c555f241a6
Use GITHUB_TOKEN
2025-11-11 18:35:37 +00:00
742408cf7f
Remove github.token
2025-11-11 18:26:29 +00:00
d6c644b97e
Add actions: read permissions to be able to access artifacts for other workflow run
2025-11-11 18:24:47 +00:00
2ef8a3a2fc
try another way of setting token
2025-11-11 18:13:33 +00:00
68f019947d
Use secrets.GITHUB_TOKEN
2025-11-11 17:00:17 +00:00
99ec56b8bd
syntax
2025-11-11 16:53:30 +00:00
fc1ebb1a3f
Remove invalid workflow input
2025-11-11 16:52:15 +00:00
d7091e5e8e
Add workflow_dispatch trigger
2025-11-11 16:50:44 +00:00
4a71e2ebae
syntax
2025-11-11 16:49:05 +00:00
53e74a3ac0
Hardcode values for testing
2025-11-11 16:46:45 +00:00
b572cd15de
Try to use docker-builds artifacts
2025-11-11 16:24:56 +00:00
ada16deeaa
Name of artifact must be unique
2025-11-11 15:29:52 +00:00
a937421281
Upload artifacts named for each matrix element
2025-11-11 15:25:55 +00:00
e314ea03c3
Upload artifacts for docker-builds
2025-11-11 15:03:11 +00:00
b89f0103a4
Check if caching benefits are seen on test runners
2025-11-11 02:06:44 +00:00
e02e639c8c
Restore code to skip ghcr.io step if not push
2025-11-11 00:43:10 +00:00
4ba1846d8a
Pull ghcr.io image and tag it as ECR image
2025-11-11 00:34:06 +00:00
42dd417bc7
Run ghcr.io push step but don't actually push
2025-11-10 21:14:12 +00:00
1e34a45463
Merge branch 'main' into add_rocm_docker_caching
2025-11-10 15:06:10 -06:00
6e2c62ac17
Use ghcr.io image to see if docker pull time decreases
2025-11-10 21:02:06 +00:00
fe0bb7cf60
[export, 3.14] handle patching methods with functools.partial correctly in non-strict export ( #167396 )
...
Note: dynamo is not affected by this since patching class methods are not supported right now.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/167396
Approved by: https://github.com/angelayi
ghstack dependencies: #167382 , #167383 , #167384 , #167387
2025-11-10 20:52:05 +00:00
cf63b212e3
[3.14, dataloader] handle forkserver default mp start method in 3.14 ( #167387 )
...
Pull Request resolved: https://github.com/pytorch/pytorch/pull/167387
Approved by: https://github.com/malfet
ghstack dependencies: #167382 , #167383 , #167384
2025-11-10 20:52:05 +00:00
17e70ae459
[dynamo, 3.14] enable dynamo in 3.14 ( #167384 )
...
dynamo tests are passing in the CI PR above - so we could probably just enable dynamo right now.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/167384
Approved by: https://github.com/Skylion007 , https://github.com/mlazos
ghstack dependencies: #167382 , #167383
2025-11-10 20:52:05 +00:00
ad7db3617e
[inductor, 3.14] catch pickle.PicklingError exceptions ( #167383 )
...
Pull Request resolved: https://github.com/pytorch/pytorch/pull/167383
Approved by: https://github.com/aorenste
ghstack dependencies: #167382
2025-11-10 20:52:04 +00:00
5320ca3725
[inductor, 3.14] fix itertools.product pickle error in test_cpu_repro ( #167382 )
...
`inductor/test_cpu_cpp_wrapper` was failing since it was attempting to pickle`itertools.product`, and that is no longer picklable in 3.14. We work around by eagerly generating a list.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/167382
Approved by: https://github.com/atalman , https://github.com/malfet
2025-11-10 20:52:04 +00:00
3e4faca130
[torch.export] Refactor placeholder_naming_pass to reduce CCN ( #166600 )
...
