[NVFuser]Benchmark minor update (#46778)
Summary:
This is a tiny PR for two minor fixes:
1. Added `torch._C._jit_set_texpr_fuser_enabled(False)` to enable shape inference on nv fuser runs.
2. Renamed dynamic benchmark module names to avoid multiple matching. i.e. `simple_element` with `dynamic_simple_element`. I guess it'd be much easier if the pattern matching was based on `startswith`. Would be happy to update that if agreed.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/46778
Reviewed By: zhangguanheng66
Differential Revision: D24516911
Pulled By: bertmaher
fbshipit-source-id: 839f9a3e058f9d7aca17b2e6eb8b558e0e48e8f4
