Skip to content

Add torch.backends.cuda.math_sdp.fp32_precision #2841

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
wants to merge 133 commits into from
Closed

Add torch.backends.cuda.math_sdp.fp32_precision #2841

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
133 commits
Select commit Hold shift + click to select a range
2923244
[dynamo] Remove redundant _nonvar_fields assignments in UserDefinedOb…
hinriksnaer Nov 24, 2025
dfdf024
Revert "[DTensor] compute shape and offset for arbitrary _StridedShar…
pytorchmergebot Nov 24, 2025
bec6e68
[dynamo] Skip optree tests when optree isn't installed (#168931)
jansel Nov 23, 2025
821047d
[dynamo] Fix local test failures for test/dynamo/test_after_aot.py (#…
jansel Nov 23, 2025
55b10d7
[dynamo] Fix local test failures for test_compiler_bisector.py (#168915)
jansel Nov 23, 2025
627f6c7
[dynamo] Fix more cases of tests leaking config changes (#168924)
jansel Nov 23, 2025
7dcdb3c
[dynamo] Isolate test/dynamo/test_aot_compile.py tests in subproc (#1…
jansel Nov 23, 2025
fd14c10
[dynamo] Make test custom op naming unique (#168926)
jansel Nov 23, 2025
35944cb
Revert "Checking if the input is finite before calculation in lowerin…
pytorchmergebot Nov 24, 2025
33d4cf4
Revert "Move CUDAEvent to c10 (#158219)"
pytorchmergebot Nov 24, 2025
8989130
Fix local_map default partitioner issue (#168396)
soulitzer Nov 22, 2025
0ecdf68
[CI][CUDA][B200][Smoke Test] Add one more B200 smoke test. Change per…
nWEIdia Nov 24, 2025
09de09e
[AOTI] Skip emit_multi_arch_kernel when CUDA version is lower than 12…
desertfire Nov 24, 2025
3e12e0f
[CD] [aarch64] unify the build.sh to build for aarch64 wheel (#166044)
tinglvv Nov 24, 2025
4ddca70
Add human-readable type name comment to TYPE_MATCH guards for debuggi…
manigkrish Nov 24, 2025
43e23ee
[3.14] Fix nn.Module annotations lookup (#168325)
rtimpe Nov 21, 2025
9ff3bd7
[precompile] set strict_autograd_cache=True when precompiling (#168989)
bobrenjc93 Nov 24, 2025
aa2cf78
Refactor call_function (#168932)
aorenste Nov 23, 2025
40733d7
Revert "[DebugMode] wait before hashing collectives by default (#1681…
pytorchmergebot Nov 24, 2025
f6572c3
[BE] Delete Pytorch-circleci-labels (#169003)
malfet Nov 24, 2025
93fef4b
[inductor] fix picking wrong contiguous node (#168371)
shunting314 Nov 21, 2025
83bb24c
[simplefsdp] fix dsv3 estimation (#168199)
ruisizhang123 Nov 24, 2025
9708e04
[hoo] Invoke subgraph + effect (#167231)
angelayi Nov 21, 2025
561c1eb
[invoke_subgraph] Don't run the graph twice when autograd enabled (#1…
angelayi Nov 21, 2025
53b2f62
[hoo] Fix unlift of effects with invoke_subgraph (#167363)
angelayi Nov 21, 2025
9e88d50
[3/N] Use context managers (#167788)
cyyever Nov 24, 2025
3aeb7b0
inductor: fix failure in test_flex_decoding on <sm80 (#165404)
jsmolic Nov 24, 2025
42ab53d
Fix take_along_dim negative index handling (#146211) (#152161)
KaaustaaubShankar Nov 25, 2025
6936e33
Adding check for step size=0 in unfold backward to avoid divide by 0 …
mansiag05 Nov 25, 2025
b5a4cde
[6/N] Use key in dict for existence checks (#168350)
cyyever Nov 25, 2025
a2413b9
[precompile] move strict_autograd_cache=True decorator to aot_compile…
bobrenjc93 Nov 24, 2025
7c35036
