OSSForks/pytorch
1
0
Fork 0
mirror of https://github.com/zebrajr/pytorch.git synced 2025-12-06 00:20:18 +01:00
Code Issues Actions 38 Packages Projects Releases Wiki Activity
94,579 Commits 8,304 Branches 150 Tags 1.5 GiB
b2953f5643
Commit Graph

94579 Commits

This Branch
This Branch
All Branches
Author SHA1 Message Date
Yuanyuan Chen
b2953f5643 [9/N] Apply ruff UP035 rule (#165515) 
This is follow-up of #165214 to continue applying ruff UP035 rule to the code base.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/165515
Approved by: https://github.com/Lucaskabela
 2025-10-17 00:09:51 +00:00
PyTorch MergeBot
470e2f61c3 Revert "[Fix] Use sys.executable instead of hardcoded python (#165633)" 
This reverts commit 37f3ba274a.

Reverted https://github.com/pytorch/pytorch/pull/165633 on behalf of https://github.com/malfet due to Looks like it broke test_collect_callgrind in slow workflows, see e0fe37fa68/1 ([comment](https://github.com/pytorch/pytorch/pull/165633#issuecomment-3413290813))
 2025-10-17 00:06:40 +00:00
Kurt Mohler
e0fe37fa68 [MPS] Move torch.cat impl to Metal (#165373) 
After this change, all of the cases tested in [this performance measurement script](10de64c5ac/cat/perf0.py) take either roughly the same runtime or less.

Before:

```
idx: cpu time, mps time, speedup, op, args, kwargs
-----------------------------------------
0: 0.000857 ms, 0.016098 ms, 0.05, cat, [[tensor(shape[5, 5]), tensor(shape[5, 5])]], {'dim': -1}
1: 0.000858 ms, 0.014861 ms, 0.06, cat, [[tensor(shape[5, 5]), tensor(shape[5, 5])]], {'dim': 1}
2: 0.000806 ms, 0.015145 ms, 0.05, cat, [[tensor(shape[10, 5]), tensor(shape[5, 5])]], {'dim': 0}
3: 0.000829 ms, 0.015355 ms, 0.05, cat, [[tensor(shape[1, 2, 3]), tensor(shape[1, 2, 3])]], {'dim': -2}
4: 0.000591 ms, 0.000582 ms, 1.02, cat, [[tensor(shape[0]), tensor(shape[0])]], {'dim': 0}
5: 0.001076 ms, 0.022387 ms, 0.05, cat, [[tensor(shape[0]), tensor(shape[5, 5])]], {'dim': 1}
6: 0.000708 ms, 0.022300 ms, 0.03, cat, [[tensor(shape[0, 5]), tensor(shape[5, 5])]], {'dim': 0}
7: 0.000640 ms, 0.014367 ms, 0.04, cat, [[tensor(shape[1]), tensor(shape[1])]], {}
8: 0.000777 ms, 0.027506 ms, 0.03, cat, [[tensor(shape[2, 2, 2, 2])], 1], {}
9: 0.003383 ms, 0.269277 ms, 0.01, cat, "[[tensor(shape[3, 1, 2]), tensor(shape[3, 2, 2]), tensor(shape[3, 3, 2]), tensor(shape[3, 1, 2]), te...", {'dim': 1}
10: 0.526138 ms, 0.650852 ms, 0.81, cat, "[[tensor(shape[3, 1, 2]), tensor(shape[3, 2, 2]), tensor(shape[3, 3, 2]), tensor(shape[3, 1, 2]), te...", {'dim': 1}
11: 0.444091 ms, 0.628630 ms, 0.71, cat, "[[tensor(shape[1, 3, 2]), tensor(shape[2, 3, 2]), tensor(shape[3, 3, 2]), tensor(shape[1, 3, 2]), te...", {'dim': 0}
12: 2.011870 ms, 0.989525 ms, 2.03, cat, [[tensor(shape[1000000, 3, 2]), tensor(shape[1000000, 3, 2])]], {'dim': 0}
13: 3.100653 ms, 0.948178 ms, 3.27, cat, [[tensor(shape[3, 1000000, 2]), tensor(shape[3, 1000000, 2])]], {'dim': 1}
14: 3.112174 ms, 0.954174 ms, 3.26, cat, [[tensor(shape[3, 2, 1000000]), tensor(shape[3, 2, 1000000])]], {'dim': 2}
```

