google/gemma-4-31B-it-qat-q4_0-unquantized
Full fine-tune vs google/gemma-4-31B-it. Fine-tune: 98% of the weights changed, across 60 of 60 layers.
Tensors changed: Embeddings 2/4, Norms 241/523, other 50/62, MLP 180/261, Attention 230/338
28 derivatives checked
- google/gemma-4-31B-it-qat-w4a16-ct: Quantization
- unsloth/gemma-4-31B-it-qat-GGUF: Not comparable yet
- google/gemma-4-31B-it-qat-q4_0-gguf: Quantization (fewer layers)
- lmstudio-community/gemma-4-31B-it-QAT-GGUF: Quantization (fewer layers)
- cyankiwi/gemma-4-31B-it-qat-AWQ-INT4: Quantization
- melcheikh/gemma-4-31B-it-qat-NVFP4-Blackwell: Quantization
- huihui-ai/Huihui-gemma-4-31B-it-qat-q4_0-unquantized-abliterated-GGUF: Abliterated (fewer layers)
- mradermacher/gemma-4-31B-it-qat-q4_0-unquantized-i1-GGUF: Quantization (fewer layers)
- mlx-community/gemma-4-31B-it-qat-OptiQ-4bit: Quantization
- JMingo/gemma-4-31B-it-qat-GGUF: Same weights, converted (fewer layers)
- unsloth/gemma-4-31B-it-qat-w4a16: Quantization
- TMA-1/gemma-4-31B-it-qat-GGUF: Not comparable yet
- Alberto-Codes/gemma-4-31B-it-fit24gib-GGUF: Quantization (fewer layers)
- mradermacher/gemma-4-31B-it-qat-q4_0-unquantized-GGUF: Quantization (fewer layers)
- xbill9/gemma-4-31B-it-qat-q4_0-w4a16-ct-text-emb4: Quantization
- unigilby/gemma-4-31B-it-qat-oQ8e: Quantization
- xbill9/gemma-4-31B-it-qat-q4_0-w4a16-ct-text: Quantization
- sakamakismile/Huihui-gemma-4-31B-it-qat-abliterated-MTP-NVFP4: Quantization
- melcheikh/gemma-4-31B-it-qat-NVFP4-mse-Blackwell: Quantization
- FreedomAISVR/Gemma-4-31B-it-QAT-NVFP4-GGUF: Quantization (fewer layers)
- wazimondo/gemma-4-31B-it-qat-Q4_0_V2: Quantization (fewer layers)
- aifeifei798/gemma-4-31B-Queen-it-qat-q4_0-unquantized: LoRA-style
- FreedomAISVR/Gemma-4-31B-it-QAT-MXFP4-GGUF: Quantization (fewer layers)
- OsaurusAI/gemma-4-31B-it-qat-JANG_4M: Quantization
- prithivMLmods/gemma-4-31B-it-qat-FP8: Quantization
- llmfan46/gemma-4-31B-it-qat-q4_0-unquantized-uncensored-heretic: Abliterated
- coder3101/gemma-4-31B-it-qat-q4_0-unquantized-heretic: Abliterated
- aifeifei798/Couture-Engine-31B-v1.0-lora: Adapter
2 other repos hold byte-for-byte identical weight files.