Avoid using PyTorch fork for FlexAttention benchmark

[anmyachev]

intel-xpu-backend-for-triton/.github/workflows/triton-benchmarks.yml

Line 277 in 423d734

# Install Pytorch with FlexAttention XPU support enabled

We need to start merging changes from pytorch/pytorch#143553 into main PyTorch.

[anmyachev]

First: pytorch/pytorch#151012

[vlad-penkin]

@anmyachev does this issue duplicate:

• [Flex Attention] Remove forked pytorch usage from benchmarks workflow #3887

?

[anmyachev]

@anmyachev does this issue duplicate:

• [Flex Attention] Remove forked pytorch usage from benchmarks workflow #3887

?

Oops, yes. Feel free to close this issue.