[Pyspeech] avoid .item call in feature_processing #4015

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Open

ynonaolga wants to merge 1 commit into pytorch:main from ynonaolga:export-D79451615

+22 −3

ynonaolga commented Aug 4, 2025

Summary:
tensor.item() causes CPU_GPU sync and degradates performance.
Avoid using .item() and rewrite mask_along_axis_iid to treat mask_value as a tensor

comparison https://fburl.com/mlhub/37l2ufrp
{F1980844270}

It increases SM U by 1%
{F1980890373}
{F1980890377}

Differential Revision: D79451615

ynonaolga requested a review from a team as a code owner

August 4, 2025 13:29

pytorch-bot bot commented Aug 4, 2025 •

edited

Loading

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/audio/4015

📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

❌ 8 New Failures, 2 Unrelated Failures

As of commit 02d6815 with merge base c819dbe ():

NEW FAILURES - The following jobs have failed:

.github/workflows/bandit.yml (gh)
.github/workflows/integration-test.yml (gh)
.github/workflows/lint.yml (gh)
.github/workflows/pr-labels.yml (gh)
.github/workflows/unittest-macos-cpu.yml (gh)
.github/workflows/unittest-windows-cpu.yml (gh)
.github/workflows/unittest-windows-gpu.yml (gh)
Build documentation / build / Build doc (gh)
At least one of the pre-conditions you specified did not hold

BROKEN TRUNK - The following jobs failed but were present on the merge base:

👉 Rebase onto the `viable/strict` branch to avoid these failures

Build Aarch64 Linux Wheels / pytorch/audio / build-wheel-py3_9-cuda-aarch6412_9-aarch64 (gh) (trunk failure)
ImportError: /lib64/libm.so.6: version GLIBC_2.29' not found (required by /usr/local/cuda/lib64/libnvshmem_host.so.3)`
Build Aarch64 Linux Wheels / pytorch/audio / upload / upload-wheel-py3_9-cuda-aarch6412_9-aarch64 (gh) (trunk failure)
Unable to download artifact(s): Artifact not found for name: pytorch_audio__3.9_cu129_aarch64

This comment was automatically generated by Dr. CI and updates every 15 minutes.

meta-cla bot added the CLA Signed label

Contributor

facebook-github-bot commented Aug 4, 2025

This pull request was exported from Phabricator. Differential Revision: D79451615

facebook-github-bot added the fb-exported label

ynonaolga added a commit to ynonaolga/audio that referenced this pull request


          [Pyspeech] avoid .item call in feature_processing (pytorch#4015)

e2e7642

Summary:

tensor.item() causes CPU_GPU sync and degradates performance.
Avoid using .item() and rewrite mask_along_axis_iid to treat mask_value as a tensor

comparison https://fburl.com/mlhub/37l2ufrp
{F1980844270}

It increases SM U by 1%
{F1980890373}
{F1980890377}

Differential Revision: D79451615

ynonaolga force-pushed the export-D79451615 branch from 83a6780 to e2e7642 Compare

August 4, 2025 13:36

ynonaolga added a commit to ynonaolga/audio that referenced this pull request


          [Pyspeech] avoid .item call in feature_processing (pytorch#4015)

2ba91d0

Summary:
Pull Request resolved: pytorch#4015

tensor.item() causes CPU_GPU sync and degradates performance.
Avoid using .item() and rewrite mask_along_axis_iid to treat mask_value as a tensor

comparison https://fburl.com/mlhub/37l2ufrp
{F1980844270}

It increases SM U by 1%
{F1980890373}
{F1980890377}

Differential Revision: D79451615

Contributor

facebook-github-bot commented Aug 4, 2025

This pull request was exported from Phabricator. Differential Revision: D79451615

ynonaolga force-pushed the export-D79451615 branch from e2e7642 to 2ba91d0 Compare

August 4, 2025 13:40

ynonaolga added a commit to ynonaolga/audio that referenced this pull request


          [Torchaudio] allow masked_value to be tensor (pytorch#4015)

6c3e0fd

Summary:

Add support for mask_value to be tensor: for example, if we want to fill with mean values of featured.

when mask_value is tensor masked_fill creates additional D2D copy, rewriting with torch.where to avoid it

Differential Revision: D79451615

ynonaolga force-pushed the export-D79451615 branch from 2ba91d0 to 6c3e0fd Compare

August 4, 2025 13:46

Contributor

facebook-github-bot commented Aug 4, 2025

This pull request was exported from Phabricator. Differential Revision: D79451615

ynonaolga added a commit to ynonaolga/audio that referenced this pull request


          [Torchaudio] allow masked_value to be tensor (pytorch#4015)

abef5cc

Summary:

