Skip to content

[Bug]: sorting does not works in ASR train #1722

Description

@starcell

Describe the bug

When training with mult-GPU.
Setting sorting: random and sorting: ascending in hparams/*.yaml file show the same activity.
In case sorting setted ascending, batch dataset processed randomly in training.

Expected behaviour

When I set sorting=ascending, shorter data trained earlier.

To Reproduce

In yaml file, set as below
sorting: ascending

then ASR train!
It takes the same time when set sorting: random

Versions

v0.5.13

With v0.5.11, no problem appeared.
I can train with short data at first in v0.5.11.

Relevant log output

No response

Additional context

I tested and come to know there are some probelms in line 1000 of core.py.

1000 with tqdm(
train_set,
initial=self.step,
dynamic_ncols=True,
disable=not enable,
) as t:
for batch in t:

the train_set is sorted befor tqdm, but batch data is not sorted after tqdm().

Metadata

Metadata

Assignees

No one assigned

    Labels

    bugSomething isn't working

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions