google/gemma-3-4b-it-qat-q4_0-unquantized
Full fine-tune vs google/gemma-3-4b-it. Fine-tune: 90% of the weights changed, across 34 of 34 layers.
Tensors changed: Embeddings 1/4, Norms 201/316, MLP 102/210, Attention 136/352, other 1/1
10 derivatives checked
- unsloth/gemma-3-4b-it-qat-GGUF: Same weights, converted (fewer layers)
- bartowski/google_gemma-3-4b-it-qat-GGUF: Same weights, converted (fewer layers)
- mlabonne/gemma-3-4b-it-qat-abliterated-GGUF: Quantization (fewer layers)
- axiom-of-choice/gemma-3-4b-es-reasoning-gguf: Quantization (fewer layers)
- unsloth/gemma-3-4b-it-qat: Identical copy
- unsloth/gemma-3-4b-it-qat-unsloth-bnb-4bit: Quantization
- NexesQuants/google_gemma-3-4b-it-qat-q4_0-unquantized-iMat-NXS-GGUF: Same weights, converted (fewer layers)
- soob3123/amoral-gemma3-4B-v2-qat: LoRA-style
- mlabonne/gemma-3-4b-it-qat-abliterated: LoRA-style (fewer layers)
- aifeifei798/DrakIdol-Roleplayer-1.0: Identical copy
6 other repos hold byte-for-byte identical weight files.