Summary: Reduced CCN from 37 to 28 of placeholder_naming_pass method
Test Plan: Existing tests
Differential Revision: D85820388
Pull Request resolved: https://github.com/pytorch/pytorch/pull/166600
Approved by: https://github.com/angelayi
2025-11-10 20:44:18 +00:00
0c2f206ded
Typo fix - baddbmm_strategy ( #166963 )
...
This is called by registration with decorator, so function not called directly. For clarity, add the "b" for "batch" in function name.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/166963
Approved by: https://github.com/janeyx99
2025-11-10 20:35:42 +00:00
6cf21fa331
Fix -ffunction-sections, -fdata-sections not being added on aarch64. ( #166407 )
...
Preferred solution to #166380
Changes:
- Moved summary print to bottom of CMakeLists.txt
- Fix the problem 'add_compile_options' should be called before targets defined, so opted for `append_cxx_flag_if_supported` and `append_c_flag_if_supported` ( new ).
- Added extra verbosity so it can be seen when linker script added.
( unfortunately linker script has to be added per-target rather than globally due to ninja/cmake depdendency tracking ).
Also move summary print to bottom of CMakeLists.txt and improve logging
Pull Request resolved: https://github.com/pytorch/pytorch/pull/166407
Approved by: https://github.com/Aidyn-A , https://github.com/atalman
2025-11-10 20:32:08 +00:00
cdc8460f2c
Use c7i.2xlarge for H100 build ( #167466 )
...
The build system maybe oversized for what is necessary. Reduce the size to optimize costs. The default workflow runner is linux.c7i.2xlarge so we are just removing the runner definition in the workflow so that it uses the default.
Relates to pytorch/test-infra#7175 .
Pull Request resolved: https://github.com/pytorch/pytorch/pull/167466
Approved by: https://github.com/seemethere
2025-11-10 20:20:54 +00:00
86130aa2ca
Fix flaky memory profiler test [2] ( #167268 )
...
Fixes #167037
Move the module definition outside of the unit test so when we run the unit test multiple times, the module is not re-compiled.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/167268
Approved by: https://github.com/angelayi
2025-11-10 19:51:38 +00:00
9491830c79
move subgraph_has_impure_ops from node.is_impure into const_fold to unblock production ( #167443 )
...
Summary:
https://github.com/pytorch/pytorch/pull/166609 updates `node.is_impure` to consider a submodule as impure if submodule contains impure node. This in turn changes `graph.eliminate_dead_code()` function behavior, which does not eliminate nodes with side effects, see [pytorch documentation](https://docs.pytorch.org/docs/stable/fx.html#torch.fx.Graph.eliminate_dead_code )
> Remove all dead code from the graph, based on each node’s number of users, and whether the nodes have any side effects.
While this is correct that a submodule containing side-effectful ops is side-effectful and should not be dead code eliminated, some customers rely on the dead code elimination to eliminate submodules that contain impure ops which is the behavior before #166609 fix.
Due to production environment constraints, we have to revert https://github.com/pytorch/pytorch/pull/166609 and move the side-effectful submodule check logic to `const_fold.py`, which will correctly **not** const-fold a submodule that contains impure ops.
NOTE other call sites that use `node.is_impure()` to make decisions are still incorrectly eliminating side-effectful submodules, but we can't safely change that today.
## This pr
- move `_subgraph_has_impure_op` into `fx/experimental/const_fold.py`, check and prevent const-folding an impure submodule
- added a note in `node.is_impure` to highlight the incorrect behavior and context in case people go looking in the future.
Test Plan: run test_fx_const_fold and all tests pass
Differential Revision: D86641994
Pull Request resolved: https://github.com/pytorch/pytorch/pull/167443
Approved by: https://github.com/jfix71
2025-11-10 19:29:54 +00:00
04a85b4c21
[compile-on-one-rank] Step 1: DeviceId ( #166680 )
...
Add a "--virtual-local-rank" mode to torchrun. When used instead of passing the
local rank in LOCAL_RANK it uses a LOCAL_RANK of "0" and adjusts
CUDA_VISIBLE_DEVICES to reflect the desired GPU index.
