id int64 2.74B 3.05B | title stringlengths 1 255 | user stringlengths 2 26 | state stringclasses 2
values | labels listlengths 0 24 | comments int64 0 206 | author_association stringclasses 4
values | body stringlengths 7 62.5k ⌀ | is_title bool 1
class |
|---|---|---|---|---|---|---|---|---|
3,031,972,782 | Use swap_tensors path in nn.Module.to for all subclasses that override __torch_dispatch__ | mikaylagawarecki | open | [
"Merged",
"Reverted",
"ciflow/trunk",
"release notes: nn",
"ci-no-td"
] | 9 | CONTRIBUTOR | Fixes https://github.com/pytorch/pytorch/issues/148977
Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom):
* __->__ #152539
| true |
3,031,972,618 | Disable SLEEF implementation of vec::maximum in vec128_float_neon.h | Accelerate aten::hardtanh_ by 21x | Rohanjames1997 | closed | [
"module: cpu",
"triaged",
"open source",
"Merged",
"ciflow/trunk",
"topic: not user facing"
] | 4 | CONTRIBUTOR | The `has_inf_nan` implementation in `vec::maximum` is scalar, and it slows down certain activations like `tanh` by almost 20 times. Additionally, the `vec::minimum` function simply uses NEON intrinsics and not SLEEF. This PR makes the two fns similar in implementation.
Besides, the SLEEF function `Sleef_fmaxf4` ultimately invokes the `vmaxq_f32` NEON intrinsic through [vmax_vf_vf_vf](https://github.com/shibatch/sleef/blob/d28232a309e06bcb75e9fb0f6262d9251739fd1e/src/arch/helperadvsimd.h#L253).
From a single threaded profile of mobilenet on an Arm Neoverse-V2 machine (code below), the `aten::hardtanh_` takes **5.653ms** per function call while using the current PyTorch 2.7 wheel, whereas it takes **266.096us** per function call while simply using `vmaxq_f32` - a 21x speedup, and overall inference is 1.8x faster.
___
Run the below script: `OMP_NUM_THREADS=1 python profile_mobilenet.py --iterations 10`
<details >
<summary>profile_mobilenet.py</summary>
```
import torch
import torchvision.models as models
from torch.profiler import profile, record_function, ProfilerActivity
import argparse
torch.manual_seed(42)
def load_mobilenet():
model = models.mobilenet_v2(pretrained=True)
model.eval()
return model
def generate_sample_input(batch_size=8):
return torch.randn(batch_size, 3, 224, 224)
def warmup(model, sample_input, num_warmup=10):
with torch.inference_mode():
for _ in range(num_warmup):
_ = model(sample_input)
def parse_args():
parser = argparse.ArgumentParser()
parser.add_argument('--batch_size', type=int, default=8)
parser.add_argument('--iterations', type=int, default=100)
return parser.parse_args()
def main():
args = parse_args()
model = load_mobilenet()
sample_input = generate_sample_input(args.batch_size)
print("Warming up...")
warmup(model, sample_input)
print("Warmup complete.")
with profile(activities=[ProfilerActivity.CPU], record_shapes=True) as prof:
with torch.inference_mode():
for i in range(args.iterations):
with record_function("model_inference"):
outputs = model(sample_input)
print(prof.key_averages().table(sort_by="cpu_time_total"))
print(f"Throughput: {(args.iterations * args.batch_size / (prof.profiler.self_cpu_time_total / 1e6)):.3f} images/s")
if __name__ == "__main__":
main()
```
</details>
<details>
<summary>Profiler output using the current Pytorch 2.7 wheel </summary>
```
-------------------------------- ------------ ------------ ------------ ------------ ------------ ------------
Name Self CPU % Self CPU CPU total % CPU total CPU time avg # of Calls
-------------------------------- ------------ ------------ ------------ ------------ ------------ ------------
model_inference 2.39% 101.839ms 100.00% 4.254s 425.437ms 10
aten::hardtanh_ 0.02% 905.454us 46.50% 1.978s 5.653ms 350
aten::hardtanh 0.03% 1.239ms 46.48% 1.977s 5.650ms 350
aten::clamp 46.45% 1.976s 46.45% 1.976s 5.646ms 350
aten::conv2d 0.06% 2.468ms 43.89% 1.867s 3.591ms 520
aten::convolution 0.06% 2.491ms 43.83% 1.865s 3.586ms 520
aten::_convolution 0.13% 5.546ms 43.77% 1.862s 3.581ms 520
aten::thnn_conv2d 0.04% 1.658ms 24.13% 1.027s 3.019ms 340
aten::_slow_conv2d_forward 23.99% 1.021s 24.09% 1.025s 3.014ms 340
aten::mkldnn_convolution 14.42% 613.285ms 19.51% 829.885ms 4.610ms 180
aten::batch_norm 0.06% 2.368ms 6.89% 292.928ms 563.323us 520
aten::_batch_norm_impl_index 0.11% 4.600ms 6.83% 290.560ms 558.769us 520
aten::native_batch_norm 6.60% 280.762ms 6.69% 284.567ms 547.244us 520
aten::contiguous 0.01% 623.099us 5.01% 213.152ms 1.184ms 180
aten::clone 0.02% 988.729us 5.00% 212.529ms 1.181ms 180
aten::copy_ 4.94% 210.315ms 4.94% 210.315ms 1.052ms 200
aten::linear 0.00% 58.347us 0.18% 7.659ms 765.905us 10
aten::addmm 0.17% 7.373ms 0.18% 7.483ms 748.309us 10
aten::empty 0.17% 7.161ms 0.17% 7.161ms 1.790us 4000
aten::add 0.11% 4.742ms 0.11% 4.742ms 47.419us 100
aten::empty_like 0.03% 1.315ms 0.09% 3.890ms 5.557us 700
aten::view 0.05% 1.933ms 0.05% 1.933ms 2.801us 690
aten::as_strided_ 0.04% 1.599ms 0.04% 1.599ms 8.885us 180
aten::resize_ 0.04% 1.493ms 0.04% 1.493ms 2.871us 520
aten::adaptive_avg_pool2d 0.00% 55.360us 0.04% 1.491ms 149.051us 10
aten::mean 0.00% 116.997us 0.03% 1.435ms 143.515us 10
aten::sum 0.02% 935.980us 0.02% 992.121us 99.212us 10
aten::detach 0.02% 707.217us 0.02% 707.217us 2.080us 340
aten::div_ 0.00% 161.473us 0.01% 326.035us 32.604us 10
aten::to 0.00% 178.193us 0.01% 321.253us 0.892us 360
aten::_nnpack_available 0.01% 302.835us 0.01% 302.835us 0.891us 340
aten::_to_copy 0.00% 63.170us 0.00% 143.060us 14.306us 10
aten::t 0.00% 49.759us 0.00% 117.621us 11.762us 10
aten::transpose 0.00% 40.637us 0.00% 67.862us 6.786us 10
aten::flatten 0.00% 42.634us 0.00% 58.867us 5.887us 10
aten::fill_ 0.00% 56.141us 0.00% 56.141us 5.614us 10
aten::expand 0.00% 42.687us 0.00% 48.930us 4.893us 10
aten::empty_strided 0.00% 40.589us 0.00% 40.589us 4.059us 10
aten::as_strided 0.00% 33.468us 0.00% 33.468us 1.673us 20
aten::resolve_conj 0.00% 9.066us 0.00% 9.066us 0.453us 20
aten::dropout 0.00% 5.782us 0.00% 5.782us 0.578us 10
-------------------------------- ------------ ------------ ------------ ------------ ------------ ------------
Self CPU time total: 4.254s
Throughput: 18.804 images/s
```
</details>
<details>
<summary>Profiler output after this PR's changes </summary>
```
-------------------------------- ------------ ------------ ------------ ------------ ------------ ------------
Name Self CPU % Self CPU CPU total % CPU total CPU time avg # of Calls
-------------------------------- ------------ ------------ ------------ ------------ ------------ ------------
model_inference 4.43% 104.484ms 100.00% 2.359s 235.883ms 10
aten::conv2d 0.10% 2.313ms 79.19% 1.868s 3.592ms 520
aten::convolution 0.10% 2.293ms 79.09% 1.866s 3.588ms 520
aten::_convolution 0.23% 5.436ms 78.99% 1.863s 3.583ms 520
aten::thnn_conv2d 0.08% 1.799ms 44.29% 1.045s 3.072ms 340
aten::_slow_conv2d_forward 44.03% 1.039s 44.21% 1.043s 3.067ms 340
aten::mkldnn_convolution 24.91% 587.584ms 34.47% 812.992ms 4.517ms 180
aten::batch_norm 0.10% 2.350ms 11.83% 279.113ms 536.757us 520
aten::_batch_norm_impl_index 0.20% 4.788ms 11.73% 276.764ms 532.238us 520
aten::native_batch_norm 11.30% 266.660ms 11.46% 270.420ms 520.038us 520
aten::contiguous 0.02% 575.723us 9.41% 222.080ms 1.234ms 180
aten::clone 0.04% 1.061ms 9.39% 221.504ms 1.231ms 180
aten::copy_ 9.29% 219.131ms 9.29% 219.131ms 1.096ms 200
aten::hardtanh_ 0.04% 917.669us 3.95% 93.133ms 266.096us 350
aten::hardtanh 0.05% 1.130ms 3.91% 92.216ms 263.474us 350
aten::clamp 3.85% 90.894ms 3.86% 91.086ms 260.246us 350
aten::linear 0.00% 68.681us 0.33% 7.899ms 789.945us 10
aten::addmm 0.32% 7.598ms 0.33% 7.707ms 770.673us 10
aten::empty 0.30% 7.176ms 0.30% 7.176ms 1.794us 4000
aten::add 0.20% 4.627ms 0.20% 4.627ms 46.268us 100
aten::empty_like 0.06% 1.316ms 0.17% 3.973ms 5.676us 700
aten::view 0.08% 2.001ms 0.08% 2.001ms 2.899us 690
aten::adaptive_avg_pool2d 0.00% 53.745us 0.07% 1.548ms 154.791us 10
aten::resize_ 0.06% 1.533ms 0.06% 1.533ms 2.948us 520
aten::as_strided_ 0.06% 1.521ms 0.06% 1.521ms 8.450us 180
aten::mean 0.00% 117.637us 0.06% 1.494ms 149.417us 10
aten::sum 0.04% 973.291us 0.04% 1.013ms 101.342us 10
aten::detach 0.03% 652.224us 0.03% 652.224us 1.918us 340
aten::div_ 0.01% 195.077us 0.02% 363.103us 36.310us 10
aten::to 0.01% 212.758us 0.02% 359.655us 0.999us 360
aten::_nnpack_available 0.01% 295.235us 0.01% 295.235us 0.868us 340
aten::_to_copy 0.00% 68.726us 0.01% 146.897us 14.690us 10
aten::t 0.00% 53.873us 0.01% 124.033us 12.403us 10
aten::transpose 0.00% 42.512us 0.00% 70.160us 7.016us 10
aten::flatten 0.00% 44.040us 0.00% 66.631us 6.663us 10
aten::expand 0.00% 44.632us 0.00% 51.177us 5.118us 10
aten::fill_ 0.00% 40.134us 0.00% 40.134us 4.013us 10
aten::empty_strided 0.00% 35.291us 0.00% 35.291us 3.529us 10
aten::as_strided 0.00% 34.193us 0.00% 34.193us 1.710us 20
aten::resolve_conj 0.00% 8.594us 0.00% 8.594us 0.430us 20
aten::dropout 0.00% 6.758us 0.00% 6.758us 0.676us 10
-------------------------------- ------------ ------------ ------------ ------------ ------------ ------------
Self CPU time total: 2.359s
Throughput: 33.915 images/s
```
</details>
___
Using torchbench, the models `mobilenet_v2` and `mobilenet_v3_large` showed improvements as expected too.
Before -> After (latency in ms)
```
"mobilenet_v3_large-eval_latency": 1207.212 -> 844.902
"mobilenet_v2-eval_latency": 1029.834 -> 662.476
```
cc @jgong5 @mingfeima @XiaobingSuper @sanchitintel @ashokei @jingxu10 @jerryzh168 | true |
3,031,970,557 | [CI] Use cmake from pip instead of conda in CI docker images | clee2000 | open | [
"Merged",
"Reverted",
"ciflow/trunk",
"topic: not user facing",
"ciflow/inductor",
"ci-no-td"
] | 13 | CONTRIBUTOR | As in title
idk how the install_cmake script is used because I see it being called with 3.18 but when I look at the build jobs some say 3.18 and others 3.31
Just make everything install cmake via the requirements-ci.txt. I don't know if the comment at https://github.com/pytorch/pytorch/blob/5d36485b4aa5823bb9ee5eb070548c4ee355f3b2/.ci/docker/common/install_conda.sh#L78 still holds, but pretty much every build has CONDA_CMAKE set to true, so I'm just defaulting to installing through pip
Also defaulting to 4.0.0 everywhere except the executorch docker build because executorch reinstalls 3.31.something | true |
3,031,883,217 | [PT2] Port replace_lce_with_matmul / replace_first_lce_with_fused_matmul_lce to PT2 pre_grad passes (#152450) | kqfu | closed | [
"fb-exported",
"Merged",
"ciflow/trunk",
"topic: not user facing",
"module: inductor",
"ciflow/inductor"
] | 7 | CONTRIBUTOR | Summary:
Same with D71358949, but removing newly added log to avoid test failures.
Port over replace_lce_with_matmul and replace_first_lce_with_fused_matmul_lce to PT2 pre_grad pass.
Original dper pass diffs: D67884534, D68123479, D68384238
Test Plan:
Test 1. Covers replace_lce_with_matmul and case 1 of replace_first_lce_with_fused_matmul_lce
```
CUDA_VISIBLE_DEVICES=6 TORCH_LOGS=+inductor,aot TORCH_COMPILE_DEBUG=1 TORCHINDUCTOR_MAX_AUTOTUNE=1 buck2 run mode/opt-split-dwarf mode/inplace -c fbcode.platform010_cuda_version=12 -c fbcode.nvcc_arch=h100 caffe2/torch/fb/model_transform/experimental/benchmark:mts_gpu_benchmark -- --model-path=manifold://ads_storage_fblearner/tree/user/facebook/fblearner/predictor/669809193/0/gpu_lowering/input.predictor.disagg.gpu.merge --lower-backend="AOT_INDUCTOR" --add_passes="use_matmul_fuse_lce_replace_first_LCE,use_contiguous_linear_reduction_replace_linear_reduction" --batch-size=3072 --gpu-trace --disable_acc_tracer=true 2>&1 | tee ~/logs/disable_acc_tracer/aoti_cmf_ctr_triton_669809193_0_diable_acc.log
```
Log: P1798246938
Test 2. Covers replace_lce_with_matmul and case 2 of replace_first_lce_with_fused_matmul_lce
```
CUDA_VISIBLE_DEVICES=7 TORCH_LOGS=+inductor,aot TORCH_COMPILE_DEBUG=1 TORCHINDUCTOR_MAX_AUTOTUNE=1 buck2 run mode/opt-split-dwarf mode/inplace -c fbcode.platform010_cuda_version=12 -c fbcode.nvcc_arch=h100 caffe2/torch/fb/model_transform/experimental/benchmark:mts_gpu_benchmark -- --model-path=manifold://ads_storage_fblearner/tree/user/facebook/fblearner/predictor/677734158/9/gpu_lowering/input.predictor.disagg.gpu.merge --lower-backend="AOT_INDUCTOR" --add_passes="use_matmul_fuse_lce_replace_first_LCE,use_matmul_lce_replace_normal_LCE" --batch-size=3072 --gpu-trace --disable_acc_tracer=true 2>&1 | tee ~/logs/disable_acc_tracer/aoti_cmf_ctr_triton_677734158_9_diable_acc.log
```
Log: P1798246675
Seeing logs like
`[Pre grad(predispatch IR)] Apply use_matmul_fuse_lce_replace_first_LCE pass, save before/after graph to /tmp/tmp8lyzoh79, graph before/after are the same = False`
Differential Revision: D73934142
cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @ipiszy @chenyang78 @kadeng @muchulee8 @amjames @chauhang @aakhundov | true |
3,031,815,308 | optree package status in PyTorch | zou3519 | open | [
"high priority",
"triaged",
"module: pytree",
"dependency issue"
] | 5 | CONTRIBUTOR | ## Motivation
"optree >= 0.13.0" is an optional dependency for PyTorch. There's not a good way to specify this in the package metadata, so we have code checks doing this. We ran into some issues with the PyTorch 2.7 release where the "optional dependency checking" code broken and we ended up (1) crashing on optree < 0.13.0 and (2) hard requiring optree >= 0.13.0. Furthermore, I submitted two fixes for this, where the first fix didn't work and the second one actually worked.
We should fix this so that future releases (e.g. 2.8.0) do not run into the risk of this problem.
## Pitch
I see three high-level solutions:
1. We continue to have optree as an optional dependency. We would need to beef up our CI. In particular, we want to test three configurations: optree == 0.13.0, no optree, and optree < 0.13.0
2. We take the optree pypi package as a mandatory dependency.
3. We take optree as a required pytorch submodule and build it into PyTorch.
From discussions with @malfet, @seemethere, @atalman and @albanD it sounded like we preferred (3). The risk of doing (2) is that we get into dependency hell: if a third-party library pins optree and pytorch pins optree then we end up in a not-so-good-place. The direction our project is going in is taking optree as a mandatory requirement somehow, so (1) seems not worth it.
Thoughts? @XuehaiPan @angelayi
cc @ezyang @gchanan @kadeng @msaroufim @XuehaiPan | true |
3,031,617,655 | AsyncCollectiveTensor doesn't trigger wait upon dtype cast | lw | closed | [
"oncall: distributed",
"triaged"
] | 9 | CONTRIBUTOR | ### 🐛 Describe the bug
This repro fails:
```py
import tempfile
import torch
import torch.distributed as dist
import torch.distributed._functional_collectives as funcol
from torch.distributed.device_mesh import DeviceMesh
from torch.distributed.tensor.experimental import local_map
from torch.distributed.tensor import DTensor, Replicate, Shard
WORKAROUND = False
@torch.library.custom_op(
"custom_ns::custom_op_fwd",
mutates_args=(),
device_types="cuda",
)
def custom_op_fwd(
input_: torch.Tensor,
) -> torch.Tensor:
input_ = funcol.all_gather_tensor(input_, gather_dim=0, group=dist.group.WORLD)
print(f"type after funcol: {type(input_)=}")
if WORKAROUND:
input_ = input_.mul(1)
print(f"type after mul (no-op): {type(input_)=}")
input_ = input_.float()
print(f"type after cast to float: {type(input_)=}")
return input_.sum(dim=-1)
@torch.library.custom_op(
"custom_ns::custom_op_bwd",
mutates_args=(),
device_types="cuda",
)
def custom_op_bwd(
input_: torch.Tensor,
grad_output: torch.Tensor,
) -> torch.Tensor:
grad_output = grad_output.view((-1, 1)).expand((-1, input_.shape[1])).contiguous()
grad_output = funcol.reduce_scatter_tensor(
grad_output, "avg", scatter_dim=0, group=dist.group.WORLD
)
print(f"type after funcol: {type(grad_output)=}")
if WORKAROUND:
grad_output = grad_output.mul(1)
print(f"type after mul (no-op): {type(grad_output)=}")
grad_output = grad_output.bfloat16()
print(f"type after cast to bfloat16: {type(grad_output)=}")
return grad_output
def custom_op_setup_context(
ctx: torch.autograd.function.FunctionCtx,
inputs: tuple[torch.Tensor],
output: torch.Tensor,
) -> None:
ctx.save_for_backward(*inputs)
def custom_op_bwd_bridge(
ctx: torch.autograd.function.FunctionCtx,
grad_output: torch.Tensor,
) -> tuple[torch.Tensor]:
input_, = ctx.saved_tensors
grad_input = custom_op_bwd(input_, grad_output)
return grad_input
torch.library.register_autograd(
"custom_ns::custom_op_fwd",
custom_op_bwd_bridge,
setup_context=custom_op_setup_context,
)
def reference(input_: torch.Tensor) -> torch.Tensor:
return input_.float().sum(-1)
def our(input_: torch.Tensor) -> torch.Tensor:
return local_map(custom_op_fwd, out_placements=[Replicate(),])(input_)
def run_repro(rank: int, world_size: int, rdv_dir: str) -> None:
torch.manual_seed(0)
torch.cuda.set_device(rank)
input_ = torch.randn((16384, 4096), dtype=torch.bfloat16, device="cuda")
grad_output = torch.randn((16384,), dtype=torch.bfloat16, device="cuda")
ref_output, ref_grad_input_ = torch.autograd.functional.vjp(
reference, input_, grad_output
)
dist.init_process_group(
backend="nccl",
rank=rank,
world_size=world_size,
init_method=f"file://{rdv_dir}/rdv",
)
device_mesh = DeviceMesh.from_group(
dist.group.WORLD, device_type="cuda", mesh_dim_names=("tp",)
)
d_input_ = DTensor.from_local(input_.tensor_split(world_size)[rank], device_mesh, (Shard(0),))
d_grad_output = DTensor.from_local(grad_output, device_mesh, (Replicate(),))
our_output, our_grad_input_ = torch.autograd.functional.vjp(
our, d_input_, d_grad_output
)
torch.testing.assert_close(our_output.full_tensor(), ref_output)
torch.testing.assert_close(our_grad_input_.full_tensor(), ref_grad_input_)
if __name__ == "__main__":
world_size = torch.cuda.device_count()
with tempfile.TemporaryDirectory() as rdv_dir:
torch.multiprocessing.spawn(
run_repro,
args=(world_size, rdv_dir),
nprocs=world_size,
)
```
The issue seems to be that for an AsyncCollectiveTensor `t`, invoking `t.float()` does _not_ trigger the `wait_tensor`, in which case it would return a regular `torch.Tensor`, but instead it returns a new AsyncCollectiveTensor with garbage data. Inserting a dummy `.mul(1)` triggers the wait and makes everything work again.
### Versions
The above repro fails under PyTorch 2.6.0 (with CUDA 12.6, on a H100 machine).
The original issue I had was also failing under similar conditions, but was _passing_ with a nightly build of PyTorch, hence it's possible that somehow the issue has been fixed in main.
cc @H-Huang @awgu @wanchaol @fegin @fduwjj @wz337 @wconstab @d4l3k @ezyang @gchanan @zou3519 @kadeng @msaroufim | true |
3,031,611,336 | Do not check out nccl when not building it | mgorny | closed | [
"triaged",
"open source",
"Merged",
"ciflow/trunk",
"topic: not user facing"
] | 7 | CONTRIBUTOR | Add additional conditions to `build_pytorch_libs.py` to avoid fetching NCCL when `USE_CUDA` or `USE_NCCL` are disabled. While at it, adjust the existing condition for `USE_SYSTEM_NCCL` to use the utility function.
| true |
3,031,471,033 | The 2.7.0 release tarball is missing `.ci/docker/ci_commit_pins/nccl-cu12.txt` required for building | mgorny | open | [
"oncall: package/deploy"
] | 1 | CONTRIBUTOR | ### 🐛 Describe the bug
When trying to build PyTorch 2.7.0 from the `.tar.gz` attached to [the release](https://github.com/pytorch/pytorch/releases/tag/v2.7.0), I'm seeing the following error:
```pytb
Traceback (most recent call last):
File "/var/tmp/conda-bld/work/setup.py", line 1503, in <module>
main()
File "/var/tmp/conda-bld/work/setup.py", line 1170, in main
build_deps()
File "/var/tmp/conda-bld/work/setup.py", line 490, in build_deps
build_pytorch(
File "/var/tmp/conda-bld/work/tools/build_pytorch_libs.py", line 122, in build_pytorch
checkout_nccl()
File "/var/tmp/conda-bld/work/tools/build_pytorch_libs.py", line 102, in checkout_nccl
release_tag = read_nccl_pin()
^^^^^^^^^^^^^^^
File "/var/tmp/conda-bld/work/tools/build_pytorch_libs.py", line 97, in read_nccl_pin
with open(nccl_pin_path) as f:
^^^^^^^^^^^^^^^^^^^
FileNotFoundError: [Errno 2] No such file or directory: '/var/tmp/conda-bld/work/.ci/docker/ci_commit_pins/nccl-cu12.txt'
```
And indeed, the top-level `.ci` directory is entirely missing from the tarball. However, the build code unconditionally reads it:
https://github.com/pytorch/pytorch/blob/134179474539648ba7dee1317959529fbd0e7f89/tools/build_pytorch_libs.py#L101-L122
### Versions
n/a | true |
3,031,031,190 | [inductor][triton] Inductor is not compatible with the latest upstream Triton | xuzhao9 | closed | [
"oncall: pt2",
"upstream triton"
] | 1 | CONTRIBUTOR | ### 🐛 Describe the bug
Upstream https://github.com/triton-lang/triton/commit/850525276426fb9814399a8e0ee8fdf744229b02 removes `launch_enter_hook` from triton.compiler.CompiledKernel and https://github.com/triton-lang/triton/commit/efa87747342dc50b3c87a357f13288168d6e6e1c renames it to `triton.knobs.runtime.launch_enter_hook`.
