llmfan46/gemma-4-E2B-it-ultra-uncensored-heretic
Abliterated vs google/gemma-4-E2B-it. Looks like abliteration (refusal removal): only the output projections (o_proj) changed, in 7 of 35 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: MLP 0/585, Norms 0/453, other 0/240, Attention 7/727, Embeddings 0/6
9 derivatives checked
- llmfan46/gemma-4-E2B-it-ultra-uncensored-heretic-GGUF: Same weights, converted (fewer layers)
- mradermacher/gemma-4-E2B-it-ultra-uncensored-heretic-i1-GGUF: Quantization (fewer layers)
- minas2025/Gemma-4-E2B-Conversational-Uncensored: Quantization (fewer layers)
- JMingo/gemma-4-E2B-it-ultra-uncensored-heretic-UD-GGUF: Same weights, converted (fewer layers)
- mradermacher/gemma-4-E2B-it-ultra-uncensored-heretic-GGUF: Same weights, converted (fewer layers)
- taddymason/UMBASA-gemma-4-E2B-it-ultra-uncensored-heretic-GGUF: Same weights, converted (fewer layers)
- princepppp/gemma-4-E2B-it-ultra-uncensored-heretic-GGUF: Same weights, converted (fewer layers)
- minas2025/Gemma-4-E2B-Coder-Uncensored: Quantization (fewer layers)
- Dzluck/gemma-4-E2B-it-ultra-uncensored-heretic-GGUF: Same weights, converted (fewer layers)