Output None as attention when layer is skipped

[jonghwanhyeon]

What does this PR do?

Fixes an out of index error caused by skipped layers when output_attentions is True in training mode on WavLM model.

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you read the contributor guideline,
  Pull Request section?
• Was this discussed/approved via a Github issue or the forum? Please add a link
  to it if that's the case.
• Did you make sure to update the documentation with your changes? Here are the
  documentation guidelines, and
  here are tips on formatting docstrings.
• Did you write any new necessary tests?

Who can review?

Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.

@sanchit-gandhi @patrickvonplaten

[amyeroberts]

Hi @jonghwanhyeon, thanks for adding this! Could you add a test which passes with this fix and fails on main?

cc @kamilakesbi for first review

[jonghwanhyeon]

Hi @jonghwanhyeon, thanks for adding this! Could you add a test which passes with this fix and fails on main?

cc @kamilakesbi for first review

Added a test case

[kamilakesbi]

Hi @jonghwanhyeon,

Thank you very much for working on this!

Could you provide your working environnement (by running transformers-cli env) as well as a minimal reproducer of the bug?

I've run the test you suggested both withlayer_outputs = (None,None,None) and layer_outputs = (None,None), and the test seems to pass in both cases ( I'm using transformers version 4.41.0.dev0).

Thanks!

cc @sanchit-gandhi

[jonghwanhyeon]

Oh, that is because the value of model.config.layerdrop is so low. Please check the following code and updated commits:

import torch

from transformers import WavLMConfig, WavLMModel

model = WavLMModel(WavLMConfig())
model.config.layerdrop = 1.0
model.train()
outputs = model(torch.randn(1, 8_000), output_attentions=True)

For your information, my environment is as below (with the modifications of (None, None, None)):

• transformers version: 4.41.0.dev0
• Platform: Linux-5.4.0-174-generic-x86_64-with-glibc2.35
• Python version: 3.11.9
• Huggingface_hub version: 0.22.2
• Safetensors version: 0.4.3
• Accelerate version: 0.29.2
• Accelerate config: not found
• PyTorch version (GPU?): 2.2.2+cu118 (True)
• Tensorflow version (GPU?): not installed (NA)
• Flax version (CPU?/GPU?/TPU?): not installed (NA)
• Jax version: not installed
• JaxLib version: not installed
• Using GPU in script?:
• Using distributed or parallel set-up in script?:

[kamilakesbi]

Ok, thanks a lot @jonghwanhyeon!

LGTM - I can reproduce the bug and your code seems to fix it. I've tried your test and it passes know.

CC @amyeroberts for a final check!

[amyeroberts]

Thanks for fixing and adding a test!