After:

```
idx: cpu time, mps time, speedup, op, args, kwargs
-----------------------------------------
0: 0.000790 ms, 0.013111 ms, 0.06, cat, [[tensor(shape[5, 5]), tensor(shape[5, 5])]], {'dim': -1}
1: 0.000800 ms, 0.014419 ms, 0.06, cat, [[tensor(shape[5, 5]), tensor(shape[5, 5])]], {'dim': 1}
2: 0.000748 ms, 0.010019 ms, 0.07, cat, [[tensor(shape[10, 5]), tensor(shape[5, 5])]], {'dim': 0}
3: 0.000767 ms, 0.010063 ms, 0.08, cat, [[tensor(shape[1, 2, 3]), tensor(shape[1, 2, 3])]], {'dim': -2}
4: 0.000591 ms, 0.000591 ms, 1.00, cat, [[tensor(shape[0]), tensor(shape[0])]], {'dim': 0}
5: 0.001220 ms, 0.009763 ms, 0.12, cat, [[tensor(shape[0]), tensor(shape[5, 5])]], {'dim': 1}
6: 0.000739 ms, 0.006203 ms, 0.12, cat, [[tensor(shape[0, 5]), tensor(shape[5, 5])]], {'dim': 0}
7: 0.000647 ms, 0.009905 ms, 0.07, cat, [[tensor(shape[1]), tensor(shape[1])]], {}
8: 0.000753 ms, 0.007818 ms, 0.10, cat, [[tensor(shape[2, 2, 2, 2])], 1], {}
9: 0.003823 ms, 0.192723 ms, 0.02, cat, "[[tensor(shape[3, 1, 2]), tensor(shape[3, 2, 2]), tensor(shape[3, 3, 2]), tensor(shape[3, 1, 2]), te...", {'dim': 1}
10: 0.576564 ms, 0.733920 ms, 0.79, cat, "[[tensor(shape[3, 1, 2]), tensor(shape[3, 2, 2]), tensor(shape[3, 3, 2]), tensor(shape[3, 1, 2]), te...", {'dim': 1}
11: 0.462957 ms, 0.692799 ms, 0.67, cat, "[[tensor(shape[1, 3, 2]), tensor(shape[2, 3, 2]), tensor(shape[3, 3, 2]), tensor(shape[1, 3, 2]), te...", {'dim': 0}
12: 2.017181 ms, 0.968345 ms, 2.08, cat, [[tensor(shape[1000000, 3, 2]), tensor(shape[1000000, 3, 2])]], {'dim': 0}
13: 3.203508 ms, 0.986382 ms, 3.25, cat, [[tensor(shape[3, 1000000, 2]), tensor(shape[3, 1000000, 2])]], {'dim': 1}
14: 3.181249 ms, 1.007773 ms, 3.16, cat, [[tensor(shape[3, 2, 1000000]), tensor(shape[3, 2, 1000000])]], {'dim': 2}
```

Fixes #165350
Pull Request resolved: https://github.com/pytorch/pytorch/pull/165373
Approved by: https://github.com/kulinseth, https://github.com/malfet
 2025-10-17 00:03:04 +00:00
PyTorch MergeBot
d2c82bafb7 Revert "158232 Fix autocast cache incorrectly retaining no_grad state (#165068)" 
This reverts commit 5daef30b26.

Reverted https://github.com/pytorch/pytorch/pull/165068 on behalf of https://github.com/jeffdaily due to This broke ROCm CI. test/test_transformers.py::TestTransformersCUDA::test_transformerencoder_fastpath_use_torchscript_False_enable_nested_tensor_True_use_autocast_True_d_model_256_cuda [GH job link](https://github.com/pytorch/pytorch/actions/runs/18572589089/job/52952074008) [HUD commit link](5daef30b26) ([comment](https://github.com/pytorch/pytorch/pull/165068#issuecomment-3413184445))
 2025-10-16 23:08:27 +00:00
Colin L Reliability Rice
98a488c9aa Start recording inductor provenance (#162669) 
Summary:
This stores information on where fx graphs come from, which makes it
significantly easier to debug.

One outstanding question

1) I only stored the kernel stack traces, do we also want the node mappings?

