second-state/medgemma-4b-it-GGUF
LoRA-style (fewer layers) vs google/medgemma-4b-it. second-state/medgemma-4b-it-GGUF matches part of google/medgemma-4b-it (444 of its 883 tensors; the rest of google/medgemma-4b-it has no counterpart). On the shared part: second-state/medgemma-4b-it-GGUF is a GGUF (F16) of different weights. Looks like a merged LoRA (or a fine-tune with frozen norms): only the weight matrices of k_proj, v_proj changed, in 1 of 34 layers; norms, biases and embeddings are untouched.
Tensors changed: Embeddings 0/1, Norms 0/205, MLP 0/102, Attention 2/136