KatterMobile/Tinfield-1-EXL3
LoRA-style vs badtheorylabs/Tinfield-1. Looks like a merged LoRA (or a fine-tune with frozen norms): only the weight matrices of in_proj_a, in_proj_b, shared_expert_gate changed, in 48 of 48 layers; norms, biases and embeddings are untouched.
Tensors changed: Norms 0/302, other 0/7, Attention 72/357, MLP 48/274, Embeddings 0/4