Bump PyTorch pin to 2025-03-25

[swolchok]

Found release from https://hud2.pytorch.org/hud/pytorch/pytorch/nightly/0 .

[swolchok]

Stack from ghstack (oldest at bottom):

• Add optimized ELU implementation #9521
• -> Bump PyTorch pin to 2025-03-25 #9578

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/9578

• 📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

⏳ 1 Pending, 1 Unrelated Failure

As of commit e221434 with merge base c70c208 ():

UNSTABLE - The following job is marked as unstable, possibly due to flakiness on trunk:

• trunk / test-qnn-model (fp32, mb) / linux-job (gh) (#9878)
  RuntimeError: Command docker exec -t 66ae97aab271c8e21b947287ce734ce0550f86ba4938b409416170751b9ed439 /exec failed with exit code 1

This comment was automatically generated by Dr. CI and updates every 15 minutes.

[mergennachin]

Can we do this after our branch cut on Wednesday 6pm?

The reason is that 3-12 is the RC candidate for PT2.7 - anything after that is in the PT2.8

Otherwise, we would have to go back and forth, instead of monotonically increasing the nightly.

[mergennachin]

Unfortunately, this is the downside of delaying the branch cut and diverging from pytorch timeline

[swolchok]

@mergennachin the branch is cut now, right?

[mergennachin]

Thank you for your patience, we can move forward now that its post branch cut

[mergennachin]

cc @metascroy

[swolchok]

looks like unittest-arm failure are new, CC @digantdesai

[swolchok]

resnapshotted the ARM test. we'll see if it builds.

[swolchok]

pinged @digantdesai for advice on how to proceed with ARM failures

[zingo]

We have looked into this and have some fixes in the Arm delegate in the works to handle this, hopefully it will be ready tomorrow.

[zingo]

Fix that hopefully handled the Arm backend problems merged to main e.g. #9834

[swolchok]

Fix that hopefully handled the Arm backend problems merged to main e.g. #9834

@zingo still one failure on unittest-arm:

FAILED backends/arm/test/models/test_torch_functions.py::test_torch_fns_BI[t] - RuntimeError: Unknown operator <EdgeOpOverload: quantized_decomposed.quantize_per_tensor.default>: schema = quantized_decomposed::quantize_per_tensor(Tensor input, float scale, int zero_point, int quant_min, int quant_max, ScalarType dtype) -> Tensor for TOSA : TOSA-0.80+BI

[zingo]

Fix that hopefully handled the Arm backend problems merged to main e.g. #9834

@zingo still one failure on unittest-arm:

FAILED backends/arm/test/models/test_torch_functions.py::test_torch_fns_BI[t] - RuntimeError: Unknown operator <EdgeOpOverload: quantized_decomposed.quantize_per_tensor.default>: schema = quantized_decomposed::quantize_per_tensor(Tensor input, float scale, int zero_point, int quant_min, int quant_max, ScalarType dtype) -> Tensor for TOSA : TOSA-0.80+BI

Sorry we missed that, for some reason that one didn't give us problem in the internal testing. There is a new fix cooking that hopefully be uploaded tomorrow.

[zingo]

Just merged #9898

[swolchok]

test-qnn-model is failing on main, test-llava-runner-linux failure looks like a flake, retrying