Skip to content

Add support for copy_ for plain layout and tensor core tiled layout #1791

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.

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

Merged
merged 5 commits into from
Feb 28, 2025

Conversation

jerryzh168
Copy link
Contributor

@jerryzh168 jerryzh168 commented Feb 27, 2025

Summary:
att, only support copy_ from AQT to another AQT with same metadata (shapes etc.)

Tested int4wo, int8wo, int8dq

This is to support the param.data.copy_ usage in vllm/sglang

Test Plan:
python test/dtypes/test_affine_quantized.py -k test_copy_

Reviewers:

Subscribers:

Tasks:

Tags:

Copy link

pytorch-bot bot commented Feb 27, 2025

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/ao/1791

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

⏳ No Failures, 1 Pending

As of commit 0885a49 with merge base 1ab1b77 (image):
💚 Looks good so far! There are no failures yet. 💚

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

@facebook-github-bot facebook-github-bot added the CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. label Feb 27, 2025
@jerryzh168 jerryzh168 requested a review from drisspg February 27, 2025 04:06
@jerryzh168 jerryzh168 added the topic: improvement Use this tag if this PR is an improvement (doesn't fit into any of the other categories) label Feb 27, 2025
Copy link
Contributor

@drisspg drisspg left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Looks good just the one note on the test

Summary:
att, only support copy_ from AQT to another AQT with same metadata (shapes etc.)

Tested int4wo, int8wo, int8dq

Test Plan:
python test/dtypes/test_affine_quantized.py -k test_copy_

Reviewers:

Subscribers:

Tasks:

Tags:
@jerryzh168 jerryzh168 merged commit 79e3366 into pytorch:main Feb 28, 2025
16 of 17 checks passed
@jithunnair-amd
Copy link
Collaborator

jithunnair-amd commented Feb 28, 2025

@jerryzh168 This broke ROCm CI: https://hud.pytorch.org/pytorch/ao/commit/79e3366e273dcc50e0300384a1d0d6b1cc8d5e1f
Noticed on torchao HUD: https://hud.pytorch.org/hud/pytorch/ao/3219318ac3aae9770338177eede24201424795f6/1?per_page=50&name_filter=linux.rocm.gpu.torchao&mergeLF=true

I'm not sure if we plan to follow the same approach as we do in the PyTorch project of reverting and relanding PRs that break ROCm CI. But if we do, please revert this PR and add the ciflow/rocm label to kick off ROCm CI on this PR so you can ensure it's green before relanding.

cc @msaroufim @supriyar @pruthvistony @petrex

drisspg pushed a commit that referenced this pull request Feb 28, 2025
…1791)

* Add support for copy_ for plain layout and tensor core tiled layout

Summary:
att, only support copy_ from AQT to another AQT with same metadata (shapes etc.)

Tested int4wo, int8wo, int8dq

Test Plan:
python test/dtypes/test_affine_quantized.py -k test_copy_

Reviewers:

Subscribers:

Tasks:

Tags:

* remove print

* add metadata mismatch test

* rebase and add float8

* cutlass int4 support
drisspg pushed a commit that referenced this pull request Feb 28, 2025
…1791)

* Add support for copy_ for plain layout and tensor core tiled layout

Summary:
att, only support copy_ from AQT to another AQT with same metadata (shapes etc.)

Tested int4wo, int8wo, int8dq

Test Plan:
python test/dtypes/test_affine_quantized.py -k test_copy_

Reviewers:

Subscribers:

Tasks:

Tags:

* remove print

* add metadata mismatch test

* rebase and add float8

* cutlass int4 support

stack-info: PR: #1804, branch: drisspg/stack/40
drisspg pushed a commit that referenced this pull request Mar 4, 2025
…1791)

* Add support for copy_ for plain layout and tensor core tiled layout

Summary:
att, only support copy_ from AQT to another AQT with same metadata (shapes etc.)

Tested int4wo, int8wo, int8dq

Test Plan:
python test/dtypes/test_affine_quantized.py -k test_copy_

Reviewers:

Subscribers:

Tasks:

Tags:

* remove print

* add metadata mismatch test

* rebase and add float8

* cutlass int4 support

stack-info: PR: #1804, branch: drisspg/stack/40
drisspg pushed a commit that referenced this pull request Mar 5, 2025
…1791)

* Add support for copy_ for plain layout and tensor core tiled layout

Summary:
att, only support copy_ from AQT to another AQT with same metadata (shapes etc.)

Tested int4wo, int8wo, int8dq

Test Plan:
python test/dtypes/test_affine_quantized.py -k test_copy_

Reviewers:

Subscribers:

Tasks:

Tags:

* remove print

* add metadata mismatch test

* rebase and add float8

* cutlass int4 support

stack-info: PR: #1804, branch: drisspg/stack/40
drisspg pushed a commit that referenced this pull request Mar 5, 2025
…1791)

* Add support for copy_ for plain layout and tensor core tiled layout

Summary:
att, only support copy_ from AQT to another AQT with same metadata (shapes etc.)

Tested int4wo, int8wo, int8dq

Test Plan:
python test/dtypes/test_affine_quantized.py -k test_copy_

Reviewers:

Subscribers:

Tasks:

Tags:

* remove print

* add metadata mismatch test

* rebase and add float8

* cutlass int4 support

stack-info: PR: #1804, branch: drisspg/stack/40
drisspg added a commit that referenced this pull request Mar 5, 2025
…1791) (#1804)

* Add support for copy_ for plain layout and tensor core tiled layout

Summary:
att, only support copy_ from AQT to another AQT with same metadata (shapes etc.)

Tested int4wo, int8wo, int8dq

Test Plan:
python test/dtypes/test_affine_quantized.py -k test_copy_

Reviewers:

Subscribers:

Tasks:

Tags:

* remove print

* add metadata mismatch test

* rebase and add float8

* cutlass int4 support

stack-info: PR: #1804, branch: drisspg/stack/40

Co-authored-by: Jerry Zhang <[email protected]>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. topic: improvement Use this tag if this PR is an improvement (doesn't fit into any of the other categories)
Projects
None yet
Development

Successfully merging this pull request may close these issues.

4 participants