google/gemma-4-E2B-it
Full fine-tune vs google/gemma-4-E2B. Fine-tune: 90% of the weights changed, across 35 of 35 layers.
Tensors changed: MLP 105/585, Norms 177/453, other 105/240, Attention 100/727, Embeddings 4/6
247 derivatives checked
- unsloth/gemma-4-E2B-it-GGUF: Not comparable yet
- bartowski/google_gemma-4-E2B-it-GGUF: Same weights, converted (fewer layers)
- lmstudio-community/gemma-4-E2B-it-MLX-4bit: Quantization
- lmstudio-community/gemma-4-E2B-it-MLX-8bit: Quantization
- lmstudio-community/gemma-4-E2B-it-MLX-5bit: Quantization
- lmstudio-community/gemma-4-E2B-it-MLX-6bit: Quantization
- lmstudio-community/gemma-4-E2B-it-GGUF: Quantization (fewer layers)
- unsloth/gemma-4-E2B-it-unsloth-bnb-4bit: Quantization
- unsloth/gemma-4-E2B-it: Identical copy
- ggml-org/gemma-4-E2B-it-GGUF: Same weights, converted (fewer layers)
- cyankiwi/gemma-4-E2B-it-AWQ-INT4: Quantization
- ente-ai/gemma-4-E2B-it-GGUF: Not comparable yet
- google/gemma-4-E2B-it-qat-q4_0-unquantized: Full fine-tune
- mlx-community/gemma-4-e2b-it-4bit: Quantization
- batiai/gemma-4-E2B-it-GGUF: Quantization (fewer layers)
- TrevorJS/gemma-4-E2B-it-uncensored: Abliterated
- unsloth/gemma-4-E2B-it-NVFP4: Quantization
- yemeni-ai-lab/gemma-4-e2b-yemeni-arabic-assistant-GGUF: LoRA-style (fewer layers)
- AtomicChat/gemma-4-E2B-it-GGUF: Quantization (fewer layers)
- huihui-ai/Huihui-gemma-4-E2B-it-abliterated: Abliterated
- google/gemma-4-E2B-it-qat-mobile-transformers: Quantization
- unsloth/gemma-4-E2B-it-qat-q4_0-unquantized: Full fine-tune
- NobodyWho/Google_Gemma4-E2B-GGUF: Same weights, converted (fewer layers)
- giladgd/gemma-4-E2B-it-GGUF: Same weights, converted (fewer layers)
- aisingapore/Gemma-SEA-LION-v4.5-E2B-IT: Full fine-tune
- atakankutalp/gemma-4-E2B-it-heretic: Abliterated
- arjhinety/onebee-gf-dpo-v1-4epoch: Partial fine-tune
- arjhinety/onebee-gf-distill-v1: Partial fine-tune
- arjhinety/onebee-gf-dpo-v1-scale: Partial fine-tune
- WhiskyAKM/Gemma-4-E2B-it-qat-GGUF: Full fine-tune (fewer layers)
99 other repos hold byte-for-byte identical weight files.