ntnu-smil/Phi-4-multimodal-instruct_QA_NoImage_0325
LoRA-style vs microsoft/Phi-4-multimodal-instruct. Looks like a merged LoRA (or a fine-tune with frozen norms): only the weight matrices of speech changed, in 32 of 32 layers; norms, biases and embeddings are untouched.
Tensors changed: Embeddings 0/1342, Norms 0/65, MLP 128/320, Attention 128/320