Yvthyvq/Liujgoj-Cantonese-Qwen2.5-Omni-7B-SFT-V2
LoRA-style vs Yvthyvq/Liujgoj-Cantonese-Qwen2.5-Omni-7B-CPT-V2. Looks like a merged LoRA (or a fine-tune with frozen norms): only the weight matrices of down_proj, gate_proj, k_proj, o_proj, q_proj, up_proj, v_proj changed, in 4 of 32 layers; norms, biases and embeddings are untouched.
Tensors changed: other 0/549, MLP 84/480, Norms 0/347, Attention 112/1020, Embeddings 0/51, lm_head 0/1