Test Plan:
I wrote a explicit logging test which makes a module, fx traces it, compiles it, and makes sure the logging infomration shows up.

```
clr@devvm17763 ~/fbsource/fbcode/caffe2/test/dynamo
 % buck2 test @//mode/opt fbcode//caffe2/test/dynamo:test_dynamo -- test_utils

File changed: fbsource//xplat/caffe2/test/dynamo/test_utils.py
File changed: fbcode//caffe2/test/dynamo/test_utils.py
Buck UI: https://www.internalfb.com/buck2/528dea32-2416-4a62-a1ec-39f3c0efdd2e
Test UI: https://www.internalfb.com/intern/testinfra/testrun/13229324015574003
Network: Up: 0B  Down: 0B
Executing actions. Remaining     0/2
Command: test.
Time elapsed: 17.3s
Tests finished: Pass 16. Fail 0. Fatal 0. Skip 0. Build failure 0
```

Rollback Plan:

Differential Revision: D82037582

Pull Request resolved: https://github.com/pytorch/pytorch/pull/162669
Approved by: https://github.com/yushangdi
 2025-10-16 23:05:31 +00:00
Shivam Raikundalia
5b3ea75895 [Mem Snapshot] Add Metadata Field (#165490) 
Summary:
The implementation adds the ability to:

Set custom metadata strings that will be attached to all subsequent allocations
Clear or change the metadata at any point
View the metadata in memory snapshots via _dump_snapshot()

Test Plan: Added test in test_cuda.py and check manually in snapshot to see that metadata was added.

Differential Revision: D84654933

Pull Request resolved: https://github.com/pytorch/pytorch/pull/165490
Approved by: https://github.com/yushangdi
 2025-10-16 22:54:27 +00:00
Pian Pawakapan
556fc09a9f [DebugMode][1/N] refactor logs into _DebugCalls (#165376) 
Pull Request resolved: https://github.com/pytorch/pytorch/pull/165376
Approved by: https://github.com/SherlockNoMad
 2025-10-16 22:43:52 +00:00
Nikita Shulga
ce109b3f79 Add torch.backends.mkldnn.is_acl_available() method (#165678) 
That tells whether or not PyTorch was compiled with Arm Compute Library
Pull Request resolved: https://github.com/pytorch/pytorch/pull/165678
Approved by: https://github.com/Skylion007, https://github.com/atalman, https://github.com/albanD
ghstack dependencies: #165583, #165584, #165676
 2025-10-16 22:34:21 +00:00
Nikita Shulga
4d833f859b [BE] [CI] Fix aarch64 arch checks (#165676) 
Instead of relying on `TEST_CONFIG` environment variable  to contain `aarch64`, which is prone to errors,  use output of  `$(uname -m)` that is equal to `aarch64` on Linux ARM systems
Pull Request resolved: https://github.com/pytorch/pytorch/pull/165676
Approved by: https://github.com/huydhn, https://github.com/atalman
ghstack dependencies: #165583, #165584
 2025-10-16 22:19:53 +00:00
Wei Wang
d7e275d4b4 [CI][CUDA] Add periodic b200 distributed job (#159323) 
1. Run distributed job with B200 runner, periodically.
2. discovered generic distributed test issue that certain unit test hard-coded ranks, calling for require_exact_world_size(world_size) API instead of require_world_size(world_size).

Pull Request resolved: https://github.com/pytorch/pytorch/pull/159323
Approved by: https://github.com/eqy

Co-authored-by: Aidyn-A <aidyn.b.aitzhan@gmail.com>
 2025-10-16 21:54:04 +00:00
Jithun Nair
d5db3aee0d [CI] Use 1-GPU runners for rocm-mi355.yml (#165658) 
Should only need 1-GPU runners for rocm-mi355.yml since it runs `default` test config which only needs 1 GPU

Pull Request resolved: https://github.com/pytorch/pytorch/pull/165658
Approved by: https://github.com/jeffdaily
 2025-10-16 21:53:22 +00:00
Maggie Moss
5641de7b6b Add suppressions for _inductor/codegen (#165659) 
Adds suppressions to pyrefly will typecheck clean: https://github.com/pytorch/pytorch/issues/163283

Test plan:
dmypy restart && python3 scripts/lintrunner.py -a
pyrefly check

step 1: delete lines in the pyrefly.toml file from the project-excludes field
step 2: run pyrefly check
step 3: add suppressions, clean up unused suppressions
before: https://gist.github.com/maggiemoss/4b3bf2037014e116bc00706a16aef199

after:
INFO 0 errors (6,884 ignored)

Pull Request resolved: https://github.com/pytorch/pytorch/pull/165659
Approved by: https://github.com/oulgen
 2025-10-16 21:37:37 +00:00
Nicolas De Carli
cbc08c8993 Add NEON acceleration for Vectorized<int[8|16|32|64> (#165273) 
Summary:
Adding NEON specializations of Vectorized<T> for int8, int16, int32 and int64.

