DeepSeek-R1-Distill-Llama-70B-SAE-l48

Model Information

A SAE (Sparse Autoencoder) for deepseek-ai/DeepSeek-R1-Distill-Llama-70B.

It is trained specifically on layer 48 of DeepSeek-R1-Distill-Llama-70B and achieves a final L0 of 36.

This model is used to decompose Llama's activations into interpretable features.

The SAE weights are released under Apache, however DeepSeek-R1-Distill-Llama-70B is to be used under Meta's Llama 3 License.

How to use

A Jupyter Notebook is provided to test the model

<a target="_blank" href="https://colab.research.google.com/github/qrsch/SAE/blob/main/SAE.ipynb"> <img src="https://colab.research.google.com/assets/colab-badge.svg" alt="Open In Colab", width="200px"/> </a>

Training

Our SAE was trained using LMSYS-Chat-1M dataset as well as PrimeIntellect/NuminaMath-QwQ-CoT-5M.

Acknowledgements

This release wouldn't have been possible without the work of Goodfire and Anthropic

A huge thank goes to runpod, who generously sponsored the compute for this run!

                                       .x+=:.                                                             
                                      z`    ^%                                                  .uef^"    
               .u    .                   .   <k                           .u    .             :d88E       
    .u@u     .d88B :@8c       .u       .@8Ned8"      .u          u      .d88B :@8c        .   `888E       
 .zWF8888bx ="8888f8888r   ud8888.   .@^%8888"    ud8888.     us888u.  ="8888f8888r  .udR88N   888E .z8k  
.888  9888    4888>'88"  :888'8888. x88:  `)8b. :888'8888. .@88 "8888"   4888>'88"  <888'888k  888E~?888L 
I888  9888    4888> '    d888 '88%" 8888N=*8888 d888 '88%" 9888  9888    4888> '    9888 'Y"   888E  888E 
I888  9888    4888>      8888.+"     %8"    R88 8888.+"    9888  9888    4888>      9888       888E  888E 
I888  9888   .d888L .+   8888L        @8Wou 9%  8888L      9888  9888   .d888L .+   9888       888E  888E 
`888Nx?888   ^"8888*"    '8888c. .+ .888888P`   '8888c. .+ 9888  9888   ^"8888*"    ?8888u../  888E  888E 
 "88" '888      "Y"       "88888%   `   ^"F      "88888%   "888*""888"     "Y"       "8888P'  m888N= 888> 
       88E                  "YP'                   "YP'     ^Y"   ^Y'                  "P'     `Y"   888  
       98>                                                                                          J88"  
       '8                                                                                           @%    
        `                                                                                         :"