFT (LoRA, QLoRA, DoRA, P-/Prefix-Tuning), GRPO, DPO, ORPO, KTO & PPO; composable correctness/format rewards + LLM-as-a-Judge evals (DeepEval, Evidently AI) across math, multi-hop, medical & general QA on Llama 3, Mistral, Phi-4, Gemma & Qwen3. Built on TRL, PEFT & Unsloth.