To train the model, some configuration should be as follows: python "your path"/clevr_qwen2vl.py
--data_root "your path"/custom_dataset
--output_dir qwen2vl_clevr_e5_bs1_lr2e5
--do_train
--epochs 5
--batch_size 1
--grad_accum 4
--lr 2e-5
--submission_path submission_e5_bs1_lr2e5_t01_ens5.csv
--ensemble_size 5
--gen_max_new_tokens 128
--gen_temperature 0.1
--gen_top_p 0.8