vaghawan/gemma-4-31b-it-lora-last
Adapter vs unsloth/gemma-4-31B-it. vaghawan/gemma-4-31b-it-lora-last is a LoRA adapter, not full weights: rank 128, on down_proj, gate_proj, k_proj, o_proj, q_proj, up_proj, v_proj in 60 layers (410 adapted matrices). It modifies unsloth/gemma-4-31B-it at load time.