Skip to content

✨[Feature] Performance optimization of PyTorch + TRT subgraphs #3277

New issue

Have a question about this project? # for a free GitHub account to open an issue and contact its maintainers and the community.

#

By clicking “#”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? # to your account

Jump to bottom
Closed
Tracked by #3274
keehyuna opened this issue Nov 4, 2024 · 0 comments · Fixed by #3270
Closed
Tracked by #3274

✨[Feature] Performance optimization of PyTorch + TRT subgraphs #3277

keehyuna opened this issue Nov 4, 2024 · 0 comments · Fixed by #3270
Assignees
@keehyuna
Labels
feature request New feature or request

Comments

@keehyuna
Copy link
Collaborator

keehyuna commented Nov 4, 2024

Is your feature request related to a problem? Please describe.

If there is graph breaks due to unsupported ops, overhead between TRT and torch module is observed.

Describe the solution you'd like

Entire subgraphs are capture/replayed by cuda graphs in wrapper runtime module

Describe alternatives you've considered

cuda graph can be applied to torch and trt module individually but it's not ideal to reduce cpu overhead.

Additional context
# for free to join this conversation on GitHub. Already have an account? # to comment
Assignees

@keehyuna keehyuna

Labels
feature request New feature or request
Projects
None yet
Milestone
No milestone
Development

Successfully merging a pull request may close this issue.

Wrapper module around TRT + pytorch subgraphs keehyuna/TensorRT
2 participants
@narendasan @keehyuna