CCYChongyanChen
commented
3 years ago The "checkpoint-2000000" can be downloaded at this link: "https://github.com/microsoft/Oscar/blob/master/VinVL_DOWNLOAD.md".
Have no idea with "tokenizer_name".The demo code does not have that arg. Maybe you could try without having this arg.
python oscar/run_vqa.py -j 4 --img_feature_dim 2054 --max_img_seq_length 50 --data_label_type mask --img_feature_type faster_r-cnn --data_dir vinvl/datasets/vqa --model_type bert --model_name_or_path vinvl/model_ckpts/vqa/base/checkpoint-2000000 --task_name vqa_text --do_train --do_lower_case --max_seq_length 128 --per_gpu_eval_batch_size 256 --per_gpu_train_batch_size 32 --learning_rate 5e-05 --num_train_epochs 25 --output_dir results --label_file datasets/vqa/cache/trainval_ans2label.pkl --save_epoch 1 --seed 88 --evaluate_during_training --logging_steps 4000 --drop_out 0.3 --weight_decay 0.05 --warmup_steps 0 --loss_type bce --img_feat_format pt --classifier linear --cls_hidden_scale 3 --txt_data_dir vinvl/datasets/vqa
jiuge-sun
Hi Oscar teams: I read your code and find that some parser_argument may loss, like "--model_name_or_path vinvl/model_ckpts/vqa/base/checkpoint-2000000" and "--tokenizer_name". Can you provide this file? Thanks.