verl-rl-training - RL Training for LLMs with verl
Train large language models with reinforcement learning algorithms using the verl library
Tags
Updated: 2026-06-30Capabilities
Typical Inputs
Typical Outputs
What this skill does
- implement RLHF
- configure GRPO
- run PPO
- train distributed models
- train vision-language models
- configure multi-turn tools
Inputs
- training dataset
- base model
- configuration file
- reward function
Outputs
- training logs
- model checkpoint
- performance metrics
- evaluation results
Requirements
- GPU cluster
- PyTorch
- Ray
- vLLM
- transformers
- verl