Correcness has been checked using test_ops.py and the comprehensive torch test

operator_benchmark_test.py has been enhanced by adding cases of bitwise operations, boolean ops and integer ops.
The benchmark, which uses the PyTorch API, shows significant enhancements in a wide variety of operations:

Before:

bitwise xor: 779.882us
boolean any: 636.209us
boolean all: 538.621us
integer mul: 304.457us
integer asr: 447.997us

After:

bitwise xor: 680.221us ---> 15% higher throughput
boolean any: 391.468us ---> 63% higher throughput
boolean all: 390.189us ---> 38% higher throughput
integer mul: 193.532us ---> 57% higher throughput
integer asr: 179.929us---> 149% higher throughput

Test Plan:
Correctness:

buck2 test @mode/opt //caffe2/test:test_ops
buck2 test @mode/opt //caffe2/test:torch
buck2 test @mode/opt //caffe2/test/distributed/launcher/fb:fb_run_test

Performance:

buck2 run mode/opt //caffe2/benchmarks/operator_benchmark/fb:operator_benchmark_test

Differential Revision: D84424638

Pull Request resolved: https://github.com/pytorch/pytorch/pull/165273
Approved by: https://github.com/malfet
 2025-10-16 21:35:13 +00:00
Yiming Zhou
1a54d3333d [easy] Fix graph_capture in aot_joint_with_descriptors test (#165660) 
when `with_export=True`, `aot_export_joint_with_descriptors` should take the graph produced by `_dynamo_graph_capture_for_export`

```
python test/functorch/test_aot_joint_with_descriptors.py -k test_preserve_annotate_simple
python test/functorch/test_aot_joint_with_descriptors.py -k test_preserve_annotate_flex_attention
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/165660
Approved by: https://github.com/yushangdi
 2025-10-16 21:10:11 +00:00
Aaron Orenstein
4c1c341fa0 FakeTensorMode shouldn't cache syms when tracing (#164718) 
Improve FakeTensor cache to handle SymNode and tracing properly.

For now, when we're proxy tracing just don't bother caching operations that contain SymNodes in the output. The problem is that the proxy tracer relies on SymNode identity and our cache doesn't preserve that. It can be fixed (and I left some notes in _validate_symbolic_output_for_caching() how) but it's not worth it for now.

If we aren't proxy tracing then caching is fine.

Thus these changes:

1. Our cache key needs to include whether we were actively tracing or not - this way if we create a cache entry when we weren't tracing and then we try to use it when we ARE tracing it gets rerun.

2. If there's a SymNode in the output then bypass tracing.

3. Some general cleanup of the output validation - we were unnecessarily doing it as a two-step process when it could just be a single step (it's still two parts internally but only a single outer try/except).

Pull Request resolved: https://github.com/pytorch/pytorch/pull/164718
Approved by: https://github.com/bobrenjc93
ghstack dependencies: #165266, #164717
 2025-10-16 20:57:07 +00:00
Aaron Orenstein
5f21cc786a Teach ProxyTorchDispatchMode how to decompose sympy.Expr into known inputs (#164717) 
In a training library we hit a weird conflict between dtensor, dynamic shapes, and proxy tensor.

The problem is occuring because in sharding_prop we use FakeTensors to compute an operation size (so we don't have to  use the full "real" data). We turn off proxy tracing while we're doing that because we don't want the FakeTensor ops to end up in the graph.  We then use that size when doing later operations.

Normally this is no problem - but when those sizes are dynamic shapes then we have a problem - the proxy tracer wants to track the provenance of all shape operations (`s1*s2`) but since tracing is disabled it doesn't see the operation and when we then use the result shape later on the proxy tracer gets all confused (because the SymNode appeared out of nowhere).

At first we were thinking to never disable shape tracing - but that caused a slew of other downstream problems (lots of code that actually needs the shape tracing to be disabled) so instead we enable having a "sym tracing override" and surgically when we disable proxy tracing we leave shape tracing enabled.

After this change the dtensor embedding is "fixed" but then runs afoul of a FakeTensor cache bug - which is fixed in the next PR.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/164717
Approved by: https://github.com/bobrenjc93, https://github.com/ezyang
ghstack dependencies: #165266
