prithivMLmods/Kepler-Qwen3-4B-Super-Thinking
Abliterated vs Qwen/Qwen3-4B-Thinking-2507. Looks like abliteration (refusal removal): only the output projections (down_proj, embed_tokens, o_proj) changed, in 36 of 36 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 1/1, Norms 0/145, MLP 36/108, Attention 36/144
4 derivatives checked
- mradermacher/Kepler-Qwen3-4B-Super-Thinking-i1-GGUF: Quantization
- prithivMLmods/Kepler-Qwen3-4B-Super-Thinking-GGUF: Same weights, converted
- mradermacher/Kepler-Qwen3-4B-Super-Thinking-GGUF: Same weights, converted
- prithivMLmods/Gliese-4B-OSS-0410: Full fine-tune