s444415/donut
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

random choice for sequences in streaming dataset

Browse Source
This commit is contained in:
michal.kozlowski 2023-03-17 14:38:19 +01:00
parent 30369a7885
commit dd39efb763
1 changed files with 1 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
utils/donut_dataset_stream.py
View File

@ -68,7 +68,7 @@ class DonutDatasetStream:
pixel_values = pixel_values.squeeze()
# targets
target_sequence = self.gt_token_sequences # can be more than one, e.g., DocVQA Task 1
target_sequence = random.choice(self.gt_token_sequences) # can be more than one, e.g., DocVQA Task 1
input_ids = self.processor.tokenizer(
target_sequence,
add_special_tokens=False,