openrlhf-training
Train large language models with OpenRLHF.
What it does
Guides distributed reinforcement-learning and preference training of large language models with OpenRLHF.
When to use it
Use for large-model reinforcement learning when distributed GPU training, vLLM acceleration, or multiple training algorithms are needed.
How to use it
Give it a large-model training request; it uses OpenRLHF training workflows that save trained models into the project.
Uses
Good to know
- cannot be undone: xgboost; transformer_engine; flash_attn; pynvml
What you need · 8
The installation workflow uses Docker with the NVIDIA runtime and an NVIDIA PyTorch container; the hardware section specifies container version 25.02+.
The hardware requirements specify eight A100 40GB GPUs for a 7B model with the Hybrid Engine, or 48 A100 80GB GPUs for a 70B model.
Access · 7
This skill
xgboost
Removes installed xgboost (irreversible)
transformer_engine
Removes installed transformer_engine (irreversible)