Proposal to improve performance

No response

Report of performance regression

According to the command in docs/source/user-guide/pd-disaggregation/1p1d.md, it used --enforce-eager to disable CUDA graph. I found the performance significantly regressed with --enforce-eager.
And when I remove --enforce-eager option, UCM is speed up and everything is correct, so why the command in the 1p1d document disabled CUDA graph?

Misc discussion on performance

No response

Your current environment (if you think it is necessary)

The output of `python collect_env.py`