bf16 scale/bias for INT4 (#5595)#5595
Closed
jeetkanjani7 wants to merge 1 commit intopytorch:mainfrom
Closed
Conversation
Contributor
|
@jeetkanjani7 has exported this pull request. If you are a Meta employee, you can view the originating Diff in D95859348. |
meta-codesync
Bot
changed the title
jeetkanjani7
force-pushed
the
export-D95859348
branch
from
962a486 to
c9a4ef8
Compare
jeetkanjani7
added a commit
to jeetkanjani7/FBGEMM-1
that referenced
this pull request
Apr 13, 2026
Summary: Add bf16 scale/bias support for INT4/INT2 fused N-bit rowwise quantization in FBGEMM and SilverTorch. Previously, fused N-bit rowwise quantization only supported fp16 scale/bias - storing scale/bias in bf16 avoids precision loss from fp16 truncation during quantization round-trips. X-link: facebookresearch/FBGEMM#2551 Differential Revision: D95859348
jeetkanjani7
added a commit
to jeetkanjani7/FBGEMM-1
that referenced
this pull request
Apr 13, 2026
Summary: Add bf16 scale/bias support for INT4/INT2 fused N-bit rowwise quantization in FBGEMM and SilverTorch. Previously, fused N-bit rowwise quantization only supported fp16 scale/bias - storing scale/bias in bf16 avoids precision loss from fp16 truncation during quantization round-trips. X-link: facebookresearch/FBGEMM#2551 Differential Revision: D95859348
jeetkanjani7
force-pushed
the
export-D95859348
branch
from
c9a4ef8 to
1b262b9
Compare
jeetkanjani7
added a commit
to jeetkanjani7/FBGEMM-1
that referenced
this pull request
Apr 13, 2026
Summary: Add bf16 scale/bias support for INT4/INT2 fused N-bit rowwise quantization in FBGEMM and SilverTorch. Previously, fused N-bit rowwise quantization only supported fp16 scale/bias - storing scale/bias in bf16 avoids precision loss from fp16 truncation during quantization round-trips. X-link: https://github.com/facebookresearch/FBGEMM/pull/2551 Pull Request resolved: pytorch#5595 Differential Revision: D95859348
jeetkanjani7
force-pushed
the
export-D95859348
branch
2 times, most recently
from
39d74bc to
b65c26e
Compare
jeetkanjani7
added a commit
to jeetkanjani7/FBGEMM-1
that referenced
this pull request
Apr 13, 2026
Summary: Add bf16 scale/bias support for INT4/INT2 fused N-bit rowwise quantization in FBGEMM and SilverTorch. Previously, fused N-bit rowwise quantization only supported fp16 scale/bias - storing scale/bias in bf16 avoids precision loss from fp16 truncation during quantization round-trips. X-link: https://github.com/facebookresearch/FBGEMM/pull/2551 Pull Request resolved: pytorch#5595 Differential Revision: D95859348
jeetkanjani7
force-pushed
the
export-D95859348
branch
from
b65c26e to
f222b33
Compare
jeetkanjani7
added a commit
to jeetkanjani7/FBGEMM-1
that referenced
this pull request
Apr 21, 2026
Summary: Add bf16 scale/bias support for INT4/INT2 fused N-bit rowwise quantization in FBGEMM and SilverTorch. Previously, fused N-bit rowwise quantization only supported fp16 scale/bias - storing scale/bias in bf16 avoids precision loss from fp16 truncation during quantization round-trips. X-link: facebookresearch/FBGEMM#2551 Differential Revision: D95859348
jeetkanjani7
force-pushed
the
export-D95859348
branch
from
f222b33 to
1a0f21b
Compare
jeetkanjani7
added a commit
to jeetkanjani7/FBGEMM-1
that referenced
this pull request
Apr 23, 2026
Summary: Add bf16 scale/bias support for INT4/INT2 fused N-bit rowwise quantization in FBGEMM and SilverTorch. Previously, fused N-bit rowwise quantization only supported fp16 scale/bias - storing scale/bias in bf16 avoids precision loss from fp16 truncation during quantization round-trips. X-link: facebookresearch/FBGEMM#2551 Differential Revision: D95859348