[xpu][fix] Refine memory pool logic when expandable segement enabled …
guangyey Nov 25, 2025
a928c9d
Remove useless parent method delegation (#168355)
cyyever Nov 25, 2025
b3bc797
[Inductor][Quant]Support qconv_pointwise.tensor and qconv2d_pointwise…
jiayisunx Nov 19, 2025
ca3e8b3
[1/N] Use TYPE_CHECKING (#165852)
cyyever Nov 25, 2025
a5436a5
[CPU] add onednn context cache for qlinear to improve performance (#1…
Xia-Weiwen Nov 25, 2025
5a607fe
Back out "Make PT2 compile backprop through custom op without autogra…
ezyang Nov 27, 2025
99024de
Revert "Removed deprecated `split_cat_fx_passes` (#167738)"
pytorchmergebot Nov 27, 2025
9844fbe
Revert "[DTensor] update redistribute_cost, add disable_graph_based_t…
pytorchmergebot Nov 27, 2025
2f9b7da
Revert "Support AC in default partitioner when functionalization is e…
pytorchmergebot Nov 27, 2025
088048f
Remove cudaProfilerInitialize (#168918)
cyyever Nov 28, 2025
9cd055e
[2/N] Remove unused header inclusion (#165831)
cyyever Nov 28, 2025
a0f3937
Remove unused inplace loop in test_conv2d_clamp of test_jit_llga_fuse…
cyyever Nov 28, 2025
a20f775
[Quant][CPU] fix fp8 qconv (#167611)
jiayisunx Nov 19, 2025
f1076f5
[dynamo] Support functools.partial as dict key (#169016)
anijain2305 Nov 27, 2025
9a296e6
Remove unused thrust inclusion (#169051)
cyyever Nov 28, 2025
7bc2a66
[CUDA][BugFix] fix truncated error messages (#168942)
lingebeng Nov 28, 2025
7c64850
[8/N] Use Python 3.10 typing (#168334)
cyyever Nov 28, 2025
2985667
Fix an unsafe indexing in fx exception handling (#169140)
gmagogsfm Nov 28, 2025
f47dd0d
Enable SIM118 (#167399)
cyyever Nov 28, 2025
9f8ef88
Unset LD_LIBRARY_PATH for binary wheel checks (#168349)
robert-hardwick Nov 27, 2025
6658a04
[Code Clean] Better error handling in aten/src/ATen/native/* and ate…
zhudada0120 Nov 28, 2025
b39813b
[ROCm][CI] Use hash for download-artifact action instead of tag (#169…
jithunnair-amd Nov 28, 2025
b7f6b9a
[cuDNN] Leak `BenchmarkCache` to avoid teardown segfault, correct com…
eqy Nov 28, 2025
4c24667
Revert "Fix local_map default partitioner issue (#168396)"
pytorchmergebot Nov 28, 2025
409a5fe
[pytree][dynamo] make `tree_*` functions accept both Python and C++ `…
XuehaiPan Nov 28, 2025
1ccb743
[BE][4/5] fix typos in aten/ (aten/src/ATen/native/) (#157553)
XuehaiPan Nov 28, 2025
f4dedf7
[BE][5/5] fix typos in aten/ (aten/src/ATen/) (#157554)
XuehaiPan Nov 28, 2025
6864e30
[BE][1/6] fix typos in test/ (#157635)
XuehaiPan Nov 28, 2025
a2973fb
[9/N] Use Python 3.10 typing (#167806)
cyyever Nov 29, 2025
d9d5e91
[dynamo, 3.14] Ensure `typing.Union` is correctly traced in Dynamo. (…
guilhermeleobas Nov 29, 2025
90b27e7
[dynamo] Skip Dynamo wrapped tests on cpython tests (#169233)
anijain2305 Nov 29, 2025
f7c0d03
[Durin] Bump Kineto Submodule to latest (#169098)
firoz1905 Nov 29, 2025
79d7b17
[openreg] Expand autocast test coverage for custom device (#169029)
hipudding Nov 29, 2025
4bebc82
Add `torch.Tensor.__annotate__` to the `testing_ignore` list (#169013)
guilhermeleobas Nov 29, 2025
8414958
[dynamo][dicts] Decentralize and Improve key hash implementation for …
anijain2305 Nov 29, 2025
076e7b1
[3.14] Skip broken numpy test (#169030)
rtimpe Nov 28, 2025