 2025-10-16 20:57:06 +00:00
Aaron Orenstein
e86942f422 minor proxy_tensor reorg (#165266) 
Moving some code around in proxy_tensor in preparation for the next PR. There we
no actual changes (other than simple relabeling such as `self.tracer` ->
`tracer`):

- Move _compute_proxy() out of ProxyTorchDispatchMode.

- Give `sympy_expr_tracker` a structured type instead of `object`.

- Split SymNode registration out of ProxyTorchDispatchMode.__sym_dispatch__() so
  it can be reused.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/165266
Approved by: https://github.com/ezyang, https://github.com/mlazos
 2025-10-16 20:57:06 +00:00
Dzmitry Huba
2cd5fd1588 Enable local tensor mode on DTensor view ops test (#165596) 
While enabling this test discovered lack of support for sub meshes. Added limited support
for sub meshes by properly computing rank coordinates for a given sub mesh. The implementation
follows similar approach to collectives. We infer all sub meshes for the given dimensions and
compute each rank's coordinates with respect to is sub mesh.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/165596
Approved by: https://github.com/ezyang
 2025-10-16 20:52:06 +00:00
Oguz Ulgen
7d0f872cb3 Use union syntax in torch/_inductor runtime and fx_passes (#165652) 
Pull Request resolved: https://github.com/pytorch/pytorch/pull/165652
Approved by: https://github.com/aorenste
 2025-10-16 20:51:59 +00:00
PyTorch MergeBot
fb06e49ce8 Revert "[inductor] print 0.0 as 0 for triton (#164291)" 
This reverts commit 99b32a6750.

Reverted https://github.com/pytorch/pytorch/pull/164291 on behalf of https://github.com/malfet due to Broke slow job, see aba8c43594/1  ([comment](https://github.com/pytorch/pytorch/pull/164291#issuecomment-3412768915))
 2025-10-16 20:44:29 +00:00
PyTorch MergeBot
27a98e6ae9 Revert "[DeviceMesh] Prefer using _layout over _mesh for all sorts of things (#165554)" 
This reverts commit d61a9b88cf.

Reverted https://github.com/pytorch/pytorch/pull/165554 on behalf of https://github.com/malfet due to Looks like it broke serialization test, see aba8c43594/1 ([comment](https://github.com/pytorch/pytorch/pull/165554#issuecomment-3412765681))
 2025-10-16 20:41:37 +00:00
PyTorch MergeBot
b10f463b1a Revert "[DeviceMesh] Introduce private constructor instead of _create_mesh_from_ranks (#165555)" 
This reverts commit 99097b6d89.

Reverted https://github.com/pytorch/pytorch/pull/165555 on behalf of https://github.com/malfet due to Looks like it broke serialization test, see aba8c43594/1 ([comment](https://github.com/pytorch/pytorch/pull/165554#issuecomment-3412765681))
 2025-10-16 20:41:37 +00:00
PyTorch MergeBot
431c13cf61 Revert "[DeviceMesh] Simplify unflatten method (#165556)" 
This reverts commit 86fd4fc23e.

Reverted https://github.com/pytorch/pytorch/pull/165556 on behalf of https://github.com/malfet due to Looks like it broke serialization test, see aba8c43594/1 ([comment](https://github.com/pytorch/pytorch/pull/165554#issuecomment-3412765681))
 2025-10-16 20:41:37 +00:00
Ketan Ambati
aead9270f5 12/n : Remove fbandroid_compiler_flags (#165558) 
Summary:
Currently `get_c2_fbandroid_xplat_compiler_flags()` is reading the `caffe2.strip_glog` buckconfig which we want to get rid of.
This diff removes the `fbandroid_compiler_flags` arg and merges it with compiler_flags with a nested select and the select version of the method

The goal is to get rid of all the usages of `get_c2_fbandroid_xplat_compiler_flags()` so that we can get rid of the `caffe2.strip_glog` buckconfig

Test Plan: CI

bifferential Revision: D84626885

Pull Request resolved: https://github.com/pytorch/pytorch/pull/165558
Approved by: https://github.com/malfet
 2025-10-16 20:41:24 +00:00
Janani Sriram
9bf5b38c14 [Inductor][Triton][FP8] Refactor scaled_mm template to accept scaling mode (#164318) 
Summary: Refactor `scaled_mm` Inductor template to support template choice based on scaling mode. This modification sets up the infrastructure for adding new templates based on new scaling modes, such as deepseek-style scaling (a follow-up diff), as new scaling modes (deepseek, block, group) scale before the accumulation (as opposed to per-tensor and per-row scaling, which apply scaling after accumulation). This modification also further enables Inductor to infer a scaling type based on the shape of the scaling tensors, which makes existing infrastructure more extensible to new scaling modes.

