Add support for bmm and to for fbgemm Tensor
#2337
Open
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/ao/2337
Note: Links to docs will display an error until the docs builds have been completed. ❌ 1 New FailureAs of commit 06211ee with merge base 4235837 ( NEW FAILURE - The following job has failed:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
facebook-github-bot
added
the
CLA Signed
jerryzh168
added
the
topic: improvement
drisspg
reviewed
Jun 8, 2025
drisspg
reviewed
Jun 8, 2025
|
|
||
| # not used | ||
| num_tokens = torch.empty([input_tensor.size(0)], device=input_tensor.device) | ||
| xq, x_scale = torch.ops.fbgemm.quantize_fp8_per_row( |
There was a problem hiding this comment.
This ot use num_tokens feels weird, maybe make an issue on fbgemm? or update the op to not need
There was a problem hiding this comment.
yeah I checked with @jiawenliu64 and this arg is indeed only used in internal use cases, he was recommending to use the triton op, although I found the triton op is a bit slower, maybe it requires some tuning. I'll double check
drisspg
reviewed
Jun 8, 2025
drisspg
reviewed
Jun 8, 2025
jerryzh168
force-pushed
the
add-bmm
branch
3 times, most recently
from
59bc6cf to
a02edc9
Compare
Summary: att, this PR adds support for running quantized bmm, the quantized bmm kernel for int4 and fp8 (with dynamic activation quantization) requires transpose of weights in order to run, so added transpose_input to the convert function to transpose the weights first Test Plan: python test/dtypes/test_fbgemm_fp8.py -k test_bmm python test/dtypes/test_fbgemm_int4.py -k test_bmm Reviewers: Subscribers: Tasks: Tags:
drisspg
approved these changes
Jun 9, 2025
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Summary:
att, this PR adds support for running quantized bmm, the quantized bmm kernel for int4 and fp8 (with dynamic activation quantization) requires transpose of weights in order to run, so added transpose_input to the convert function to transpose the weights first
Test Plan:
python test/dtypes/test_fbgemm_fp8.py -k test_bmm
python test/dtypes/test_fbgemm_int4.py -k test_bmm
Reviewers:
Subscribers:
Tasks:
Tags: