-
Notifications
You must be signed in to change notification settings - Fork 389
Pull requests: pytorch/ao
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
[mxfp8 moe training] add CUDA kernel for per group blocked layout with groups along M
CLA Signed
This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
#3546
opened Dec 25, 2025 by
danielvegamyhre
Loading…
update version compatibility to allow torch 2.11.0.dev
CLA Signed
This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
#3545
opened Dec 25, 2025 by
danielvegamyhre
Loading…
[ROCM] Add MI350 support for MXFP8 colwise quantization.
CLA Signed
This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
module: rocm
#3544
opened Dec 25, 2025 by
xiaobochen-amd
Loading…
[CPU] Optimize FP8 QSDPA
CLA Signed
This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
topic: not user facing
Use this tag if you don't want this PR to show up in release notes
#3541
opened Dec 25, 2025 by
Valentine233
Loading…
[rocm] scaled_grouped_mm support gfx942 fp8 data type
CLA Signed
This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
module: rocm
#3540
opened Dec 25, 2025 by
xiaobochen-amd
Loading…
[Fix][CI] Fix the CI import issue in test_quant_api.py
CLA Signed
This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
topic: for developers
Use this tag if this PR is mainly developer facing
#3539
opened Dec 25, 2025 by
Stonepia
Loading…
[CPU][PT2E] Add inductor lowering path for quantize_affine_float8_non_decomposed
CLA Signed
This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
topic: not user facing
Use this tag if you don't want this PR to show up in release notes
#3538
opened Dec 24, 2025 by
Xia-Weiwen
Loading…
Separately control the activation quantization granularity
CLA Signed
This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
[mxfp8 moe training] update readme with kernel microbenchmarks for dsv3
CLA Signed
This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
moe
mx
topic: documentation
Use this tag if this PR adds or improves documentation
#3521
opened Dec 20, 2025 by
danielvegamyhre
Loading…
Remove support for quant_llm_linear
CLA Signed
This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
topic: bc-breaking
Use this tag if this PR breaks backward compatibility
topic: deprecation
Use this tag if this PR deprecates a feature
#3520
opened Dec 20, 2025 by
howardzhang-cv
Loading…
[DO NOT MERGE] Update workflow to use amdfaa/test-infra
ciflow/rocm
CLA Signed
This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
module: rocm
#3519
opened Dec 19, 2025 by
amdfaa
Loading…
[fix] Load model layer by layer to avoid OOM
CLA Signed
This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
Add GPTQ to prototype
accuracy
Accuracy related
CLA Signed
This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
topic: improvement
Use this tag if this PR is an improvement (doesn't fit into any of the other categories)
#3517
opened Dec 19, 2025 by
jcaip
Loading…
Remove ::kernels for namespacing
CLA Signed
This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
topic: improvement
Use this tag if this PR is an improvement (doesn't fit into any of the other categories)
#3515
opened Dec 18, 2025 by
syedshazli
Loading…
Remove dType Field
CLA Signed
This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
topic: deprecation
Use this tag if this PR deprecates a feature
#3514
opened Dec 18, 2025 by
syedshazli
Loading…
deprecate v1 of This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
topic: bc-breaking
Use this tag if this PR breaks backward compatibility
topic: deprecation
Use this tag if this PR deprecates a feature
Int4WeightOnlyConfig
CLA Signed
#3513
opened Dec 18, 2025 by
vkuzo
Loading…
deprecate v1 of This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
topic: bc-breaking
Use this tag if this PR breaks backward compatibility
topic: deprecation
Use this tag if this PR deprecates a feature
IntxWeightOnlyConfig
CLA Signed
#3512
opened Dec 18, 2025 by
vkuzo
Loading…
deprecate v1 of This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
topic: bc-breaking
Use this tag if this PR breaks backward compatibility
topic: deprecation
Use this tag if this PR deprecates a feature
Int8DynamicActivationIntxWeightConfig
CLA Signed
#3511
opened Dec 18, 2025 by
vkuzo
Loading…
[mxfp8 moe training] test floor vs rceil in 3d quantization benchmarks
CLA Signed
This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
moe
mx
topic: not user facing
Use this tag if you don't want this PR to show up in release notes
#3507
opened Dec 17, 2025 by
danielvegamyhre
Loading…
[mxfp8 moe training] update benchmark plots to use new kernel
CLA Signed
This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
moe
mx
topic: not user facing
Use this tag if you don't want this PR to show up in release notes
#3506
opened Dec 17, 2025 by
danielvegamyhre
Loading…
[mxfp8 moe training] integrate new cuda kernel for blocked layout for groups along K
CLA Signed
This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
moe
mx
topic: not user facing
Use this tag if you don't want this PR to show up in release notes
#3505
opened Dec 17, 2025 by
danielvegamyhre
Loading…
[mxfp8 moe training] add CUDA kernel for per-group conversion of scale factors to blocked layout
CLA Signed
This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
moe
mx
topic: not user facing
Use this tag if you don't want this PR to show up in release notes
#3504
opened Dec 17, 2025 by
danielvegamyhre
Loading…
add performance and accuracy eval of flux-1.schnell
CLA Signed
This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
topic: not user facing
Use this tag if you don't want this PR to show up in release notes
#3502
opened Dec 17, 2025 by
vkuzo
Loading…
[xpu][test] Port test_int8_tensor.py UT file to intel XPU
ciflow/xpu
label used to trigger xpu CI jobs
CLA Signed
This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
topic: for developers
Use this tag if this PR is mainly developer facing
xpu
Intel XPU related features
#3500
opened Dec 17, 2025 by
zxd1997066
Loading…
[ZENDNN] Adding support for slice in opaque
CLA Signed
This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
#3494
opened Dec 16, 2025 by
Priyjain-amd
Loading…
Previous Next
ProTip!
Exclude everything labeled
bug with -label:bug.