Disable remote caching when calling compile_fx

[zou3519]

The problem is as follows:

• vLLM requires its monkeypatched functions to run (e.g.

  vllm/vllm/compilation/compiler_interface.py

  Line 251 in 7b5ecf7

  def hijack_compiled_fx_graph_hash(*args, **kwargs):

  )
• These functions may not run if (1) a user has torch.compile remote cache set up and (2) there is a remote cache hit.
• When the monkeypatched/hijacked functions fail to run, we get some assertions:

  vllm/vllm/compilation/compiler_interface.py

  Lines 299 to 302 in 7b5ecf7

  	assert hash_str is not None, (
  	"failed to get the hash of the compiled graph")
  	assert file_path is not None, (
  	"failed to get the file path of the compiled graph")

This PR disables torch.compile remote caching for vLLM compile.

Test Plan:

• tested locally with vllm serve "meta-llama/Llama-4-Scout-17B-16E-Instruct" -tp 8 --max_ model_len 1000 --override-generation-config='{"attn_temperature_tuning": true}'

[github-actions]

👋 Hi! Thank you for contributing to the vLLM project.

💬 Join our developer Slack at https://slack.vllm.ai to discuss your PR in #pr-reviews, coordinate on features in #feat- channels, or join special interest groups in #sig- channels.

Just a reminder: PRs would not trigger full CI run by default. Instead, it would only run fastcheck CI which starts running only a small and essential subset of CI tests to quickly catch errors. You can run other CI tests on top of those by going to your fastcheck build on Buildkite UI (linked in the PR checks section) and unblock them. If you do not have permission to unblock, ping simon-mo or khluu to add you in our Buildkite org.

Once the PR is approved and ready to go, your PR reviewer(s) can run CI to test the changes comprehensively before merging.

To run CI, PR reviewers can either: Add ready label to the PR or enable auto-merge.

🚀

[houseroad]

Thanks for the fix, let me play around locally.

[zou3519]

This PR is mostly to fix a meta-internal bug I noticed (where we do have the torch.compile remote caches on), but I think this is generally applicable to vLLM so we should ship it

[houseroad]

Looks good.