-
Notifications
You must be signed in to change notification settings - Fork 177
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
fix: enable compilation with ONNX 1.13 #1835
Conversation
Let me tag @benjaminhuth and @Corentin-Allaire. Thoughts? |
Very good to stay compatible with newer ONNX versions. However, I see two points before this gets merged:
|
I have done those exact same change (plus some other) in my own branch. I general I think we should discuss how Onnx is implemented in Acts right now (with the current implementation you need to create a new class that inherit from OnnxRuntimeBase each time you import a new network but we can discuss this once I open a PR for the ambiguity resolver |
Good point about Exa.Trk, tagging @xju2 -- do you know if your pipeline works with OnnxRuntime >= 1.12.0? |
@Corentin-Allaire Do you mean that in your ambi. solver work, you rely on changing the current implementation? If so, maybe it'd be worth it to merge that part in sooner (or start the discussion as you suggest) so as to minimize potential divergence? just my .02$ |
@gagnonlg My main change for now but it is not definitive was to switch |
I see, seems like that would be a backwards compatible change if that decision is made so there's no hurry, I guess
Oh, it's literally "my 2 cents", an idiom that basically means "in my opinion", i.e. feel free to disagree |
I got the 2 cents afterward, I knew the expression but it this the first I see it written that way ! |
Codecov Report
@@ Coverage Diff @@
## main #1835 +/- ##
=======================================
Coverage 49.50% 49.50%
=======================================
Files 407 407
Lines 22646 22646
Branches 10334 10334
=======================================
Hits 11210 11210
Misses 4247 4247
Partials 7189 7189 📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more |
📊 Physics performance monitoring for a63f605Full report VertexingSeedingCKFAmbiguity resolutionTruth tracking (Kalman Filter)Truth tracking (GSF) |
We have a new image now, if you replace |
@benjaminhuth awesome, let's try it now |
@benjaminhuth are you sure it's v31? Looks like OnnxRuntime 1.10 is still being picked up |
Yeah I also noticed it. I'll check what I messed up |
Ah I mixed the numbers. It is release 34, not 31. Sorry! |
No worries, it's an easy fix! |
Hmm this time the error is
Do we need to reduce somehow the size of the image @paulgessinger ? |
No I guess the disk is simply full on the VM. In principle I clear the image storage regularly but you can try rerunning which hopefully ends up on another VM. |
Hi @gagnonlg , the ONNX models were not working because some operations were not supported by that time. Maybe the new version will fix those. ;-) |
@benjaminhuth from the ci-bridge:
|
Nice that it builds without errors! The test workflow still runs on the |
@benjaminhuth of course, I need to bump the test image as well 🤡 |
All fine now, I think this can go in. @benjaminhuth can you approve? |
Simple PR to fix a compilation error against ONNX runtime 1.13.1. Basically, the `Ort::Session::Get{Input,Output}Name` functions got removed starting from 1.13, and we have the use the `Ort::Session::Get{Input,Output}NameAllocated` versions which were introduced in 1.12. One implication is that this PR introduces a lower bound of 1.12.0 on the ONNX runtime version.
Simple PR to fix a compilation error against ONNX runtime 1.13.1. Basically, the
Ort::Session::Get{Input,Output}Name
functions got removed starting from 1.13, and we have the use theOrt::Session::Get{Input,Output}NameAllocated
versions which were introduced in 1.12.One implication is that this PR introduces a lower bound of 1.12.0 on the ONNX runtime version.