45d310a
[3.14] Fix dynamo error on np.broadcast_shapes (#168888)
rtimpe Nov 28, 2025
d503895
Avoid std::tie and returning value constructions in qconv_unpack.cpp …
cyyever Dec 1, 2025
42e9005
Remove unused TVM code in dynamo (#169247)
cyyever Dec 1, 2025
94ca8d5
[xpu][feature] enable Sycl CPP extension on Windows (#162579)
tbohutyn Dec 1, 2025
c8210e7
[Accelerator] Add Accelerator Capabilities API (#165631)
can-gaa-hou Dec 1, 2025
539ba71
add device generalization support for distributed tests (#165067)
harikodali Dec 1, 2025
9fb52ef
[user-streams] Handle the record_stream problem (#168061)
mlazos Dec 1, 2025
55c4ab5
Revert "[dynamo][dicts] Decentralize and Improve key hash implementat…
pytorchmergebot Dec 1, 2025
7d2a33e
Revert "[Accelerator] Add Accelerator Capabilities API (#165631)"
pytorchmergebot Dec 1, 2025
29e5455
[inductor] Add option to transpose tensor descriptors on load / store…
kundaMwiza Dec 1, 2025
ded9bcd
Remove the CUPTI CMake check for kineto (#161370)
cyyever Dec 1, 2025
5f0030b
[xpu][fix] Support xpu custom raw_alloc/delete in caching allocator (…
guangyey Nov 28, 2025
481e5ab
Replace vscode recommendation for type checker (#169021)
lolpack Dec 1, 2025
1ee32a8
[reland][inductor] fix the decision of inner reduction (#168391)
shunting314 Dec 1, 2025
a7dc6da
bump transformer pin to 4.57.3 (#169226)
BoyuanFeng Dec 1, 2025
cddec65
[mergebot] Set header for workflow when calling Dr.CI (#169037)
huydhn Dec 1, 2025
f49d32d
bucketing compile time improve (#168122)
eellison Dec 1, 2025
90f0139
Track overlap per-pg (#169019)
eellison Dec 1, 2025
9f7fceb
[BE] Add missing method docstrings for pytorch quantization classes (…
jiannanWang Dec 1, 2025
b1aed4e
[BE][MPS] Add out-of-bounds checks for embedding_bag (#168930)
malfet Nov 30, 2025
641cdb6
Revert "Avoid std::tie and returning value constructions in qconv_unp…
pytorchmergebot Dec 1, 2025
b8c4ba3
[xpu][feature][1/N] Enable SDPA XPU FlashAttention backend with SYCL-…
LuFinch Dec 2, 2025
518c2b1
[10/N] Use Python 3.10 typing (#169229)
cyyever Dec 2, 2025
1902edd
[precompile] support serde for torch.nn.attention.SDPBackend (#168988)
bobrenjc93 Dec 1, 2025
587d63a
[precompile] generate nonce key if enable_aot_compile is enabled (#16…
bobrenjc93 Dec 1, 2025
ce5e7e3
[CI][CUDA][Distributed] Update NCCL to 2.28.9 for CUDA13 (#168091)
nWEIdia Dec 2, 2025
2d1f78f
[DebugMode] Fix hash for 0 ele tensor; Add more tests (#169027)
yushangdi Dec 2, 2025
503b264
[AOTI] Fix a GPU memory leak caused by reference circle (#168063)
desertfire Dec 1, 2025
3cd98b4
[Inductor] support masked vectorization for the tail_loop for integer…
jiayisunx Dec 1, 2025
d9cb8a7
[AOTI] Set device info for subgraphs (#169001)
desertfire Dec 1, 2025
1e526fb
Parity of rng offset compute and ranks subset support for Local Tenso…
dzmitry-huba Dec 1, 2025
c04e2c6
[dynamo][dicts] Decentralize and Improve key hash implementation for …
anijain2305 Dec 1, 2025
d766976
[Inductor] [Triton] Capture Timeout errors without crashing the job (…
njriasan Dec 2, 2025
bb30341
Avoid std::tie and returning value constructions in qconv_unpack.cpp …
cyyever Dec 2, 2025
fb5be22
Refactor: Remove unnecessary ConstantVariable wrapping in raise_obser…
