wangzhang/gemma-4-31B-it-abliterated
LoRA-style vs google/gemma-4-31B-it. Looks like a merged LoRA (or a fine-tune with frozen norms): only the weight matrices of k_proj, o_proj, q_proj, v_proj changed, in 40 of 60 layers; norms, biases and embeddings are untouched.
Tensors changed: Embeddings 0/4, Norms 0/523, other 0/62, MLP 0/261, Attention 94/338
11 derivatives checked
- mradermacher/gemma-4-31B-it-abliterated-GGUF: Quantization (fewer layers)
- mradermacher/gemma-4-31B-it-abliterated-i1-GGUF: Quantization (fewer layers)
- wangzhang/gemma-4-31B-it-abliterated-GGUF: LoRA-style (fewer layers)
- lhca521/gemma-4-31B-it-abliterated-i1-GGUF: Quantization (fewer layers)
- NeuralNet-Hub/gemma-4-31B-it-abliterated-uncensored-NVFP4: Quantization
- zecanard/gemma-4-31B-it-uncensored-abliterix-MLX-8bit-mxfp8: Quantization
- zecanard/gemma-4-31B-it-uncensored-abliterix-MLX-8bit-int8-affine: Quantization
- lhca521/gemma-4-31B-it-abliterated-GGUF: Quantization (fewer layers)
- zecanard/gemma-4-31B-it-uncensored-abliterix-MLX-6bit-int6-affine: Quantization
- McG-221/gemma-4-31B-it-abliterated-8bit: Quantization
- zecanard/gemma-4-31B-it-uncensored-abliterix-MLX-4bit-mixed_4_6: Quantization
2 other repos hold byte-for-byte identical weight files.