Refactor stacked version of FP8 Grouped Gemm for reduced overhead #3699
Conversation
This file contains 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
|
This pull request was exported from Phabricator. Differential Revision: D69544396 |
✅ Deploy Preview for pytorch-fbgemm-docs ready!
To edit notification comments on pull requests, go to your Netlify site configuration. |
|
This pull request was exported from Phabricator. Differential Revision: D69544396 |
jwfromm
force-pushed
the
export-D69544396
branch
from
e8865f2 to
5e0b36c
Compare
jwfromm
added a commit
to jwfromm/FBGEMM
that referenced
this pull request
Feb 17, 2025
…torch#3699) Summary: Pull Request resolved: pytorch#3699 X-link: facebookresearch/FBGEMM#780 Currently, the stacked version of FP8 grouped gemm accepts lists of tensor inputs and produces a single tensor output. This reduces quite a bit of overhead when cuda graphs are used, but still requires splitting input tensors in prefill which can be costly. This diff updates the input types of stacked grouped gemm to support single tensors. Notably, since M varies across group and we do no padding, this change requires that we provide a new input tensor called `M_offsets` that indicates the row that each group begins at within in the first input. We create M_offsets by taking the cumulative sum of M for each group, which we may be able to further optimize. This diff also includes a long overdue refactor of grouped gemm setup for nvidia such that we only launch a single kernel rather than one per group. This should reduce overhead by quite a bit in some cases. Differential Revision: D69544396
|
This pull request was exported from Phabricator. Differential Revision: D69544396 |
jwfromm
added a commit
to jwfromm/FBGEMM
that referenced
this pull request
Feb 17, 2025
…torch#3699) Summary: Pull Request resolved: pytorch#3699 X-link: facebookresearch/FBGEMM#780 Currently, the stacked version of FP8 grouped gemm accepts lists of tensor inputs and produces a single tensor output. This reduces quite a bit of overhead when cuda graphs are used, but still requires splitting input tensors in prefill which can be costly. This diff updates the input types of stacked grouped gemm to support single tensors. Notably, since M varies across group and we do no padding, this change requires that we provide a new input tensor called `M_offsets` that indicates the row that each group begins at within in the first input. We create M_offsets by taking the cumulative sum of M for each group, which we may be able to further optimize. This diff also includes a long overdue refactor of grouped gemm setup for nvidia such that we only launch a single kernel rather than one per group. This should reduce overhead by quite a bit in some cases. Differential Revision: D69544396
jwfromm
force-pushed
the
export-D69544396
branch
from
5e0b36c to
f5c437a
Compare
…torch#3699) Summary: Pull Request resolved: pytorch#3699 X-link: facebookresearch/FBGEMM#780 Currently, the stacked version of FP8 grouped gemm accepts lists of tensor inputs and produces a single tensor output. This reduces quite a bit of overhead when cuda graphs are used, but still requires splitting input tensors in prefill which can be costly. This diff updates the input types of stacked grouped gemm to support single tensors. Notably, since M varies across group and we do no padding, this change requires that we provide a new input tensor called `M_offsets` that indicates the row that each group begins at within in the first input. We create M_offsets by taking the cumulative sum of M for each group, which we may be able to further optimize. This diff also includes a long overdue refactor of grouped gemm setup for nvidia such that we only launch a single kernel rather than one per group. This should reduce overhead by quite a bit in some cases. Differential Revision: D69544396
|
This pull request was exported from Phabricator. Differential Revision: D69544396 |
jwfromm
force-pushed
the
export-D69544396
branch
from
f5c437a to
d8a3a6f
Compare
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:
X-link: https://github.com/facebookresearch/FBGEMM/pull/780
Currently, the stacked version of FP8 grouped gemm accepts lists of tensor inputs and produces a single tensor output. This reduces quite a bit of overhead when cuda graphs are used, but still requires splitting input tensors in prefill which can be costly. This diff updates the input types of stacked grouped gemm to support single tensors. Notably, since M varies across group and we do no padding, this change requires that we provide a new input tensor called
M_offsetsthat indicates the row that each group begins at within in the first input. We create M_offsets by taking the cumulative sum of M for each group, which we may be able to further optimize.Differential Revision: D69544396