Add support for mask_value to be tensor: for example, if we want to fill with mean values of featured.

when mask_value is tensor masked_fill creates additional D2D copy, rewriting with torch.where to avoid it

Differential Revision: D79451615

ynonaolga force-pushed the export-D79451615 branch from 6c3e0fd to abef5cc Compare

August 7, 2025 10:19

Contributor

facebook-github-bot commented Aug 7, 2025

This pull request was exported from Phabricator. Differential Revision: D79451615

ynonaolga added a commit to ynonaolga/audio that referenced this pull request


          [Torchaudio] allow masked_value to be tensor (pytorch#4015)

563efe3

Summary:

Add support for mask_value to be tensor: for example, if we want to fill with mean values of featured.

when mask_value is tensor masked_fill creates additional D2D copy, rewriting with torch.where to avoid it

Differential Revision: D79451615

ynonaolga force-pushed the export-D79451615 branch from abef5cc to 563efe3 Compare

August 7, 2025 11:22

Contributor

facebook-github-bot commented Aug 7, 2025

This pull request was exported from Phabricator. Differential Revision: D79451615

ynonaolga added a commit to ynonaolga/audio that referenced this pull request


          [Torchaudio] allow masked_value to be tensor (pytorch#4015)

e22d0f5

Summary:
Pull Request resolved: pytorch#4015

Add support for mask_value to be tensor: for example, if we want to fill with mean values of featured.

when mask_value is tensor masked_fill creates additional D2D copy, rewriting with torch.where to avoid it

Differential Revision: D79451615

ynonaolga force-pushed the export-D79451615 branch from 563efe3 to e22d0f5 Compare

August 7, 2025 11:26

ynonaolga added a commit to ynonaolga/audio that referenced this pull request


          [Torchaudio] allow masked_value to be tensor (pytorch#4015)

afa7d75

Summary:

Add support for mask_value to be tensor: for example, if we want to fill with mean values of featured.

when mask_value is tensor masked_fill creates additional D2D copy, rewriting with torch.where to avoid it

Differential Revision: D79451615

ynonaolga force-pushed the export-D79451615 branch from e22d0f5 to afa7d75 Compare

August 7, 2025 12:36

Contributor

facebook-github-bot commented Aug 7, 2025

This pull request was exported from Phabricator. Differential Revision: D79451615

ynonaolga added a commit to ynonaolga/audio that referenced this pull request


          [Torchaudio] allow masked_value to be tensor (pytorch#4015)

acb94a7

Summary:
Pull Request resolved: pytorch#4015

Add support for mask_value to be tensor: for example, if we want to fill with mean values of featured.

when mask_value is tensor masked_fill creates additional D2D copy, rewriting with torch.where to avoid it

Differential Revision: D79451615

ynonaolga force-pushed the export-D79451615 branch 2 times, most recently from acb94a7 to 3196abd Compare

August 8, 2025 09:37

ynonaolga added a commit to ynonaolga/audio that referenced this pull request


          [Torchaudio] allow masked_value to be tensor (pytorch#4015)

3196abd

Summary:

Add support for mask_value to be tensor: for example, if we want to fill with mean values of featured.

when mask_value is tensor masked_fill creates additional D2D copy, rewriting with torch.where to avoid it

Differential Revision: D79451615

Contributor

facebook-github-bot commented Aug 8, 2025

This pull request was exported from Phabricator. Differential Revision: D79451615

ynonaolga added a commit to ynonaolga/audio that referenced this pull request


          [Torchaudio] allow masked_value to be tensor (pytorch#4015)

7971f16

Summary:
Pull Request resolved: pytorch#4015

Add support for mask_value to be tensor: for example, if we want to fill with mean values of featured.

when mask_value is tensor masked_fill creates additional D2D copy, rewriting with torch.where to avoid it

Differential Revision: D79451615

ynonaolga force-pushed the export-D79451615 branch from 3196abd to 7971f16 Compare

August 8, 2025 09:41

ynonaolga added a commit to ynonaolga/audio that referenced this pull request


          [Torchaudio] allow masked_value to be tensor (pytorch#4015)

862bc0c

Summary:

Add support for mask_value to be tensor: for example, if we want to fill with mean values of featured.

when mask_value is tensor masked_fill creates additional D2D copy, rewriting with torch.where to avoid it

