franciscobdl/Granite-Estigia-MultilanguageV3
LoRA-style vs ibm-granite/granite-3.1-3b-a800m-instruct. Looks like a merged LoRA (or a fine-tune with frozen norms): only the weight matrices of k_proj, o_proj, q_proj, v_proj changed, in 32 of 32 layers; norms, biases and embeddings are untouched.
Tensors changed: Embeddings 0/1, other 0/96, Norms 0/65, Attention 128/128
1 derivatives checked
- mradermacher/Granite-Estigia-MultilanguageV3-GGUF: Same weights, converted