Inductor is using the old API: https://github.com/pytorch/pytorch/blob/36acaaae3fb008955320484a8650761e31ce97ad/torch/_inductor/runtime/static_cuda_launcher.py#L48
cc @chauhang @penguinwu @bertmaher @int3 @davidberard98 @nmacchioni @chenyang78 @embg @peterbell10 @aakhundov
### Error logs
https://github.com/pytorch-labs/tritonbench/actions/runs/14735416820/job/41360153735
### Versions
Latest Nightly | true |
3,030,865,350 | NotImplementedError: Operator aten.view.dtype does not have a sharding strategy registered. | neverix | closed | [] | 1 | CONTRIBUTOR | ### 🐛 Describe the bug
As stated in the title, the dtype transmute operator doesn't work for DTensor.
This code fails:
```
import torch
import torch.distributed as dist
import torch.distributed.tensor as dtensor
from torch.distributed.tensor.device_mesh import init_device_mesh
import os
rank = int(os.environ.get("LOCAL_RANK"))
dist.init_process_group(
device_id=torch.device(rank),
)
dist.barrier()
tp = 1
world_size = dist.get_world_size()
mesh = init_device_mesh(
"cuda",
(world_size // tp, tp),
mesh_dim_names=("dp", "tp"),
)
dist.barrier()
x = torch.randn(2, 4, dtype=torch.float32).to(rank)
x = dtensor.DTensor.from_local(
x,
mesh,
(dtensor.Shard(0), dtensor.Shard(1)),
)
x = x.view(dtype=torch.bfloat16)
dist.barrier()
dist.destroy_process_group()
```
With the error:
`NotImplementedError: Operator aten.view.dtype does not have a sharding strategy registered.`
### Versions
```
[pip3] torch==2.6.0
[pip3] triton==3.2.0
``` | true |
3,030,850,004 | Add methods for checking Triton availability to the device interface | galexite | closed | [
"triaged",
"open source",
"Merged",
"ciflow/trunk",
"topic: not user facing",
"module: dynamo"
] | 7 | CONTRIBUTOR | Adds the `is_triton_capable` and `raise_if_triton_unavailable` class methods to the device interface, to allow device types to run their own checks for Triton _capability_ (which means a device can actually support Triton in the first place) and _availability_ (if the correct backend of Triton is installed and is functional for the device).
Using the device interface allows us to do these checks in a device-agnostic way, allow external backends to attest their Triton support by simply implementing those methods. The intention is for this to back things like the `has_triton` utility method.
This has been split from #139171.
cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @chenyang78 @kadeng @chauhang @amjames | true |
3,030,777,501 | flex attention does not leverage masking, memory error | Bastien-mva | closed | [
"triaged",
"module: flex attention"
] | 1 | NONE | ### 🐛 Describe the bug
I use the `flex_attention` for very large time series, and want to compute a sliding window attention. I create a `block_mask` to avoid computing the not needed outer products of classical attention. However, it seems that `flex_attention` computes this outer product between query and key. This gives a memory error for large time series. I tried both on GPU and CPU.
Here is a minimal example:
```
import torch
from torch.nn.attention.flex_attention import flex_attention, create_block_mask
WINDOW = 5
batch_size = 16
nhead = 3
seq_len = 25000
head_dim = 32
queries = torch.randn(batch_size, nhead, seq_len, head_dim).cuda()
values = torch.randn(batch_size, nhead, seq_len, head_dim).cuda()
keys = torch.randn(batch_size, nhead, seq_len, head_dim).cuda()
def sliding_window_mask(b, h, q_idx, kv_idx):
return q_idx - kv_idx <= WINDOW
block_mask = create_block_mask(
sliding_window_mask,
B=None,
H=None,
Q_LEN=seq_len,
KV_LEN=seq_len,
device=queries.device,
_compile=True,
)
flex_attention(queries, keys, values, None, block_mask)
```
I thought the whole point of using a mask was not to compute the useless outer products that will not be used after. Am I missing something here ? This should be O(seq_len*window) in memory no ?
### Versions
Collecting environment information...
PyTorch version: 2.7.0+rocm6.3
Is debug build: False
CUDA used to build PyTorch: N/A
ROCM used to build PyTorch: 6.3.42131-fa1d09cbd
OS: Ubuntu 24.04.2 LTS (x86_64)
GCC version: (Ubuntu 13.3.0-6ubuntu2~24.04) 13.3.0
Clang version: Could not collect
CMake version: Could not collect
Libc version: glibc-2.39
Python version: 3.12.3 (main, Feb 4 2025, 14:48:35) [GCC 13.3.0] (64-bit runtime)
Python platform: Linux-6.11.0-24-generic-x86_64-with-glibc2.39
Is CUDA available: True
CUDA runtime version: Could not collect
CUDA_MODULE_LOADING set to: LAZY
GPU models and configuration: AMD Radeon Graphics (gfx1035)
Nvidia driver version: Could not collect
cuDNN version: Could not collect
HIP runtime version: 6.3.42131
MIOpen runtime version: 3.3.0
Is XNNPACK available: True
CPU:
Architecture: x86_64
CPU op-mode(s): 32-bit, 64-bit
Address sizes: 48 bits physical, 48 bits virtual
Byte Order: Little Endian
CPU(s): 16
On-line CPU(s) list: 0-15
Vendor ID: AuthenticAMD
Model name: AMD Ryzen 7 7735HS with Radeon Graphics
CPU family: 25
Model: 68
Thread(s) per core: 2
Core(s) per socket: 8
Socket(s): 1
Stepping: 1
Frequency boost: enabled
CPU(s) scaling MHz: 41%
CPU max MHz: 4829.0000
CPU min MHz: 400.0000
BogoMIPS: 6387.63
Flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt pdpe1gb rdtscp lm constant_tsc rep_good nopl xtopology nonstop_tsc cpuid extd_apicid aperfmperf rapl pni pclmulqdq monitor ssse3 fma cx16 sse4_1 sse4_2 x2apic movbe popcnt aes xsave avx f16c rdrand lahf_lm cmp_legacy svm extapic cr8_legacy abm sse4a misalignsse 3dnowprefetch osvw ibs skinit wdt tce topoext perfctr_core perfctr_nb bpext perfctr_llc mwaitx cpb cat_l3 cdp_l3 hw_pstate ssbd mba ibrs ibpb stibp vmmcall fsgsbase bmi1 avx2 smep bmi2 erms invpcid cqm rdt_a rdseed adx smap clflushopt clwb sha_ni xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local user_shstk clzero irperf xsaveerptr rdpru wbnoinvd cppc arat npt lbrv svm_lock nrip_save tsc_scale vmcb_clean flushbyasid decodeassists pausefilter pfthreshold avic v_vmsave_vmload vgif v_spec_ctrl umip pku ospke vaes vpclmulqdq rdpid overflow_recov succor smca fsrm debug_swap
Virtualization: AMD-V
L1d cache: 256 KiB (8 instances)
L1i cache: 256 KiB (8 instances)
L2 cache: 4 MiB (8 instances)
L3 cache: 16 MiB (1 instance)
NUMA node(s): 1
NUMA node0 CPU(s): 0-15
Vulnerability Gather data sampling: Not affected
Vulnerability Itlb multihit: Not affected
Vulnerability L1tf: Not affected
Vulnerability Mds: Not affected
Vulnerability Meltdown: Not affected
Vulnerability Mmio stale data: Not affected
Vulnerability Reg file data sampling: Not affected
Vulnerability Retbleed: Not affected
Vulnerability Spec rstack overflow: Mitigation; Safe RET
Vulnerability Spec store bypass: Mitigation; Speculative Store Bypass disabled via prctl
Vulnerability Spectre v1: Mitigation; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2: Mitigation; Retpolines; IBPB conditional; IBRS_FW; STIBP always-on; RSB filling; PBRSB-eIBRS Not affected; BHI Not affected
Vulnerability Srbds: Not affected
Vulnerability Tsx async abort: Not affected
Versions of relevant libraries:
[pip3] mypy_extensions==1.1.0
[pip3] numpy==2.2.5
[pip3] nvidia-cublas-cu12==12.6.4.1
[pip3] nvidia-cuda-cupti-cu12==12.6.80
[pip3] nvidia-cuda-nvrtc-cu12==12.6.77
[pip3] nvidia-cuda-runtime-cu12==12.6.77
[pip3] nvidia-cudnn-cu12==9.5.1.17
[pip3] nvidia-cufft-cu12==11.3.0.4
[pip3] nvidia-curand-cu12==10.3.7.77
[pip3] nvidia-cusolver-cu12==11.7.1.2
[pip3] nvidia-cusparse-cu12==12.5.4.2
[pip3] nvidia-cusparselt-cu12==0.6.3
[pip3] nvidia-nccl-cu12==2.26.2
[pip3] nvidia-nvjitlink-cu12==12.6.85
[pip3] nvidia-nvtx-cu12==12.6.77
[pip3] pytorch-triton-rocm==3.3.0
[pip3] torch==2.7.0+rocm6.3
[pip3] torchaudio==2.7.0+rocm6.3
[pip3] torchvision==0.22.0+rocm6.3
[pip3] triton==3.3.0
[conda] Could not collect
cc @chauhang @penguinwu @zou3519 @ydwu4 @bdhirsh @Chillee @drisspg @yanboliang @BoyuanFeng | true |
3,030,729,300 | can't reconstruct the communication group using PyTorch. | nannaer | open | [
"oncall: distributed",
"triaged",
"module: c10d"
] | 2 | NONE | ### 🐛 Describe the bug
I attempted to use PyTorch to establish communication groups, but I encountered an error. The code flow is as follows: First, use two processes, 0 and 1, as the first communication group to perform an all-reduce operation, and use three processes, 2, 3, and 4, as the second communication group to perform an all-reduce operation. Then, add a pre - launched process (i.e., process 2) to the first communication group. At this point, there should be two new communication groups. The first communication group consists of three processes, 0, 1, and 2, performing an all-reduce operation, and the second communication group consists of two processes, 3 and 4, performing an all-reduce operation.
```python
import os
import torch
import torch.distributed as dist
import torch.multiprocessing as mp
import time
def setup(rank, world_size):
os.environ['MASTER_ADDR'] = 'localhost'
os.environ['MASTER_PORT'] = '30030'
print(f"Rank {rank}: Initializing the distributed environment...")
dist.init_process_group(
backend="nccl",
rank=rank,
world_size=world_size
)
print(f"Rank {rank}: Distributed environment initialization completed")
def test_loop(rank, world_size):
setup(rank, world_size)
dist.barrier()
# ================== Phase 1: Initial Grouping ==================
print(f"\nRank {rank}: ===== Entering Phase 1: Initial Grouping =====")
initial_groups = {}
# Create initial communication groups
if rank in [0, 1]:
print(f"Rank {rank}: Creating initial group1...")
group = dist.new_group(ranks=[0, 1])
initial_groups['group1'] = group
print(f"Rank {rank}: Initial group1 created")
elif rank in [2, 3, 4]:
print(f"Rank {rank}: Creating initial group2...")
group = dist.new_group(ranks=[2, 3, 4])
initial_groups['group2'] = group
print(f"Rank {rank}: Initial group2 created")
dist.barrier()
# Perform all-reduce within the group
tensor = torch.tensor([rank], dtype=torch.float32)
if 'group1' in initial_groups:
print(f"Rank {rank}: [Phase 1] Starting all-reduce for group1")
dist.all_reduce(tensor, op=dist.ReduceOp.SUM, group=initial_groups['group1'])
print(f"Rank {rank}: [Phase 1] Completed all-reduce for group1, result: {tensor}")
elif 'group2' in initial_groups:
print(f"Rank {rank}: [Phase 1] Starting all-reduce for group2")
dist.all_reduce(tensor, op=dist.ReduceOp.SUM, group=initial_groups['group2'])
print(f"Rank {rank}: [Phase 1] Completed all-reduce for group2, result: {tensor}")
# ================== Synchronous Wait ==================
print(f"\nRank {rank}: Waiting for all processes to complete Phase 1...")
dist.barrier()
print(f"Rank {rank}: All processes have completed Phase 1")
# ================== Phase 2: Reconstruct Communication Groups ==================
print(f"\nRank {rank}: ===== Entering Phase 2: Reconstruct Communication Groups =====")
new_groups = {}
# Create new communication groups
if rank in [0, 1, 2]:
print(f"Rank {rank}: Starting to create new_group1")
start_time = time.time()
new_group = dist.new_group(ranks=[0, 1, 2])
end_time = time.time()
new_groups['new_group1'] = new_group
print(f"Rank {rank}: Completed creating new_group1 (time taken: {(end_time - start_time)*1000:.2f}ms)")
elif rank in [3, 4]:
print(f"Rank {rank}: Starting to create new_group2")
start_time = time.time()
new_group = dist.new_group(ranks=[3, 4])
end_time = time.time()
new_groups['new_group2'] = new_group
print(f"Rank {rank}: Completed creating new_group2 (time taken: {(end_time - start_time)*1000:.2f}ms)")
# Perform all-reduce within the new group
tensor = torch.tensor([rank], dtype=torch.float32)
if 'new_group1' in new_groups:
print(f"Rank {rank}: [Phase 2] Starting all-reduce for new_group1")
dist.all_reduce(tensor, op=dist.ReduceOp.SUM, group=new_groups['new_group1'])
print(f"Rank {rank}: [Phase 2] Completed all-reduce for new_group1, result: {tensor}")
elif 'new_group2' in new_groups:
print(f"Rank {rank}: [Phase 2] Starting all-reduce for new_group2")
dist.all_reduce(tensor, op=dist.ReduceOp.SUM, group=new_groups['new_group2'])
print(f"Rank {rank}: [Phase 2] Completed all-reduce for new_group2, result: {tensor}")
# Clean up
print(f"\nRank {rank}: Cleaning up resources...")
dist.destroy_process_group()
print(f"Rank {rank}: Resource cleanup completed")
if __name__ == "__main__":
num_processes = 5
torch.multiprocessing.spawn(
test_loop,
args=(num_processes,),
nprocs=num_processes,
join=True
)
```
The error message is as follows.
```python
W0430 17:30:06.737000 145845 site-packages/torch/multiprocessing/spawn.py:169] Terminating process 146342 via signal SIGTERM
W0430 17:30:06.737000 145845 site-packages/torch/multiprocessing/spawn.py:169] Terminating process 146343 via signal SIGTERM
W0430 17:30:06.738000 145845 site-packages/torch/multiprocessing/spawn.py:169] Terminating process 146344 via signal SIGTERM
W0430 17:30:06.738000 145845 site-packages/torch/multiprocessing/spawn.py:169] Terminating process 146346 via signal SIGTERM
Traceback (most recent call last):
File "/nvme1/chenjiefei/project/vllm/tests/distributed/test_raw_gloo.py", line 198, in <module>
mp.spawn(
File "/nvme1/chenjiefei/miniconda3/envs/vllm-pp/lib/python3.12/site-packages/torch/multiprocessing/spawn.py", line 340, in spawn
return start_processes(fn, args, nprocs, join, daemon, start_method="spawn")
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/nvme1/chenjiefei/miniconda3/envs/vllm-pp/lib/python3.12/site-packages/torch/multiprocessing/spawn.py", line 296, in start_processes
while not context.join():
^^^^^^^^^^^^^^
File "/nvme1/chenjiefei/miniconda3/envs/vllm-pp/lib/python3.12/site-packages/torch/multiprocessing/spawn.py", line 215, in join
raise ProcessRaisedException(msg, error_index, failed_process.pid)
torch.multiprocessing.spawn.ProcessRaisedException:
-- Process 3 terminated with the following error:
Traceback (most recent call last):
File "/nvme1/chenjiefei/miniconda3/envs/vllm-pp/lib/python3.12/site-packages/torch/multiprocessing/spawn.py", line 90, in _wrap
fn(i, *args)
File "/nvme1/chenjiefei/project/vllm/tests/distributed/test_raw_gloo.py", line 151, in test_loop
dist.all_reduce(tensor, op=dist.ReduceOp.SUM, group=initial_groups['group2'])
File "/nvme1/chenjiefei/miniconda3/envs/vllm-pp/lib/python3.12/site-packages/torch/distributed/c10d_logger.py", line 81, in wrapper
return func(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^
File "/nvme1/chenjiefei/miniconda3/envs/vllm-pp/lib/python3.12/site-packages/torch/distributed/distributed_c10d.py", line 2806, in all_reduce
work = group.allreduce([tensor], opts)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
torch.distributed.DistBackendError: NCCL error in: /pytorch/torch/csrc/distributed/c10d/NCCLUtils.hpp:268, unhandled system error (run with NCCL_DEBUG=INFO for details), NCCL version 2.21.5
ncclSystemError: System call (e.g. socket, malloc) or external library call failed or device error.
Last error:
socketStartConnect: Connect to 10.130.8.138<40441> failed : Software caused connection abort
```
### Versions
you can reproduce the error by run the script above
cc @H-Huang @awgu @wanchaol @fegin @fduwjj @wz337 @wconstab @d4l3k | true |
3,030,571,422 | Use std::apply for CPU code | cyyever | open | [
"module: cpu",
"triaged",
"open source",
"topic: not user facing"
] | 3 | COLLABORATOR | The supported compilers are recent enough to enable std::apply in C++17.
cc @jgong5 @mingfeima @XiaobingSuper @sanchitintel @ashokei @jingxu10 @jerryzh168 | true |
3,030,567,310 | elastic: do not shutdown rendezvous on leaving workers | georgkaleido | open | [
"oncall: distributed",
"triaged",
"open source",
"release notes: distributed (torchelastic)"
] | 1 | NONE | In #117066, shutdown of the rendezvous was added if a worker shuts down. This is incorrect, because the rendezvous is actually shutdown in [this file](https://github.com/pytorch/pytorch/blob/fa6f9eb2be07f6289d2ab4e781077f7fc75dbe55/torch/distributed/launcher/api.py#L290) but should not be shutdown if a signal is received. See also [this pull request](https://github.com/pytorch/pytorch/pull/67749).
#124819 then tried to remediate the situation by fixing the faulty shutdown for the restart case. But this is only triggered if the agent restarts the training, but not if the shutdown of the rendezvous happened before.
Removing both these changes restores the original behavior. The rendezvous should only be shutdown if a run completes or fails, not for a single worker leaving.
Fixes #150916
Fixes #147064
cc @H-Huang @awgu @wanchaol @fegin @fduwjj @wz337 @wconstab @d4l3k | true |
3,030,410,049 | Release torch with CUDA12.1 for 2.6 and even latest version | zh794390558 | closed | [
"oncall: releng"
] | 1 | NONE | ### 🚀 The feature, motivation and pitch
Release torch with CUDA12.1 for 2.6 and even latest version
### Alternatives
_No response_
### Additional context
_No response_ | true |
3,030,367,093 | [compile async] [cache] testing | ChuanqiXu9 | open | [
"open source",
"topic: not user facing",
"module: inductor"
] | 7 | CONTRIBUTOR | The comment (https://github.com/pytorch/pytorch/blob/5a52e050248c71dd6e84f51d25cbd17a88555800/torch/_inductor/compile_fx_subproc.py#L70-L87) say it is problematic to not clean the cache for subprocess compile.
But I have some problems to test this locally. I followed the suggestion here https://dev-discuss.pytorch.org/t/how-to-test-properly/2492 to test it.
Would anyone like to approve this to test this? Thanks in ahead.
cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @ipiszy @chenyang78 @kadeng @muchulee8 @amjames @chauhang @aakhundov @aorenste
cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @ipiszy @chenyang78 @kadeng @muchulee8 @amjames @chauhang @aakhundov | true |
3,030,312,358 | Remove the unnecessary cuda/Tensor.cpp | FFFrog | closed | [
"open source",
"Merged",
"ciflow/trunk",
"topic: not user facing",
"skip-pr-sanity-checks"
] | 4 | COLLABORATOR | Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom):
* __->__ #152522
* #152521
* #152513
* #152512
As the title stated.
**Question:**
I have carefully looked through all the .h files in Tensor.cpp and from my perspective this file does not make sense. Does anyone know what the background is for doing this?
cc @albanD | true |
3,030,312,126 | Make torch/csrc/utils.h to be device-agnostic | FFFrog | closed | [
"open source",
"Merged",
"ciflow/trunk",
"topic: not user facing",
"ciflow/rocm"
] | 6 | COLLABORATOR | Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom):
* #152522
* __->__ #152521
* #152513
* #152512
`torch/csrc/utils.h` should be device-independent. Currently, it contains CUDA-related implementations, which indirectly causes the [failure of ROCm testing](https://github.com/pytorch/pytorch/pull/151914#issuecomment-2839691038) (The reason is that the ROCm test environment shouldn`t expose HIP-related header files, which causes the JIT compilation to fail during testing)
Therefore, move CUDA-related implementations to `torch/csrc/cuda/utils.h`.
**Question:**
This change may introduce BC-breack.
I searched for this function globally on github and I think the impact is very small. | true |
3,030,309,032 | DISABLED test_comprehensive_lu_cuda_float32 (__main__.TestInductorOpInfoCUDA) | pytorch-bot[bot] | open | [
"high priority",
"triaged",
"module: flaky-tests",
"skipped",
"oncall: pt2",
"module: inductor"
] | 4 | NONE | Platforms: inductor
This test was disabled because it is failing in CI. See [recent examples](https://hud.pytorch.org/flakytest?name=test_comprehensive_lu_cuda_float32&suite=TestInductorOpInfoCUDA&limit=100) and the most recent trunk [workflow logs](https://github.com/pytorch/pytorch/runs/41392207876).
Over the past 3 hours, it has been determined flaky in 4 workflow(s) with 4 failures and 4 successes.
**Debugging instructions (after clicking on the recent samples link):**
DO NOT ASSUME THINGS ARE OKAY IF THE CI IS GREEN. We now shield flaky tests from developers so CI will thus be green but it will be harder to parse the logs.
