Update PyTorch pin

[anmyachev]

No description provided.

[anmyachev]

This will be blocked until vllm-xpu-kernels is switched to C++20:

• CI failure: https://github.com/intel/intel-xpu-backend-for-triton/actions/runs/30745366090/job/91489923118?pr=7622#step:13:2415:
• Most likely solution: cmake: mandate c++20 for xe kernels vllm-project/vllm-xpu-kernels#495

@quinnlp please ensure that your vLLM update includes this change and all other related changes.

[quinnlp]

@quinnlp please ensure that your vLLM update includes this change and all other related changes.

Is it ok to wait for vllm-project/vllm-xpu-kernels#495 to be merged, then update the vllm-xpu-kernels pin?

[quinnlp]

Is it ok to wait for vllm-project/vllm-xpu-kernels#495 to be merged, then update the vllm-xpu-kernels pin?

vllm-project/vllm-xpu-kernels#495 was merged in vllm-project/vllm-xpu-kernels@5085fdd. #7539 updates the vllm-xpu-kernels pin to the hash.