nandan2003 Dec 2, 2025
166efda
[MTIAGraph][Pytorch] Add the graph_pool_handle api (#169283)
andyanwang Dec 2, 2025
62d3ccd
[audio hash update] update the pinned audio hash (#169198)
pytorchupdatebot Dec 2, 2025
7d1bbaf
Add sum support for qlinear_binary templated implementation (#163249)
CaoE Dec 2, 2025
66004b9
[dynamo] Fix test state leakage for test_modes.py (#168928)
jansel Dec 2, 2025
b9c8f3a
[dynamo] Fix local test failures for logging (#168927)
jansel Dec 2, 2025
4cfb47f
[CP] Refactor CP sharding rules into separate module and only registe…
fegin Dec 1, 2025
f2d6a75
bump timm pin (#169227)
BoyuanFeng Dec 2, 2025
b555c39
[pallas backend] Add multi-output support to Pallas backend (#169323)
oulgen Dec 1, 2025
afdff7f
[pallas backend] support bitcast (#169324)
oulgen Dec 1, 2025
a951a9c
[inductor] Increase tolerance for test_emulate_precision_casts_mean_r…
jansel Dec 2, 2025
556375b
[inductor] Increase tolerance for test_conv3d_binary_broadcast_shapes…
jansel Dec 2, 2025
1e34fb2
Revert "Refactor: Remove unnecessary ConstantVariable wrapping in rai…
pytorchmergebot Dec 2, 2025
4917316
[Dynamo][Guards]Fix TLParse CPP guard message with sorting get_leaf_g…
fxdawnn Dec 2, 2025
5778f6f
[strobelight][gpusnoop] compress aoti stack (#169291)
sonntex Dec 2, 2025
892640e
Hide all symbols (except stable/headeronly/shim) if TORCH_STABLE_ONLY…
mikaylagawarecki Dec 1, 2025
70d797a
[dist] add reduce_scatter_out (#168260)
IvanKobzarev Dec 2, 2025
285779b
[Accelerator] Add Accelerator Capabilities API (#165631)
can-gaa-hou Dec 2, 2025
174272c
[precompile] Include inductor-generated Dynamo guards in AOT (#169239)
zhxchen17 Dec 2, 2025
9ff4a2e
Support AC in default partitioner when functionalization is enabled (…
soulitzer Dec 1, 2025
44ac693
Add DCE pass to remove unused intermediates (#169131)
tugsbayasgalan Dec 1, 2025
b4cc132
[BE][Typing][Dynamo] Type torch/_dynamo/variables/nn_module.py (#167342)
Lucaskabela Dec 2, 2025
32d0f1e
Add torch.backends.cuda.math_sdp.fp32_precision
xinyazhang Nov 5, 2025
e0c7cbb
Make torch.backends.cuda.math_sdp.fp32_precision effective for math_sdp
xinyazhang Nov 5, 2025
e7c76da
torch/testing: add ctx mananger math_sdp_precision
xinyazhang Nov 5, 2025
8654659
test/test_transformers: decorate all tests that uses fp32 math as gol…
xinyazhang Nov 5, 2025
0e4f9a3
fix build error
xinyazhang Nov 5, 2025
fe067e7
test/test_transformers: sanity check of golden tensor
xinyazhang Nov 5, 2025
19fde78
fix Fp32PrecisonGuard
xinyazhang Nov 5, 2025
6e6c0c6
more documentation
xinyazhang Nov 5, 2025
8284287
fix lint
xinyazhang Nov 6, 2025
b996c2e
Follow the practice of cuBLASModule and ignore MathSDPModule in docs …
xinyazhang Dec 2, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
The table of contents is too big for display.
Diff view
Diff view
  •  
  •  
  •  
19 changes: 0 additions & 19 deletions .ci/aarch64_linux/README.md
View file
Open in desktop

This file was deleted.

53 changes: 0 additions & 53 deletions .ci/aarch64_linux/aarch64_ci_build.sh
View file
Open in desktop

This file was deleted.

21 changes: 0 additions & 21 deletions .ci/aarch64_linux/aarch64_ci_setup.sh
View file
Open in desktop

This file was deleted.

Loading