Test Plan:
```
TORCHINDUCTOR_CACHE_DIR=~/personal/cache_dir_inductor CUDA_LAUNCH_BLOCKING=1 TORCH_USE_CUDA_DSA=1 TRITON_PRINT_AUTOTUNING=1 TRITON_ALWAYS_COMPILE=1 TORCH_LOGS=+inductor TORCHINDUCTOR_FORCE_DISABLE_CACHES=1 ENABLE_PERSISTENT_TMA_MATMUL=1 TORCHINDUCTOR_MAX_AUTOTUNE_GEMM=1 buck2 run mode/{opt,inplace} pytorch/tritonbench:run -- --op fp8_gemm --only torch_fp8_gemm,pt2_fp8_gemm --metrics tflops,accuracy --m 256 --n 768 --k 512 --output="/home/jananisriram/personal/random_bench.csv" --scaling_rowwise --atol=20 --rtol=2 2>&1 | tee ~/personal/random.log
```

bifferential Revision: D83591083

Pull Request resolved: https://github.com/pytorch/pytorch/pull/164318
Approved by: https://github.com/drisspg, https://github.com/slayton58
 2025-10-16 20:40:45 +00:00
Tristan Trouwen
aba8c43594 Register var for MTIA (#165382) 
Summary: Registers variance kernel

Reviewed By: srsuryadev

Differential Revision: D84546250

Pull Request resolved: https://github.com/pytorch/pytorch/pull/165382
Approved by: https://github.com/malfet
 2025-10-16 20:35:15 +00:00
linhaifeng
37f3ba274a [Fix] Use sys.executable instead of hardcoded python (#165633) 
Replace hardcoded "python" string with sys.executable to ensure correct Python interpreter is used. This fixes failures on systems with multiple Python runtimes or where "python" is not in PATH.

Similar to pytorch/pytorch#155918

Fixes #ISSUE_NUMBER

Pull Request resolved: https://github.com/pytorch/pytorch/pull/165633
Approved by: https://github.com/Skylion007
 2025-10-16 20:26:10 +00:00
IvanKobzarev
585b9dbb5e [async_tp] Support ag+mm with gather_dim lastdim of mat_A (#163068) 
Adding ag+mm support for the case, when gather_dim is last dim of matmul (reduction dim).

When we decompose matmul by reduction dimension we result in partials that needs additional reduction,
we allocate memory for accumulator.

Decomposition should not produce small (thin) mms that can not efficiently load the GPU. Limiting for minimal size of the shard 1024 (found empirically by testing in torchtitan).

scaled_mm is not supported yet for this case.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/163068
Approved by: https://github.com/ngimel
 2025-10-16 20:14:39 +00:00
Maggie Moss
d795fb225a [RFC] Add pyrefly to lintrunner (#165179) 
This will add pyrefly to lint runner as a warning only - and allow us to collect feedback about the tool before switching to pyrefly as the main type checker.

References the steps outlined here: : https://github.com/pytorch/pytorch/issues/163283:

test plan:
`lintrunner init`
`lintrunner`
confirm when pyrefly errors are present results look like: https://gist.github.com/maggiemoss/e6cb2d015dd1ded560ae1329098cf33f

Pull Request resolved: https://github.com/pytorch/pytorch/pull/165179
Approved by: https://github.com/ezyang
 2025-10-16 20:07:09 +00:00
tvukovic-amd
7df9aca529 [ROCm][Windows] Enable AOTriton runtime compile on Windows (#165538) 
AOTriton uses prebuilt runtime binaries if the user's ROCm version matches the ones used to generate the prebuilt runtime. However, since there's no prebuilt runtime available for Windows, this check needs to be bypassed for Windows. This PR enables it by changing condition to always build AOTriton runtime from source on Windows.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/165538
Approved by: https://github.com/xinyazhang, https://github.com/jeffdaily
 2025-10-16 19:51:43 +00:00
Shangdi Yu
d4a713cd9c Change forkserver test to only run below 3.13.8 (#165667) 
A multiprocessing bug is fixed in 3.13.8, see [https://docs.python.org/3.13/whatsnew/changelog.html](https://l.workplace.com/l.php?u=https%3A%2F%2Fdocs.python.org%2F3.13%2Fwhatsnew%2Fchangelog.html&h=AT0qUhHJq5c2UJvQaq9_MrSo0mVhwn1VOfq1nDQl2C1UOhDI80RMbzVayhG7LSAT1uYHKtkftKnBDwiGMhbw0YRvQLe5vwE01qejpPFautHvU3LXeOE1KChPykqz3qnCRzk7czu_iNzQ05shR4F1N_qYOzR5YxejA52ZZQ), [gh-126631](https://github.com/python/cpython/issues/126631)

So this test will fail when we update to python 3.13.8
Pull Request resolved: https://github.com/pytorch/pytorch/pull/165667
Approved by: https://github.com/malfet
 2025-10-16 19:34:10 +00:00
Sean McGovern
5daef30b26 158232 Fix autocast cache incorrectly retaining no_grad state (#165068) 
Fixes #158232
