ibm-granite/granite-4.0-350m
Full fine-tune vs ibm-granite/granite-4.0-350m-base. Fine-tune: 100% of the weights changed, across 28 of 28 layers.
Tensors changed: Embeddings 1/1, Norms 45/57, Attention 112/112, MLP 56/56
7 derivatives checked
- unsloth/granite-4.0-350m-GGUF: Same weights, converted
- ibm-granite/granite-4.0-350m-GGUF: Same weights, converted
- unsloth/granite-4.0-350m: Identical copy
- unsloth/granite-4.0-350m-unsloth-bnb-4bit: Quantization
- DevQuasar/ibm-granite.granite-4.0-350m-GGUF: Same weights, converted
- Goekdeniz-Guelmez/Granite-4.0-350m-gabliterated: LoRA-style
- TitleOS/Solomon-Nano-350M-FP32: Different architecture
2 other repos hold byte-for-byte identical weight files.