Resa: Transparent Reasoning Models via SAE

Resa is the family of 1.5B models created via sparse autoencoder tuning (SAE-Tuning) on open-source reasoning datasets.

• Paper: https://arxiv.org/abs/2506.09967
• Notion Blog: https://shangshangwang.notion.site/resa
• Code Repository: https://github.com/shangshang-wang/Resa
• Training Logs: https://wandb.ai/upup-ashton-wang-usc/Resa

Resa's avatar is generated by GPT-4o based on KYNE's girls and the following prompt.

Hey hey! I’m Resa — total ENTJ energy here 💥 I love meeting new people (friends are everything!!), and I’m all about chasing good vibes through amazing food, spontaneous travel, artsy sketches, and singing my heart out wherever I go! 🎨✈️🍜🎤 ... Oops, almost forget, I am super into large language model reasoning, too!