The autocast caching heuristic in `aten/src/ATen/autocast_mode.cpp:139` did not account for gradient mode state when deciding whether to cache. FSDP2 is not directly related.

~~This PR adds `GradMode::is_enabled()` check to caching condition. Caching is now disabled in `no_grad()` contexts to prevent storing tensors with incorrect gradient state. Ensures correctness at the cost of using cache.~~
This PR proposes separate caches for gradient-enabled and gradient-disabled modes.
Adds tests.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/165068
Approved by: https://github.com/ngimel, https://github.com/janeyx99
 2025-10-16 19:32:01 +00:00
Huy Do
6dedd34c31 [CD] Skip 12.9 build on Windows (#165665) 
Per title

Pull Request resolved: https://github.com/pytorch/pytorch/pull/165665
Approved by: https://github.com/Camyll, https://github.com/malfet
 2025-10-16 19:11:27 +00:00
Shunting Zhang
a303d6dda9 [inductor] don't try to reorder loops for template (#165601) 
fix https://github.com/pytorch/pytorch/issues/165579

Pull Request resolved: https://github.com/pytorch/pytorch/pull/165601
Approved by: https://github.com/yushangdi
 2025-10-16 19:05:21 +00:00
Jagadish Krishnamoorthy
7669ac9402 [ROCm] Add scaled_mm v2 support. (#165528) 
Add mx fp4 support in Blas.cpp.
Updated the scale_kernel_dispatch array and ScaledGemmImplementation enum to include MXFP4 support.
Modify the tests under test_scaled_matmul_cuda accordingly.

PYTORCH_TEST_WITH_ROCM=1 python test/test_scaled_matmul_cuda.py -v -k test_blockwise
115 test passed.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/165528
Approved by: https://github.com/jeffdaily
 2025-10-16 18:36:41 +00:00
Luca Wehrstedt
86fd4fc23e [DeviceMesh] Simplify unflatten method (#165556) 
By adding a few small helpers (e.g., a `splice` method to `_MeshLayout`, and making `_init_process_groups` static and thus stateless) we can substantially shorten the definition of the unflatten method, and help readability.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/165556
Approved by: https://github.com/fduwjj
ghstack dependencies: #165554, #165555
 2025-10-16 18:36:16 +00:00
Luca Wehrstedt
99097b6d89 [DeviceMesh] Introduce private constructor instead of _create_mesh_from_ranks (#165555) 
The refactoring of DeviceMesh is heavily constrained by the signature of its constructor, which is a public API which contains some "legacy" concepts which we'd love to get rid of, such as an explicit/materialized `mesh` Tensor.

In other languages the solution to this would be to add a private overload of the constructor. Python doesn't natively allow this, but in this PR I managed to build something that approximates it.

This new private constructor basically only takes `_layout`, `_global_rank_permutation`, and `mesh_dim_names`.

With such a constructor we can effectively simplify a lot of callsites and get rid of the `_create_mesh_from_ranks` helper method. That's a good thing because it was instantiating many DeviceMeshes in a for loop, which always felt unnecessary.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/165555
Approved by: https://github.com/fduwjj, https://github.com/fegin
ghstack dependencies: #165554
 2025-10-16 18:36:16 +00:00
eqy
a214371008 [FP8] Add other Blackwell compute-capabiilities to expected fail test_honor_sm_carveout (#165159) 
CUTLASS SM hint also isn't working for other Blackwells, need green context for carveout

Pull Request resolved: https://github.com/pytorch/pytorch/pull/165159
Approved by: https://github.com/Skylion007
 2025-10-16 18:35:06 +00:00
IvanKobzarev
7d87d7052e [inductor][bucketing] Fx collectives bucketing of multiple dtypes (#162470) 
Bucketing of multiple dtypes to be processed in one bucketed collective.

First target is to bucket bf16 and f32, but already can be used with other dtypes.

For now multidtype bucketing is only supported with "custom_ops" mode.
Non custom_ops needs additional work on inductor side.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/162470
Approved by: https://github.com/eellison
 2025-10-16 18:31:43 +00:00
arkadip-maitra
1a34ff4e04 Fixing get_local_rank() variable missing when compiled (#165432) 
Fixes #165215

Pull Request resolved: https://github.com/pytorch/pytorch/pull/165432
Approved by: https://github.com/bdhirsh
 2025-10-16 18:20:34 +00:00
Angel Li
fe5ccb1a74 bf16 support for per tensor backward (#165362) 
Adding bf16 for the backward pass of `torch._fake_quantize_learnable_per_tensor_affine()`.