To find relevant log snippets:
1. Click on the workflow logs linked above
2. Click on the Test step of the job so that it is expanded. Otherwise, the grepping will not work.
3. Grep for `test_comprehensive_lu_cuda_float32`
4. There should be several instances run (as flaky tests are rerun in CI) from which you can study the logs.
<details><summary>Sample error message</summary>
```
Traceback (most recent call last):
File "/opt/conda/envs/py_3.12/lib/python3.12/site-packages/torch/testing/_internal/common_device_type.py", line 1131, in test_wrapper
return test(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^
File "/opt/conda/envs/py_3.12/lib/python3.12/site-packages/torch/testing/_internal/common_device_type.py", line 1430, in only_fn
return fn(self, *args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^
File "/opt/conda/envs/py_3.12/lib/python3.12/site-packages/torch/testing/_internal/common_utils.py", line 2291, in wrapper
fn(*args, **kwargs)
File "/opt/conda/envs/py_3.12/lib/python3.12/site-packages/torch/testing/_internal/common_device_type.py", line 1211, in dep_fn
return fn(slf, *args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^
File "/opt/conda/envs/py_3.12/lib/python3.12/site-packages/torch/testing/_internal/common_device_type.py", line 1211, in dep_fn
return fn(slf, *args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^
File "/opt/conda/envs/py_3.12/lib/python3.12/site-packages/torch/testing/_internal/common_device_type.py", line 1211, in dep_fn
return fn(slf, *args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^
File "/opt/conda/envs/py_3.12/lib/python3.12/site-packages/torch/testing/_internal/common_utils.py", line 1612, in wrapper
fn(*args, **kwargs)
File "/opt/conda/envs/py_3.12/lib/python3.12/site-packages/torch/testing/_internal/common_utils.py", line 1534, in wrapper
fn(*args, **kwargs)
File "/opt/conda/envs/py_3.12/lib/python3.12/unittest/mock.py", line 1396, in patched
return func(*newargs, **newkeywargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/opt/conda/envs/py_3.12/lib/python3.12/contextlib.py", line 81, in inner
return func(*args, **kwds)
^^^^^^^^^^^^^^^^^^^
File "/opt/conda/envs/py_3.12/lib/python3.12/contextlib.py", line 81, in inner
return func(*args, **kwds)
^^^^^^^^^^^^^^^^^^^
File "/opt/conda/envs/py_3.12/lib/python3.12/contextlib.py", line 81, in inner
return func(*args, **kwds)
^^^^^^^^^^^^^^^^^^^
File "/var/lib/jenkins/workspace/test/inductor/test_torchinductor_opinfo.py", line 962, in inner
raise e
File "/var/lib/jenkins/workspace/test/inductor/test_torchinductor_opinfo.py", line 954, in inner
fn(self, device, dtype, op)
File "/var/lib/jenkins/workspace/test/inductor/test_torchinductor_opinfo.py", line 1207, in test_comprehensive
raise e
File "/var/lib/jenkins/workspace/test/inductor/test_torchinductor_opinfo.py", line 1182, in test_comprehensive
self.check_model_gpu(
File "/opt/conda/envs/py_3.12/lib/python3.12/contextlib.py", line 81, in inner
return func(*args, **kwds)
^^^^^^^^^^^^^^^^^^^
File "/var/lib/jenkins/workspace/test/inductor/test_torchinductor.py", line 657, in check_model_gpu
check_model(
File "/var/lib/jenkins/workspace/test/inductor/test_torchinductor.py", line 608, in check_model
actual_grad = compute_grads(example_inputs, kwargs, actual, grads)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/var/lib/jenkins/workspace/test/inductor/test_torchinductor.py", line 406, in compute_grads
return torch.autograd.grad(
^^^^^^^^^^^^^^^^^^^^
File "/opt/conda/envs/py_3.12/lib/python3.12/site-packages/torch/autograd/__init__.py", line 503, in grad
result = _engine_run_backward(
^^^^^^^^^^^^^^^^^^^^^
File "/opt/conda/envs/py_3.12/lib/python3.12/site-packages/torch/autograd/graph.py", line 824, in _engine_run_backward
return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/opt/conda/envs/py_3.12/lib/python3.12/site-packages/torch/autograd/function.py", line 307, in apply
return user_fn(self, *args)
^^^^^^^^^^^^^^^^^^^^
File "/opt/conda/envs/py_3.12/lib/python3.12/site-packages/torch/_functorch/_aot_autograd/runtime_wrappers.py", line 2163, in backward
return impl_fn()
^^^^^^^^^
File "/opt/conda/envs/py_3.12/lib/python3.12/site-packages/torch/_functorch/_aot_autograd/runtime_wrappers.py", line 2149, in impl_fn
out = CompiledFunction._backward_impl(ctx, all_args)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/opt/conda/envs/py_3.12/lib/python3.12/site-packages/torch/_functorch/_aot_autograd/runtime_wrappers.py", line 2241, in _backward_impl
CompiledFunction.compiled_bw = aot_config.bw_compiler(
^^^^^^^^^^^^^^^^^^^^^^^
File "/opt/conda/envs/py_3.12/lib/python3.12/site-packages/torch/_functorch/aot_autograd.py", line 483, in __call__
return self.compiler_fn(gm, example_inputs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/opt/conda/envs/py_3.12/lib/python3.12/site-packages/torch/_dynamo/backends/common.py", line 73, in _wrapped_bw_compiler
disable(
File "/opt/conda/envs/py_3.12/lib/python3.12/site-packages/torch/_dynamo/eval_frame.py", line 856, in _fn
return fn(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^
File "/opt/conda/envs/py_3.12/lib/python3.12/site-packages/torch/_utils_internal.py", line 97, in wrapper_function
return function(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^
File "/opt/conda/envs/py_3.12/lib/python3.12/site-packages/torch/_inductor/compile_fx.py", line 2201, in bw_compiler
return inner_compile(
^^^^^^^^^^^^^^
File "/opt/conda/envs/py_3.12/lib/python3.12/site-packages/torch/_inductor/compile_fx.py", line 726, in compile_fx_inner
return wrap_compiler_debug(_compile_fx_inner, compiler_name="inductor")(
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/opt/conda/envs/py_3.12/lib/python3.12/site-packages/torch/_dynamo/repro/after_aot.py", line 124, in debug_wrapper
inner_compiled_fn = compiler_fn(gm, example_inputs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/opt/conda/envs/py_3.12/lib/python3.12/site-packages/torch/_inductor/compile_fx.py", line 862, in _compile_fx_inner
raise InductorError(e, currentframe()).with_traceback(
File "/opt/conda/envs/py_3.12/lib/python3.12/site-packages/torch/_inductor/compile_fx.py", line 846, in _compile_fx_inner
mb_compiled_graph = fx_codegen_and_compile(
^^^^^^^^^^^^^^^^^^^^^^^
File "/opt/conda/envs/py_3.12/lib/python3.12/site-packages/torch/_inductor/compile_fx.py", line 1460, in fx_codegen_and_compile
return scheme.codegen_and_compile(gm, example_inputs, inputs_to_check, graph_kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/opt/conda/envs/py_3.12/lib/python3.12/site-packages/torch/_inductor/compile_fx.py", line 1347, in codegen_and_compile
compiled_module = graph.compile_to_module()
^^^^^^^^^^^^^^^^^^^^^^^^^
File "/opt/conda/envs/py_3.12/lib/python3.12/site-packages/torch/_inductor/graph.py", line 2219, in compile_to_module
return self._compile_to_module()
^^^^^^^^^^^^^^^^^^^^^^^^^
File "/opt/conda/envs/py_3.12/lib/python3.12/site-packages/torch/_inductor/graph.py", line 2266, in _compile_to_module
mod = PyCodeCache.load_by_key_path(
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/opt/conda/envs/py_3.12/lib/python3.12/site-packages/torch/_inductor/codecache.py", line 3006, in load_by_key_path
mod = _reload_python_module(key, path, set_sys_modules=in_toplevel)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/opt/conda/envs/py_3.12/lib/python3.12/site-packages/torch/_inductor/runtime/compile_tasks.py", line 31, in _reload_python_module
exec(code, mod.__dict__, mod.__dict__)
File "/tmp/tmp4uk8am0s/gc/cgcy5el4nhw3okrgdi6gx5q7lckdlzhwobgat6d45umgkdklwgb3.py", line 243, in <module>
async_compile.wait(globals())
File "/opt/conda/envs/py_3.12/lib/python3.12/site-packages/torch/_inductor/async_compile.py", line 448, in wait
self._wait_futures(scope)
File "/opt/conda/envs/py_3.12/lib/python3.12/site-packages/torch/_inductor/async_compile.py", line 468, in _wait_futures
scope[key] = result.result()
^^^^^^^^^^^^^^^
File "/opt/conda/envs/py_3.12/lib/python3.12/site-packages/torch/_inductor/codecache.py", line 3508, in result
return self.result_fn()
^^^^^^^^^^^^^^^^
File "/opt/conda/envs/py_3.12/lib/python3.12/site-packages/torch/_inductor/async_compile.py", line 343, in get_result
kernel.precompile(
File "/opt/conda/envs/py_3.12/lib/python3.12/site-packages/torch/_inductor/runtime/triton_heuristics.py", line 322, in precompile
self._make_launchers()
File "/opt/conda/envs/py_3.12/lib/python3.12/site-packages/torch/_inductor/runtime/triton_heuristics.py", line 479, in _make_launchers
launchers.append(result.make_launcher())
^^^^^^^^^^^^^^^^^^^^^^
File "/opt/conda/envs/py_3.12/lib/python3.12/site-packages/torch/_inductor/runtime/triton_heuristics.py", line 1276, in make_launcher
self.reload_cubin_path()
File "/opt/conda/envs/py_3.12/lib/python3.12/site-packages/torch/_inductor/runtime/triton_heuristics.py", line 1268, in reload_cubin_path
raise RuntimeError(
torch._inductor.exc.InductorError: RuntimeError: ('Cubin file saved by TritonBundler not found at %s', '/tmp/tmpg6ru0zop/triton/RX3R4CGSAX6ULKAUAPP2RQK3HJWYNPF3X27PHUDKXPUFUALGM24A/triton_poi_fused_cat_2.cubin')
The above exception was the direct cause of the following exception:
Traceback (most recent call last):
File "/opt/conda/envs/py_3.12/lib/python3.12/site-packages/torch/testing/_internal/common_utils.py", line 3154, in wrapper
method(*args, **kwargs)
File "/opt/conda/envs/py_3.12/lib/python3.12/site-packages/torch/testing/_internal/common_utils.py", line 3154, in wrapper
method(*args, **kwargs)
File "/opt/conda/envs/py_3.12/lib/python3.12/site-packages/torch/testing/_internal/common_device_type.py", line 426, in instantiated_test
result = test(self, **param_kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/opt/conda/envs/py_3.12/lib/python3.12/site-packages/torch/testing/_internal/common_device_type.py", line 1211, in dep_fn
return fn(slf, *args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^
File "/opt/conda/envs/py_3.12/lib/python3.12/site-packages/torch/testing/_internal/common_utils.py", line 1612, in wrapper
fn(*args, **kwargs)
File "/opt/conda/envs/py_3.12/lib/python3.12/site-packages/torch/testing/_internal/common_device_type.py", line 1143, in test_wrapper
raise e_tracked from e
Exception: Caused by sample input at index 0: SampleInput(input=Tensor[size=(3, 5), device="cuda:0", dtype=torch.float32], args=(True,True), kwargs={}, broadcasts_input=False, name='')
To execute this test, run the following from the base repo dir:
PYTORCH_OPINFO_SAMPLE_INPUT_INDEX=0 python test/inductor/test_torchinductor_opinfo.py TestInductorOpInfoCUDA.test_comprehensive_lu_cuda_float32
This message can be suppressed by setting PYTORCH_PRINT_REPRO_ON_FAILURE=0
```
</details>
Test file path: `inductor/test_torchinductor_opinfo.py`
cc @ezyang @gchanan @zou3519 @kadeng @msaroufim @clee2000 @chauhang @penguinwu @voznesenskym @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @ipiszy @chenyang78 @muchulee8 @amjames @aakhundov | true |
3,030,274,206 | update README | TheHiddenLayer | closed | [
"open source",
"topic: not user facing"
] | 2 | NONE | Fixes #ISSUE_NUMBER
| true |
3,030,243,769 | Remove redundant line in partitioner | fmassa | closed | [
"fb-exported",
"Merged",
"ciflow/trunk",
"topic: not user facing",
"ciflow/inductor"
] | 5 | MEMBER | Summary: This is a cleanup from https://github.com/pytorch/pytorch/pull/152264, which contained a line which was a vestige from a previous implementation.
Test Plan: Let CI run
Differential Revision: D73904636
| true |
3,030,230,515 | second test to fix after added logging | wdvr | closed | [
"topic: not user facing",
"module: dynamo",
"ciflow/inductor"
] | 1 | CONTRIBUTOR | Fixes #ISSUE_NUMBER
cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @chenyang78 @kadeng @chauhang @amjames | true |
3,030,210,866 | [MPS] Migrate mul to TensorIterator | malfet | closed | [
"Merged",
"topic: not user facing",
"release notes: mps",
"ciflow/mps"
] | 4 | CONTRIBUTOR | Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom):
* #152743
* #152737
* __->__ #152515
What initially supposed to be a very straightforward change resulted in small refactor of binary op tensor generators when invoked for mixed dtype, which surfaced via `test_output_grad_match_sinc_mps_float16` test failure.
If operands are of different dtype (in particular float16 tensor and float32 scalar), one must perform an operation with `opmath_t` (or `TensorIterator::common_dtype()`) precision, rather than casting both operands to output dtype and performing it then, which can be demonstrated via the following example:
```
>>> torch.tensor([-1.8633, 6.2031, -2.2500, -3.3926, 8.5938, 5.9766], dtype=torch.half).mul(torch.pi)
tensor([ -5.8555, 19.4844, -7.0703, -10.6562, 27.0000, 18.7812],
dtype=torch.float16)
>>> torch.tensor([-1.8633, 6.2031, -2.2500, -3.3926, 8.5938, 5.9766], dtype=torch.half).mul(torch.tensor(torch.pi, dtype=torch.float16))
tensor([ -5.8516, 19.4844, -7.0664, -10.6562, 26.9844, 18.7656],
dtype=torch.float16)
```
Solve this problem for now, but introducing `REGISTER_OPMATH_BINARY_OP` that indicates that operands must be cast to opmath_t, before performing the computation. | true |
3,030,177,025 | [MPS][BE] Migrate `lerp.Scalar.out` to tensor iterator | malfet | closed | [
"Merged",
"release notes: mps",
"ciflow/mps"
] | 4 | CONTRIBUTOR | Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom):
* #152515
* __->__ #152514
| true |
3,030,145,839 | Remove unnecessary __STDC_FORMAT_MACROS macro | FFFrog | closed | [
"open source",
"Merged",
"ciflow/trunk",
"topic: not user facing"
] | 5 | COLLABORATOR | Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom):
* #152522
* #152521
* __->__ #152513
* #152512
As the title stated. | true |
3,030,136,915 | Remove unnecessary condition compilation macro | FFFrog | closed | [
"open source",
"Merged",
"ciflow/trunk",
"topic: not user facing"
] | 4 | COLLABORATOR | Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom):
* #152522
* #152521
* #152513
* __->__ #152512
As the title stated. | true |
3,030,129,039 | DISABLED test_inductor_debug (__main__.LoggingTests) | malfet | closed | [
"high priority",
"triage review",
"module: flaky-tests",
"skipped"
] | 2 | CONTRIBUTOR | This test was disabled because it is failing on main branch ([recent examples](https://torch-ci.com/failure?failureCaptures=%5B%22dynamo%2Ftest_logging.py%3A%3ALoggingTests%3A%3Atest_inductor_debug%22%5D)).
cc @ezyang @gchanan @zou3519 @kadeng @msaroufim @clee2000 | true |
3,030,027,552 | [BE] Migrate all add/sub ops to Metal kernels | malfet | closed | [
"Merged",
"ciflow/trunk",
"topic: not user facing",
"release notes: mps",
"ciflow/mps"
] | 5 | CONTRIBUTOR | Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom):
* __->__ #152510
As typecasting harness shoudl take care of all permutations
Fix bug in `exec_binary_kernel` where it was not properly downcasting CPU double/complexDouble scalars to floats
Fixes https://github.com/pytorch/pytorch/issues/152582 | true |
3,029,974,976 | [2/N] Deprecate c10::string_view and at::string | cyyever | open | [
"open source",
"ciflow/trunk",
"topic: not user facing"
] | 3 | COLLABORATOR | Fixes #ISSUE_NUMBER
| true |
3,029,962,779 | [quantizer] Fix quantizer tests after dq conv is enabled | mcr229 | closed | [
"fb-exported",
"ciflow/trunk",
"release notes: quantization",
"release notes: AO frontend"
] | 2 | CONTRIBUTOR | Summary:
X-link: https://github.com/pytorch/executorch/pull/10569
Fixing some broken tests after we enabled dq convs
Test Plan: CI
Differential Revision: D73898719
| true |
3,029,932,451 | [inductor] [compile async] Don't compile in eager | ChuanqiXu9 | open | [
"triaged",
"open source",
"topic: not user facing",
"module: inductor"
] | 7 | CONTRIBUTOR | Previously we will compile in eager mode.
This looks not intentional according to the test. There is a check to check the number of compilations (in current process) to be 0. But maybe due to an oversight, the number it checks is always a zero.
In _InProcessFxCompile and _SerializedFxCompile, we increment the number of `codegen_and_compile` by `self`, which is a member variable attached to the instance. But in test, we check the number of `codegen_and_compile` by the class. I think we should increment the number of `codegen_and_compile` by the class. Then the test will fail now.
See torch/_inductor/compile_fx_async.py for the fix.
CC @aorenste
cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @ipiszy @chenyang78 @kadeng @muchulee8 @amjames @chauhang @aakhundov | true |
3,029,906,972 | [Hierarchical Compile] Take into account mutation deps in cycle detection | mlazos | open | [
"module: dynamo",
"ciflow/inductor",
"release notes: dynamo"
] | 3 | CONTRIBUTOR | Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom):
* #152589
* #152572
* #152570
* __->__ #152506
* #152410
* #152505
* #152389
cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @chenyang78 @kadeng @chauhang @amjames | true |
3,029,906,894 | [Hierarchical Compilation] Use universal flatten APIs | mlazos | open | [
"module: dynamo",
"ciflow/inductor",
"release notes: dynamo"
] | 3 | CONTRIBUTOR | Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom):
* #152589
* #152572
* #152570
* #152506
* #152410
* __->__ #152505
* #152389
cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @chenyang78 @kadeng @chauhang @amjames | true |
3,029,900,600 | [Metal] Extend typecasted op support to complex dtypes | malfet | closed | [
"Merged",
"topic: not user facing",
"release notes: mps",
"ciflow/mps"
] | 3 | CONTRIBUTOR | Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom):
* #152514
* #152510
* #152485
* __->__ #152504
First of all, by extending `c10::metal::cast_to` to work correctly with complex dtypes, by introducing two more specializations: one that casts complex to scalar, and another that casts scalar to complex (as default metal typecast will turn `float x` into `float2(x, x)`)
Add ComplexHalf and ComplexFloat enum values to `c10::metal::ScalarTypes` and handle them in `val_at_offs(ptr, offs, type)` | true |
3,029,884,179 | Return ConstantVariable(None) from WithExitFunctionVariable.exit to prevent NoneType crash inside autocast exception path | jansel | closed | [
"Merged",
"ciflow/trunk",
"module: dynamo",
"ciflow/inductor",
"release notes: dynamo"
] | 3 | CONTRIBUTOR | Copy of #152013 with PR time benchmarks updated (regressions seem unrelated)
cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @chenyang78 @kadeng @chauhang @amjames | true |
3,029,879,537 | Refactor nested benchmark functions in AlgorithmSelectorCache | masnesral | closed | [
"topic: not user facing",
"module: inductor",
"module: dynamo",
"ciflow/inductor"
] | 1 | CONTRIBUTOR | Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom):
* __->__ #152502
* #152147
Summary: The motivation is make AlgorithmSelectorCache.benchmark_in_current_process() a toplevel method so that I can leverage it for remote auto-tune. Currently, it's nested inside make_benchmark_fn(). Just for consistency, I also made AlgorithmSelectorCache.benchmark_in_sub_process a toplevel staticmethod. That method technically didn't need to move, but IMO refactoring makes the code a bit more readable.
Test Plan: existing unit tests
cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @ipiszy @chenyang78 @kadeng @muchulee8 @amjames @chauhang @aakhundov | true |
3,029,872,798 | [Metal] Extend typecasted op support to complex dtypes | malfet | closed | [
"topic: bug fixes",
"release notes: mps",
"ciflow/mps"
] | 2 | CONTRIBUTOR | Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom):
* (to be filled)
| true |
3,029,852,966 | DISABLED test_comprehensive_repeat_cuda_float64 (__main__.TestInductorOpInfoCUDA) | pytorch-bot[bot] | open | [
"high priority",
"triaged",
"module: flaky-tests",
"skipped",
"oncall: pt2",
"module: inductor"
] | 2 | NONE | Platforms: inductor
This test was disabled because it is failing in CI. See [recent examples](https://hud.pytorch.org/flakytest?name=test_comprehensive_repeat_cuda_float64&suite=TestInductorOpInfoCUDA&limit=100) and the most recent trunk [workflow logs](https://github.com/pytorch/pytorch/runs/41381960290).
Over the past 3 hours, it has been determined flaky in 3 workflow(s) with 3 failures and 3 successes.
**Debugging instructions (after clicking on the recent samples link):**
DO NOT ASSUME THINGS ARE OKAY IF THE CI IS GREEN. We now shield flaky tests from developers so CI will thus be green but it will be harder to parse the logs.
