slime-rl-training
orchestra-research893 installs
Train language models with reinforcement learning.
What it does
Guides reinforcement-learning post-training of language models with slime, including training setup and custom generation workflows.
When to use it
Use when training needs Megatron-LM with SGLang inference or custom data generation and buffering.
How to use it
Give it a model checkpoint and JSONL training data; it configures and runs reinforcement-learning training with checkpoint saving.
What you provide
- Your files
Latest review
What you need · 6
The recommended installation path requires Docker capable of launching a container with GPU access.
Training requires GPUs; the quick-start configuration assigns four GPUs to the actor and four to rollout.
Access · 3
This skill
slimerl/slime:latest
Execute
Starts the slime GPU container
slime
Execute
Installs slime inside the container