Testing:
(tweaked run_train.sh to use `--log-dir`)
```
export NGPU=8
export CONFIG_FILE="./torchtitan/models/llama3/train_configs/debug_model.toml"
with-proxy ./run_train.sh --model.name compiler_toolkit.llama3 --compile.enable --parallelism.data_parallel_shard_degree=2 --parallelism.tensor_parallel_degree=4
```
And then comparing ranks:
Without --virtual-local-rank gives a lot of differences like:
```
[rank#]: mul_1: "f32[8, 512, 256]" = torch.ops.aten.mul.Tensor(mul, view_9); mul = None
-[rank#]: _to_copy_3: "bf16[8, 512, 256]" = torch.ops.aten._to_copy.default(mul_1, dtype = torch.bfloat16, layout = torch.strided, device = device(type='cuda', index=0)); mul_1 = None
+[rank#]: _to_copy_3: "bf16[8, 512, 256]" = torch.ops.aten._to_copy.default(mul_1, dtype = torch.bfloat16, layout = torch.strided, device = device(type='cuda', index=1)); mul_1 = None
[rank#]: detach: "f32[8, 512, 1]" = torch.ops.aten.detach.default(rsqrt); rsqrt = None
```
With --virtual-local-rank makes those differences go away.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/166680
Approved by: https://github.com/ezyang
2025-11-10 18:47:31 +00:00
a4437d76f0
Add some labeler rules that used to be in the autolabel bot ( #167330 )
...
See https://github.com/pytorch/test-infra/pull/7446 for the paths
Pull Request resolved: https://github.com/pytorch/pytorch/pull/167330
Approved by: https://github.com/huydhn
2025-11-10 18:38:42 +00:00
3ea829a337
Fix torch.cond HOP device in inductor ( #167354 )
...
Fixes #166918
The output device may not be on the same device as the predicate device.
```
python test/inductor/test_control_flow.py -k test_output_on_different_device
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/167354
Approved by: https://github.com/ydwu4 , https://github.com/zou3519
2025-11-10 18:19:38 +00:00
3966b5ad05
[BE] Fix out-of-bounds index_put in test_mps.py ( #167444 )
...
Discovered while enabling assertions on out-of-bounds accesses. Otherwise test fails with
```
ERROR: test_sdpa_mask_fp16_L6_S17_NH23_HS121 (__main__.TestSDPA.test_sdpa_mask_fp16_L6_S17_NH23_HS121)
----------------------------------------------------------------------
Traceback (most recent call last):
File "/Users/malfet/git/pytorch/pytorch/torch/testing/_internal/common_utils.py", line 3334, in wrapper
method(*args, **kwargs)
~~~~~~^^^^^^^^^^^^^^^^^
File "/Users/malfet/git/pytorch/pytorch/build/../test/test_mps.py", line 9494, in test_sdpa_mask_fp16_L6_S17_NH23_HS121
self._test_sdpa_mask(torch.float16, 7, 17, 23, 121)
~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/Users/malfet/git/pytorch/pytorch/build/../test/test_mps.py", line 9478, in _test_sdpa_mask
y_ref = F.scaled_dot_product_attention(q.cpu(), k.cpu(), v.cpu(), attn_mask=mask.cpu(), dropout_p=0.0, is_causal=False)
~~~~~^^
torch.AcceleratorError: index out of range
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/167444
Approved by: https://github.com/Skylion007 , https://github.com/manuelcandales
2025-11-10 18:19:28 +00:00
f6a79b2a4a
[inductor] Wrap pallas_call in jax.jit ( #167441 )
...
My understanding is this is needed for performance.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/167441
Approved by: https://github.com/oulgen
2025-11-10 17:29:56 +00:00
2fcf41dd8e
Add the ruff rule and skip everything for now ( #167360 )
...
Part of https://github.com/pytorch/pytorch/issues/164878
We can start narrowing the skips and remove them as PRs keep landing.
This PR is just to setup the scaffolding, fix will be in follow up
Pull Request resolved: https://github.com/pytorch/pytorch/pull/167360
Approved by: https://github.com/janeyx99
2025-11-10 17:10:15 +00:00
31ccd8f13e
[AOTI] Fix a mixed-device bug for scatter_add ( #167341 )
...