To find relevant log snippets:
1. Click on the workflow logs linked above
2. Click on the Test step of the job so that it is expanded. Otherwise, the grepping will not work.
3. Grep for `test_comprehensive_repeat_cuda_float64`
4. There should be several instances run (as flaky tests are rerun in CI) from which you can study the logs.
<details><summary>Sample error message</summary>
```
Traceback (most recent call last):
File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/torch/testing/_internal/common_device_type.py", line 1131, in test_wrapper
return test(*args, **kwargs)
File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/torch/testing/_internal/common_device_type.py", line 1430, in only_fn
return fn(self, *args, **kwargs)
File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/torch/testing/_internal/common_utils.py", line 2291, in wrapper
fn(*args, **kwargs)
File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/torch/testing/_internal/common_device_type.py", line 1211, in dep_fn
return fn(slf, *args, **kwargs)
File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/torch/testing/_internal/common_device_type.py", line 1211, in dep_fn
return fn(slf, *args, **kwargs)
File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/torch/testing/_internal/common_device_type.py", line 1211, in dep_fn
return fn(slf, *args, **kwargs)
File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/torch/testing/_internal/common_utils.py", line 1612, in wrapper
fn(*args, **kwargs)
File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/torch/testing/_internal/common_utils.py", line 1534, in wrapper
fn(*args, **kwargs)
File "/opt/conda/envs/py_3.10/lib/python3.10/unittest/mock.py", line 1379, in patched
return func(*newargs, **newkeywargs)
File "/opt/conda/envs/py_3.10/lib/python3.10/contextlib.py", line 79, in inner
return func(*args, **kwds)
File "/opt/conda/envs/py_3.10/lib/python3.10/contextlib.py", line 79, in inner
return func(*args, **kwds)
File "/opt/conda/envs/py_3.10/lib/python3.10/contextlib.py", line 79, in inner
return func(*args, **kwds)
File "/var/lib/jenkins/workspace/test/inductor/test_torchinductor_opinfo.py", line 962, in inner
raise e
File "/var/lib/jenkins/workspace/test/inductor/test_torchinductor_opinfo.py", line 954, in inner
fn(self, device, dtype, op)
File "/var/lib/jenkins/workspace/test/inductor/test_torchinductor_opinfo.py", line 1207, in test_comprehensive
raise e
File "/var/lib/jenkins/workspace/test/inductor/test_torchinductor_opinfo.py", line 1182, in test_comprehensive
self.check_model_gpu(
File "/opt/conda/envs/py_3.10/lib/python3.10/contextlib.py", line 79, in inner
return func(*args, **kwds)
File "/var/lib/jenkins/workspace/test/inductor/test_torchinductor.py", line 657, in check_model_gpu
check_model(
File "/var/lib/jenkins/workspace/test/inductor/test_torchinductor.py", line 498, in check_model
actual = run(*example_inputs, **kwargs)
File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/torch/_dynamo/eval_frame.py", line 675, in _fn
raise e.remove_dynamo_frames() from None # see TORCHDYNAMO_VERBOSE=1
File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/torch/_inductor/compile_fx.py", line 862, in _compile_fx_inner
raise InductorError(e, currentframe()).with_traceback(
File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/torch/_inductor/compile_fx.py", line 846, in _compile_fx_inner
mb_compiled_graph = fx_codegen_and_compile(
File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/torch/_inductor/compile_fx.py", line 1460, in fx_codegen_and_compile
return scheme.codegen_and_compile(gm, example_inputs, inputs_to_check, graph_kwargs)
File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/torch/_inductor/compile_fx.py", line 1347, in codegen_and_compile
compiled_module = graph.compile_to_module()
File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/torch/_inductor/graph.py", line 2219, in compile_to_module
return self._compile_to_module()
File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/torch/_inductor/graph.py", line 2266, in _compile_to_module
mod = PyCodeCache.load_by_key_path(
File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/torch/_inductor/codecache.py", line 3006, in load_by_key_path
mod = _reload_python_module(key, path, set_sys_modules=in_toplevel)
File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/torch/_inductor/runtime/compile_tasks.py", line 31, in _reload_python_module
exec(code, mod.__dict__, mod.__dict__)
File "/tmp/tmpfhhehv7f/bx/cbxzzc6lurule353wyo3f3furudihsrzj4ahuph247vpj4syk4rk.py", line 74, in <module>
async_compile.wait(globals())
File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/torch/_inductor/async_compile.py", line 479, in wait
self._wait_futures(scope)
File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/torch/_inductor/async_compile.py", line 499, in _wait_futures
kernel = result.result()
File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/torch/_inductor/codecache.py", line 3508, in result
return self.result_fn()
File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/torch/_inductor/async_compile.py", line 368, in get_result
kernel.precompile(
File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/torch/_inductor/runtime/triton_heuristics.py", line 324, in precompile
self._make_launchers()
File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/torch/_inductor/runtime/triton_heuristics.py", line 481, in _make_launchers
launchers.append(result.make_launcher())
File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/torch/_inductor/runtime/triton_heuristics.py", line 1278, in make_launcher
self.reload_cubin_path()
File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/torch/_inductor/runtime/triton_heuristics.py", line 1270, in reload_cubin_path
raise RuntimeError(
torch._inductor.exc.InductorError: RuntimeError: ('Cubin file saved by TritonBundler not found at %s', '/tmp/tmpz8tgkuit/triton/EHMJ6KW36AZDNSE22KDAQKRF7EKJ7K3IB5GJE32YJSHDQMMBRGIA/triton_poi_fused_repeat_0.cubin')
Set TORCHDYNAMO_VERBOSE=1 for the internal stack trace (please do this especially if you're reporting a bug to PyTorch). For even more developer context, set TORCH_LOGS="+dynamo"
The above exception was the direct cause of the following exception:
Traceback (most recent call last):
File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/torch/testing/_internal/common_utils.py", line 3154, in wrapper
method(*args, **kwargs)
File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/torch/testing/_internal/common_utils.py", line 3154, in wrapper
method(*args, **kwargs)
File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/torch/testing/_internal/common_device_type.py", line 426, in instantiated_test
result = test(self, **param_kwargs)
File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/torch/testing/_internal/common_utils.py", line 1612, in wrapper
fn(*args, **kwargs)
File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/torch/testing/_internal/common_device_type.py", line 1143, in test_wrapper
raise e_tracked from e
Exception: Caused by sample input at index 20: SampleInput(input=Tensor[size=(), device="cuda:0", dtype=torch.float64], args=((3,1,1)), kwargs={}, broadcasts_input=False, name='')
To execute this test, run the following from the base repo dir:
PYTORCH_OPINFO_SAMPLE_INPUT_INDEX=20 python test/inductor/test_torchinductor_opinfo.py TestInductorOpInfoCUDA.test_comprehensive_repeat_cuda_float64
This message can be suppressed by setting PYTORCH_PRINT_REPRO_ON_FAILURE=0
```
</details>
Test file path: `inductor/test_torchinductor_opinfo.py`
cc @ezyang @gchanan @zou3519 @kadeng @msaroufim @clee2000 @chauhang @penguinwu @voznesenskym @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @ipiszy @chenyang78 @muchulee8 @amjames @aakhundov | true |
3,029,846,567 | [ez][export] suggest torch._checks only for booleans | pianpwk | closed | [
"Merged",
"ciflow/trunk",
"fx",
"ciflow/inductor",
"release notes: export"
] | 3 | CONTRIBUTOR | We were doing this when the error was coming from int/float casts, suggesting fixes like `torch._check(zuf0), torch._check(~zuf0)`
cc @ezyang @SherlockNoMad @EikanWang @jgong5 @wenzhe-nrv | true |
3,029,816,723 | [cutlass backend] Add addmm dynamic support | henrylhtsang | closed | [
"fb-exported",
"Merged",
"ciflow/trunk",
"topic: not user facing",
"module: inductor",
"ciflow/inductor"
] | 4 | CONTRIBUTOR | Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom):
* __->__ #152498
Differential Revision: [D73893133](https://our.internmc.facebook.com/intern/diff/D73893133/)
cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @ipiszy @chenyang78 @kadeng @muchulee8 @amjames @chauhang @aakhundov | true |
3,029,814,717 | [export] support SymInt minlength for torch.bincount() | pianpwk | closed | [
"Merged",
"ciflow/trunk",
"release notes: export"
] | 4 | CONTRIBUTOR | null | true |
3,029,803,932 | Expose NCCL communicator from ProcessGroupNCCL via an unsafe API | GD06 | closed | [
"oncall: distributed",
"fb-exported",
"Merged",
"ciflow/trunk",
"release notes: distributed (c10d)",
"ciflow/autoformat"
] | 14 | CONTRIBUTOR | Differential Revision: D73892691
cc @H-Huang @awgu @wanchaol @fegin @fduwjj @wz337 @wconstab @d4l3k | true |
3,029,790,604 | [export] Refactor pt2 save/load | angelayi | open | [
"ciflow/trunk",
"module: inductor",
"ciflow/inductor",
"release notes: export"
] | 3 | CONTRIBUTOR | Refactor the pt2 archive saving to consolidate the format of torch.export.save and torch._inductor.package.package_aoti.
This PR adds the following functions, which torch.export.save and AOTI packaging calls into:
```python
package_pt2(
f: FileLike,
*,
exported_programs: Optional[Union[ExportedProgram, dict[str, ExportedProgram]]] = None,
aoti_files: Optional[Union[list[str], dict[str, list[str]]]] = None,
extra_files: Optional[dict[str, Any]] = None,
) -> FileLike
@dataclass
class PT2ArchiveContents:
exported_programs: dict[str, ExportedProgram]
aoti_runners: dict[str, AOTICompiledModel]
extra_files: dict[str, Any]
load_pt2(f: FileLike) -> PT2ArchiveContents
```
Power users directly call into these APIs if they want to bundle multiple exported programs, aoti files, or extra metadata.
This is how the pt2 archive looks like ([spec](https://docs.google.com/document/d/1RQ4cmywilnFUT1VE-4oTGxwXdc8vowCSZsrRgo3wFA8/edit?tab=t.0)):
```
├── archive_format
├── version
├── .data
├── data
│ ├── aotinductor
│ │ └── model1
│ │ ├── model1.cpp
│ │ ├── model1.so # currently AOTI automatically moves weights in here, TODO to move it out
│ │ ├── cg7domx3woam3nnliwud7yvtcencqctxkvvcafuriladwxw4nfiv.cubin
│ │ └── cubaaxppb6xmuqdm4bej55h2pftbce3bjyyvljxbtdfuolmv45ex.cubin
│ ├── weights
│ │ ├── model1.pt # TODO to dedup weights between model1/model2
│ │ └── model2.pt
│ └── constants
│ │ ├── model1.pt # TODO to dedup weights between model1/model2
│ │ └── model2.pt
│ └── sample_inputs
│ ├── model1.pt # TODO to dedup weights between model1/model2
│ └── model2.pt
├── extra
│ └── user_metadata.txt
└── models
├── model1.json
└── model2.json
```
Future todos:
- unbundle the weights -- instead of .pt, we can use bin files, which will also allow us to dedup weights if we store multiple models
- update aoti_compile_and_package to also save the exported program
- integrate TNR with this packaging flow
cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @ipiszy @chenyang78 @kadeng @muchulee8 @amjames @chauhang @aakhundov | true |
3,029,758,654 | [inductor][invoke_subgraph] Free the buffers before the subgraph call | anijain2305 | closed | [
"Merged",
"Reverted",
"ciflow/trunk",
"topic: not user facing",
"module: inductor",
"ciflow/inductor",
"ci-no-td",
"ciflow/pull"
] | 19 | CONTRIBUTOR | Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom):
* #152675
* __->__ #152494
Before
After
tlparse - https://manifold.edge.x2p.facebook.net/v0/read/tree/logs/.tmph5dwWt/index.html?bucketName=tlparse_reports&apiKey=tlparse_reports-key&withPayload=1&timeoutMsec=10000
cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @ipiszy @chenyang78 @kadeng @muchulee8 @amjames @chauhang @aakhundov | true |
3,029,756,050 | fix tests broken after #152450 | wdvr | open | [
"Merged",
"Reverted",
"ciflow/trunk",
"topic: not user facing",
"module: dynamo",
"ciflow/inductor",
"ci-no-td"
] | 14 | CONTRIBUTOR | Updating test expected value after #152450
cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @chenyang78 @kadeng @chauhang @amjames | true |
3,029,750,573 | [ROCm] add almalinux images | jeffdaily | closed | [
"module: rocm",
"open source",
"Merged",
"topic: not user facing",
"ciflow/rocm"
] | 3 | COLLABORATOR | Fixes #ISSUE_NUMBER
cc @sunway513 @jithunnair-amd @pruthvistony @ROCmSupport @dllehr-amd @jataylo @hongxiayang @naromero77amd | true |
3,029,734,938 | [CUDA][SDPA] Bump python `fused_attention_vs_math_ref_grads` `fudge_factor` for `sm120` | eqy | closed | [
"module: cuda",
"open source",
"Merged",
"ciflow/trunk",
"topic: not user facing",
"module: sdpa"
] | 3 | COLLABORATOR | 🍦
cc @ptrblck @msaroufim @jerryzh168 | true |
3,029,727,452 | [invoke_subgraph] Simplify output code for subgraph output node | anijain2305 | closed | [
"Merged",
"Reverted",
"ciflow/trunk",
"topic: not user facing",
"module: inductor",
"ciflow/inductor",
"ci-no-td"
] | 8 | CONTRIBUTOR | Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom):
* #152675
* #152494
* __->__ #152490
* #152383
Before - [manifold.edge.x2p.facebook.net/v0/read/tree/logs/.tmppQg3F8/index.html?bucketName=tlparse_reports&apiKey=tlparse_reports-key&withPayload=1&timeoutMsec=10000](https://manifold.edge.x2p.facebook.net/v0/read/tree/logs/.tmppQg3F8/index.html?bucketName=tlparse_reports&apiKey=tlparse_reports-key&withPayload=1&timeoutMsec=10000)
After fix - https://manifold.edge.x2p.facebook.net/v0/read/tree/logs/.tmp9a5EM0/index.html?bucketName=tlparse_reports&apiKey=tlparse_reports-key&withPayload=1&timeoutMsec=10000
cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @ipiszy @chenyang78 @kadeng @muchulee8 @amjames @chauhang @aakhundov | true |
3,029,727,054 | UNSTABLE Lint / Lint URLs / linux-job | huydhn | open | [
"module: ci",
"triaged",
"unstable"
] | 1 | CONTRIBUTOR | This is added recently, mark it as unstable for now
cc @seemethere @malfet @pytorch/pytorch-dev-infra | true |
3,029,707,668 | [ROCm] Use almalinux docker files for building Magma | jeffdaily | closed | [
"module: rocm",
"open source",
"Merged",
"topic: not user facing",
"ciflow/rocm"
] | 3 | COLLABORATOR | Fixes #151707 for ROCm Magma builds. See also #152358. Depends on #152492.
cc @sunway513 @jithunnair-amd @pruthvistony @ROCmSupport @dllehr-amd @jataylo @hongxiayang @naromero77amd | true |
3,029,706,699 | [inductor][BE] Add more debug logs for why fx graph cache doesn't happen | henrylhtsang | closed | [
"Merged",
"Reverted",
"ciflow/trunk",
"topic: not user facing",
"module: inductor",
"module: dynamo",
"ciflow/inductor",
"ci-no-td"
] | 10 | CONTRIBUTOR | Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom):
* __->__ #152487
cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @ipiszy @chenyang78 @kadeng @muchulee8 @amjames @chauhang @aakhundov | true |
3,029,694,822 | Change unsafe_marked_cacheable_functions to a dictionary, so that you can specify a static cache key | jamesjwu | open | [
"ciflow/trunk",
"topic: not user facing",
"module: inductor",
"module: dynamo",
"ciflow/inductor",
"merging"
] | 11 | CONTRIBUTOR | Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom):
* __->__ #152486
Fixes https://github.com/pytorch/pytorch/issues/152434
cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @ipiszy @chenyang78 @kadeng @muchulee8 @amjames @chauhang @aakhundov | true |
3,029,684,105 | [MPS][BE] Remove `exec_binary_alpha_kernel` | malfet | closed | [
"Merged",
"topic: not user facing",
"release notes: mps",
"ciflow/mps"
] | 3 | CONTRIBUTOR | Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom):
* #152515
* #152514
* #152510
* __->__ #152485
Which was almost a complete copy-n-paste from exec_binary_kernel anyway
Just add `Scalar` as an optional argument and figure out kernel name during the invocation rather than in executor | true |
3,029,681,192 | Fix flaky test in test_custom_ops | angelayi | open | [
"Merged",
"Reverted",
"ciflow/trunk",
"topic: not user facing",
"merging",
"ci-no-td",
"ciflow/pull"
] | 17 | CONTRIBUTOR | Hopefully fixes https://github.com/pytorch/pytorch/issues/151301, https://github.com/pytorch/pytorch/issues/151281 by making the ops have different names | true |
3,029,661,503 | [inductor][BE] cleanup and improve precompilation loggings | henrylhtsang | closed | [
"Merged",
"ciflow/trunk",
"topic: not user facing",
"module: inductor",
"ciflow/inductor"
] | 3 | CONTRIBUTOR | Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom):
* __->__ #152483
cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @ipiszy @chenyang78 @kadeng @muchulee8 @amjames @chauhang @aakhundov | true |
3,029,655,481 | Update CODEOWNERS (torch/utils/data/) | divyanshk | closed | [
"Merged",
"ciflow/trunk",
"topic: not user facing"
] | 4 | CONTRIBUTOR | Updating codeowners for dataloading
| true |
3,029,654,460 | [nativert] port enumerate from folly to c10::utill | dolpm | closed | [
"fb-exported",
"Merged",
"ciflow/trunk",
"topic: not user facing"
] | 17 | CONTRIBUTOR | Summary:
nativert RFC: https://github.com/zhxchen17/rfcs/blob/master/RFC-0043-torch-native-runtime.md
To land the runtime into PyTorch core, we will gradually land logical parts of the code into the Github issue and get each piece properly reviewed.
This diff ports an enumeration util from folly into c10.
Test Plan: CI
Differential Revision: D73881042
| true |
3,029,650,141 | [dynamo] Try tracing into einops | StrongerXi | open | [
"triaged",
"oncall: pt2",
"module: dynamo"
] | 0 | CONTRIBUTOR | ### 🐛 Describe the bug
This will allow us to
1. get rid of https://github.com/pytorch/pytorch/blob/accffef504b9162718c535201c263070565f28fa/torch/_dynamo/decorators.py#L704-L735
2. handle `einops` in a better way for AOTDispatcher cache: #152369.
### Error logs
_No response_
### Versions
main
cc @chauhang @penguinwu @voznesenskym @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @chenyang78 @kadeng @amjames | true |
3,029,650,004 | [MPS] Fix lerp for complex numbers | malfet | closed | [
"Merged",
"topic: bug fixes",
"release notes: mps",
"ciflow/mps"
] | 3 | CONTRIBUTOR | Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom):
* #152510
* #152485
* #152504
* __->__ #152479
* #152466
As well as `.add`/`.sub` with complex alpha
Before this change `python3 -c "import torch;print(torch.rand(10, device='mps', dtype=torch.complex64).add(torch.rand(10, device='mps', dtype=torch.complex64), alpha=.5j))"` used to fail with
```
RuntimeError: value cannot be converted to type double without overflow
``` | true |
3,029,637,430 | [ONNX] Suggest users setting dynamo=True when exporting | titaiwangms | closed | [
"open source",
"Merged",
"ciflow/trunk",
"release notes: onnx"
] | 7 | COLLABORATOR | Fixes #152025
| true |
3,029,633,929 | [DO NOT REVIEW] Attempt a mixed precision fused adam | janeyx99 | open | [
"ciflow/trunk",
"release notes: foreach_frontend"
] | 2 | CONTRIBUTOR | Non ghstack version of stack at #147653 for easy import
| true |
3,029,627,570 | [inductor][BE] cleanup and improve precompilation loggings | henrylhtsang | closed | [
"topic: not user facing",
"module: inductor",
"ciflow/inductor"
] | 2 | CONTRIBUTOR | Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom):
* __->__ #152476
cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @ipiszy @chenyang78 @kadeng @muchulee8 @amjames @chauhang @aakhundov | true |
3,029,613,684 | [nativert] Move TensorMeta to pytorch core | yiming0416 | closed | [
"fb-exported",
"Merged",
"ciflow/trunk",
"topic: not user facing"
] | 20 | CONTRIBUTOR | Summary:
Torch Native Runtime RFC: https://github.com/pytorch/rfcs/pull/72
This diff moves `TensorMeta.cpp` and `TensorMeta.h` to PyTorch core under `torch/nativert/graph/`
Existing `torch::_export::TensorMeta` in `torch/csrc/utils/generated_serialization_types.h` is auto-generated from the export serde schema and therefore only containing the most basic serializable types. We need the newly added `TensorMeta.cpp` to deserialize the metadata into a in-memory class with c10 types so that it can be consumed by the runtime later.
Test Plan:
Added test under `test/cpp/nativert/test_tensor_meta.cpp`
Differential Revision: D73820548
| true |
3,029,604,114 | [CUDA] Fix `test_multi_device_context_manager` on CUDA | eqy | closed | [
"module: multi-gpu",
"module: cuda",
"open source",
"Merged",
"ciflow/trunk",
"topic: not user facing"
] | 3 | COLLABORATOR | Seems there was a typo where `set_device` was called when the intent was to use `current_device`
As-is the test will fail on multigpu systems with
`TypeError: set_device() missing 1 required positional argument: 'device'`
cc @ptrblck @msaroufim @jerryzh168 | true |
3,029,602,292 | [dynamo] Dynamo fails to run torch.cat() with FakeTensors because it can't confirm 's0 + s1*u0' is nonzero | Adam27X | open | [
"triaged",
"oncall: pt2",
"module: fakeTensor",
"module: dynamic shapes",
"module: dynamo"
] | 1 | NONE | ### 🐛 Describe the bug
Here's a fully reproducible test case. I found this issue using PyTorch 2.5.0 but was able to reproduce it with PyTorch 2.7.0:
```
import torch
from typing import List
from torch._dynamo.backends.common import aot_autograd
from functorch.compile import make_boxed_func
print(torch.__version__)
device = torch.device("cuda")
class Model(torch.nn.Module):
def __init__(self):
self.vsite_count = 8
super().__init__()
def forward(self, ml_atoms):
ml_atoms_vs = ml_atoms[:, ml_atoms[0]].repeat(1, self.vsite_count) #Edit for FxGraph generation with PyTorch 2.5-2.7
ml_atoms = torch.cat((ml_atoms, ml_atoms_vs), dim=1)
return ml_atoms
ml_atoms_in = torch.zeros([1,4296], device=device, dtype=torch.bool)
ml_atoms_in[:,0:6] = True
model = Model()
torch._dynamo.config.capture_dynamic_output_shape_ops = True
torch._dynamo.config.capture_scalar_outputs = True
def test_compiler(gm: torch.fx.GraphModule, example_inputs: List[torch.Tensor]):
graph_modules.append(gm)
return make_boxed_func(gm.forward)
test_compiler = aot_autograd(fw_compiler=test_compiler,bw_compiler=test_compiler, decompositions=torch._decomp.core_aten_decompositions())
prototype = torch.compile(model, fullgraph=True, dynamic=True, backend=test_compiler)
ml_atoms_cat = prototype(ml_atoms_in)
```
I think the issue here is that torch.cat has a check that the size of the concatenated dimension is nonzero and for some reason dynamo can't figure out that `s0 + s1*u0` must be nonzero because `s0`, `s1`, and `u0` are all symints. Note that this is something that worked with PyTorch 2.2.0. Perhaps this check was added since then.
