💻Github Repo 🖨️arXiv Paper
The official SFT and PRM training data for "MedS3: Towards Medical Slow Thinking with Self-Evolved Soft Dual-sided Process Supervision"
The data is a synthetic dataset built from a 8k seed dataset, covering 16 datasets with 5 diverse medical tasks.
This dataset is evolved using Monte-Carlo Tree Search, aimed for provided SFT data and PRM data with high quality.
This dataset draws from a diverse array of text domains… See the full description on the dataset page:
https://huggingface.co/datasets/pixas/MedSSS-data.
