Shypd.aiThe largest open index of AI toolsverl is an open-source reinforcement learning (RL) training library for large language models (LLMs), initiated by ByteDance Seed team. It offers a flexible and efficient framework for post-training LLMs.
Trending
About
verl, short for Volcano Engine Reinforcement Learning for LLMs, is an open-source RL training library designed for large language models. Initiated by ByteDance Seed team and maintained by the verl community, it provides a flexible, efficient, and production-ready framework for post-training. Key features include easy extension of diverse RL algorithms through its hybrid-controller programming model, seamless integration with existing LLM infrastructures like FSDP and Megatron-LM, and flexible device mapping for efficient resource utilization. verl is known for its state-of-the-art throughput and efficient actor model resharding with 3D-HybridEngine, significantly reducing memory redundancy and communication overhead. It supports various RL algorithms such as PPO, GRPO, and DAPO, and is compatible with popular Hugging Face and Modelscope Hub models.
Capabilities
Pricing & Plans
Open Source
Free
FAQs
Trending
Also listed in
