ibm-granite/granite-4.0-h-350m
Full fine-tune vs ibm-granite/granite-4.0-h-350m-base. Fine-tune: 100% of the weights changed, across 32 of 32 layers.
Tensors changed: Embeddings 1/1, Norms 89/93, other 186/196, MLP 64/64, Attention 16/16
9 derivatives checked
- ibm-granite/granite-4.0-h-350m-GGUF: Not comparable yet
- unsloth/granite-4.0-h-350m-GGUF: Not comparable yet
- mlx-community/granite-4.0-h-350m-8bit: Quantization
- unsloth/granite-4.0-h-350m: Identical copy
- DevQuasar/ibm-granite.granite-4.0-h-350m-GGUF: Not comparable yet
- mlx-community/granite-4.0-h-350m-5bit: Quantization
- Etherll/Tashkeel-350M-v2: Full fine-tune
- mlx-community/granite-4.0-h-350m-4bit: Quantization
- glogwa68/granite-4.0-h-350m-DISTILL-gemini-think: Full fine-tune
4 other repos hold byte-for-byte identical weight files.