Float8Tensor per row quantization pass bias to fbgemm kernel #2884
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/2884
Note: Links to docs will display an error until the docs builds have been completed. ❗ 1 Active SEVsThere are 1 currently active SEVs. If your PR is affected, please view them below: ✅ No FailuresAs of commit 4d70152 with merge base fbe3df9 ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
jerryzh168
added a commit
that referenced
this pull request
Aug 26, 2025
Summary: Previously bias is not passed to fbgemm kernel for float8 per row quant, this PR adds it. Difference is we should have a faster float8 per row quantized kernel, without changing numerics or other things. Test Plan: ``` python test/dtypes/test_affine_quantized_float.py -k test_expected_kernels_on_gpu python test/quantization/quantize_/workflows/float8/test_float8_tensor.py ``` Reviewers: Subscribers: Tasks: Tags: stack-info: PR: #2884, branch: jerryzh168/stack/60
jerryzh168
force-pushed
the
jerryzh168/stack/59
branch
from
6935cc8 to
bacbe8c
Compare
jerryzh168
force-pushed
the
jerryzh168/stack/60
branch
from
2c1e5da to
613585d
Compare
meta-cla
bot
added
the
CLA Signed
jerryzh168
added
the
topic: improvement
jerryzh168
added a commit
that referenced
this pull request
Aug 26, 2025
Summary: Previously bias is not passed to fbgemm kernel for float8 per row quant, this PR adds it. Difference is we should have a faster float8 per row quantized kernel, without changing numerics or other things. Test Plan: ``` python test/dtypes/test_affine_quantized_float.py -k test_expected_kernels_on_gpu python test/quantization/quantize_/workflows/float8/test_float8_tensor.py ``` Reviewers: Subscribers: Tasks: Tags: stack-info: PR: #2884, branch: jerryzh168/stack/60
jerryzh168
force-pushed
the
jerryzh168/stack/60
branch
from
613585d to
b98440e
Compare
jerryzh168
force-pushed
the
jerryzh168/stack/60
branch
from
b98440e to
6acc102
Compare
jerryzh168
added a commit
that referenced
this pull request
Aug 27, 2025
Summary: Previously bias is not passed to fbgemm kernel for float8 per row quant, this PR adds it. Difference is we should have a faster float8 per row quantized kernel, without changing numerics or other things. Test Plan: ``` python test/dtypes/test_affine_quantized_float.py -k test_expected_kernels_on_gpu python test/quantization/quantize_/workflows/float8/test_float8_tensor.py ``` Reviewers: Subscribers: Tasks: Tags: stack-info: PR: #2884, branch: jerryzh168/stack/60
Contributor
|
looks reasonable, can we add a test which covers the new path? |
jerryzh168
added a commit
that referenced
this pull request
Aug 28, 2025
Summary: Previously bias is not passed to fbgemm kernel for float8 per row quant, this PR adds it. Difference is we should have a faster float8 per row quantized kernel, without changing numerics or other things. Test Plan: ``` python test/dtypes/test_affine_quantized_float.py -k test_expected_kernels_on_gpu python test/quantization/quantize_/workflows/float8/test_float8_tensor.py ``` Reviewers: Subscribers: Tasks: Tags: stack-info: PR: #2884, branch: jerryzh168/stack/60
jerryzh168
force-pushed
the
jerryzh168/stack/60
branch
from
6acc102 to
d5465ea
Compare
jerryzh168
added a commit
that referenced
this pull request
Aug 28, 2025
Summary: Previously bias is not passed to fbgemm kernel for float8 per row quant, this PR adds it. Difference is we should have a faster float8 per row quantized kernel, without changing numerics or other things. Test Plan: ``` python test/dtypes/test_affine_quantized_float.py -k test_expected_kernels_on_gpu python test/quantization/quantize_/workflows/float8/test_float8_tensor.py ``` Reviewers: Subscribers: Tasks: Tags: stack-info: PR: #2884, branch: jerryzh168/stack/60
jerryzh168
force-pushed
the
jerryzh168/stack/60
branch
from
d5465ea to
951a49c
Compare
jerryzh168
added a commit
that referenced
this pull request
Aug 28, 2025
Summary: Previously bias is not passed to fbgemm kernel for float8 per row quant, this PR adds it. Difference is we should have a faster float8 per row quantized kernel, without changing numerics or other things. Test Plan: ``` python test/dtypes/test_affine_quantized_float.py -k test_expected_kernels_on_gpu python test/quantization/quantize_/workflows/float8/test_float8_tensor.py ``` Reviewers: Subscribers: Tasks: Tags: stack-info: PR: #2884, branch: jerryzh168/stack/60
jerryzh168
force-pushed
the
jerryzh168/stack/60
branch
from
951a49c to
5860000
Compare
jerryzh168
added a commit
that referenced
this pull request
Aug 28, 2025
Summary: Previously bias is not passed to fbgemm kernel for float8 per row quant, this PR adds it. Difference is we should have a faster float8 per row quantized kernel, without changing numerics or other things. Test Plan: ``` python test/dtypes/test_affine_quantized_float.py -k test_expected_kernels_on_gpu python test/quantization/quantize_/workflows/float8/test_float8_tensor.py ``` Reviewers: Subscribers: Tasks: Tags: stack-info: PR: #2884, branch: jerryzh168/stack/60
jerryzh168
force-pushed
the
jerryzh168/stack/60
branch
from
d94d663 to
0c70dee
Compare
jerryzh168
added a commit
that referenced
this pull request
Aug 28, 2025