### Error logs
```
2.7.0+cu126
I0429 17:30:24.974000 2078076 torch/fx/experimental/symbolic_shapes.py:3334] [0/0] create_env
I0429 17:30:25.044000 2078076 torch/fx/experimental/symbolic_shapes.py:4606] [0/0] create_symbol s0 = 4296 for L['ml_atoms'].size()[1] [2, int_oo] ml_atoms_vs = ml_atoms[:, ml_atoms[0]].repeat(1, self.vsite_count) #Garand edit for FxGraph generation with PyTorch 2.5-2.7 # ./sandbox/torch_compile_cat_dynamic_shapes/test.py:17 in forward (_dynamo/variables/builder.py:3033 in <lambda>), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_CREATE_SYMBOL="s0" or to suppress this message run with TORCHDYNAMO_EXTENDED_ADVICE="0"
I0429 17:30:25.052000 2078076 torch/fx/experimental/symbolic_shapes.py:4276] [0/0] create_unbacked_symint u0 [-int_oo, int_oo] ml_atoms_vs = ml_atoms[:, ml_atoms[0]].repeat(1, self.vsite_count) #Garand edit for FxGraph generation with PyTorch 2.5-2.7 # ./sandbox/torch_compile_cat_dynamic_shapes/test.py:17 in forward (_subclasses/fake_impls.py:465 in nonzero)
I0429 17:30:25.053000 2078076 torch/fx/experimental/symbolic_shapes.py:7162] [0/0] constrain_symbol_range u0 [0, 9223372036854775806]
I0429 17:30:25.071000 2078076 torch/fx/experimental/symbolic_shapes.py:1130] [0/0] compute_unbacked_bindings [u0]
I0429 17:30:25.074000 2078076 torch/fx/experimental/symbolic_shapes.py:4606] [0/0] create_symbol s1 = 8 for L['self'].vsite_count [-int_oo, int_oo] ml_atoms_vs = ml_atoms[:, ml_atoms[0]].repeat(1, self.vsite_count) #Garand edit for FxGraph generation with PyTorch 2.5-2.7 # ./sandbox/torch_compile_cat_dynamic_shapes/test.py:17 in forward (_dynamo/variables/builder.py:2010 in wrap_symint), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_CREATE_SYMBOL="s1" or to suppress this message run with TORCHDYNAMO_EXTENDED_ADVICE="0"
I0429 17:30:25.087000 2078076 torch/fx/experimental/symbolic_shapes.py:6630] [0/0] runtime_assert s1*u0 >= 0 [guard added] ml_atoms_vs = ml_atoms[:, ml_atoms[0]].repeat(1, self.vsite_count) #Garand edit for FxGraph generation with PyTorch 2.5-2.7 # ./sandbox/torch_compile_cat_dynamic_shapes/test.py:17 in forward (_refs/__init__.py:4796 in new_empty), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_GUARD_ADDED="s1*u0 >= 0"
I0429 17:30:25.118000 2078076 torch/fx/experimental/symbolic_shapes.py:6630] [0/0] runtime_assert s0 + s1*u0 >= 0 [guard added] ml_atoms = torch.cat((ml_atoms, ml_atoms_vs), dim=1) # ./sandbox/torch_compile_cat_dynamic_shapes/test.py:18 in forward (_prims_common/__init__.py:620 in validate_dim_length), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_GUARD_ADDED="s0 + s1*u0 >= 0"
W0429 17:30:25.138000 2078076 torch/fx/experimental/symbolic_shapes.py:6679] [0/0] failed during evaluate_expr(Eq(s0 + s1*u0, 0), hint=None, size_oblivious=True, forcing_spec=False
E0429 17:30:25.139000 2078076 torch/fx/experimental/recording.py:299] [0/0] failed while running evaluate_expr(*(Eq(s0 + s1*u0, 0), None, False, True), **{})
E0429 17:30:25.139000 2078076 torch/fx/experimental/recording.py:299] [0/0] Traceback (most recent call last):
E0429 17:30:25.139000 2078076 torch/fx/experimental/recording.py:299] [0/0] File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/fx/experimental/recording.py", line 263, in wrapper
E0429 17:30:25.139000 2078076 torch/fx/experimental/recording.py:299] [0/0] return retlog(fn(*args, **kwargs))
E0429 17:30:25.139000 2078076 torch/fx/experimental/recording.py:299] [0/0] File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/fx/experimental/symbolic_shapes.py", line 6671, in evaluate_expr
E0429 17:30:25.139000 2078076 torch/fx/experimental/recording.py:299] [0/0] return self._evaluate_expr(
E0429 17:30:25.139000 2078076 torch/fx/experimental/recording.py:299] [0/0] File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/fx/experimental/symbolic_shapes.py", line 6894, in _evaluate_expr
E0429 17:30:25.139000 2078076 torch/fx/experimental/recording.py:299] [0/0] raise self._make_data_dependent_error(
E0429 17:30:25.139000 2078076 torch/fx/experimental/recording.py:299] [0/0] torch.fx.experimental.symbolic_shapes.GuardOnDataDependentSymNode: Could not guard on data-dependent expression Eq(8*u0 + 4296, 0) (unhinted: Eq(s0 + s1*u0, 0)). (Size-like symbols: u0)
E0429 17:30:25.139000 2078076 torch/fx/experimental/recording.py:299] [0/0]
E0429 17:30:25.139000 2078076 torch/fx/experimental/recording.py:299] [0/0] Caused by: ml_atoms = torch.cat((ml_atoms, ml_atoms_vs), dim=1) # ./sandbox/torch_compile_cat_dynamic_shapes/test.py:18 in forward (_ops.py:756 in __call__)
E0429 17:30:25.139000 2078076 torch/fx/experimental/recording.py:299] [0/0] For more information, run with TORCH_LOGS="dynamic"
E0429 17:30:25.139000 2078076 torch/fx/experimental/recording.py:299] [0/0] For extended logs when we create symbols, also add TORCHDYNAMO_EXTENDED_DEBUG_CREATE_SYMBOL="u0"
E0429 17:30:25.139000 2078076 torch/fx/experimental/recording.py:299] [0/0] If you suspect the guard was triggered from C++, add TORCHDYNAMO_EXTENDED_DEBUG_CPP=1
E0429 17:30:25.139000 2078076 torch/fx/experimental/recording.py:299] [0/0] For more debugging help, see https://docs.google.com/document/d/1HSuTTVvYH1pTew89Rtpeu84Ht3nQEFTYhAX3Ypa_xJs/edit?usp=sharing
E0429 17:30:25.139000 2078076 torch/fx/experimental/recording.py:299] [0/0]
E0429 17:30:25.139000 2078076 torch/fx/experimental/recording.py:299] [0/0] User Stack (most recent call last):
E0429 17:30:25.139000 2078076 torch/fx/experimental/recording.py:299] [0/0] (snipped, see stack below for prefix)
E0429 17:30:25.139000 2078076 torch/fx/experimental/recording.py:299] [0/0] File "/u/pgh/mclaugha/./sandbox/torch_compile_cat_dynamic_shapes/test.py", line 18, in forward
E0429 17:30:25.139000 2078076 torch/fx/experimental/recording.py:299] [0/0] ml_atoms = torch.cat((ml_atoms, ml_atoms_vs), dim=1)
E0429 17:30:25.139000 2078076 torch/fx/experimental/recording.py:299] [0/0]
E0429 17:30:25.139000 2078076 torch/fx/experimental/recording.py:299] [0/0] For C++ stack trace, run with TORCHDYNAMO_EXTENDED_DEBUG_CPP=1
Traceback (most recent call last):
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_dynamo/utils.py", line 3284, in run_node
return node.target(*args, **kwargs)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/utils/_stats.py", line 27, in wrapper
return fn(*args, **kwargs)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_subclasses/fake_tensor.py", line 1282, in __torch_dispatch__
return self.dispatch(func, types, args, kwargs)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_subclasses/fake_tensor.py", line 1823, in dispatch
return self._cached_dispatch_impl(func, types, args, kwargs)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_subclasses/fake_tensor.py", line 1393, in _cached_dispatch_impl
output = self._dispatch_impl(func, types, args, kwargs)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_subclasses/fake_tensor.py", line 2333, in _dispatch_impl
decomposition_table[func](*args, **kwargs)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_prims_common/wrappers.py", line 308, in _fn
result = fn(*args, **kwargs)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_compile.py", line 51, in inner
return disable_fn(*args, **kwargs)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_dynamo/eval_frame.py", line 838, in _fn
return fn(*args, **kwargs)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_prims_common/wrappers.py", line 149, in _fn
result = fn(**bound.arguments)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_refs/__init__.py", line 2863, in cat
return prims.cat(filtered, dim).clone(memory_format=memory_format)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_ops.py", line 756, in __call__
return self._op(*args, **kwargs)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/utils/_stats.py", line 27, in wrapper
return fn(*args, **kwargs)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_subclasses/fake_tensor.py", line 1282, in __torch_dispatch__
return self.dispatch(func, types, args, kwargs)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_subclasses/fake_tensor.py", line 1823, in dispatch
return self._cached_dispatch_impl(func, types, args, kwargs)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_subclasses/fake_tensor.py", line 1393, in _cached_dispatch_impl
output = self._dispatch_impl(func, types, args, kwargs)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_subclasses/fake_tensor.py", line 2355, in _dispatch_impl
func.prim_meta_impl(*args, **kwargs)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_prims/__init__.py", line 1794, in _cat_meta
return TensorMeta(
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_prims/__init__.py", line 264, in TensorMeta
return torch.empty_strided(shape, strides, dtype=dtype, device=device)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/utils/_stats.py", line 27, in wrapper
return fn(*args, **kwargs)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_subclasses/fake_tensor.py", line 1282, in __torch_dispatch__
return self.dispatch(func, types, args, kwargs)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_subclasses/fake_tensor.py", line 1823, in dispatch
return self._cached_dispatch_impl(func, types, args, kwargs)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_subclasses/fake_tensor.py", line 1384, in _cached_dispatch_impl
output = self._dispatch_impl(func, types, args, kwargs)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_subclasses/fake_tensor.py", line 2397, in _dispatch_impl
op_impl_out = op_impl(self, func, *args, **kwargs)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_subclasses/fake_impls.py", line 189, in constructors
r = func(*args, **new_kwargs)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_ops.py", line 756, in __call__
return self._op(*args, **kwargs)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/fx/experimental/sym_node.py", line 588, in guard_size_oblivious
r = self.evaluate(size_oblivious=True)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/fx/experimental/sym_node.py", line 510, in evaluate
return self.shape_env.evaluate_sym_node(self, size_oblivious)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/fx/experimental/symbolic_shapes.py", line 6655, in evaluate_sym_node
return self.evaluate_expr(
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/fx/experimental/recording.py", line 263, in wrapper
return retlog(fn(*args, **kwargs))
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/fx/experimental/symbolic_shapes.py", line 6671, in evaluate_expr
return self._evaluate_expr(
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/fx/experimental/symbolic_shapes.py", line 6894, in _evaluate_expr
raise self._make_data_dependent_error(
torch.fx.experimental.symbolic_shapes.GuardOnDataDependentSymNode: Could not guard on data-dependent expression Eq(8*u0 + 4296, 0) (unhinted: Eq(s0 + s1*u0, 0)). (Size-like symbols: u0)
Caused by: ml_atoms = torch.cat((ml_atoms, ml_atoms_vs), dim=1) # ./sandbox/torch_compile_cat_dynamic_shapes/test.py:18 in forward (_ops.py:756 in __call__)
For more information, run with TORCH_LOGS="dynamic"
For extended logs when we create symbols, also add TORCHDYNAMO_EXTENDED_DEBUG_CREATE_SYMBOL="u0"
If you suspect the guard was triggered from C++, add TORCHDYNAMO_EXTENDED_DEBUG_CPP=1
For more debugging help, see https://docs.google.com/document/d/1HSuTTVvYH1pTew89Rtpeu84Ht3nQEFTYhAX3Ypa_xJs/edit?usp=sharing
User Stack (most recent call last):
(snipped, see stack below for prefix)
File "/u/pgh/mclaugha/./sandbox/torch_compile_cat_dynamic_shapes/test.py", line 18, in forward
ml_atoms = torch.cat((ml_atoms, ml_atoms_vs), dim=1)
For C++ stack trace, run with TORCHDYNAMO_EXTENDED_DEBUG_CPP=1
The above exception was the direct cause of the following exception:
Traceback (most recent call last):
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_dynamo/utils.py", line 3127, in get_fake_value
ret_val = wrap_fake_exception(
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_dynamo/utils.py", line 2641, in wrap_fake_exception
return fn()
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_dynamo/utils.py", line 3128, in <lambda>
lambda: run_node(tx.output, node, args, kwargs, nnmodule)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_dynamo/utils.py", line 3325, in run_node
raise RuntimeError(make_error_message(e)).with_traceback(
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_dynamo/utils.py", line 3284, in run_node
return node.target(*args, **kwargs)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/utils/_stats.py", line 27, in wrapper
return fn(*args, **kwargs)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_subclasses/fake_tensor.py", line 1282, in __torch_dispatch__
return self.dispatch(func, types, args, kwargs)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_subclasses/fake_tensor.py", line 1823, in dispatch
return self._cached_dispatch_impl(func, types, args, kwargs)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_subclasses/fake_tensor.py", line 1393, in _cached_dispatch_impl
output = self._dispatch_impl(func, types, args, kwargs)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_subclasses/fake_tensor.py", line 2333, in _dispatch_impl
decomposition_table[func](*args, **kwargs)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_prims_common/wrappers.py", line 308, in _fn
result = fn(*args, **kwargs)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_compile.py", line 51, in inner
return disable_fn(*args, **kwargs)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_dynamo/eval_frame.py", line 838, in _fn
return fn(*args, **kwargs)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_prims_common/wrappers.py", line 149, in _fn
result = fn(**bound.arguments)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_refs/__init__.py", line 2863, in cat
return prims.cat(filtered, dim).clone(memory_format=memory_format)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_ops.py", line 756, in __call__
return self._op(*args, **kwargs)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/utils/_stats.py", line 27, in wrapper
return fn(*args, **kwargs)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_subclasses/fake_tensor.py", line 1282, in __torch_dispatch__
return self.dispatch(func, types, args, kwargs)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_subclasses/fake_tensor.py", line 1823, in dispatch
return self._cached_dispatch_impl(func, types, args, kwargs)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_subclasses/fake_tensor.py", line 1393, in _cached_dispatch_impl
output = self._dispatch_impl(func, types, args, kwargs)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_subclasses/fake_tensor.py", line 2355, in _dispatch_impl
func.prim_meta_impl(*args, **kwargs)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_prims/__init__.py", line 1794, in _cat_meta
return TensorMeta(
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_prims/__init__.py", line 264, in TensorMeta
return torch.empty_strided(shape, strides, dtype=dtype, device=device)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/utils/_stats.py", line 27, in wrapper
return fn(*args, **kwargs)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_subclasses/fake_tensor.py", line 1282, in __torch_dispatch__
return self.dispatch(func, types, args, kwargs)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_subclasses/fake_tensor.py", line 1823, in dispatch
return self._cached_dispatch_impl(func, types, args, kwargs)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_subclasses/fake_tensor.py", line 1384, in _cached_dispatch_impl
output = self._dispatch_impl(func, types, args, kwargs)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_subclasses/fake_tensor.py", line 2397, in _dispatch_impl
op_impl_out = op_impl(self, func, *args, **kwargs)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_subclasses/fake_impls.py", line 189, in constructors
r = func(*args, **new_kwargs)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_ops.py", line 756, in __call__
return self._op(*args, **kwargs)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/fx/experimental/sym_node.py", line 588, in guard_size_oblivious
r = self.evaluate(size_oblivious=True)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/fx/experimental/sym_node.py", line 510, in evaluate
return self.shape_env.evaluate_sym_node(self, size_oblivious)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/fx/experimental/symbolic_shapes.py", line 6655, in evaluate_sym_node
return self.evaluate_expr(
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/fx/experimental/recording.py", line 263, in wrapper
return retlog(fn(*args, **kwargs))
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/fx/experimental/symbolic_shapes.py", line 6671, in evaluate_expr
return self._evaluate_expr(
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/fx/experimental/symbolic_shapes.py", line 6894, in _evaluate_expr
raise self._make_data_dependent_error(
RuntimeError: Dynamo failed to run FX node with fake tensors: call_function <built-in method cat of type object at 0x7fa772b7df40>(*((FakeTensor(..., device='cuda:0', size=(1, s0), dtype=torch.bool), FakeTensor(..., device='cuda:0', size=(1, s1*u0), dtype=torch.bool)),), **{'dim': 1}): got GuardOnDataDependentSymNode('Could not guard on data-dependent expression Eq(8*u0 + 4296, 0) (unhinted: Eq(s0 + s1*u0, 0)). (Size-like symbols: u0)\n\nCaused by: ml_atoms = torch.cat((ml_atoms, ml_atoms_vs), dim=1) # ./sandbox/torch_compile_cat_dynamic_shapes/test.py:18 in forward (_ops.py:756 in __call__)\nFor more information, run with TORCH_LOGS="dynamic"\nFor extended logs when we create symbols, also add TORCHDYNAMO_EXTENDED_DEBUG_CREATE_SYMBOL="u0"\nIf you suspect the guard was triggered from C++, add TORCHDYNAMO_EXTENDED_DEBUG_CPP=1\nFor more debugging help, see https://docs.google.com/document/d/1HSuTTVvYH1pTew89Rtpeu84Ht3nQEFTYhAX3Ypa_xJs/edit?usp=sharing\n\nUser Stack (most recent call last):\n (snipped, see stack below for prefix)\n File "/u/pgh/mclaugha/./sandbox/torch_compile_cat_dynamic_shapes/test.py", line 18, in forward\n ml_atoms = torch.cat((ml_atoms, ml_atoms_vs), dim=1)\n\nFor C++ stack trace, run with TORCHDYNAMO_EXTENDED_DEBUG_CPP=1')
During handling of the above exception, another exception occurred:
Traceback (most recent call last):
File "/u/pgh/mclaugha/./sandbox/torch_compile_cat_dynamic_shapes/test.py", line 37, in <module>
ml_atoms_cat = prototype(ml_atoms_in)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1751, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1762, in _call_impl
return forward_call(*args, **kwargs)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_dynamo/eval_frame.py", line 655, in _fn
return fn(*args, **kwargs)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1751, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1762, in _call_impl
return forward_call(*args, **kwargs)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_dynamo/convert_frame.py", line 1432, in __call__
return self._torchdynamo_orig_callable(
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_dynamo/convert_frame.py", line 598, in __call__
return _compile(
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_dynamo/convert_frame.py", line 1059, in _compile
guarded_code = compile_inner(code, one_graph, hooks, transform)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_utils_internal.py", line 97, in wrapper_function
return function(*args, **kwargs)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_dynamo/convert_frame.py", line 761, in compile_inner
return _compile_inner(code, one_graph, hooks, transform)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_dynamo/convert_frame.py", line 797, in _compile_inner
out_code = transform_code_object(code, transform)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_dynamo/bytecode_transformation.py", line 1422, in transform_code_object
transformations(instructions, code_options)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_dynamo/convert_frame.py", line 257, in _fn
return fn(*args, **kwargs)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_dynamo/convert_frame.py", line 715, in transform
tracer.run()
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_dynamo/symbolic_convert.py", line 3500, in run
super().run()
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_dynamo/symbolic_convert.py", line 1337, in run
while self.step():
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_dynamo/symbolic_convert.py", line 1246, in step
self.dispatch_table[inst.opcode](self, inst)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_dynamo/symbolic_convert.py", line 819, in wrapper
return inner_fn(self, inst)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_dynamo/symbolic_convert.py", line 2278, in CALL_FUNCTION_KW
self.call_function(fn, args, kwargs)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_dynamo/symbolic_convert.py", line 1170, in call_function
self.push(fn.call_function(self, args, kwargs)) # type: ignore[arg-type]
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_dynamo/variables/torch.py", line 1181, in call_function
tensor_variable = wrap_fx_proxy(
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_dynamo/variables/builder.py", line 2302, in wrap_fx_proxy
return wrap_fx_proxy_cls(target_cls=TensorVariable, **kwargs)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_dynamo/variables/builder.py", line 2368, in wrap_fx_proxy_cls
return _wrap_fx_proxy(
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_dynamo/variables/builder.py", line 2464, in _wrap_fx_proxy
example_value = get_fake_value(proxy.node, tx, allow_non_graph_fake=True)
File "/u/pgh/mclaugha/.venv/lib/python3.10/site-packages/torch/_dynamo/utils.py", line 3214, in get_fake_value
raise UserError( # noqa: B904
torch._dynamo.exc.UserError: Could not guard on data-dependent expression Eq(8*u0 + 4296, 0) (unhinted: Eq(s0 + s1*u0, 0)). (Size-like symbols: u0)
Caused by: ml_atoms = torch.cat((ml_atoms, ml_atoms_vs), dim=1) # ./sandbox/torch_compile_cat_dynamic_shapes/test.py:18 in forward (_ops.py:756 in __call__)
For more information, run with TORCH_LOGS="dynamic"
For extended logs when we create symbols, also add TORCHDYNAMO_EXTENDED_DEBUG_CREATE_SYMBOL="u0"
If you suspect the guard was triggered from C++, add TORCHDYNAMO_EXTENDED_DEBUG_CPP=1
For more debugging help, see https://docs.google.com/document/d/1HSuTTVvYH1pTew89Rtpeu84Ht3nQEFTYhAX3Ypa_xJs/edit?usp=sharing
User Stack (most recent call last):
(snipped, see stack below for prefix)
File "/u/pgh/mclaugha/./sandbox/torch_compile_cat_dynamic_shapes/test.py", line 18, in forward
ml_atoms = torch.cat((ml_atoms, ml_atoms_vs), dim=1)
For C++ stack trace, run with TORCHDYNAMO_EXTENDED_DEBUG_CPP=1
For more information about this error, see: https://pytorch.org/docs/main/generated/exportdb/index.html#constrain-as-size-example
from user code:
File "/u/pgh/mclaugha/./sandbox/torch_compile_cat_dynamic_shapes/test.py", line 18, in forward
ml_atoms = torch.cat((ml_atoms, ml_atoms_vs), dim=1)
```
### Versions
2.7.0
cc @chauhang @penguinwu @eellison @ezyang @bobrenjc93 @voznesenskym @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @chenyang78 @kadeng @amjames @zou3519 @bdhirsh | true |
3,029,602,252 | [CUDAGraph Trees] support memory allocation on side stream | BoyuanFeng | closed | [
"Merged",
"Reverted",
"module: cuda graphs",
"ciflow/trunk",
"topic: not user facing",
"module: inductor",
"module: dynamo",
"ciflow/inductor",
"ci-no-td"
] | 13 | CONTRIBUTOR | I tried `beginAllocateToPool` instead of `_cuda_beginAllocateCurrentStreamToPool` and the error in #151199 does not happen any more.
However, this approach is unsafe for multithreading. When multiple run_eager happens concurrently, we expect memory allocation to different mem_pool. Since beginAllocateToPool does not check stream, these memory allocation may happen on the same mem_pool.
So, I use `_cuda_beginAllocateCurrentThreadToPool` to direct all memory allocation on the same thread to a given mem_pool. In particular, `_cuda_beginAllocateCurrentThreadToPool` records the launching thread id, and during runtime checks if the current thread id matches the launching thread id.
Fixes #151199
cc @mcarilli @ezyang @eellison @penguinwu @voznesenskym @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @ipiszy @chenyang78 @kadeng @muchulee8 @amjames @chauhang @aakhundov | true |
3,029,600,038 | add device generalisation support for distributed tests | harikodali | open | [
"oncall: distributed",
"triaged",
"open source",
"release notes: distributed (c10d)",
"ciflow/xpu"
] | 1 | NONE | Modified following files for device generalization
test/distributed/optim/test_zero_redundancy_optimizer.py
test/distributed/test_c10d_logger.py
test/distributed/test_compute_comm_reordering.py
torch/testing/_internal/common_distributed.py
device info is retrieved and used for tests, via get_devtype() from torch.testing._internal.common_fsdp.
DistributedTestBase is used instead of MultiProcessTestCase, to make use of helper functions.
cc @H-Huang @awgu @wanchaol @fegin @fduwjj @wz337 @wconstab @d4l3k | true |
3,029,599,825 | DISABLED test_comprehensive_polygamma_polygamma_n_1_cuda_float16 (__main__.TestInductorOpInfoCUDA) | pytorch-bot[bot] | open | [
"high priority",
"triaged",
"module: flaky-tests",
"skipped",
"oncall: pt2",
"module: inductor"
] | 5 | NONE | Platforms: inductor
This test was disabled because it is failing in CI. See [recent examples](https://hud.pytorch.org/flakytest?name=test_comprehensive_polygamma_polygamma_n_1_cuda_float16&suite=TestInductorOpInfoCUDA&limit=100) and the most recent trunk [workflow logs](https://github.com/pytorch/pytorch/runs/41372514673).
Over the past 3 hours, it has been determined flaky in 3 workflow(s) with 3 failures and 3 successes.
**Debugging instructions (after clicking on the recent samples link):**
DO NOT ASSUME THINGS ARE OKAY IF THE CI IS GREEN. We now shield flaky tests from developers so CI will thus be green but it will be harder to parse the logs.
To find relevant log snippets:
1. Click on the workflow logs linked above
2. Click on the Test step of the job so that it is expanded. Otherwise, the grepping will not work.
3. Grep for `test_comprehensive_polygamma_polygamma_n_1_cuda_float16`
4. There should be several instances run (as flaky tests are rerun in CI) from which you can study the logs.
<details><summary>Sample error message</summary>
```
Traceback (most recent call last):
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/testing/_internal/common_device_type.py", line 1131, in test_wrapper
return test(*args, **kwargs)
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/testing/_internal/common_device_type.py", line 1430, in only_fn
return fn(self, *args, **kwargs)
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/testing/_internal/common_utils.py", line 2291, in wrapper
fn(*args, **kwargs)
~~^^^^^^^^^^^^^^^^^
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/testing/_internal/common_device_type.py", line 1211, in dep_fn
return fn(slf, *args, **kwargs)
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/testing/_internal/common_device_type.py", line 1211, in dep_fn
return fn(slf, *args, **kwargs)
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/testing/_internal/common_device_type.py", line 1211, in dep_fn
return fn(slf, *args, **kwargs)
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/testing/_internal/common_utils.py", line 1612, in wrapper
fn(*args, **kwargs)
~~^^^^^^^^^^^^^^^^^
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/testing/_internal/common_utils.py", line 1534, in wrapper
fn(*args, **kwargs)
~~^^^^^^^^^^^^^^^^^
File "/opt/conda/envs/py_3.13/lib/python3.13/unittest/mock.py", line 1424, in patched
return func(*newargs, **newkeywargs)
File "/opt/conda/envs/py_3.13/lib/python3.13/contextlib.py", line 85, in inner
return func(*args, **kwds)
File "/opt/conda/envs/py_3.13/lib/python3.13/contextlib.py", line 85, in inner
return func(*args, **kwds)
File "/opt/conda/envs/py_3.13/lib/python3.13/contextlib.py", line 85, in inner
return func(*args, **kwds)
File "/var/lib/jenkins/workspace/test/inductor/test_torchinductor_opinfo.py", line 962, in inner
raise e
File "/var/lib/jenkins/workspace/test/inductor/test_torchinductor_opinfo.py", line 954, in inner
fn(self, device, dtype, op)
~~^^^^^^^^^^^^^^^^^^^^^^^^^
File "/var/lib/jenkins/workspace/test/inductor/test_torchinductor_opinfo.py", line 1207, in test_comprehensive
raise e
File "/var/lib/jenkins/workspace/test/inductor/test_torchinductor_opinfo.py", line 1182, in test_comprehensive
self.check_model_gpu(
~~~~~~~~~~~~~~~~~~~~^
fn,
^^^
...<2 lines>...