Differential Revision: D79451615

ynonaolga force-pushed the export-D79451615 branch from 7971f16 to 862bc0c Compare

August 8, 2025 10:46

Contributor

facebook-github-bot commented Aug 8, 2025

This pull request was exported from Phabricator. Differential Revision: D79451615

ynonaolga added a commit to ynonaolga/audio that referenced this pull request


          [Torchaudio] allow masked_value to be tensor (pytorch#4015)

717c8bd

Summary:

Add support for mask_value to be tensor: for example, if we want to fill with mean values of featured.

when mask_value is tensor masked_fill creates additional D2D copy, rewriting with torch.where to avoid it

Differential Revision: D79451615

ynonaolga force-pushed the export-D79451615 branch from 862bc0c to 717c8bd Compare

August 8, 2025 13:33

Contributor

facebook-github-bot commented Aug 8, 2025

This pull request was exported from Phabricator. Differential Revision: D79451615

ynonaolga added a commit to ynonaolga/audio that referenced this pull request


          [Torchaudio] allow masked_value to be tensor (pytorch#4015)

2541ba5

Summary:
Pull Request resolved: pytorch#4015

Add support for mask_value to be tensor: for example, if we want to fill with mean values of featured.

when mask_value is tensor masked_fill creates additional D2D copy, rewriting with torch.where to avoid it

Differential Revision: D79451615

ynonaolga force-pushed the export-D79451615 branch from 717c8bd to 2541ba5 Compare

August 8, 2025 13:36

ynonaolga added a commit to ynonaolga/audio that referenced this pull request


          [Torchaudio] allow masked_value to be tensor (pytorch#4015)

Summary:

Add support for mask_value to be tensor: for example, if we want to fill with mean values of featured.

when mask_value is tensor masked_fill creates additional D2D copy, rewriting with torch.where to avoid it

Differential Revision: D79451615

ynonaolga force-pushed the export-D79451615 branch from 2541ba5 to 9882707 Compare

August 8, 2025 13:38

Contributor

facebook-github-bot commented Aug 8, 2025

This pull request was exported from Phabricator. Differential Revision: D79451615

ynonaolga added a commit to ynonaolga/audio that referenced this pull request


          [Torchaudio] allow masked_value to be tensor (pytorch#4015)

7cb48af

Summary:

Add support for mask_value to be tensor: for example, if we want to fill with mean values of featured.

when mask_value is tensor masked_fill creates additional D2D copy, rewriting with torch.where to avoid it

Differential Revision: D79451615

ynonaolga force-pushed the export-D79451615 branch from 9882707 to 7cb48af Compare

August 8, 2025 14:37

Contributor

facebook-github-bot commented Aug 8, 2025

This pull request was exported from Phabricator. Differential Revision: D79451615

ynonaolga added a commit to ynonaolga/audio that referenced this pull request


          [Torchaudio] allow masked_value to be tensor (pytorch#4015)

3b4661e

Summary:

Add support for mask_value to be tensor: for example, if we want to fill with mean values of featured.

when mask_value is tensor masked_fill creates additional D2D copy, rewriting with torch.where to avoid it

Differential Revision: D79451615

ynonaolga force-pushed the export-D79451615 branch from 7cb48af to 3b4661e Compare

August 8, 2025 17:16


          [Torchaudio] allow masked_value to be tensor (pytorch#4015)

02d6815

Summary:
Pull Request resolved: pytorch#4015

Add support for mask_value to be tensor: for example, if we want to fill with mean values of featured.

when mask_value is tensor masked_fill creates additional D2D copy, rewriting with torch.where to avoid it

Differential Revision: D79451615

ynonaolga added a commit to ynonaolga/audio that referenced this pull request


          [Torchaudio] allow masked_value to be tensor (pytorch#4015)

78f5754

Summary:

Add support for mask_value to be tensor: for example, if we want to fill with mean values of featured.

when mask_value is tensor masked_fill creates additional D2D copy, rewriting with torch.where to avoid it

Differential Revision: D79451615

ynonaolga force-pushed the export-D79451615 branch from 3b4661e to 78f5754 Compare

August 8, 2025 17:19

Contributor

facebook-github-bot commented Aug 8, 2025

This pull request was exported from Phabricator. Differential Revision: D79451615

1 similar comment

Contributor

facebook-github-bot commented Aug 8, 2025

This pull request was exported from Phabricator. Differential Revision: D79451615

ynonaolga force-pushed the export-D79451615 branch from 78f5754 to 02d6815 Compare

August 8, 2025 17:22

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

CLA Signed fb-exported