guanxuyu/visual-jev-4b-answer-sft
Adapter vs Qwen/Qwen3-VL-4B-Instruct. guanxuyu/visual-jev-4b-answer-sft is a LoRA adapter, not full weights: rank 16, on down_proj, gate_proj, k_proj, o_proj, q_proj, up_proj, v_proj in 36 layers (252 adapted matrices). It modifies Qwen/Qwen3-VL-4B-Instruct at load time.