Pinkstack/Parm-2-CoT-14B-16k-o1-QwQ
Abliterated vs microsoft/phi-4. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 40 of 40 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/1, Norms 0/81, MLP 40/40, Attention 40/40, lm_head 0/1
9 derivatives checked
- mradermacher/Parm-2-CoT-14B-16k-o1-QwQ-i1-GGUF: Quantization
- mradermacher/SuperThoughts-CoT-14B-16k-o1-QwQ-i1-GGUF: Quantization
- mradermacher/SuperThoughts-CoT-14B-16k-o1-QwQ-PyTorch-GGUF: Quantization
- mradermacher/SuperThoughts-CoT-14B-16k-o1-QwQ-GGUF: Quantization
- mradermacher/Parm-2-CoT-14B-16k-o1-QwQ-GGUF: Quantization
- Pinkstack/SuperThoughts-CoT-14B-16k-o1-QwQ-GGUF: Full fine-tune
- MaziyarPanahi/SuperThoughts-CoT-14B-16k-o1-QwQ-GGUF: Same weights, converted
- bunnycore/Phi-4-Model-Stock-v2: LoRA-style
- bunnycore/Phi-4-Model-Stock-v3: LoRA-style