Skip to content

Convert scalar to tensor before quantizer annoate#2958

Closed
cccclai wants to merge 2 commits intogh/cccclai/2/basefrom
gh/cccclai/2/head
Closed

Convert scalar to tensor before quantizer annoate#2958
cccclai wants to merge 2 commits intogh/cccclai/2/basefrom
gh/cccclai/2/head

Conversation

@cccclai
Copy link
Copy Markdown
Contributor

@cccclai cccclai commented Apr 10, 2024 •
edited
Loading

@pytorch-bot
Copy link
Copy Markdown

pytorch-bot Bot commented Apr 10, 2024 •
edited
Loading

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/2958

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

❌ 2 New Failures, 1 Unrelated Failure

As of commit de04834 with merge base e733f2d (image):

NEW FAILURES - The following jobs have failed:

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

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

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 Apr 10, 2024
@cccclai cccclai mentioned this pull request Apr 10, 2024
Closed
@facebook-github-bot
Copy link
Copy Markdown
Contributor

facebook-github-bot commented Apr 10, 2024

This pull request was exported from Phabricator. Differential Revision: D55946527

cccclai added a commit that referenced this pull request Apr 10, 2024
@cccclai
Convert scalar to tensor before quantizer annoate
b66e850 
Differential Revision: [D55946527](https://our.internmc.facebook.com/intern/diff/D55946527/)

ghstack-source-id: 221922774
Pull Request resolved: #2958
@cccclai cccclai requested a review from chunit-quic April 10, 2024 00:02
@cccclai cccclai added the partner: qualcomm For backend delegation, kernels, demo, etc. from the 3rd-party partner, Qualcomm label Apr 10, 2024
@chiwwang
Copy link
Copy Markdown
Contributor

chiwwang commented Apr 10, 2024

@shewu-quic @haowhsu-quic

@facebook-github-bot
Copy link
Copy Markdown
Contributor

facebook-github-bot commented Apr 10, 2024

This pull request was exported from Phabricator. Differential Revision: D55946527

cccclai added a commit that referenced this pull request Apr 10, 2024
@cccclai
Convert scalar to tensor before quantizer annoate
e992555 
Pull Request resolved: #2958


ghstack-source-id: 222058912
@exported-using-ghexport

In case like
```
def f(x, 2):
    return x * 2
```
Since 2 is a scalar, we'd need to convert them to tensor so quantizer knows how to handle

Differential Revision: [D55946527](https://our.internmc.facebook.com/intern/diff/D55946527/)
@shewu-quic
Copy link
Copy Markdown
Collaborator

shewu-quic commented Apr 11, 2024

It seems to convert the scalar node of the binary op into a tensor node and allow the node to be quantized.
Maybe the following two passes are also to achieve this thing.

executorch/backends/qualcomm/passes/annotate_and_quant_scalar.py

Line 19 in d761f99

class AnnotateAndQuantScalar(ExportPass):

executorch/backends/qualcomm/passes/convert_binary_op_with_scalar.py

Line 15 in d761f99

class ConvertBinaryOpsWithScalar(ExportPass):

shewu-quic
shewu-quic reviewed Apr 11, 2024
View reviewed changes
Comment thread backends/qualcomm/quantizer/quantizer.py
if non_const_arg is None or const_arg is None:
continue

tensor_constant = torch.tensor([const_arg], dtype=torch.float32)
Copy link
Copy Markdown
Collaborator

@shewu-quic shewu-quic Apr 11, 2024

Choose a reason for hiding this comment

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

Maybe we will want to infer data type from data.

@cccclai
Copy link
Copy Markdown
Contributor Author

cccclai commented Apr 12, 2024

It seems to convert the scalar node of the binary op into a tensor node and allow the node to be quantized. Maybe the following two passes are also to achieve this thing.

executorch/backends/qualcomm/passes/annotate_and_quant_scalar.py

Line 19 in d761f99

class AnnotateAndQuantScalar(ExportPass):

executorch/backends/qualcomm/passes/convert_binary_op_with_scalar.py

Line 15 in d761f99

class ConvertBinaryOpsWithScalar(ExportPass):

Oh I see what's going. The convert_binary_op_with_scalar pass runs after quantization process, however we need to run this pass before the quantizer starts annotating, otherwise the scalar may not have an accurate quantization params.

@cccclai
Copy link
Copy Markdown
Contributor Author

cccclai commented Jun 6, 2025

Stale PR

1 similar comment
@cccclai
Copy link
Copy Markdown
Contributor Author

cccclai commented Jun 6, 2025

Stale PR

@cccclai cccclai closed this Jun 6, 2025
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@chunit-quic chunit-quic Awaiting requested review from chunit-quic

1 more reviewer

@shewu-quic shewu-quic shewu-quic left review comments

Reviewers whose approvals may not affect merge requirements

Assignees

No one assigned

Labels

CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. fb-exported partner: qualcomm For backend delegation, kernels, demo, etc. from the 3rd-party partner, Qualcomm

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

4 participants

@cccclai @facebook-github-bot @chiwwang @shewu-quic