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

[TensorRT EP] revert oss parser to 10.7-GA #23710

Open
wants to merge 11 commits into
base: main
Choose a base branch
from
Open

[TensorRT EP] revert oss parser to 10.7-GA #23710

wants to merge 11 commits into from

Conversation

yf711
Copy link
Contributor

@yf711 yf711 commented Feb 14, 2025

Description

  • Revert oss parser version as 10.8-GA rely on onnx 1.18, which is not available yet.

Motivation and Context

  • OSS 10.8-GA can't work with ONNX 1.17, which will cause ORT build failure when enabling --use_tensorrt_oss_parser

  • Action needed to adapt latest 10.7-GA or 10.8-GA oss parser branch to fix scatterND attribute issue

@yf711 yf711 requested a review from chilo-ms February 14, 2025 23:14
@chilo-ms
Copy link
Contributor

The TRT CI has building issue.

@yf711
Copy link
Contributor Author

yf711 commented Feb 18, 2025

onnx-tensorrt introduced a dependency plugin.h, which is available by tarball/zip installation (underTensorRT-tarball/python/include/impl/) but not available by linux package installation. OSS parser users who use Windows/Linux tarball installation are not affected.
Updating the logic to make sure builtin parser users are not blockedThis logic is not needed, as onnx-tensorrt is able to locate the dependency by itself

@jywu-msft
Copy link
Member

this is a bit confusing. if we publish packages built against TRT 10.8 but stay with10.7 parser?

@chilo-ms
Copy link
Contributor

chilo-ms commented Feb 19, 2025

onnx-tensorrt introduced a dependency plugin.h, which is available by tarball/zip installation (underTensorRT-tarball/python/include/impl/) but not available by linux package installation. OSS parser users who use Windows/Linux tarball installation are not affected.
Updating the logic to make sure builtin parser users are not blocked

Why previously when our CI was running with TRT 10.6 built-in parser didn't complain about this plugin.h depedency?

@yf711
Copy link
Contributor Author

yf711 commented Feb 19, 2025

onnx-tensorrt introduced a dependency plugin.h, which is available by tarball/zip installation (underTensorRT-tarball/python/include/impl/) but not available by linux package installation. OSS parser users who use Windows/Linux tarball installation are not affected.
Updating the logic to make sure builtin parser users are not blocked

Why previously when our CI was running with TRT 10.6 built-in parser didn't complain about this plugin.h depedency?

This dependency is only needed by oss parser. CIs that use Bulit-in parser have no issue even though there's no plugin.h

@chilo-ms
Copy link
Contributor

Updating the logic to make sure builtin parser users are not blocked

You mention "Updating the logic to make sure builtin parser users are not blocked", it should be oss parser, right?

endif()

# Check TRT version >= requirement
function(check_trt_version req_major req_minor req_patch out_var)
Copy link
Contributor

@chilo-ms chilo-ms Feb 19, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This function is much complex than the original check.
In additions to check 10.0.1, do we need to check other TRT version?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

So far it's not needed. I will clean this

@yf711
Copy link
Contributor Author

yf711 commented Feb 19, 2025

Updating the logic to make sure builtin parser users are not blocked

You mention "Updating the logic to make sure builtin parser users are not blocked", it should be oss parser, right?

I didn't mention that clearly.
The original code I added (search plugin.h without if check) was a blocker. This part is now removed, as it's not needed.

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