arguments/model.yaml

model_id: qwen3-vl-2b
model_local_path: Qwen/Qwen3-VL-2B-Instruct
use_doc_attention: false
use_full_attention: false
use_multilayer_features: false
use_up_attention: false