**adjusted_kwargs,
^^^^^^^^^^^^^^^^^^
)
^
File "/opt/conda/envs/py_3.13/lib/python3.13/contextlib.py", line 85, in inner
return func(*args, **kwds)
File "/var/lib/jenkins/workspace/test/inductor/test_torchinductor.py", line 657, in check_model_gpu
check_model(
~~~~~~~~~~~^
self,
^^^^^
...<13 lines>...
output_process_fn_grad=output_process_fn_grad,
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
)
^
File "/var/lib/jenkins/workspace/test/inductor/test_torchinductor.py", line 608, in check_model
actual_grad = compute_grads(example_inputs, kwargs, actual, grads)
File "/var/lib/jenkins/workspace/test/inductor/test_torchinductor.py", line 406, in compute_grads
return torch.autograd.grad(
~~~~~~~~~~~~~~~~~~~^
flat_diff_results,
^^^^^^^^^^^^^^^^^^
...<3 lines>...
retain_graph=True,
^^^^^^^^^^^^^^^^^^
)
^
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/autograd/__init__.py", line 503, in grad
result = _engine_run_backward(
outputs,
...<5 lines>...
accumulate_grad=False,
)
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/autograd/graph.py", line 824, in _engine_run_backward
return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
t_outputs, *args, **kwargs
^^^^^^^^^^^^^^^^^^^^^^^^^^
) # Calls into the C++ engine to run the backward pass
^
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/autograd/function.py", line 307, in apply
return user_fn(self, *args)
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/_functorch/_aot_autograd/runtime_wrappers.py", line 2163, in backward
return impl_fn()
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/_functorch/_aot_autograd/runtime_wrappers.py", line 2149, in impl_fn
out = CompiledFunction._backward_impl(ctx, all_args)
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/_functorch/_aot_autograd/runtime_wrappers.py", line 2241, in _backward_impl
CompiledFunction.compiled_bw = aot_config.bw_compiler(
~~~~~~~~~~~~~~~~~~~~~~^
copy.deepcopy(bw_module), placeholder_list
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
)
^
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/_functorch/aot_autograd.py", line 483, in __call__
return self.compiler_fn(gm, example_inputs)
~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/_dynamo/backends/common.py", line 73, in _wrapped_bw_compiler
disable(
~~~~~~~~
bw_compiler_fn, reason="do not trace backward compiler function"
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
)(*args, **kwargs),
~^^^^^^^^^^^^^^^^^
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/_dynamo/eval_frame.py", line 856, in _fn
return fn(*args, **kwargs)
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/_utils_internal.py", line 97, in wrapper_function
return function(*args, **kwargs)
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/_inductor/compile_fx.py", line 2201, in bw_compiler
return inner_compile(
gm,
...<5 lines>...
boxed_forward_device_index=forward_device,
)
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/_inductor/compile_fx.py", line 726, in compile_fx_inner
return wrap_compiler_debug(_compile_fx_inner, compiler_name="inductor")(
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^
gm,
^^^
example_inputs,
^^^^^^^^^^^^^^^
**kwargs,
^^^^^^^^^
)
^
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/_dynamo/repro/after_aot.py", line 124, in debug_wrapper
inner_compiled_fn = compiler_fn(gm, example_inputs)
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/_inductor/compile_fx.py", line 862, in _compile_fx_inner
raise InductorError(e, currentframe()).with_traceback(
e.__traceback__
) from None
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/_inductor/compile_fx.py", line 846, in _compile_fx_inner
mb_compiled_graph = fx_codegen_and_compile(
gm, example_inputs, inputs_to_check, **graph_kwargs
)
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/_inductor/compile_fx.py", line 1460, in fx_codegen_and_compile
return scheme.codegen_and_compile(gm, example_inputs, inputs_to_check, graph_kwargs)
~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/_inductor/compile_fx.py", line 1347, in codegen_and_compile
compiled_module = graph.compile_to_module()
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/_inductor/graph.py", line 2219, in compile_to_module
return self._compile_to_module()
~~~~~~~~~~~~~~~~~~~~~~~^^
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/_inductor/graph.py", line 2266, in _compile_to_module
mod = PyCodeCache.load_by_key_path(
key,
...<2 lines>...
attrs={**self.constants, **self.torchbind_constants},
)
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/_inductor/codecache.py", line 3006, in load_by_key_path
mod = _reload_python_module(key, path, set_sys_modules=in_toplevel)
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/_inductor/runtime/compile_tasks.py", line 31, in _reload_python_module
exec(code, mod.__dict__, mod.__dict__)
~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/tmp/tmpfcfyaga1/le/clerbfl5cvgzsi7mue3yy6hq2346emksh4svebdjys63r4nrnkef.py", line 76, in <module>
async_compile.wait(globals())
~~~~~~~~~~~~~~~~~~^^^^^^^^^^^
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/_inductor/async_compile.py", line 448, in wait
self._wait_futures(scope)
~~~~~~~~~~~~~~~~~~^^^^^^^
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/_inductor/async_compile.py", line 468, in _wait_futures
scope[key] = result.result()
~~~~~~~~~~~~~^^
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/_inductor/codecache.py", line 3508, in result
return self.result_fn()
~~~~~~~~~~~~~~^^
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/_inductor/async_compile.py", line 343, in get_result
kernel.precompile(
~~~~~~~~~~~~~~~~~^
warm_cache_only=False,
^^^^^^^^^^^^^^^^^^^^^^
reload_kernel=reload_kernel_in_parent,
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
static_triton_bundle_key=CompiledTritonKernels.key(source_code),
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
)
^
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/_inductor/runtime/triton_heuristics.py", line 322, in precompile
self._make_launchers()
~~~~~~~~~~~~~~~~~~~~^^
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/_inductor/runtime/triton_heuristics.py", line 479, in _make_launchers
launchers.append(result.make_launcher())
~~~~~~~~~~~~~~~~~~~~^^
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/_inductor/runtime/triton_heuristics.py", line 1276, in make_launcher
self.reload_cubin_path()
~~~~~~~~~~~~~~~~~~~~~~^^
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/_inductor/runtime/triton_heuristics.py", line 1268, in reload_cubin_path
raise RuntimeError(
"Cubin file saved by TritonBundler not found at %s", cubin_location
)
torch._inductor.exc.InductorError: RuntimeError: ('Cubin file saved by TritonBundler not found at %s', '/tmp/tmp31qnqqmw/triton/OPN4Y6JTPZRK7W3V3VNIMP4B6FRUQAWHOA3KTF3NPIVJSIQC3SUA/triton_poi_fused_mul_0.cubin')
The above exception was the direct cause of the following exception:
Traceback (most recent call last):
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/testing/_internal/common_utils.py", line 3154, in wrapper
method(*args, **kwargs)
~~~~~~^^^^^^^^^^^^^^^^^
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/testing/_internal/common_utils.py", line 3154, in wrapper
method(*args, **kwargs)
~~~~~~^^^^^^^^^^^^^^^^^
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/testing/_internal/common_device_type.py", line 426, in instantiated_test
result = test(self, **param_kwargs)
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/testing/_internal/common_utils.py", line 1612, in wrapper
fn(*args, **kwargs)
~~^^^^^^^^^^^^^^^^^
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/testing/_internal/common_device_type.py", line 1143, in test_wrapper
raise e_tracked from e
Exception: Caused by sample input at index 0: SampleInput(input=Tensor[size=(5, 5), device="cuda:0", dtype=torch.float16], args=(1), kwargs={}, broadcasts_input=False, name='')
To execute this test, run the following from the base repo dir:
PYTORCH_OPINFO_SAMPLE_INPUT_INDEX=0 python test/inductor/test_torchinductor_opinfo.py TestInductorOpInfoCUDA.test_comprehensive_polygamma_polygamma_n_1_cuda_float16
This message can be suppressed by setting PYTORCH_PRINT_REPRO_ON_FAILURE=0
```
</details>
Test file path: `inductor/test_torchinductor_opinfo.py`
ResponseTimeoutError: Response timeout for 5000ms, GET https://raw.githubusercontent.com/pytorch/pytorch/main/test/inductor/test_torchinductor_opinfo.py 200 (connected: true, keepalive socket: false, socketHandledRequests: 1, socketHandledResponses: 1)
headers: {"connection":"keep-alive","content-length":"46058","cache-control":"max-age=300","content-security-policy":"default-src 'none'; style-src 'unsafe-inline'; sandbox","content-type":"text/plain; charset=utf-8","etag":"\"af763533c4c8961a66cc9a4318e1cda0b3b2d5420172f31f6b1a6dbe29ece83d\"","strict-transport-security":"max-age=31536000","x-content-type-options":"nosniff","x-frame-options":"deny","x-xss-protection":"1; mode=block","x-github-request-id":"DEAA:1F7751:11F4B4C:14FC1CB:68114782","accept-ranges":"bytes","date":"Tue, 29 Apr 2025 21:41:22 GMT","via":"1.1 varnish","x-served-by":"cache-sjc10070-SJC","x-cache":"HIT","x-cache-hits":"1","x-timer":"S1745962882.468956,VS0,VE212","vary":"Authorization,Accept-Encoding,Origin","access-control-allow-origin":"*","cross-origin-resource-policy":"cross-origin","x-fastly-request-id":"dfbef9f3d638ff993eb11574bbe19b3acf91acca","expires":"Tue, 29 Apr 2025 21:46:22 GMT","source-age":"0"}
cc @ezyang @gchanan @zou3519 @kadeng @msaroufim @clee2000 @chauhang @penguinwu @voznesenskym @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @ipiszy @chenyang78 @muchulee8 @amjames @aakhundov | true |
3,029,599,820 | DISABLED test_comprehensive_polygamma_polygamma_n_0_cuda_float16 (__main__.TestInductorOpInfoCUDA) | pytorch-bot[bot] | open | [
"high priority",
"triaged",
"module: flaky-tests",
"skipped",
"oncall: pt2",
"module: inductor"
] | 6 | NONE | Platforms: inductor
This test was disabled because it is failing in CI. See [recent examples](https://hud.pytorch.org/flakytest?name=test_comprehensive_polygamma_polygamma_n_0_cuda_float16&suite=TestInductorOpInfoCUDA&limit=100) and the most recent trunk [workflow logs](https://github.com/pytorch/pytorch/runs/41372364802).
Over the past 3 hours, it has been determined flaky in 3 workflow(s) with 3 failures and 3 successes.
**Debugging instructions (after clicking on the recent samples link):**
DO NOT ASSUME THINGS ARE OKAY IF THE CI IS GREEN. We now shield flaky tests from developers so CI will thus be green but it will be harder to parse the logs.
To find relevant log snippets:
1. Click on the workflow logs linked above
2. Click on the Test step of the job so that it is expanded. Otherwise, the grepping will not work.
3. Grep for `test_comprehensive_polygamma_polygamma_n_0_cuda_float16`
4. There should be several instances run (as flaky tests are rerun in CI) from which you can study the logs.
<details><summary>Sample error message</summary>
```
Traceback (most recent call last):
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/testing/_internal/common_device_type.py", line 1131, in test_wrapper
return test(*args, **kwargs)
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/testing/_internal/common_device_type.py", line 1430, in only_fn
return fn(self, *args, **kwargs)
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/testing/_internal/common_utils.py", line 2291, in wrapper
fn(*args, **kwargs)
~~^^^^^^^^^^^^^^^^^
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/testing/_internal/common_device_type.py", line 1211, in dep_fn
return fn(slf, *args, **kwargs)
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/testing/_internal/common_device_type.py", line 1211, in dep_fn
return fn(slf, *args, **kwargs)
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/testing/_internal/common_device_type.py", line 1211, in dep_fn
return fn(slf, *args, **kwargs)
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/testing/_internal/common_utils.py", line 1612, in wrapper
fn(*args, **kwargs)
~~^^^^^^^^^^^^^^^^^
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/testing/_internal/common_utils.py", line 1534, in wrapper
fn(*args, **kwargs)
~~^^^^^^^^^^^^^^^^^
File "/opt/conda/envs/py_3.13/lib/python3.13/unittest/mock.py", line 1424, in patched
return func(*newargs, **newkeywargs)
File "/opt/conda/envs/py_3.13/lib/python3.13/contextlib.py", line 85, in inner
return func(*args, **kwds)
File "/opt/conda/envs/py_3.13/lib/python3.13/contextlib.py", line 85, in inner
return func(*args, **kwds)
File "/opt/conda/envs/py_3.13/lib/python3.13/contextlib.py", line 85, in inner
return func(*args, **kwds)
File "/var/lib/jenkins/workspace/test/inductor/test_torchinductor_opinfo.py", line 962, in inner
raise e
File "/var/lib/jenkins/workspace/test/inductor/test_torchinductor_opinfo.py", line 954, in inner
fn(self, device, dtype, op)
~~^^^^^^^^^^^^^^^^^^^^^^^^^
File "/var/lib/jenkins/workspace/test/inductor/test_torchinductor_opinfo.py", line 1207, in test_comprehensive
raise e
File "/var/lib/jenkins/workspace/test/inductor/test_torchinductor_opinfo.py", line 1182, in test_comprehensive
self.check_model_gpu(
~~~~~~~~~~~~~~~~~~~~^
fn,
^^^
...<2 lines>...
**adjusted_kwargs,
^^^^^^^^^^^^^^^^^^
)
^
File "/opt/conda/envs/py_3.13/lib/python3.13/contextlib.py", line 85, in inner
return func(*args, **kwds)
File "/var/lib/jenkins/workspace/test/inductor/test_torchinductor.py", line 657, in check_model_gpu
check_model(
~~~~~~~~~~~^
self,
^^^^^
...<13 lines>...
output_process_fn_grad=output_process_fn_grad,
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
)
^
File "/var/lib/jenkins/workspace/test/inductor/test_torchinductor.py", line 608, in check_model
actual_grad = compute_grads(example_inputs, kwargs, actual, grads)
File "/var/lib/jenkins/workspace/test/inductor/test_torchinductor.py", line 406, in compute_grads
return torch.autograd.grad(
~~~~~~~~~~~~~~~~~~~^
flat_diff_results,
^^^^^^^^^^^^^^^^^^
...<3 lines>...
retain_graph=True,
^^^^^^^^^^^^^^^^^^
)
^
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/autograd/__init__.py", line 503, in grad
result = _engine_run_backward(
outputs,
...<5 lines>...
accumulate_grad=False,
)
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/autograd/graph.py", line 824, in _engine_run_backward
return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
t_outputs, *args, **kwargs
^^^^^^^^^^^^^^^^^^^^^^^^^^
) # Calls into the C++ engine to run the backward pass
^
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/autograd/function.py", line 307, in apply
return user_fn(self, *args)
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/_functorch/_aot_autograd/runtime_wrappers.py", line 2163, in backward
return impl_fn()
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/_functorch/_aot_autograd/runtime_wrappers.py", line 2149, in impl_fn
out = CompiledFunction._backward_impl(ctx, all_args)
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/_functorch/_aot_autograd/runtime_wrappers.py", line 2241, in _backward_impl
CompiledFunction.compiled_bw = aot_config.bw_compiler(
~~~~~~~~~~~~~~~~~~~~~~^
copy.deepcopy(bw_module), placeholder_list
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
)
^
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/_functorch/aot_autograd.py", line 483, in __call__
return self.compiler_fn(gm, example_inputs)
~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/_dynamo/backends/common.py", line 73, in _wrapped_bw_compiler
disable(
~~~~~~~~
bw_compiler_fn, reason="do not trace backward compiler function"
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
)(*args, **kwargs),
~^^^^^^^^^^^^^^^^^
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/_dynamo/eval_frame.py", line 856, in _fn
return fn(*args, **kwargs)
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/_utils_internal.py", line 97, in wrapper_function
return function(*args, **kwargs)
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/_inductor/compile_fx.py", line 2201, in bw_compiler
return inner_compile(
gm,
...<5 lines>...
boxed_forward_device_index=forward_device,
)
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/_inductor/compile_fx.py", line 726, in compile_fx_inner
return wrap_compiler_debug(_compile_fx_inner, compiler_name="inductor")(
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^
gm,
^^^
example_inputs,
^^^^^^^^^^^^^^^
**kwargs,
^^^^^^^^^
)
^
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/_dynamo/repro/after_aot.py", line 124, in debug_wrapper
inner_compiled_fn = compiler_fn(gm, example_inputs)
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/_inductor/compile_fx.py", line 862, in _compile_fx_inner
raise InductorError(e, currentframe()).with_traceback(
e.__traceback__
) from None
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/_inductor/compile_fx.py", line 846, in _compile_fx_inner
mb_compiled_graph = fx_codegen_and_compile(
gm, example_inputs, inputs_to_check, **graph_kwargs
)
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/_inductor/compile_fx.py", line 1460, in fx_codegen_and_compile
return scheme.codegen_and_compile(gm, example_inputs, inputs_to_check, graph_kwargs)
~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/_inductor/compile_fx.py", line 1347, in codegen_and_compile
compiled_module = graph.compile_to_module()
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/_inductor/graph.py", line 2219, in compile_to_module
return self._compile_to_module()
~~~~~~~~~~~~~~~~~~~~~~~^^
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/_inductor/graph.py", line 2266, in _compile_to_module
mod = PyCodeCache.load_by_key_path(
key,
...<2 lines>...
attrs={**self.constants, **self.torchbind_constants},
)
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/_inductor/codecache.py", line 3006, in load_by_key_path
mod = _reload_python_module(key, path, set_sys_modules=in_toplevel)
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/_inductor/runtime/compile_tasks.py", line 31, in _reload_python_module
exec(code, mod.__dict__, mod.__dict__)
~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/tmp/tmp8r24cgp0/4q/c4qwyn7lpefsq4v7psi2ipa66uaugapjlmpblytcooo3sdy2igx2.py", line 79, in <module>
async_compile.wait(globals())
~~~~~~~~~~~~~~~~~~^^^^^^^^^^^
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/_inductor/async_compile.py", line 448, in wait
self._wait_futures(scope)
~~~~~~~~~~~~~~~~~~^^^^^^^
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/_inductor/async_compile.py", line 468, in _wait_futures
scope[key] = result.result()
~~~~~~~~~~~~~^^
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/_inductor/codecache.py", line 3508, in result
return self.result_fn()
~~~~~~~~~~~~~~^^
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/_inductor/async_compile.py", line 343, in get_result
kernel.precompile(
~~~~~~~~~~~~~~~~~^
warm_cache_only=False,
^^^^^^^^^^^^^^^^^^^^^^
reload_kernel=reload_kernel_in_parent,
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
static_triton_bundle_key=CompiledTritonKernels.key(source_code),
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
)
^
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/_inductor/runtime/triton_heuristics.py", line 322, in precompile
self._make_launchers()
~~~~~~~~~~~~~~~~~~~~^^
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/_inductor/runtime/triton_heuristics.py", line 479, in _make_launchers
launchers.append(result.make_launcher())
~~~~~~~~~~~~~~~~~~~~^^
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/_inductor/runtime/triton_heuristics.py", line 1276, in make_launcher
self.reload_cubin_path()
~~~~~~~~~~~~~~~~~~~~~~^^
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/_inductor/runtime/triton_heuristics.py", line 1268, in reload_cubin_path
raise RuntimeError(
"Cubin file saved by TritonBundler not found at %s", cubin_location
)
torch._inductor.exc.InductorError: RuntimeError: ('Cubin file saved by TritonBundler not found at %s', '/tmp/tmpjt7206uo/triton/A7NL4DJP7DPYVZJOUY3KZAS6IXTYSHXPX4GE3WY2YQVV744RRENQ/triton_poi_fused_mul_0.cubin')
The above exception was the direct cause of the following exception:
Traceback (most recent call last):
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/testing/_internal/common_utils.py", line 3154, in wrapper
method(*args, **kwargs)
~~~~~~^^^^^^^^^^^^^^^^^
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/testing/_internal/common_utils.py", line 3154, in wrapper
method(*args, **kwargs)
~~~~~~^^^^^^^^^^^^^^^^^
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/testing/_internal/common_device_type.py", line 426, in instantiated_test
result = test(self, **param_kwargs)
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/testing/_internal/common_utils.py", line 1612, in wrapper
fn(*args, **kwargs)
~~^^^^^^^^^^^^^^^^^
File "/opt/conda/envs/py_3.13/lib/python3.13/site-packages/torch/testing/_internal/common_device_type.py", line 1143, in test_wrapper
raise e_tracked from e
Exception: Caused by sample input at index 7: SampleInput(input=Tensor[size=(), device="cuda:0", dtype=torch.float16], args=(3), kwargs={}, broadcasts_input=False, name='')
To execute this test, run the following from the base repo dir:
PYTORCH_OPINFO_SAMPLE_INPUT_INDEX=7 python test/inductor/test_torchinductor_opinfo.py TestInductorOpInfoCUDA.test_comprehensive_polygamma_polygamma_n_0_cuda_float16
This message can be suppressed by setting PYTORCH_PRINT_REPRO_ON_FAILURE=0
```
</details>
Test file path: `inductor/test_torchinductor_opinfo.py`
ResponseTimeoutError: Response timeout for 5000ms, GET https://raw.githubusercontent.com/pytorch/pytorch/main/test/inductor/test_torchinductor_opinfo.py 200 (connected: true, keepalive socket: false, socketHandledRequests: 1, socketHandledResponses: 1)
headers: {"connection":"keep-alive","content-length":"46058","cache-control":"max-age=300","content-security-policy":"default-src 'none'; style-src 'unsafe-inline'; sandbox","content-type":"text/plain; charset=utf-8","etag":"\"af763533c4c8961a66cc9a4318e1cda0b3b2d5420172f31f6b1a6dbe29ece83d\"","strict-transport-security":"max-age=31536000","x-content-type-options":"nosniff","x-frame-options":"deny","x-xss-protection":"1; mode=block","x-github-request-id":"DEAA:1F7751:11F4B4C:14FC1CB:68114782","accept-ranges":"bytes","date":"Tue, 29 Apr 2025 21:41:22 GMT","via":"1.1 varnish","x-served-by":"cache-sjc1000114-SJC","x-cache":"MISS","x-cache-hits":"0","x-timer":"S1745962882.467390,VS0,VE210","vary":"Authorization,Accept-Encoding,Origin","access-control-allow-origin":"*","cross-origin-resource-policy":"cross-origin","x-fastly-request-id":"4bf7e97aa0bc090adc18f7ec19f0e74df6647bd6","expires":"Tue, 29 Apr 2025 21:46:22 GMT","source-age":"0"}
cc @ezyang @gchanan @zou3519 @kadeng @msaroufim @clee2000 @chauhang @penguinwu @voznesenskym @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @ipiszy @chenyang78 @muchulee8 @amjames @aakhundov | true |
3,029,593,172 | [CUDA][TF32] Account for TF32 in `compile_kernel_advanced` | eqy | closed | [
"module: cuda",
"open source",
"Merged",
"module: tf32",
"ciflow/trunk",
"topic: not user facing"
] | 3 | COLLABORATOR | Also cleanup some uses of `assert_close` in favor of `self.assertEqual`
cc @ptrblck @msaroufim @jerryzh168 @zasdfgbnm | true |
3,029,591,389 | `torch.export` fails on `InstanceNorm1d` | ar0ck | closed | [
"oncall: pt2",
"oncall: export"
] | 2 | CONTRIBUTOR | ### 🐛 Describe the bug
```python
import torch
mod = torch.nn.InstanceNorm1d(1)
args = torch.randn(1, 2),
torch.export.export(mod, args)
```
### Error logs
```python
Traceback (most recent call last):
File "bug.py", line 5, in <module>
torch.export.export(mod, args)
File ".../lib/python3.12/site-packages/torch/export/__init__.py", line 318, in export
raise e
File ".../lib/python3.12/site-packages/torch/export/__init__.py", line 285, in export
return _export(
^^^^^^^^
File ".../lib/python3.12/site-packages/torch/export/_trace.py", line 1110, in wrapper
raise e
File ".../lib/python3.12/site-packages/torch/export/_trace.py", line 1076, in wrapper
ep = fn(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^
File ".../lib/python3.12/site-packages/torch/export/exported_program.py", line 122, in wrapper
return fn(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^
File ".../lib/python3.12/site-packages/torch/export/_trace.py", line 2122, in _export
ep = _export_for_training(
^^^^^^^^^^^^^^^^^^^^^
File ".../lib/python3.12/site-packages/torch/export/_trace.py", line 1110, in wrapper
raise e
File ".../lib/python3.12/site-packages/torch/export/_trace.py", line 1076, in wrapper
ep = fn(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^
File ".../lib/python3.12/site-packages/torch/export/exported_program.py", line 122, in wrapper
return fn(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^
File ".../lib/python3.12/site-packages/torch/export/_trace.py", line 1983, in _export_for_training
export_artifact = export_func(
^^^^^^^^^^^^
File ".../lib/python3.12/site-packages/torch/export/_trace.py", line 1925, in _non_strict_export
aten_export_artifact = _to_aten_func( # type: ignore[operator]
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File ".../lib/python3.12/site-packages/torch/export/_trace.py", line 1710, in _export_to_aten_ir_make_fx
gm, graph_signature = transform(_make_fx_helper)(
^^^^^^^^^^^^^^^^^^^^^^^^^^^
File ".../lib/python3.12/site-packages/torch/export/_trace.py", line 1851, in _aot_export_non_strict
gm, sig = aot_export(wrapped_mod, args, kwargs=kwargs, **flags)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File ".../lib/python3.12/site-packages/torch/export/_trace.py", line 1639, in _make_fx_helper
buf: input_names[param_len + i]
~~~~~~~~~~~^^^^^^^^^^^^^^^
IndexError: list index out of range
```
### Versions
```
Collecting environment information...