Summary: Previously bias is not passed to fbgemm kernel for float8 per row quant, this PR adds it. Difference is we should have a faster float8 per row quantized kernel, without changing numerics or other things. Test Plan: ``` python test/dtypes/test_affine_quantized_float.py -k test_expected_kernels_on_gpu python test/quantization/quantize_/workflows/float8/test_float8_tensor.py ``` Reviewers: Subscribers: Tasks: Tags: stack-info: PR: #2884, branch: jerryzh168/stack/60
jerryzh168
force-pushed
the
jerryzh168/stack/60
branch
from
0c70dee to
2b5046b
Compare
jerryzh168
added a commit
that referenced
this pull request
Aug 28, 2025
Summary: Previously bias is not passed to fbgemm kernel for float8 per row quant, this PR adds it. Difference is we should have a faster float8 per row quantized kernel, without changing numerics or other things. Test Plan: ``` python test/quantization/quantize_/workflows/float8/test_float8_tensor.py -k test_kernel_preference_numerical_equivalence python test/quantization/quantize_/workflows/float8/test_float8_tensor.py -k test_expected_gpu_kernel_fbgemm ``` Reviewers: Subscribers: Tasks: Tags: stack-info: PR: #2884, branch: jerryzh168/stack/60
jerryzh168
force-pushed
the
jerryzh168/stack/60
branch
from
2b5046b to
091ad9e
Compare
jerryzh168
force-pushed
the
jerryzh168/stack/60
branch
from
091ad9e to
94aec1a
Compare
jerryzh168
added a commit
that referenced
this pull request
Aug 28, 2025
Summary: Previously bias is not passed to fbgemm kernel for float8 per row quant, this PR adds it. Difference is we should have a faster float8 per row quantized kernel, without changing numerics or other things. Test Plan: ``` python test/quantization/quantize_/workflows/float8/test_float8_tensor.py -k test_kernel_preference_numerical_equivalence python test/quantization/quantize_/workflows/float8/test_float8_tensor.py -k test_expected_gpu_kernel_fbgemm ``` Reviewers: Subscribers: Tasks: Tags: stack-info: PR: #2884, branch: jerryzh168/stack/60
jerryzh168
added a commit
that referenced
this pull request
Aug 29, 2025
Summary: Previously bias is not passed to fbgemm kernel for float8 per row quant, this PR adds it. Difference is we should have a faster float8 per row quantized kernel, without changing numerics or other things. Test Plan: ``` python test/quantization/quantize_/workflows/float8/test_float8_tensor.py -k test_kernel_preference_numerical_equivalence python test/quantization/quantize_/workflows/float8/test_float8_tensor.py -k test_expected_gpu_kernel_fbgemm ``` Reviewers: Subscribers: Tasks: Tags: stack-info: PR: #2884, branch: jerryzh168/stack/60
jerryzh168
force-pushed
the
jerryzh168/stack/60
branch
from
94aec1a to
5f8d5e2
Compare
drisspg
reviewed
Sep 2, 2025
Contributor
drisspg
left a comment
drisspg
left a comment
There was a problem hiding this comment.
should we have a test, this seems like it was a bug but wasn't caught?
Contributor
Author
|
@drisspg yeah test is this one:
but it's probably better to have a more exhaustive one for all options (auto and torch) as well it is a performance "bug" I think, the effect is that we are not using the most efficient path if there is a bias |
jerryzh168
commented
Sep 2, 2025
| "torch.ops.triton.quantize_fp8_row.default", 1 | ||
| ).check_count("torch.ops.fbgemm.f8f8bf16_rowwise.default", 1).run(code[0]) | ||
| ).check_count("torch.ops.fbgemm.f8f8bf16_rowwise.default", 1).check_not( | ||
| "triton_poi_fused_add_0" |
Contributor
Author
There was a problem hiding this comment.
cc @drisspg we explicitly test that triton_poi_fused_add_0 is not generated, to make sure there is no additional res + bias there in this code path
Contributor
There was a problem hiding this comment.
it's not safe to depend on inductor generated kernel names such as triton_poi_fused_add_0, I think it would be better to ensure that there are no additional kernels called after torch.ops.fbgemm.f8f8bf16_rowwise.default is called.
Summary: Previously bias is not passed to fbgemm kernel for float8 per row quant, this PR adds it. Difference is we should have a faster float8 per row quantized kernel, without changing numerics or other things. Test Plan: ``` python test/quantization/quantize_/workflows/float8/test_float8_tensor.py -k test_kernel_preference_numerical_equivalence python test/quantization/quantize_/workflows/float8/test_float8_tensor.py -k test_expected_gpu_kernel_fbgemm ``` Reviewers: Subscribers: Tasks: Tags: stack-info: PR: #2884, branch: jerryzh168/stack/60
jerryzh168
force-pushed
the
jerryzh168/stack/60
branch
from
5f8d5e2 to
4d70152
Compare
drisspg
approved these changes
Sep 5, 2025
andrewor14
pushed a commit
that referenced
this pull request
Sep 5, 2025
Summary: Previously bias is not passed to fbgemm kernel for float8 per row quant, this PR adds it. Difference is we should have a faster float8 per row quantized kernel, without changing numerics or other things. Test Plan: ``` python test/quantization/quantize_/workflows/float8/test_float8_tensor.py -k test_kernel_preference_numerical_equivalence python test/quantization/quantize_/workflows/float8/test_float8_tensor.py -k test_expected_gpu_kernel_fbgemm ``` Reviewers: Subscribers: Tasks: Tags: stack-info: PR: #2884, branch: jerryzh168/stack/60
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.
Stacked PRs:
Float8Tensor per row quantization pass bias to fbgemm kernel
Summary:
Previously bias is not passed to fbgemm kernel for float8 per row quant,
this PR adds it.
Difference is we should have a faster float8 per row quantized kernel, without changing
numerics or other things.
Test Plan:
Reviewers:
Subscribers:
Tasks:
Tags: