Can not use torch.compile with dynamic=Ture when using multi-threads

[CaoE]

🐛 Describe the bug

When running a model processed by torch.compile in a multi-threaded environment, the following error will be reported:
RuntimeError: Detected that you are using FX to symbolically trace a dynamo-optimized function. This is not supported at the moment.
There is a case where the same model is run in multiple threads:

model = torch.compile(model, dynamic=True)

thread = threading.Thread(
    target=run_sharing_model,
    args=(model, dataloader, ...)
)

Is this error expected and is it possible to support this case?

Error logs

RuntimeError: Detected that you are using FX to symbolically trace a dynamo-optimized function. This is not supported at the moment.

Minified repro

No response

Versions

torch 2.4.0.dev20240417+cpu

cc @ezyang @msaroufim @bdhirsh @anijain2305 @chauhang

[ezyang]

Probably same root cause as #118387