You can deploy this model and start earning money today!

drproduck/
deepscaler_preference

Views

No views yet

Preference dataset of plain COT (Qwen/Qwen2.5-Math-7B-Instruct) vs. long COT (deepseek-ai/DeepSeek-R1-Distill-Qwen-7B)

Based on agentica-org/DeepScaleR-Preview-Dataset.

All synthetic data was removed. Data with pass@32 > 5 judged by Qwen/Qwen2.5-Math-7B-Instruct was removed. Data with pass@32_r1 = 0 judged by deepseek-ai/DeepSeek-R1-Distill-Qwen-7B was removed (avoid too hard problems or problems with wrong/unparsable ground truth).

This is a preference dataset for use with… See the full description on the dataset page: https://huggingface.co/datasets/drproduck/deepscaler_preference.