-
Notifications
You must be signed in to change notification settings - Fork 482
[NN Clusterizer] CCDB fetching within reco workflow #14841
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
Conversation
|
REQUEST FOR PRODUCTION RELEASES: This will add The following labels are available |
|
@davidrohr
|
Please consider the following formatting changes to AliceO2Group#14841
|
Please, no intermediate files. |
|
For 1.: I do not think this will cause a problem now, but I do not really like it, since it can have side effects. But this should be easy to fix. Let me think a bit. Perhaps we can even move the ReadConfigurableParam to the constructor, to call it only once. I will have a look when I am back. Otherwise, we can also access the settings directly from a local instance.
One point we have to think about though is: Do we want to allow to change the network during the processing? I.e., what happens when the time range of the CCDB object does not cover the entire run. In that case, we will receive a new CCDB object at some point while we are processing. That also means that the old CCDB object might be gone now, so we must no longer access its memory. Finally, GPU reconstruction supports processing 2 time frames at the same time to hide GPU transfer delays. For this reason, I create a copy of all the calib objects, so when the second TF arrive, I still have the old object for the first TF. This is a bit hacky. And if we do not need to support changing the NN objects during a run, we do not need this functionality for the NN objectrs. |
|
The new version now uses the internal char* buffer to load the model. Loading is successful in the clusterization task, however it fails at some internal stage within ONNX runtime (although model loading from file works and the buffer and size that I pass are correct). I am not sure yet what the problem is. Investigating. Otherwise, is this solution feasible. Dumping to file is implemented and can be switched on or off on demand. |
Please consider the following formatting changes to AliceO2Group#14841
|
Error while checking build/O2/fullCI_slc9 for 4fed621 at 2025-11-25 01:09: Full log here. |
Please consider the following formatting changes to AliceO2Group#14841
|
Open points:
|
|
Error while checking build/O2/fullCI_slc9 for 979a8d5 at 2025-11-27 03:50: Full log here. |
This PR removes the previous CCDB fetching and adds fetching during initialisation in GPUWorkflowSpec.cxx