Skip to content
New issue

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

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

Already on GitHub? Sign in to your account

Run subgraphs with model inputs #527

Closed

Conversation

ericcraw
Copy link

@ericcraw ericcraw commented Jan 6, 2025

Description

Run subgraphs with model inputs.

Motivation and Context

Subgraphs with model inputs were omitted for an unknown reason. Including them improves performance for key use cases.

@preetha-intel
Copy link

LGTM

@sfatimar
Copy link

sfatimar commented Jan 7, 2025

@ericcraw can you test this issue with multiple ORT Optimization levels @vthaniel can we run a perf regression test with this fix.

@ericcraw
Copy link
Author

ericcraw commented Jan 7, 2025

@ericcraw can you test this issue with multiple ORT Optimization levels @vthaniel can we run a perf regression test with this fix.

The currently supported usage of OVEP is to run with optimization disabled. General question: why would we test/support any other optimization level? We should be clear on what configuration is supported and stick to it no?

Regardless, I have run with highest optimization for this particular use case with model proxies, and it runs with slightly lower performance but is seemingly accurate (as determined by just eyeballing the output).

@sfatimar
Copy link

sfatimar commented Jan 8, 2025

@ericcraw can you test this issue with multiple ORT Optimization levels @vthaniel can we run a perf regression test with this fix.

The currently supported usage of OVEP is to run with optimization disabled. General question: why would we test/support any other optimization level? We should be clear on what configuration is supported and stick to it no?
MSFT has requested graphs to be enabled for all optimization levels
Regardless, I have run with highest optimization for this particular use case with model proxies, and it runs with slightly lower performance but is seemingly accurate (as determined by just eyeballing the output).

@ericcraw
Copy link
Author

ericcraw commented Jan 9, 2025

@sfatimar/@vthaniel is there an ETA for merge? or anything to be done to accelerate this?

@sfatimar
Copy link

@sfatimar/@vthaniel is there an ETA for merge? or anything to be done to accelerate this?

Please track this through the new Pull Request #532

@sfatimar sfatimar closed this Jan 10, 2025
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants