tensorfiend/tunelm-gemma4-4b-grpo-pilot-20260928-50-best
LoRA-style vs tensorfiend/tunelm-gemma4-4b-sft-quality-20260927-86-best. Looks like a merged LoRA (or a fine-tune with frozen norms): only the weight matrices of q_proj, v_proj changed, in 42 of 42 layers; norms, biases and embeddings are untouched.
Tensors changed: MLP 0/606, Norms 0/484, other 0/261, Attention 66/719, Embeddings 0/6