huihui-ai/Qwen2.5-32B-Instruct-abliterated
Abliterated vs Qwen/Qwen2.5-32B-Instruct. Looks like abliteration (refusal removal): only the output projections (down_proj, embed_tokens, o_proj) changed, in 64 of 64 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 1/1, Norms 0/129, MLP 64/192, Attention 64/448, lm_head 0/1
5 derivatives checked
- mradermacher/Qwen2.5-32B-Instruct-abliterated-GGUF: Quantization
- mradermacher/Qwen2.5-32B-Instruct-abliterated-i1-GGUF: Quantization
- featherless-ai-quants/huihui-ai-Qwen2.5-32B-Instruct-abliterated-GGUF: Quantization
- tensorblock/Qwen2.5-32B-Instruct-abliterated-GGUF: Quantization
- kargeor/Qwen2.5-32B-Cyberpunk-Storyteller-v2: LoRA-style
1 other repos hold byte-for-byte identical weight files.