PyTorch version: 2.8.0.dev20250428+cpu
Is debug build: False
CUDA used to build PyTorch: None
ROCM used to build PyTorch: N/A
OS: Ubuntu 24.04.2 LTS (x86_64)
GCC version: (Ubuntu 13.3.0-6ubuntu2~24.04) 13.3.0
Clang version: Could not collect
CMake version: version 3.28.3
Libc version: glibc-2.39
Python version: 3.12.3 (main, Feb 4 2025, 14:48:35) [GCC 13.3.0] (64-bit runtime)
Python platform: Linux-6.8.0-45-generic-x86_64-with-glibc2.39
Is CUDA available: False
CUDA runtime version: No CUDA
CUDA_MODULE_LOADING set to: N/A
GPU models and configuration: No CUDA
Nvidia driver version: No CUDA
cuDNN version: No CUDA
HIP runtime version: N/A
MIOpen runtime version: N/A
Is XNNPACK available: True
CPU:
Architecture: x86_64
CPU op-mode(s): 32-bit, 64-bit
Address sizes: 46 bits physical, 48 bits virtual
Byte Order: Little Endian
CPU(s): 20
On-line CPU(s) list: 0-19
Vendor ID: GenuineIntel
Model name: 13th Gen Intel(R) Core(TM) i7-1370P
CPU family: 6
Model: 186
Thread(s) per core: 2
Core(s) per socket: 14
Socket(s): 1
Stepping: 2
CPU(s) scaling MHz: 16%
CPU max MHz: 5200.0000
CPU min MHz: 400.0000
BogoMIPS: 4377.60
Flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx pdpe1gb rdtscp lm constant_tsc art arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc cpuid aperfmperf tsc_known_freq pni pclmulqdq dtes64 monitor ds_cpl vmx smx est tm2 ssse3 sdbg fma cx16 xtpr pdcm pcid sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand lahf_lm abm 3dnowprefetch cpuid_fault epb ssbd ibrs ibpb stibp ibrs_enhanced tpr_shadow flexpriority ept vpid ept_ad fsgsbase tsc_adjust bmi1 avx2 smep bmi2 erms invpcid rdseed adx smap clflushopt clwb intel_pt sha_ni xsaveopt xsavec xgetbv1 xsaves split_lock_detect user_shstk avx_vnni dtherm ida arat pln pts hwp hwp_notify hwp_act_window hwp_epp hwp_pkg_req hfi vnmi umip pku ospke waitpkg gfni vaes vpclmulqdq tme rdpid movdiri movdir64b fsrm md_clear serialize pconfig arch_lbr ibt flush_l1d arch_capabilities
Virtualization: VT-x
L1d cache: 544 KiB (14 instances)
L1i cache: 704 KiB (14 instances)
L2 cache: 11.5 MiB (8 instances)
L3 cache: 24 MiB (1 instance)
NUMA node(s): 1
NUMA node0 CPU(s): 0-19
Vulnerability Gather data sampling: Not affected
Vulnerability Itlb multihit: Not affected
Vulnerability L1tf: Not affected
Vulnerability Mds: Not affected
Vulnerability Meltdown: Not affected
Vulnerability Mmio stale data: Not affected
Vulnerability Reg file data sampling: Mitigation; Clear Register File
Vulnerability Retbleed: Not affected
Vulnerability Spec rstack overflow: Not affected
Vulnerability Spec store bypass: Mitigation; Speculative Store Bypass disabled via prctl
Vulnerability Spectre v1: Mitigation; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2: Mitigation; Enhanced / Automatic IBRS; IBPB conditional; RSB filling; PBRSB-eIBRS SW sequence; BHI BHI_DIS_S
Vulnerability Srbds: Not affected
Vulnerability Tsx async abort: Not affected
Versions of relevant libraries:
[pip3] torch==2.8.0.dev20250428+cpu
[conda] Could not collect
```
cc @chauhang @penguinwu @avikchaudhuri @gmagogsfm @zhxchen17 @tugsbayasgalan @angelayi @suo @ydwu4 | true |
3,029,584,793 | [MPS][BE] Introduce `c10::metal::mul` | malfet | closed | [
"Merged",
"topic: not user facing",
"release notes: mps",
"ciflow/mps"
] | 6 | CONTRIBUTOR | Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom):
* #152510
* #152485
* #152504
* #152479
* __->__ #152466
Which multiplies two arguments for either scalar or complex data types
This allows one to get rid of bunch of complex specialization in BinaryOps | true |
3,029,536,955 | torch.export with dynamic shapes on Static Cache HF LLama model fails | peri044 | open | [
"oncall: pt2",
"oncall: export"
] | 1 | CONTRIBUTOR | ### 🐛 Describe the bug
I'm trying to export HF Llama model with Static Cache. StaticCache export is supported but only can be used with static input shapes (https://github.com/huggingface/transformers/blob/f39f4960f30e3eadd6d948e4dcb2da32eda253b5/tests/utils/test_cache_utils.py#L247-L271 and https://github.com/huggingface/transformers/blob/main/src/transformers/integrations/executorch.py#L493-L497). When I try to export it using dynamic shapes, I face the following error
```py
torch._dynamo.exc.UserError: Cannot associate shape {} specified at `dynamic_shapes['past_key_values']` to non-tensor type <class 'transformers.cache_utils.StaticCache'> at `inputs['past_key_values']` (expected None)
For more information about this error, see: https://pytorch.org/docs/main/generated/exportdb/index.html#dynamic-shapes-validation
The error above occurred when calling torch.export.export. If you would like to view some more information about this error, and get a list of all other errors that may occur in your export call, you can replace your `export()` call with `draft_export()`
```
If I change the past_key_values (in `dynamic_shapes`) to `"past_key_values" : None`, the error is
```py
ValueError: Unsupported input type <class 'transformers.cache_utils.StaticCache'>. Export only supports pytree containers of basic types (Tensor, int, float, ...) as input. To register a custom dataclass, use torch.export.register_dataclass. To register a custom container type, use torch.utils._pytree.register_pytree_node. To register a constant input, use torch.utils._pytree.register_constant
```
What is the correct way to export it with dynamic shapes ? Is this supported ? Thanks !!
cc: @angelayi
```py
import torch
import transformers
from transformers import AutoModelForCausalLM, AutoTokenizer, GenerationConfig, TorchExportableModuleWithStaticCache, StaticCache
with torch.inference_mode():
max_seq_len = 2176
DEVICE="cuda"
model_id = "meta-llama/Llama-3.2-1B-Instruct"
model = AutoModelForCausalLM.from_pretrained(
model_id,
device_map=DEVICE,
torch_dtype=torch.float16,
attn_implementation="sdpa",
num_hidden_layers=1,
generation_config=GenerationConfig(
use_cache=True,
cache_implementation="static",
max_length=max_seq_len,
cache_config={
"batch_size": 1,
"max_cache_len": max_seq_len,
"device": DEVICE,
},
),
).eval().cuda()
static_cache = StaticCache(
config=model.config,
max_batch_size=model.generation_config.cache_config.batch_size,
max_cache_len=model.generation_config.cache_config.max_cache_len,
device=model.generation_config.cache_config.device,
dtype=model.dtype,
)
for i in range(len(static_cache.key_cache)):
model.register_buffer(f"key_cache_{i}", static_cache.key_cache[i], persistent=False)
model.register_buffer(f"value_cache_{i}", static_cache.value_cache[i], persistent=False)
model.is_causal = any("CausalLM" in arch for arch in model.model.config.architectures)
if model.is_causal:
causal_mask = torch.tril(
torch.ones(
static_cache.max_cache_len,
static_cache.max_cache_len,
dtype=torch.bool,
)
)
model.register_buffer("mask", causal_mask, persistent=False)
tokenizer = AutoTokenizer.from_pretrained(model_id)
prompt = "What is parallel programming ?"
model_inputs = tokenizer(prompt, return_tensors="pt")
input_ids = model_inputs["input_ids"].to(DEVICE)
cache_position = torch.arange(input_ids.shape[1]).to(DEVICE)
seq_len = torch.export.Dim("seq_len", min=1, max=max_seq_len)
cache_len = torch.export.Dim("cache_len", min=1, max=max_seq_len)
exported_program = torch.export.export(
model,
args=(),
kwargs={"input_ids" : input_ids, "cache_position": cache_position, "past_key_values": static_cache},
dynamic_shapes={"input_ids" : {1: seq_len}, "cache_position" : {0: cache_len}, "past_key_values" : {}},
strict=False,
)
gm = exported_program.module()
print(gm.graph)
```
### Versions
>>> import torch
>>> torch.__version__
'2.8.0.dev20250423+cu128'
>>> transformers.__version__
'4.49.0'
cc @chauhang @penguinwu @avikchaudhuri @gmagogsfm @zhxchen17 @tugsbayasgalan @angelayi @suo @ydwu4 | true |
3,029,534,053 | Run link checks on modified files on push too | shoumikhin | closed | [
"Merged",
"topic: not user facing"
] | 3 | CONTRIBUTOR | https://github.com/pytorch/pytorch/issues/152439 | true |
3,029,517,587 | consolidate guard_or_x and definitely_x | laithsakka | closed | [
"Merged",
"ciflow/trunk",
"release notes: fx",
"fx",
"module: inductor",
"ciflow/inductor",
"suppress-bc-linter"
] | 12 | CONTRIBUTOR | Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom):
* __->__ #152463
definitely_true is almost same as guard_or_false, the potential differences are not meaningful to a degree that justify the
existence of both. same for definitely_false, it can be expressed with guard_or_true and guard_or_false.
cc @ezyang @SherlockNoMad @EikanWang @jgong5 @wenzhe-nrv @voznesenskym @penguinwu @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @jiayisunx @ipiszy @chenyang78 @kadeng @muchulee8 @amjames @chauhang @aakhundov | true |
3,029,492,771 | [standalone_compile] fix dynamic shapes with config_patches | zou3519 | closed | [
"Merged",
"ciflow/trunk",
"module: inductor",
"ciflow/inductor",
"release notes: inductor"
] | 3 | CONTRIBUTOR | Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom):
* __->__ #152462
compile_fx with config_patches goes down another path where we need to
propagate the kwarg...
Test Plan:
- updated test
cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @ipiszy @chenyang78 @kadeng @muchulee8 @amjames @chauhang @aakhundov | true |
3,029,454,563 | Remove dead binary_ios_build, test, upload scripts | clee2000 | closed | [
"Merged",
"topic: not user facing"
] | 3 | CONTRIBUTOR | Can't find any mentions of them in the codebase, presumably no longer used? | true |
3,029,427,049 | [pytorch][triton] flex attention fwd kernel with TMA loads (#151923) | mandroid6 | open | [
"fb-exported",
"ciflow/trunk",
"topic: not user facing",
"module: inductor",
"ciflow/inductor"
] | 6 | CONTRIBUTOR | Summary:
Device side TMA for flex_attention fwd kernel, Q K V tensors
Test Plan:
Unit test:
```
buck test 'fbcode//mode/opt' fbcode//caffe2/test/inductor:flex_attention -- test_tma_with_customer_kernel_options
```
https://www.internalfb.com/intern/testinfra/testrun/14355223891618726
Differential Revision: D71082691
cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @ipiszy @chenyang78 @kadeng @muchulee8 @amjames @chauhang @aakhundov | true |
3,029,420,061 | [IR] Input Adapter refactor prototype | felixsu2006 | open | [
"fb-exported",
"release notes: export"
] | 3 | CONTRIBUTOR | Summary:
1. Adding `input` field to `_adapt_flat_args` function
2. In `process_forward_inputs`, `reorder_kwargs` will now do nothing if no kwargs are provided (previously would error)
3. Pass `args` as input to `_adapt_flat_args`
These changes are made to update the InputAdapter
see more context in D72341439
Test Plan: see D72341439
Differential Revision: D72341439
| true |
3,029,407,184 | fix: Update padding_mode to use Literal for type checking | sujeet4010 | open | [
"triaged",
"open source",
"topic: not user facing"
] | 2 | NONE | Fixes #152280 | true |
3,029,388,614 | [inductor] Fix usage of launch_enter_hook/launch_exit_hook | danzimm | closed | [
"Merged",
"ciflow/trunk",
"topic: not user facing",
"module: inductor",
"ciflow/inductor",
"release notes: inductor"
] | 5 | CONTRIBUTOR | In https://github.com/triton-lang/triton/pull/6467 I moved where `launch_enter_hook`/`launch_exit_hook` are specified (from the kernel class to a config). This PR updates the usages to use the config module if it exists to support tip of main triton.
In https://github.com/triton-lang/triton/pull/6641 I renamed `triton.config` to `triton.knobs`, hence the second commit in this PR.
Test Plan: Setup OSS PT with tip of main triton (namely including https://github.com/triton-lang/triton/pull/6641) and run `python test/inductor/test_pad_mm.py`
cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @ipiszy @chenyang78 @kadeng @muchulee8 @amjames @chauhang @aakhundov @oulgen @jamesjwu | true |
3,029,280,080 | Fix XLA issue. | laithsakka | closed | [
"release notes: jit",
"ciflow/inductor"
] | 2 | CONTRIBUTOR | Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom):
* __->__ #152456
* #152455
| true |
3,029,279,968 | [export] add runtime assert messages to python torch checks (#150719) | laithsakka | open | [
"release notes: fx",
"fx",
"ciflow/inductor"
] | 1 | CONTRIBUTOR | Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom):
* #152456
* __->__ #152455
Summary:
~fixes https://github.com/pytorch/pytorch/issues/150063 (for python at least)
Before:
```
Runtime assertion failed for expression Eq(Mod(s16*s35, s35 - 1), 0) on node 'eq'
```
Now:
```
RuntimeError: Runtime assertion failed for expression Eq(Mod(s16*s35, s35 - 1), 0) on node 'eq'
The original traceback points to the following location and error message:
/data/users/pianpwk/pytorch/torch/_prims_common/__init__.py:954
shape '[s35 - 1, ((s16*s35)//(s35 - 1))]' is invalid for input of size s16*s35
```
cc ezyang SherlockNoMad EikanWang jgong5 wenzhe-nrv
Reviewed By: laithsakka
Differential Revision: D72483950
Pulled By: pianpwk
cc @ezyang @SherlockNoMad @EikanWang @jgong5 @wenzhe-nrv | true |
3,029,254,320 | Decorate `test_host_memory_stats` with `@serialTest` | eqy | closed | [
"open source",
"Merged",
"ciflow/trunk",
"topic: not user facing"
] | 3 | COLLABORATOR | Seems to need it as it is expecting only its allocation behavior to be visible, to address #152422 | true |
3,029,244,351 | Add epoch to fake tensor cache key | tugsbayasgalan | open | [
"fb-exported",
"ciflow/trunk",
"topic: not user facing",
"ciflow/inductor"
] | 20 | CONTRIBUTOR | Summary: This is especially necessary when the output of an op is unbacked symbol. In those cases,we need to regenerate new unbacked val and bind it to the old unbacked symbol via rebind_unbacked_symbol in downstream use.
Test Plan: CI
Differential Revision: D73869239
| true |
3,029,244,113 | Implement async manifold cache write | oulgen | closed | [
"fb-exported",
"Merged",
"ciflow/trunk",
"topic: not user facing",
"module: inductor",
"ciflow/inductor"
] | 4 | CONTRIBUTOR | Summary: This diff implements an AsyncManifoldCache class that performs cache write and update ttl operations in an async manner. Essentially we are ok with the fire and forget approach where we dont guarantee that we can observe our writes, this gives us better runtime latency.
Test Plan: added new unit test
Reviewed By: jamesjwu
Differential Revision: D73867797
cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @ipiszy @chenyang78 @kadeng @muchulee8 @amjames @chauhang @aakhundov | true |
3,029,234,271 | [dynamo] `torch.compile` prevents fsdp warning from getting generated | StrongerXi | open | [
"oncall: distributed",
"triaged",
"oncall: pt2"
] | 0 | CONTRIBUTOR | ### 🐛 Describe the bug
Repro:
```
PYTORCH_TEST_WITH_DYNAMO=1 python test/distributed/fsdp/test_wrap.py TestAutoWrap.test_frozen_params
```
### Error logs
```
Traceback (most recent call last):
File "/home/ryanguo99/repos/pytorch/torch/testing/_internal/common_utils.py", line 3154, in wrapper
method(*args, **kwargs)
File "/home/ryanguo99/repos/pytorch/test/distributed/fsdp/test_wrap.py", line 883, in test_frozen_params
self._test_frozen_params(use_orig_params, policy)
File "/home/ryanguo99/repos/pytorch/test/distributed/fsdp/test_wrap.py", line 894, in _test_frozen_params
with ctx:
AssertionError: UserWarning not triggered
To execute this test, run the following from the base repo dir:
PYTORCH_TEST_WITH_DYNAMO=1 python test/distributed/fsdp/test_wrap.py TestAutoWrap.test_frozen_params
```
### Versions
48555f19062, python 3.11
cc @H-Huang @awgu @wanchaol @fegin @fduwjj @wz337 @wconstab @d4l3k @chauhang @penguinwu | true |
3,029,230,610 | [PT2] Port replace_lce_with_matmul / replace_first_lce_with_fused_matmul_lce to PT2 pre_grad passes | kqfu | open | [
"fb-exported",
"Merged",
"Reverted",
"ciflow/trunk",
"topic: not user facing",
"module: inductor",
"ciflow/inductor",
"ci-no-td"
] | 12 | CONTRIBUTOR | Summary:
Port over replace_lce_with_matmul and replace_first_lce_with_fused_matmul_lce to PT2 pre_grad pass.
Original dper pass diffs: D67884534, D68123479, D68384238
Test Plan:
Test 1. Covers replace_lce_with_matmul and case 1 of replace_first_lce_with_fused_matmul_lce
```
CUDA_VISIBLE_DEVICES=6 TORCH_LOGS=+inductor,aot TORCH_COMPILE_DEBUG=1 TORCHINDUCTOR_MAX_AUTOTUNE=1 buck2 run mode/opt-split-dwarf mode/inplace -c fbcode.platform010_cuda_version=12 -c fbcode.nvcc_arch=h100 caffe2/torch/fb/model_transform/experimental/benchmark:mts_gpu_benchmark -- --model-path=manifold://ads_storage_fblearner/tree/user/facebook/fblearner/predictor/669809193/0/gpu_lowering/input.predictor.disagg.gpu.merge --lower-backend="AOT_INDUCTOR" --add_passes="use_matmul_fuse_lce_replace_first_LCE,use_contiguous_linear_reduction_replace_linear_reduction" --batch-size=3072 --gpu-trace --disable_acc_tracer=true 2>&1 | tee ~/logs/disable_acc_tracer/aoti_cmf_ctr_triton_669809193_0_diable_acc.log
```
Log: P1798246938
Test 2. Covers replace_lce_with_matmul and case 2 of replace_first_lce_with_fused_matmul_lce
```
CUDA_VISIBLE_DEVICES=7 TORCH_LOGS=+inductor,aot TORCH_COMPILE_DEBUG=1 TORCHINDUCTOR_MAX_AUTOTUNE=1 buck2 run mode/opt-split-dwarf mode/inplace -c fbcode.platform010_cuda_version=12 -c fbcode.nvcc_arch=h100 caffe2/torch/fb/model_transform/experimental/benchmark:mts_gpu_benchmark -- --model-path=manifold://ads_storage_fblearner/tree/user/facebook/fblearner/predictor/677734158/9/gpu_lowering/input.predictor.disagg.gpu.merge --lower-backend="AOT_INDUCTOR" --add_passes="use_matmul_fuse_lce_replace_first_LCE,use_matmul_lce_replace_normal_LCE" --batch-size=3072 --gpu-trace --disable_acc_tracer=true 2>&1 | tee ~/logs/disable_acc_tracer/aoti_cmf_ctr_triton_677734158_9_diable_acc.log
```
Log: P1798246675
Seeing logs like
`[Pre grad(predispatch IR)] Apply use_matmul_fuse_lce_replace_first_LCE pass, save before/after graph to /tmp/tmp8lyzoh79, graph before/after are the same = False`
Reviewed By: huxintong
Differential Revision: D71358949
cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @ipiszy @chenyang78 @kadeng @muchulee8 @amjames @chauhang @aakhundov | true |
3,029,230,160 | Add new profiling events to `DebugAutotuner` | exclamaforte | open | [
"module: inductor",
"ciflow/inductor"
] | 2 | CONTRIBUTOR | This PR adds support for more granular profiling events in DebugAutotuner. These profiling events are already present in CachingAutotuner, and their lack in DebugAutotuner breaks #149697 when `config.profile_bandwidth` is set.
