[Model Evaluation] Encountering IndexError When Running run_umt_pretraining.py on a New Dataset

[kaizhuanren]

I am attempting to evaluate your model's performance on a brand new dataset by running the run_umt_pretraining.py script. I have only modified the datapath and batch_size parameters to match my dataset and hardware specifications, while keeping all other parameters at their default values. However, I encountered an error during execution. The specific error message and stack trace are as follows:
Traceback (most recent call last):
File "unmasked_teacher-main/single_modality/run_umt_pretraining.py", line 346, in
main(opts)
File "unmasked_teacher-main/single_modality/run_umt_pretraining.py", line 303, in main
train_stats = train_one_epoch(
File "unmasked_teacher-main/single_modality/engines/engine_for_pretraining_umt.py", line 83, in train_one_epoch
targets_clip_vis = norm_clip[~clip_bool_masked_pos].reshape(K, B, -1, C_CLIP)
IndexError: The shape of the mask [1, 8, 1568] at index 2 does not match the shape of the indexed tensor [1, 8, 3136, 512] at index 2

norm_clip is theoretically the output after feature extraction. Why does it not match with clip_bool_masked_pos formed by attn.shape

[Andy1621]

Hi! Please give the full script you used.

[kaizhuanren]

Uploading pretraining_output.log…