UserWarning: __floordiv__ is deprecated

[ypirkani]

Dear All,

I am trying to Fine-tune XLSR for Speech Recognition "facebook/wav2vec2-base" model for Urdu Language.

However when I run the trainer, I am getting the following warning:

/home/e-team/anaconda3/lib/python3.9/site-packages/transformers/models/wav2vec2/modeling_wav2vec2.py:704: UserWarning: __floordiv__ is deprecated, and its behavior will change in a future version of pytorch. It currently rounds toward 0 (like the 'trunc' function NOT 'floor'). This results in incorrect rounding for negative values. To keep the current behavior, use torch.div(a, b, rounding_mode='trunc'), or for actual floor division, use torch.div(a, b, rounding_mode='floor').
  return (input_length - kernel_size) // stride + 1

Also Consequently my WER is 1.00 which means training is not being done correctly.
Can anyone guide ? What changes to make and where?

Torch Version: 1.10.1+cu102
Python Version: 3.9.7
CUDA Version: 11.2
Tensorflow Version: 2.7.0

Best regards,
Yasir

[NielsRogge]

The same warning is shown for LayoutLMv2, see #14577

[github-actions]

This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.

Please note that issues that do not follow the contributing guidelines are likely to be ignored.

[anton-l]

Fixed by #15180

[gautierdag]

Running in the same issue for the DetrModel model:

transformers/src/transformers/models/detr/modeling_detr.py

Line 422 in c962c2a

dim_t = self.temperature ** (2 * (dim_t // 2) / self.embedding_dim)

Happy to submit a quick PR using the torch_int_div function added in #15180:

dim_t = self.temperature ** (2 * torch_int_div(dim_t, 2) / self.embedding_dim)

EDIT: I've submitted the PR (#15702) since it was a quick fix.