[Superseded by #46523] Add GloParT inference facility #45830
Conversation
|
cms-bot internal usage |
|
+code-checks Logs: https://cmssdt.cern.ch/SDT/code-checks/cms-sw-PR-45830/41572 |
|
A new Pull Request was created by @colizz for master. It involves the following packages:
@cmsbuild, @ftorrresd, @hqucms, @jfernan2, @mandrenguyen, @vlimant can you please review it and eventually sign? Thanks. cms-bot commands are listed here |
|
enable nano |
|
test parameters: |
|
please test |
|
+1 Size: This PR adds an extra 60KB to repository The following merge commits were also included on top of IB + this PR after doing git cms-merge-topic: You can see more details here: Comparison SummarySummary:
NANO Comparison SummarySummary:
Nano size comparison Summary:
|
|
Hi all, please see below a security check on the inferred tagger scores (using the GloParT 2 model). They are consistent with the score predicted by Weaver. |
|
enable profiling |
|
please test |
|
+1 Size: This PR adds an extra 12KB to repository
Comparison SummarySummary:
|
|
+code-checks Logs: https://cmssdt.cern.ch/SDT/code-checks/cms-sw-PR-45830/41810 |
|
Pull request #45830 was updated. @cmsbuild, @ftorrresd, @hqucms, @jfernan2, @mandrenguyen can you please check and sign again. |
|
Hi @jfernan2 @tvami, many thanks! This PR requires us to finalize the upload of the model in cms-data/RecoBTag-Combined#61. XPOG has suggested that we upload our current "beta1" model (instead of a dummy model) so that performance can be tested. We've now done this and updated both PRs accordingly. Thanks very much for your test! |
|
please abort |
|
please test |
|
+1 Size: This PR adds an extra 24KB to repository The following merge commits were also included on top of IB + this PR after doing git cms-merge-topic:
You can see more details here: Comparison SummarySummary:
|
|
enable profiling,nano |
|
please test |
|
+1 Size: This PR adds an extra 12KB to repository The following merge commits were also included on top of IB + this PR after doing git cms-merge-topic:
You can see more details here: Comparison SummarySummary:
NANO Comparison SummarySummary:
Nano size comparison Summary:
|
colizz
changed the title
PR description:
This PR adds the GloParT 3 model's inference utility into CMSSW, prepared for NanoAODv15. The topic was discussed in the recent JMAR meeting [slides]
Global Particle Transformer (GloParT) 3 is an inclusive tagging model for AK8 jets that covers the entire phase space and enables resonance mass regression for each jet class. It functions as both a global tagger and a mass regression model for AK8 jets and can also be utilized as a pre-trained model. The hidden layer neurons (with dimension 256) are stored in MiniAOD, providing the capability to resume all output scores. Further details can be found in the slides.
In summary, the AK8 jet features added to MiniAOD and NanoAOD are illustrated below.
Please test this PR with cms-data/RecoBTag-Combined#61.
Note: the current model in cms-data/RecoBTag-Combined#61 is a placeholder for incorporating this PR. During our development history, GloParT 2 has already been validated to function properly within CMSSW (using this branch: https://github.com/colizz/cmssw/tree/dev-130X-glopart ), although it has not been officially integrated. We are now upgrading GloParT from version 2 to version 3 (details in slides), and the final GloParT 3 model will be ready within 2-3 weeks. We are suggested to make this infrastructure PR to cmssw first.
Adding @qliphy @cmantill @rkansal47 @jmduarte
PR validation:
The PR passes the unit test
test_globalpart_cfg.py. It is also validated under the workflow AODSIM->MINIAOD->NANOAOD.