Summary: Fix https://github.com/pytorch/pytorch/issues/166841 . AOTI incorrectly generates a call to aoti_torch_cuda_scatter_reduce_two_out while the op should actually run on CPU. Fix by using the correct device when calling _generate_scatter_fallback in the wrapper codegen.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/167341
Approved by: https://github.com/yushangdi
2025-11-10 16:59:44 +00:00
59307ca1bc
[BE] adding documentation ( #167334 )
...
`torch.ao.quantization` and `torch.fx.experimental`
<img width="833" height="518" alt="Screenshot 2025-11-07 at 3 20 54 PM" src="https://github.com/user-attachments/assets/47b72f28-29bd-4bab-b41f-24d97419e411 " />
<img width="892" height="560" alt="Screenshot 2025-11-07 at 3 20 45 PM" src="https://github.com/user-attachments/assets/129825ab-6706-41f2-964d-8774debab18c " />
Pull Request resolved: https://github.com/pytorch/pytorch/pull/167334
Approved by: https://github.com/janeyx99
2025-11-10 14:46:42 +00:00
c28475db7c
Update slow tests ( #166844 )
...
This PR is auto-generated weekly by [this action](https://github.com/pytorch/pytorch/blob/main/.github/workflows/weekly.yml ).
Update the list of slow tests.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/166844
Approved by: https://github.com/pytorchbot
2025-11-10 12:39:27 +00:00
74aec83841
[xla hash update] update the pinned xla hash ( #167452 )
...
This PR is auto-generated nightly by [this action](https://github.com/pytorch/pytorch/blob/main/.github/workflows/nightly.yml ).
Update the pinned xla hash.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/167452
Approved by: https://github.com/pytorchbot
2025-11-10 12:03:01 +00:00
52e744d68a
[DTensor] Support convert StridedShard to shard order and vice versa ( #166740 )
...
We plan to use `StridedShard` to express `shard_order`. This PR adds the function to support the conversion between `StridedShard` and `shard_order`.
I moved some test related function into torch/testing/_internal/common_utils.py. We may only care about **_dtensor_spec.py** and **test_utils.py** in this PR for the review.
### How to convert shard order to StridedShard:
Considering the example:
- placements = $[x_0, x_1, x_2, x_3, x_4]$, all $x_?$ are shard on the same tensor dim.
Let's see how the shard order will impact the split_factor (sf). We loop from right to left in the placements to construct the split_factor by assuming different shard order. Starting from $x_4$, this should be a normal shard.
Then $x_3$. There are two possibilities, $x_3$'s order can be before $x_4$. If so, $x_3$'s sf=1, because $x_3$ is before $x_4$ in the placements. Else $x_3$'s order is after $x_4$, then the $x_3$'s sf should be the mesh dim size of $x_4$, which is $T(x_4)$:
<img width="820" height="431" alt="image" src="https://github.com/user-attachments/assets/f53b4b24-2523-42cc-ad6f-41f3c280db70 " />
We can use this method to decide on the split factor for $x_2$, $x_1$ and so on.
### How to convert StridedShard to shard order:
This follows the same method above. We check all possible paths and use the real split_factor to see which path matchs the split_factor. If no such matches, the StridedShard is unable to be converted to shard order.
---
Pull Request resolved: https://github.com/pytorch/pytorch/pull/166740
Approved by: https://github.com/ezyang
2025-11-10 09:35:10 +00:00
3cfbf98ea9
[xpu][feature] Add XPU support on torch.accelerator.get_memory_info ( #162564 )
...
# Motivation
Support XPU for `torch.accelerator.get_memory_info`.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/162564
Approved by: https://github.com/albanD
ghstack dependencies: #156812
2025-11-10 05:34:49 +00:00
47db55258b
[MPS] sparse sparse mm ( #167013 )
...
Sparse sparse mm op implementation
Pull Request resolved: https://github.com/pytorch/pytorch/pull/167013
Approved by: https://github.com/malfet
2025-11-10 05:27:49 +00:00
50af6f3393
[MPS] erfinv for sparse mps ( #166711 )
...
