#431 · +86 · 1d ·GitHub
I implemented a transformer model from scratch using PyTorch, based on the paper Attention is All You Need. You can use my scripts to train your own billion or million parameter LLM using a single GPU.
Global
pth） - 支持跨不同 GPU 数量恢复（自动调整 step） - 支持 wandb 训练记录连续性（自动恢复同一个 run）
#710 · +48 · 1d ·GitHub
Comprehensive open-source library of AI research and engineering skills for any AI model. Package the skills and your claude code/codex/gemini agent will be an AI research agent with full horsepower. Maintained by Orchestra Research.
Local UI to run and train LLMs and diffusion models. 8, DeepSeek-V4, MiniMax-H3, Gemma 4, FLUX and more.
#976 · +32 · 1d ·GitHub
🤖 Agent 应用层 - ✅ Agent / Workflow - LangGraph、OpenAI Agents SDK - AutoGen、CrewAI、Pydantic AI - Dify、n8n、Flowise - ✅ 任务形态 - Research / Coding / Web Agent - Multi-Agent / Supervisor / Handoff - Computer Use / Browser Automation
> 第八章聚焦 GRPO、OPD、Search-R1 与 ReTool。更多 Agentic RL 算法、训练代码与实验实践，可以前往作者持续维护的另一个仓库 agentic-rl-lab；该仓库更新频率更高，会持续跟进新的算法与环境。
#2072 · +12 · 1d ·GitHub
🚀 An open-source, hands-on curriculum bridging the gap from basic RL concepts to LLM alignment, RLVR, and advanced Agentic systems.
verl/HybridFlow: A Flexible and Efficient RL Post-Training Framework
A comprehensive library to post-train foundation models
🍔 In addition, ms-swift integrates the latest training technologies, including Megatron parallelism techniques such as TP, PP, CP, EP to accelerate training, as well as numerous GRPO algorithm family reinforcement learning algorithms including: GRPO, DAPO, GSPO, SAPO, CISPO…
W&B Training (Serverless RL) is the first publicly available service for flexibly training models with reinforcement learning.
#3109 · +6 · 1d ·GitHub
json") to use FlashAttention-3 or xformers once installed.
Halo is an open-source framework built by White Circle for training large language and multimodal models
#6261 · +1 · 1d ·GitHub
Scalable RL in real-world settings — Agentic RL for terminal, GUI, SWE, and tool-call settings.
#6401 · +1 · 1d ·GitHub
8-Flash-Next is the foundation model developed by Qwen Team, Alibaba Group.
Meshy models every role of an RL run as an independent service. Samples flow between services through a single TransferQueue data plane, control flow is driven by data availability, and the whole topology is derived locally by each process from one declarative recipe.