You can deploy this model and start earning money today!

dyyyyyyyy/
FAPO-Reasoning-Dataset

Views

No views yet

FAPO: Flawed-Aware Policy Optimization for Efficient and Reliable Reasoning

Dataset used to train the FAPO-32B model.

train.parquet: data source from DAPO-Math-17K, with "\boxed{}" instruction format and duplicated 20 times. test.parquet: mixture of AIME24 (duplicated 32 times), AIME25 (duplicated 32 times), and GPQA-Diamond (duplicated 4 times).

Project Homepage: https://fapo-rl.github.io/ Code Implementation: https://github.com/volcengine/verl/tree/main/recipe/fapo Welcome to… See the full description on the dataset page: https://huggingface.co/datasets/dyyyyyyyy/FAPO-Reasoning-Dataset.