671B parameter DeepSeek R1Backtracking featureAnswer Quickly feature which cuts thought traceSelf-correction featureAttention Sink featurewww.goodfire.aiWe have trained the first ever sparse autoencoders (SAEs) on the 671B parameter DeepSeek R1 model and open-sourced the SAEs.https://www.goodfire.ai/blog/under-the-hood-of-a-reasoning-modelGoodfire/DeepSeek-R1-SAE-l37 · Hugging FaceWe’re on a journey to advance and democratize artificial intelligence through open source and open science.https://huggingface.co/Goodfire/DeepSeek-R1-SAE-l37Deepseek R1 Qwen 1.5b every layer mlp transcoderEleutherAI/skip-transcoder-DeepSeek-R1-Distill-Qwen-1.5B-65k · Hugging FaceWe’re on a journey to advance and democratize artificial intelligence through open source and open science.https://huggingface.co/EleutherAI/skip-transcoder-DeepSeek-R1-Distill-Qwen-1.5B-65ksaeEleutherAI/sae-DeepSeek-R1-Distill-Qwen-1.5B-65k at mainWe’re on a journey to advance and democratize artificial intelligence through open source and open science.https://huggingface.co/EleutherAI/sae-DeepSeek-R1-Distill-Qwen-1.5B-65k/tree/main