| |
Open Reproduction of DeepSeek-R1
Open R1 is a community-driven project to fully reproduce DeepSeek-R1, an advanced AI reasoning model, by building training pipelines for synthetic data generation, supervised fine-tuning, and reinforcement learning. As of May 2025, the project has completed Step 1 by releasing the Mixture-of-Thoughts dataset (350k reasoning traces) and OpenR1-Distill-7B model that matches DeepSeek's distilled model performance, with plans to progress toward replicating the pure RL pipeline and multi-stage training approaches.
Read Full Article →
← More Tech news