jeetkanjani7
added a commit
to jeetkanjani7/FBGEMM-1
that referenced
this pull request
Apr 23, 2026
Summary: Add bf16 scale/bias support for INT4/INT2 fused N-bit rowwise quantization in FBGEMM and SilverTorch. Previously, fused N-bit rowwise quantization only supported fp16 scale/bias - storing scale/bias in bf16 avoids precision loss from fp16 truncation during quantization round-trips. X-link: facebookresearch/FBGEMM#2551 Differential Revision: D95859348
jeetkanjani7
force-pushed
the
export-D95859348
branch
2 times, most recently
from
eb5b190 to
d0aa8bb
Compare
jeetkanjani7
added a commit
to jeetkanjani7/FBGEMM-1
that referenced
this pull request
Apr 24, 2026
Summary: Add bf16 scale/bias support for INT4/INT2 fused N-bit rowwise quantization in FBGEMM and SilverTorch. Previously, fused N-bit rowwise quantization only supported fp16 scale/bias - storing scale/bias in bf16 avoids precision loss from fp16 truncation during quantization round-trips. X-link: facebookresearch/FBGEMM#2551 Reviewed By: zhaozhul Differential Revision: D95859348
jeetkanjani7
added a commit
to jeetkanjani7/FBGEMM-1
that referenced
this pull request
Apr 24, 2026
Summary: Add bf16 scale/bias support for INT4/INT2 fused N-bit rowwise quantization in FBGEMM and SilverTorch. Previously, fused N-bit rowwise quantization only supported fp16 scale/bias - storing scale/bias in bf16 avoids precision loss from fp16 truncation during quantization round-trips. X-link: facebookresearch/FBGEMM#2551 Reviewed By: zhaozhul Differential Revision: D95859348
jeetkanjani7
force-pushed
the
export-D95859348
branch
from
d0aa8bb to
ec9a50e
Compare
jeetkanjani7
added a commit
to jeetkanjani7/FBGEMM-1
that referenced
this pull request
Apr 24, 2026
Summary: Add bf16 scale/bias support for INT4/INT2 fused N-bit rowwise quantization in FBGEMM and SilverTorch. Previously, fused N-bit rowwise quantization only supported fp16 scale/bias - storing scale/bias in bf16 avoids precision loss from fp16 truncation during quantization round-trips. X-link: facebookresearch/FBGEMM#2551 Reviewed By: zhaozhul Differential Revision: D95859348
jeetkanjani7
force-pushed
the
export-D95859348
branch
from
ec9a50e to
a98dd37
Compare
jeetkanjani7
added a commit
to jeetkanjani7/FBGEMM-1
that referenced
this pull request
Apr 24, 2026
Summary: Add bf16 scale/bias support for INT4/INT2 fused N-bit rowwise quantization in FBGEMM and SilverTorch. Previously, fused N-bit rowwise quantization only supported fp16 scale/bias - storing scale/bias in bf16 avoids precision loss from fp16 truncation during quantization round-trips. X-link: facebookresearch/FBGEMM#2551 Reviewed By: zhaozhul Differential Revision: D95859348
jeetkanjani7
force-pushed
the
export-D95859348
branch
from
a98dd37 to
1eef2bd
Compare
jeetkanjani7
added a commit
to jeetkanjani7/FBGEMM-1
that referenced
this pull request
Apr 24, 2026
Summary: Add bf16 scale/bias support for INT4/INT2 fused N-bit rowwise quantization in FBGEMM and SilverTorch. Previously, fused N-bit rowwise quantization only supported fp16 scale/bias - storing scale/bias in bf16 avoids precision loss from fp16 truncation during quantization round-trips. X-link: facebookresearch/FBGEMM#2551 Reviewed By: zhaozhul Differential Revision: D95859348
jeetkanjani7
force-pushed
the
export-D95859348
