search

OFA-Sys / OFA

Official repository of OFA (ICML 2022). Paper: OFA: Unifying Architectures, Tasks, and Modalities Through a Simple Sequence-to-Sequence Learning Framework
Apache License 2.0
2.39k stars 248 forks source link

"IndexError: list index out of range" on multiple GPU #421

Open bruceisme opened 10 months ago

bruceisme commented 10 months ago

Hi, when I try to run evaluate_vqa_beam.sh to generate test_dev's answers on multiple GPU, there always occurs error.

File "OFA/data/mm_data/vqa_gen_dataset.py", line 149, in getitem item = self.dataset[index] File "OFA/data/file_dataset.py", line 106, in getitem column_l = [dtype(column_l[col_id]) for col_id, dtype in zip(self.selected_col_ids, self.dtypes)] File "OFA/data/file_dataset.py", line 106, in column_l = [dtype(column_l[col_id]) for col_id, dtype in zip(self.selected_col_ids, self.dtypes)] IndexError: list index out of range

But when I run single GPUS_PER_NODE=1, the problem does not occur and could generate answers successfully.

bruceisme commented 10 months ago

By the way, the vqa_val.tsv which download from #68 could run evaluate_vqa_beam.sh on multiple GPU, but my fine only could run on single GPU.

vanloc1808 commented 5 months ago

Hi, I still met IndexError while fine-tuning at this line