Bahushruth/Qwen3.6-35B-A3B-abliterated-v4
Abliterated (fewer layers) vs Qwen/Qwen3.6-35B-A3B. Bahushruth/Qwen3.6-35B-A3B-abliterated-v4 matches part of Qwen/Qwen3.6-35B-A3B (693 of its 1045 tensors; the rest of Qwen/Qwen3.6-35B-A3B has no counterpart). On the shared part: Looks like abliteration (refusal removal): only the output projections (down_proj, embed_tokens, o_proj, out_proj) changed, in 40 of 40 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 1/1, Attention 40/280, MLP 80/280, Norms 0/131, lm_head 0/1
1 derivatives checked
- Bahushruth/Qwen3.6-35B-A3B-abliterated-v4-GGUF: Same weights, converted