Note that for testing, we modified the seed to avoid increasing tolerance due to cases where difference in Python vs CPP downcasting causes tensor mismatches. (e.g. 27.87704 vs  27.8408 before downcasting, 27.7500 vs 27.8750 after downcasting for Python vs CPP op)

Pull Request resolved: https://github.com/pytorch/pytorch/pull/165362
Approved by: https://github.com/andrewor14
 2025-10-16 17:47:01 +00:00
Thanh Ha
85586d7efc Make c7i the default for _linux-build.yml (#164747) 
Use linux.c7i.2xlarge as the default runner for the _linux-build.yml workflow. In testing we found that switching from c5 - c7i grants a 15-20% faster build times despite c7i costing 5% more. This should reduce costs of jobs using _linux-build.yml.

Relates to pytorch/test-infra#7175.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/164747
Approved by: https://github.com/atalman
 2025-10-16 17:37:51 +00:00
PyTorch MergeBot
e1d71a6b35 Revert "12/n : Remove fbandroid_compiler_flags (#165558)" 
This reverts commit d7ffa8b8a2.

Reverted https://github.com/pytorch/pytorch/pull/165558 on behalf of https://github.com/facebook-github-bot due to Diff reverted internally ([comment](https://github.com/pytorch/pytorch/pull/165558#issuecomment-3411879769))
 2025-10-16 17:18:56 +00:00
Luca Wehrstedt
d61a9b88cf [DeviceMesh] Prefer using _layout over _mesh for all sorts of things (#165554) 
The goal of this PR is to avoid storing the explicit `mesh` Tensor inside each DeviceMesh, and instead compute it on-the-fly when the end user needs it, and try to replace all of its internal usages with `_layout` and the newly-introduced `_global_rank_permutation` Tensor. The name of this attribute is up for debate. The advantage of the `_global_rank_permutation` Tensor is that it is _the same_ Tensor for the root mesh and all its children, so it doesn't need to be copied/reallocated.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/165554
Approved by: https://github.com/fduwjj
 2025-10-16 17:01:44 +00:00
Isuru Fernando
99b32a6750 [inductor] print 0.0 as 0 for triton (#164291) 
Fixes https://github.com/pytorch/pytorch/issues/164157
Fixes https://github.com/pytorch/pytorch/issues/164086

Pull Request resolved: https://github.com/pytorch/pytorch/pull/164291
Approved by: https://github.com/bobrenjc93
 2025-10-16 16:37:50 +00:00
Edward Yang
783da8b8e7 Repro for property related Dynamo graph break (#165609) 
Signed-off-by: Edward Yang <ezyang@meta.com>

Pull Request resolved: https://github.com/pytorch/pytorch/pull/165609
Approved by: https://github.com/albanD, https://github.com/gchanan, https://github.com/malfet, https://github.com/anijain2305
 2025-10-16 16:22:43 +00:00
Brian Hirsh
ed74dc054d add the option to disable functionalization in AOTDispatcher (#164577) 
I'm cleaning this PR up as a proper way of disabling functionalization via config in AOTDispatcher. I removed the non-functionalization related changes from the original version:

(1) preventing proxy mode (and functionalization) from incorrectly decomposing CIA ops (Ed has a PR for it here: https://github.com/pytorch/pytorch/pull/164939)

(2) preventing python-dispatcher-based decomps above autograd from running. I'm not doing this for now, will likely do it in a followup

Pull Request resolved: https://github.com/pytorch/pytorch/pull/164577
Approved by: https://github.com/ezyang
ghstack dependencies: #165372
 2025-10-16 15:44:11 +00:00
Brian Hirsh
f33c7e1a43 add and fix OpInfo tests for the default partitioner (#165372) 
I noticed the default partitioner was breaking in some dynamic shape tests, so prior to turning off functionalization I want to tweak it to pass all of our OpInfo tests

Pull Request resolved: https://github.com/pytorch/pytorch/pull/165372
Approved by: https://github.com/ezyang
 2025-10-16 15:44:11 +00:00
Yu, Guangye
219fb6aafc Refactor CUDAAllocatorConfig using ConfigTokenizer (#165281) 
* #165129
Pull Request resolved: https://github.com/pytorch/pytorch/pull/165281
Approved by: https://github.com/albanD
ghstack dependencies: #165129, #165131, #165135, #165136
 2025-10-16 15:26:50 +00:00
Yu, Guangye
515b5ff539 Remove unused code in CUDAAllocatorConfig (#165136) 
Pull Request resolved: https://github.com/pytorch/pytorch/pull/165136
Approved by: https://github.com/Skylion007
ghstack dependencies: #165129, #165131, #165135
 2025-10-16 15:26:50 +00:00