branch
from
1eef2bd to
f5f727c
Compare
jeetkanjani7
added a commit
to jeetkanjani7/FBGEMM-1
that referenced
this pull request
Apr 24, 2026
Summary: Add bf16 scale/bias support for INT4/INT2 fused N-bit rowwise quantization in FBGEMM and SilverTorch. Previously, fused N-bit rowwise quantization only supported fp16 scale/bias - storing scale/bias in bf16 avoids precision loss from fp16 truncation during quantization round-trips. X-link: https://github.com/facebookresearch/FBGEMM/pull/2551 Pull Request resolved: pytorch#5595 Reviewed By: zhaozhul Differential Revision: D95859348
jeetkanjani7
force-pushed
the
export-D95859348
branch
from
f5f727c to
3abc3b5
Compare
jeetkanjani7
added a commit
to jeetkanjani7/FBGEMM-1
that referenced
this pull request
Apr 24, 2026
Summary: Add bf16 scale/bias support for INT4/INT2 fused N-bit rowwise quantization in FBGEMM and SilverTorch. Previously, fused N-bit rowwise quantization only supported fp16 scale/bias - storing scale/bias in bf16 avoids precision loss from fp16 truncation during quantization round-trips. X-link: https://github.com/facebookresearch/FBGEMM/pull/2551 Pull Request resolved: pytorch#5595 Reviewed By: zhaozhul Differential Revision: D95859348
jeetkanjani7
force-pushed
the
export-D95859348
branch
from
3abc3b5 to
4b75cba
Compare
jeetkanjani7
added a commit
to jeetkanjani7/FBGEMM-1
that referenced
this pull request
Apr 24, 2026
Summary: Add bf16 scale/bias support for INT4/INT2 fused N-bit rowwise quantization in FBGEMM and SilverTorch. Previously, fused N-bit rowwise quantization only supported fp16 scale/bias - storing scale/bias in bf16 avoids precision loss from fp16 truncation during quantization round-trips. X-link: facebookresearch/FBGEMM#2551 Reviewed By: zhaozhul Differential Revision: D95859348
jeetkanjani7
force-pushed
the
export-D95859348
branch
2 times, most recently
from
4db39c2 to
3c73db9
Compare
jeetkanjani7
added a commit
to jeetkanjani7/FBGEMM-1
that referenced
this pull request
Apr 27, 2026
Summary: Add bf16 scale/bias support for INT4/INT2 fused N-bit rowwise quantization in FBGEMM and SilverTorch. Previously, fused N-bit rowwise quantization only supported fp16 scale/bias - storing scale/bias in bf16 avoids precision loss from fp16 truncation during quantization round-trips. X-link: facebookresearch/FBGEMM#2551 Reviewed By: zhaozhul Differential Revision: D95859348
Summary: Add bf16 scale/bias support for INT4/INT2 fused N-bit rowwise quantization in FBGEMM and SilverTorch. Previously, fused N-bit rowwise quantization only supported fp16 scale/bias - storing scale/bias in bf16 avoids precision loss from fp16 truncation during quantization round-trips. X-link: facebookresearch/FBGEMM#2551 Reviewed By: zhaozhul Differential Revision: D95859348
jeetkanjani7
force-pushed
the
export-D95859348
branch
from
3c73db9 to
1cc0b5c
Compare
Contributor
|
This pull request has been merged in 939f2da. |
|
This pull request has been reverted by 5e1dde6. |
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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:
Add bf16 scale/bias support for INT4/INT2 fused N-bit rowwise quantization in FBGEMM and SilverTorch.
Previously, fused N-bit rowwise quantization only supported fp16 scale/bias - storing scale/bias in bf16 avoids precision loss from fp16 truncation during quantization round-trips.
X-link: https://github.com/facebookresearch/FBGEMM/pull/2551
Reviewed By: zhaozhul
Differential Revision: D95859348