build: manually update PyTorch version

[vivekkhandelwal1]

Set PyTorch and TorchVision version to nightly release 2024-05-14.

Signed-Off By: Vivek Khandelwal vivekkhandelwal1424@gmail.com

[vivekkhandelwal1]

Hi @stellaraccident, the Roll PyTorch has been blocked on the below error since last week:

python exception: Failure while executing pass pipeline:
error: "/home/azureuser/work/torch-mlir-vivek/test/python/fx_importer/sparse_test.py":539:0: failed to legalize operation 'torch.operator'
note: "/home/azureuser/work/torch-mlir-vivek/test/python/fx_importer/sparse_test.py":539:0: see current operation: %2 = "torch.operator"(%1, %0, %0, %0) <{name = "torch.aten.to_sparse"}> : (!torch.vtensor<[2,2,2],f32>, !torch.none, !torch.none, !torch.none) -> !torch.vtensor<[2,2,2],f32>

For Torch-MLIR developers, the error can be reproduced with:
$ torch-mlir-opt -pass-pipeline='builtin.module(func.func(torch-decompose-complex-ops),torch-backend-to-linalg-on-tensors-backend-pipeline)' /tmp/UnnammedModule.mlir
Add '-mlir-print-ir-after-all -mlir-disable-threading' to get the IR dump for debugging purpose.

It seems that we need to add Torch Lowering for torch.aten.to_sparse. I don't think any such lowering exists before in the project. Can we support this? If yes, some advice on it would be helpful.

[stellaraccident]

Just create a PR to skip the test, and tag Aart on it. Either he can fix it quickly or we land it.

[vivekkhandelwal1]

Hi @stellaraccident, I have modified the failing test. But the issue is that the "Fixed Test for Nightly" is now failing for stable PyTorch.

@aartbik, do you have something in mind for this?

[vivekkhandelwal1]

Corresponding issue #3418