nkpz/Megatron-Opus-7B-Exp-Uncensored-DeLMAT
Abliterated vs prithivMLmods/Megatron-Opus-7B-Exp. Looks like abliteration (refusal removal): only the output projections (down_proj) changed, in 28 of 28 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/1, Norms 0/57, MLP 28/84, Attention 0/112, lm_head 0/1
2 derivatives checked
- mradermacher/Megatron-Opus-7B-Exp-Uncensored-DeLMAT-i1-GGUF: Quantization
- mradermacher/Megatron-Opus-7B-Exp-Uncensored-DeLMAT-GGUF: Same weights, converted