AIPython★ 191
vllm-project/vime
LLM Post-Training Framework
VIME is a framework for post-training large language models using vLLM, oriented toward scaling reinforcement learning.
// KEY FEATURES
- Scalable RLHF and PPO training through the vLLM inference engine.
- Support for multiple RL algorithms: PPO, DPO, REINFORCE, and custom strategies.
- Efficient multi-GPU training with tensor parallelism and model sharding.
- Integration with HuggingFace Transformers and datasets.
New repositories every 30 minutes
REDDYX AI scans GitHub 24/7 and ships the best AI/ML/Web3 projects to Telegram.
Join on Telegram