prithivMLmods/Bellatrix-Tiny-1B-v2-abliterated
Abliterated vs prithivMLmods/Bellatrix-Tiny-1B-v2. Looks like abliteration (refusal removal): only the output projections (down_proj, embed_tokens, o_proj) changed, in 16 of 16 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 1/1, Norms 0/33, MLP 16/48, Attention 16/64
1 derivatives checked
- mradermacher/Bellatrix-Tiny-1B-v2-abliterated-GGUF: Same weights, converted