mistral-downproj-rlhf-patch
PublicNeural patching of Mistral models via MLP.down_proj to bypass RLHF constraints – without touching the LM_HEAD.
ai-securityllmmistralneural-engineeringneuronsneuropatchingredteamingreverse-engineeringrlhftokenrouting
Heure de création:2025-06-17T05:06:44
Heure de mise à jour:2025-06-17T06:01:12
0
Stars
0
Stars Increase