NAMAA-Space/Qari-OCR-0.1-VL-2B-Instruct
Adapter vs unsloth/Qwen2-VL-2B-Instruct-unsloth-bnb-4bit. NAMAA-Space/Qari-OCR-0.1-VL-2B-Instruct is a LoRA adapter, not full weights: rank 16, on down_proj, fc1, fc2, gate_proj, k_proj, o_proj, proj, q_proj… in 32 layers (324 adapted matrices). It modifies unsloth/Qwen2-VL-2B-Instruct-unsloth-bnb-4bit at load time.