verl-rl-training
orchestra-research904 installs
Train language models with reinforcement learning.
What it does
Guides the configuration and execution of reinforcement-learning post-training for large language models with verl.
When to use it
Use when implementing RLHF, GRPO, PPO, or other reinforcement-learning algorithms for large-language-model post-training at scale.
How to use it
It changes how the agent prepares training data, reward functions, configurations, and training runs for language-model reinforcement learning.
What you provide
- An existing project
Latest review
What you need · 7
Declares verl version 0.3.0 or newer as a dependency.
Declares torch version 2.0.0 or newer as a dependency.
Access · 2
This skill
mbridge
Execute
Installs the Megatron model bridge
ray
Execute
Starts a multi-node Ray cluster