Also fixes an outstanding bug, where attempting to use the profiler to benchmark in DebugAutotuner would fail when the compilation was already in a profiler instance.
cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @ipiszy @chenyang78 @kadeng @muchulee8 @amjames @chauhang @aakhundov | true |
3,029,225,969 | add device generalisation support for distributed tests | harikodali | closed | [
"oncall: distributed",
"open source",
"topic: not user facing"
] | 1 | NONE | test/distributed/optim/test_zero_redundancy_optimizer.py
test/distributed/test_c10d_logger.py
test/distributed/test_compute_comm_reordering.py
are modified for device generic support
device info is retrieved and used for tests, via get_devtype() from torch.testing._internal.common_fsdp
DistributedTestBase is used instead of MultiProcessTestCase, to make use of helper functions
cc @H-Huang @awgu @wanchaol @fegin @fduwjj @wz337 @wconstab @d4l3k | true |
3,029,220,449 | [dynamo] guard code generation triggers attribute error on DeviceMesh object | StrongerXi | open | [
"triaged",
"oncall: pt2",
"module: dynamo"
] | 2 | CONTRIBUTOR | ### 🐛 Describe the bug
Repro:
```
PYTORCH_TEST_WITH_INDUCTOR=1 python test/distributed/tensor/debug/test_comm_mode.py TestCommMode.test_comm_mode_with_dtensor
```
### Error logs
```
Traceback (most recent call last):
File "/home/ryanguo99/repos/pytorch/torch/testing/_internal/common_utils.py", line 3154, in wrapper
method(*args, **kwargs)
File "/home/ryanguo99/repos/pytorch/test/distributed/tensor/debug/test_comm_mode.py", line 95, in test_comm_mode_with_dtensor
mesh = DeviceMesh(self.device_type, list(range(self.world_size)))
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/ryanguo99/repos/pytorch/torch/_dynamo/convert_frame.py", line 1458, in __call__
return self._torchdynamo_orig_callable(
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/ryanguo99/repos/pytorch/torch/_dynamo/convert_frame.py", line 1237, in __call__
result = self._inner_convert(
^^^^^^^^^^^^^^^^^^^^
File "/home/ryanguo99/repos/pytorch/torch/_dynamo/convert_frame.py", line 624, in __call__
return _compile(
^^^^^^^^^
File "/home/ryanguo99/repos/pytorch/torch/_dynamo/convert_frame.py", line 1133, in _compile
raise InternalTorchDynamoError(
File "/home/ryanguo99/repos/pytorch/torch/_dynamo/convert_frame.py", line 1082, in _compile
guarded_code = compile_inner(code, one_graph, hooks, transform)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/ryanguo99/repos/pytorch/torch/_utils_internal.py", line 97, in wrapper_function
return function(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/ryanguo99/repos/pytorch/torch/_dynamo/convert_frame.py", line 777, in compile_inner
return _compile_inner(code, one_graph, hooks, transform)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/ryanguo99/repos/pytorch/torch/_dynamo/convert_frame.py", line 922, in _compile_inner
check_fn = CheckFunctionManager(
^^^^^^^^^^^^^^^^^^^^^
File "/home/ryanguo99/repos/pytorch/torch/_dynamo/guards.py", line 2564, in __init__
builder, guard_manager = self.build_guards(
^^^^^^^^^^^^^^^^^^
File "/home/ryanguo99/repos/pytorch/torch/_dynamo/guards.py", line 2784, in build_guards
guard.create(builder)
File "/home/ryanguo99/repos/pytorch/torch/_guards.py", line 358, in create
return self.create_fn(builder, self)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/ryanguo99/repos/pytorch/torch/_dynamo/guards.py", line 1696, in EQUALS_MATCH
code = [f"{ref} == {val!r}"]
^^^^^^^^^^^^^^^^^^^
File "/home/ryanguo99/repos/pytorch/torch/distributed/device_mesh.py", line 629, in __repr__
if not self.mesh_dim_names
^^^^^^^^^^^^^^^^^^^
torch._dynamo.exc.InternalTorchDynamoError: AttributeError: 'DeviceMesh' object has no attribute 'mesh_dim_names'
To execute this test, run the following from the base repo dir:
PYTORCH_TEST_WITH_INDUCTOR=1 python test/distributed/tensor/debug/test_comm_mode.py TestCommMode.test_comm_mode_with_dtensor
```
### Versions
6e5e9dc321c, python 3.11
cc @chauhang @penguinwu @voznesenskym @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @chenyang78 @kadeng @amjames | true |
3,029,200,397 | [TorchDynamo] Fix failure to realize LazyVariableTracker on stack | Lucaskabela | closed | [
"module: dynamo",
"ciflow/inductor"
] | 3 | CONTRIBUTOR | Fixes a bug with install and inline for dynamo where lazy variable tracker was not realized before creating the `FakeRootModule` on [L1129](https://github.com/pytorch/pytorch/blob/a3123dd3ab936a59dc3219dd77785b71c9147019/torch/_dynamo/output_graph.py#L1129) in output_graph.py
As a result, when we try to access the graph attribute `bff`, the attribute was not found and would crash with an error like:
```
File "/data/users/lucaskabela/pytorch/torch/fx/graph_module.py", line 493, in __init__
_copy_attr(root, self, node.target)
File "/data/users/lucaskabela/pytorch/torch/fx/graph_module.py", line 238, in _copy_attr
orig = getattr(from_module, field)
^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/data/users/lucaskabela/pytorch/torch/nn/modules/module.py", line 1944, in __getattr__
raise AttributeError(
torch._dynamo.exc.InternalTorchDynamoError: AttributeError: 'FakeRootModule' object has no attribute 'L__self___bff'
```
If we realize the nested structures properly before (since they are on the [stack in L1118](https://github.com/pytorch/pytorch/blob/a3123dd3ab936a59dc3219dd77785b71c9147019/torch/_dynamo/output_graph.py#L1117)), this crash is resolved
### Test:
```
python test/export/test_export_with_inline_and_install.py InlineAndInstallStrictExportTestExport.test_constant_output_inline_and_install_strict
```
cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @chenyang78 @kadeng @chauhang @amjames | true |
3,029,188,296 | Use Literal for padding_mode in Conv2d for better type checking | sudiptap | closed | [
"oncall: distributed",
"module: cpu",
"open source",
"release notes: quantization",
"topic: not user facing",
"module: inductor",
"module: dynamo",
"release notes: distributed (checkpoint)",
"release notes: inductor (aoti)"
] | 6 | NONE | Fixes #152280 — use Literal for padding_mode to improve type checking.
cc @H-Huang @awgu @wanchaol @fegin @fduwjj @wz337 @wconstab @d4l3k @jgong5 @mingfeima @XiaobingSuper @sanchitintel @ashokei @jingxu10 @jerryzh168 @voznesenskym @penguinwu @EikanWang @Guobing-Chen @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @ipiszy @chenyang78 @kadeng @muchulee8 @amjames @chauhang @aakhundov | true |
3,029,183,923 | Log aot and idx waitcounters. | jovianjaison | closed | [
"fb-exported",
"Merged",
"Reverted",
"ciflow/trunk",
"module: inductor",
"module: dynamo",
"ciflow/inductor",
"release notes: AO frontend",
"ci-no-td"
] | 33 | CONTRIBUTOR | Summary:
Added for create_aot_dispatcher_function and compile_fx_inner.
Note:
Log wait counters flag is already set for:
1. async_compile.precompile
2. remote_fx_graph_cache_get
3. remote_fx_graph_cache_put
Test Plan: contbuild
Differential Revision: D73866124
cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @ipiszy @chenyang78 @kadeng @muchulee8 @amjames @chauhang @aakhundov | true |
3,029,167,566 | [MPS][BE] Delete unused lerp functors | malfet | closed | [
"Merged",
"Reverted",
"ciflow/trunk",
"topic: not user facing",
"release notes: mps",
"ciflow/mps",
"ci-no-td"
] | 10 | CONTRIBUTOR | Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom):
* #152504
* #152485
* #152479
* #152466
* __->__ #152443
For `lerp.Scalar_out` weight (aka alpha) is not an optional argument, so no point in having those specializations.
But move `alpha=1.0` ahead of dispatching to Metal shaders, as plain copy of tensor should still be faster https://github.com/pytorch/pytorch/blob/a1a4fee3b84b2d46f8c084f1aa9e6a45fa54b9c3/aten/src/ATen/native/mps/operations/BinaryOps.mm#L285-L290 | true |
3,029,160,787 | `torch.compile` causes assertion error in distributed checkpoint wrapper test | StrongerXi | open | [
"oncall: distributed",
"triaged",
"oncall: pt2"
] | 1 | CONTRIBUTOR | ### 🐛 Describe the bug
Repro:
```
PYTORCH_TEST_WITH_INDUCTOR=1 python test/distributed/fsdp/test_checkpoint_wrapper.py CheckpointWrapperTest.test_checkpoint_wrapper_args_kwargs
```
### Error logs
```
Traceback (most recent call last):
File "/home/ryanguo99/.conda/envs/comfyui/lib/python3.11/unittest/case.py", line 57, in testPartExecutor
yield
File "/home/ryanguo99/.conda/envs/comfyui/lib/python3.11/unittest/case.py", line 623, in run
self._callTestMethod(testMethod)
File "/home/ryanguo99/.conda/envs/comfyui/lib/python3.11/unittest/case.py", line 579, in _callTestMethod
if method() is not None:
^^^^^^^^
File "/home/ryanguo99/repos/pytorch/torch/testing/_internal/common_utils.py", line 3154, in wrapper
method(*args, **kwargs)
File "/home/ryanguo99/repos/pytorch/test/distributed/fsdp/test_checkpoint_wrapper.py", line 128, in test_checkpoint_wrapper_args_kwargs
torch.nn.Linear(1, 1),
File "/home/ryanguo99/repos/pytorch/test/distributed/fsdp/test_checkpoint_wrapper.py", line 127, in torch_dynamo_resume_in_test_checkpoint_wrapper_args_kwargs_at_128
m = checkpoint_wrapper(
File "/home/ryanguo99/repos/pytorch/torch/_dynamo/convert_frame.py", line 1458, in __call__
^^^^^^^^^
File "/home/ryanguo99/repos/pytorch/torch/_dynamo/convert_frame.py", line 1082, in _compile
guarded_code = compile_inner(code, one_graph, hooks, transform)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/ryanguo99/repos/pytorch/torch/_utils_internal.py", line 97, in wrapper_function
return function(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/ryanguo99/repos/pytorch/torch/_dynamo/convert_frame.py", line 777, in compile_inner
return _compile_inner(code, one_graph, hooks, transform)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/ryanguo99/repos/pytorch/torch/_dynamo/convert_frame.py", line 813, in _compile_inner
out_code = transform_code_object(code, transform)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/ryanguo99/repos/pytorch/torch/_dynamo/bytecode_transformation.py", line 1422, in transform_code_object
transformations(instructions, code_options)
File "/home/ryanguo99/repos/pytorch/torch/_dynamo/convert_frame.py", line 264, in _fn
return fn(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^
File "/home/ryanguo99/repos/pytorch/torch/_dynamo/convert_frame.py", line 741, in transform
tracer.run()
File "/home/ryanguo99/repos/pytorch/torch/_dynamo/symbolic_convert.py", line 3494, in run
super().run()
File "/home/ryanguo99/repos/pytorch/torch/_dynamo/symbolic_convert.py", line 1345, in run
while self.step():
^^^^^^^^^^^
File "/home/ryanguo99/repos/pytorch/torch/_dynamo/symbolic_convert.py", line 1254, in step
self.dispatch_table[inst.opcode](self, inst)
File "/home/ryanguo99/repos/pytorch/torch/_dynamo/symbolic_convert.py", line 825, in wrapper
return handle_graph_break(self, inst, speculation.reason)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/ryanguo99/repos/pytorch/torch/_dynamo/symbolic_convert.py", line 875, in handle_graph_break
self.output.compile_subgraph(self, reason=reason)
File "/home/ryanguo99/repos/pytorch/torch/_dynamo/output_graph.py", line 1239, in compile_subgraph
self.compile_and_call_fx_graph(tx, pass2.graph_output_vars(), root)
File "/home/ryanguo99/repos/pytorch/torch/_dynamo/output_graph.py", line 1500, in compile_and_call_fx_graph
compiled_fn = self.call_user_compiler(gm)
^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/ryanguo99/repos/pytorch/torch/_dynamo/output_graph.py", line 1552, in call_user_compiler
return self._call_user_compiler(gm)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/ryanguo99/repos/pytorch/torch/_dynamo/output_graph.py", line 1609, in _call_user_compiler
raise BackendCompilerFailed(
File "/home/ryanguo99/repos/pytorch/torch/_dynamo/output_graph.py", line 1584, in _call_user_compiler
compiled_fn = compiler_fn(gm, self.example_inputs())
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/ryanguo99/repos/pytorch/torch/_dynamo/repro/after_dynamo.py", line 150, in __call__
compiled_gm = compiler_fn(gm, example_inputs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/ryanguo99/repos/pytorch/torch/_dynamo/backends/inductor.py", line 23, in inductor
return compile_fx(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/ryanguo99/repos/pytorch/torch/_inductor/compile_fx.py", line 2279, in compile_fx
return aot_autograd(
^^^^^^^^^^^^^
File "/home/ryanguo99/repos/pytorch/torch/_dynamo/backends/common.py", line 106, in __call__
cg = aot_module_simplified(gm, example_inputs, **self.kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/ryanguo99/repos/pytorch/torch/_functorch/aot_autograd.py", line 1171, in aot_module_simplified
compiled_fn = AOTAutogradCache.load(
^^^^^^^^^^^^^^^^^^^^^^
File "/home/ryanguo99/repos/pytorch/torch/_functorch/_aot_autograd/autograd_cache.py", line 873, in load
compiled_fn = dispatch_and_compile()
^^^^^^^^^^^^^^^^^^^^^^
File "/home/ryanguo99/repos/pytorch/torch/_functorch/aot_autograd.py", line 1156, in dispatch_and_compile
compiled_fn, _ = create_aot_dispatcher_function(
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/ryanguo99/repos/pytorch/torch/_functorch/aot_autograd.py", line 576, in create_aot_dispatcher_function
return _create_aot_dispatcher_function(
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/ryanguo99/repos/pytorch/torch/_functorch/aot_autograd.py", line 677, in _create_aot_dispatcher_function
fw_metadata = run_functionalized_fw_and_collect_metadata(
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/ryanguo99/repos/pytorch/torch/_functorch/_aot_autograd/collect_metadata_analysis.py", line 198, in inner
flat_f_outs = f(*flat_f_args)
^^^^^^^^^^^^^^^
File "/home/ryanguo99/repos/pytorch/torch/_functorch/_aot_autograd/traced_function_transforms.py", line 899, in functional_call
out = PropagateUnbackedSymInts(mod).run(
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/ryanguo99/repos/pytorch/torch/fx/interpreter.py", line 171, in run
self.env[node] = self.run_node(node)
^^^^^^^^^^^^^^^^^^^
File "/home/ryanguo99/repos/pytorch/torch/fx/experimental/symbolic_shapes.py", line 7404, in run_node
result = super().run_node(n)
^^^^^^^^^^^^^^^^^^^
File "/home/ryanguo99/repos/pytorch/torch/fx/interpreter.py", line 240, in run_node
return getattr(self, n.op)(n.target, args, kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/ryanguo99/repos/pytorch/torch/fx/interpreter.py", line 320, in call_function
return target(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^
File "/home/ryanguo99/repos/pytorch/torch/_higher_order_ops/wrap.py", line 236, in __call__
return checkpoint(Interpreter(gmod).run, *args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/ryanguo99/repos/pytorch/torch/_compile.py", line 51, in inner
return disable_fn(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/ryanguo99/repos/pytorch/torch/_dynamo/eval_frame.py", line 856, in _fn
return fn(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^
File "/home/ryanguo99/repos/pytorch/torch/utils/checkpoint.py", line 494, in checkpoint
next(gen)
File "/home/ryanguo99/repos/pytorch/torch/utils/checkpoint.py", line 1479, in _checkpoint_without_reentrant_generator
isinstance(forward_context, TorchDispatchMode) and
torch._dynamo.exc.BackendCompilerFailed: backend='inductor' raised:
AssertionError: In torch.compile mode, `context_fn` arg passed to `torch.utils.checkpoint` must generate a tuple of two `TorchDispatchMode`s.
While executing %tag_activation_checkpoint : [num_users=1] = call_function[target=torch.ops.higher_order.tag_activation_checkpoint](args = (%wrap_body_0, %randn, %l_stack0_modules_checkpoint_wrapped_module_parameters_weight_, %l_stack0_modules_checkpoint_wrapped_module_parameters_bias_), kwargs = {use_reentrant: False})
GraphModule: class GraphModule(torch.nn.Module):
def forward(self, L_stack0_modules_checkpoint_wrapped_module_parameters_weight_: "f32[1, 1][1, 1]", L_stack0_modules_checkpoint_wrapped_module_parameters_bias_: "f32[1][1]"):
l_stack0_modules_checkpoint_wrapped_module_parameters_weight_ = L_stack0_modules_checkpoint_wrapped_module_parameters_weight_
l_stack0_modules_checkpoint_wrapped_module_parameters_bias_ = L_stack0_modules_checkpoint_wrapped_module_parameters_bias_
# File: /home/ryanguo99/repos/pytorch/test/distributed/fsdp/test_checkpoint_wrapper.py:133 in torch_dynamo_resume_in_test_checkpoint_wrapper_args_kwargs_at_127, code: m(torch.randn(2, 1)).sum().backward()
randn: "f32[2, 1][1, 1]" = torch.randn(2, 1)
# File: /home/ryanguo99/repos/pytorch/torch/distributed/algorithms/_checkpoint/checkpoint_wrapper.py:171 in forward, code: return self.checkpoint_fn( # type: ignore[misc]
wrap_body_0 = self.wrap_body_0
tag_activation_checkpoint = torch.ops.higher_order.tag_activation_checkpoint(wrap_body_0, randn, l_stack0_modules_checkpoint_wrapped_module_parameters_weight_, l_stack0_modules_checkpoint_wrapped_module_parameters_bias_, use_reentrant = False); wrap_body_0 = randn = l_stack0_modules_checkpoint_wrapped_module_parameters_weight_ = l_stack0_modules_checkpoint_wrapped_module_parameters_bias_ = None
getitem: "f32[2, 1][1, 1]" = tag_activation_checkpoint[0]; tag_activation_checkpoint = None
# File: /home/ryanguo99/repos/pytorch/test/distributed/fsdp/test_checkpoint_wrapper.py:133 in torch_dynamo_resume_in_test_checkpoint_wrapper_args_kwargs_at_127, code: m(torch.randn(2, 1)).sum().backward()
sum_1: "f32[][]" = getitem.sum(); getitem = None
return (sum_1,)
class wrap_body_0(torch.nn.Module):
def forward(self, randn: "f32[2, 1][1, 1]", l_stack0_modules_checkpoint_wrapped_module_parameters_weight_: "f32[1, 1][1, 1]", l_stack0_modules_checkpoint_wrapped_module_parameters_bias_: "f32[1][1]"):
# File: /home/ryanguo99/repos/pytorch/torch/distributed/algorithms/_checkpoint/checkpoint_wrapper.py:171 in forward, code: return self.checkpoint_fn( # type: ignore[misc]
linear: "f32[2, 1][1, 1]" = torch._C._nn.linear(randn, l_stack0_modules_checkpoint_wrapped_module_parameters_weight_, l_stack0_modules_checkpoint_wrapped_module_parameters_bias_); randn = l_stack0_modules_checkpoint_wrapped_module_parameters_weight_ = l_stack0_modules_checkpoint_wrapped_module_parameters_bias_ = None
return (linear,)
Original traceback:
File "/home/ryanguo99/repos/pytorch/test/distributed/fsdp/test_checkpoint_wrapper.py", line 133, in torch_dynamo_resume_in_test_checkpoint_wrapper_args_kwargs_at_127
m(torch.randn(2, 1)).sum().backward()
File "/home/ryanguo99/repos/pytorch/torch/distributed/algorithms/_checkpoint/checkpoint_wrapper.py", line 171, in forward
return self.checkpoint_fn( # type: ignore[misc]
Set TORCHDYNAMO_VERBOSE=1 for the internal stack trace (please do this especially if you're reporting a bug to PyTorch). For even more developer context, set TORCH_LOGS="+dynamo"
To execute this test, run the following from the base repo dir:
PYTORCH_TEST_WITH_INDUCTOR=1 python test/distributed/fsdp/test_checkpoint_wrapper.py CheckpointWrapperTest.test_checkpoint_wrapper_args_kwargs
```
### Versions
6e5e9dc321c, Python 3.11
cc @H-Huang @awgu @wanchaol @fegin @fduwjj @wz337 @wconstab @d4l3k @chauhang @penguinwu | true |
3,029,140,184 | Inductor pattern matching on mutable ops | yf225 | open | [
"triaged",
"module: functionalization",
"oncall: pt2",
"module: inductor",
"module: pt2-dispatcher"
] | 0 | CONTRIBUTOR | ### 🚀 The feature, motivation and pitch
Pattern matching on mutable ops (e.g. to replace it with custom kernels) is a common usage that shows up in libraries such as vLLM. We'd like to improve Inductor pattern matcher's support for it.
This is one use case where both `x` and `out` are non-view tensors:
```python
import torch
from torch.library import register_fake
from torch._inductor.pattern_matcher import register_replacement, fwd_only, PatternMatcherPass
@torch.library.custom_op("mylib::foo_inplace", mutates_args={"x"})
def foo_inplace(x: torch.Tensor) -> None:
x.add_(1)
# NOTE: only returning None is supported; the custom op cannot return `out`.
@torch.library.custom_op("mylib::bar", mutates_args={"out"})
def bar_out(x: torch.Tensor, out: torch.Tensor) -> None:
out.copy_(x + 2)
@register_fake("mylib::bar")
def bar_out_fake(x: torch.Tensor, out: torch.Tensor) -> None:
return None
@torch.library.custom_op("mylib::foobar_out", mutates_args={"out"})
def foobar_out(x: torch.Tensor, out: torch.Tensor) -> torch.Tensor:
x.add_(1)
out.copy_(x + 2)
return out
def pattern(x, out):
foo_inplace(x)
bar_out(x, out)
return out
def replacement(x, out):
return foobar_out(x, out)
patterns = PatternMatcherPass()
register_replacement(
search_fn=pattern,
replace_fn=replacement,
example_inputs=(torch.randn(3), torch.randn(3)),
trace_fn=fwd_only,
pass_dicts=patterns,
)
# user-function
@torch.compile(fullgraph=True)
def f(x):
x = x.clone()
out = torch.empty_like(x)
foo_inplace(x)
bar_out(x, out)
return out
x = torch.randn(3, device="cpu")
f(x)
```
which currently gives error:
```python
Traceback (most recent call last):
File "/home/local/pytorch/test/test_mutable_op_pattern_match.py", line 39, in <module>
register_replacement(
File "/home/local/miniconda3/lib/python3.12/site-packages/torch/_inductor/pattern_matcher.py", line 1563, in register_replacement
pattern.register(pass_dicts)
File "/home/local/miniconda3/lib/python3.12/site-packages/torch/_inductor/pattern_matcher.py", line 1079, in register
assert hasattr(self.pattern, "fns")
AssertionError
```
More generally, there are 4 use cases we want to support:
- [ ] `x`: non-view, `out`: non-view
- [ ] `x`: view, `out`: non-view
- [ ] `x`: non-view, `out`: view
- [ ] `x`: view, `out`: view
For view cases, we'd like to explore the `opaque_view` idea.
cc. @zou3519 @eellison
### Alternatives
_No response_
### Additional context
_No response_
cc @bdhirsh @ezyang @chauhang @penguinwu @voznesenskym @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @ipiszy @chenyang78 @kadeng @muchulee8 @amjames @aakhundov @zou3519 | true |
3,029,077,321 | Add two missing JIT tests to CMake | swolchok | closed | [
"oncall: jit",
"Merged",
"ciflow/trunk",
"topic: not user facing"
] | 3 | CONTRIBUTOR | Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom):
* __->__ #152440
Looks like I forgot to add these.
cc @EikanWang @jgong5 @wenzhe-nrv @sanchitintel | true |
3,029,071,561 | Newly added lint-urls jobs are very flaky | malfet | open | [
"module: ci",
"triaged",
"module: flaky-tests",
"module: regression"
] | 10 | CONTRIBUTOR | ### 🐛 Describe the bug
May be it's just me, but newly added lint jobs keeps intermittently failing on PRs/trunk, for example
https://github.com/pytorch/pytorch/actions/runs/14737078933/job/41365789268
### Versions
CI
cc @seemethere @pytorch/pytorch-dev-infra @clee2000 | true |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.