All Tools
S
Fine-tuningFreeOpen Source
SIMPLERL
Simple RL training for LLM reasoning
MIT
ABOUT
Many reasoning RL stacks bury the recipe under process rewards and extra models. SimpleRL keeps the loop small on verl so you can train math reasoning with outcome rewards and see whether the model actually improves.
INTEGRATION GUIDE
1. Fine-tune a math LLM with outcome-based RL on verl
2. Reproduce SimpleRL-Zoo checkpoints as a reasoning baseline
3. Compare simple RL against process-supervised recipes on the same data
4. Teach RL-for-reasoning with a readable training entrypoint
TAGS
rlreasoningmathverlresearchpytorch