Should be merged after #166708
Pull Request resolved: https://github.com/pytorch/pytorch/pull/166711
Approved by: https://github.com/Skylion007 , https://github.com/malfet
2025-11-10 05:25:31 +00:00
e545ba2d34
[DTensor] Fix Conv behavior for replicate stategy ( #167402 )
...
Pass `dim_map` to `_requires_data_exchange` and return False if both spatial and channels dimensions are replicated
Modify `test_conv1d` and `test_conv3d` to check values rather than just shape, and replicate `conv3d` across batch dimension
In general, feels like current Convolution implementation was written to work only if tensor is sharded across last dimention
Pull Request resolved: https://github.com/pytorch/pytorch/pull/167402
Approved by: https://github.com/ezyang
2025-11-10 05:13:42 +00:00
a058bbdd6f
[xpu][test] Enable profiler test for XPU ( #165423 )
...
Fixes #165130
Pull Request resolved: https://github.com/pytorch/pytorch/pull/165423
Approved by: https://github.com/EikanWang , https://github.com/atalman , https://github.com/mlazos
2025-11-10 04:02:59 +00:00
2c78080ec0
Register functorch XPU/HPU dispatch keys ( #167095 )
...
Fixes TestOperatorsXPU.test_data_write_errors_under_transform_xpu https://github.com/intel/torch-xpu-ops/issues/2237
Tests on other devices throw runtime error "_mutating directly with `.data` inside functorch transform is not allowed._", but XPU/HPU fails earlier on `_has_compatible_shallow_copy_type`. This check is not met only when calling tensor.data inside functorch call.
```cpp
bool _has_compatible_shallow_copy_type(const Tensor& self, const Tensor& from) {
return self.unsafeGetTensorImpl()->has_compatible_shallow_copy_type(
from.key_set());
}
```
### t.data
| Tensor | Device | Dispatch Keys |
|--------|---------|---------------|
| `self` | `xpu` | `XPU, ADInplaceOrView, AutogradXPU, AutocastXPU` |
| `from` | `cpu` | `CPU, ADInplaceOrView, AutogradCPU, AutocastCPU` |
### t.data inside functorch transform
| Tensor | Device | Dispatch Keys |
|--------|---------|---------------|
| `self` | `xpu` | `ADInplaceOrView, AutogradOther, FuncTorchGradWrapper` |
| `from` | `cpu` | `CPU, ADInplaceOrView, AutogradCPU, AutocastCPU, FuncTorchGradWrapper` |
### t.data inside functorch transform + XPU dispatch key
| Tensor | Device | Dispatch Keys |
|--------|---------|---------------|
| `self` | `xpu` | `XPU, ADInplaceOrView, AutogradXPU, AutocastXPU, FuncTorchGradWrapper` |
| `from` | `cpu` | `CPU, ADInplaceOrView, AutogradCPU, AutocastCPU, FuncTorchGradWrapper` |
Pull Request resolved: https://github.com/pytorch/pytorch/pull/167095
Approved by: https://github.com/guangyey , https://github.com/albanD
2025-11-10 03:10:22 +00:00
fe6615e397
Swap pallas test shard to 12.8 ( #167428 )
...
Getting some weird failures building cuda13, lets stick to what we know works
Pull Request resolved: https://github.com/pytorch/pytorch/pull/167428
Approved by: https://github.com/jansel
2025-11-10 02:42:35 +00:00
abf31db2cc
Introduce a new API torch.accelerator.get_memory_info ( #156812 )
...
# Motivation
`torch.cuda.mem_get_info` and `torch.xpu.mem_get_info` are widely used in other popular repos, such as
- 076313bd09/python/sglang/srt/utils.py (L378)7ecc2d7f39/src/accelerate/utils/modeling.py (L822)7ba34b1241/vllm/worker/worker.py (L150)https://github.com/pytorch/pytorch/pull/156812
Approved by: https://github.com/albanD
2025-11-10 01:57:39 +00:00
