Goekdeniz-Guelmez/Granite-4.0-350m-gabliterated
LoRA-style vs ibm-granite/granite-4.0-350m. Looks like a merged LoRA (or a fine-tune with frozen norms): only the weight matrices of input_linear, o_proj, output_linear changed, in 26 of 28 layers; norms, biases and embeddings are untouched.
Tensors changed: Embeddings 0/1, Norms 0/57, Attention 26/112, MLP 52/56
2 derivatives checked
- mradermacher/Granite-4.0-350m-gabliterated-i1-GGUF: Quantization
- mradermacher/Granite-4.0-350m-gabliterated-GGUF: Same weights, converted