Pinned Loading

1. Logic-RL-Lite Logic-RL-Lite Public

   Lightweight replication study of DeepSeek-R1-Zero. Interesting findings include "No Aha Moment", "Longer CoT ≠ Accuracy", and "Language Mixing in Instruct Models".

   Python 68

2. DeepEnlighten DeepEnlighten Public

   Pure RL to post-train base models for social reasoning capabilities. Lightweight replication of DeepSeek-R1-Zero with Social IQa dataset.

   Python 22