huggingface/open-r1

Open R1 is a work-in-progress, fully open reproduction effort for DeepSeek-R1, providing scripts and Makefile commands to build missing pieces of the R1 training pipeline. It includes Python scripts for supervised fine-tuning (SFT) and group relative policy optimization (GRPO), plus a data generation script that can produce synthetic reasoning traces for model training and evaluation.

A status summary will appear after the next weekly refresh.