-
Notifications
You must be signed in to change notification settings - Fork 4.3k
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
Adding DeepMET SONIC Producer to CMSSW #37963
Adding DeepMET SONIC Producer to CMSSW #37963
Conversation
The accompanying model-file PR is here: cms-data/RecoMET-METPUSubtraction#6 |
+code-checks Logs: https://cmssdt.cern.ch/SDT/code-checks/cms-sw-PR-37963/30035
|
A new Pull Request was created by @wpmccormack (Patrick McCormack) for master. It involves the following packages:
@perrotta, @clacaputo, @cmsbuild, @slava77, @jpata, @qliphy, @fabiocos, @davidlange6 can you please review it and eventually sign? Thanks. cms-bot commands are listed here |
test parameters: |
please test |
+1 Summary: https://cmssdt.cern.ch/SDT/jenkins-artifacts/pull-request-integration/PR-6f1ee4/24748/summary.html 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 Summary@slava77 comparisons for the following workflows were not done due to missing matrix map:
Summary:
|
would it make sense to define a single test workflow where all the sonic modifiers for production models are enabled? enableSonicTriton,deepMETSonicTriton,particleNetSonicTriton,... similar to what was done here: https://github.com/cms-sw/cmssw/pull/37134/files#diff-8c163203cbaa64f0e0e6aa2b16346cfab3eaf84a57a008c7fa37193e8b9b2adbR516 |
@michaelwassmer @cms-sw/jetmet-pog-l2 could you please review this from the JME side? |
Yes, I can do that. |
@jpata this workflow is already defined as
hence the modification of https://github.com/cms-sw/cmssw/blob/master/Configuration/ProcessModifiers/python/allSonicTriton_cff.py in this PR. |
test parameters:
|
@cmsbuild please test |
+code-checks Logs: https://cmssdt.cern.ch/SDT/code-checks/cms-sw-PR-37963/31030
|
Pull request #37963 was updated. @perrotta, @rappoccio, @gouskos, @clacaputo, @fabiocos, @cmsbuild, @fgolf, @jpata, @mariadalfonso, @qliphy, @davidlange6 can you please check and sign again. |
please test |
+1 Summary: https://cmssdt.cern.ch/SDT/jenkins-artifacts/pull-request-integration/PR-6f1ee4/26212/summary.html 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 Summary@slava77 comparisons for the following workflows were not done due to missing matrix map:
Summary:
|
+xpog
only rebase since last signoff |
+reconstruction
|
+1 |
This pull request is fully signed and it will be integrated in one of the next master IBs (tests are also fine). This pull request will be automatically merged. |
PR description:
This PR introduces a producer for DeepMET using SONIC for co-processor-enabled inferences-as-a-service. For more information on the status of SONIC, please refer to this presentation to JetMET DPG: https://indico.cern.ch/event/1143469/contributions/4799423/attachments/2416540/4135308/March_28_JetMET_SONIC.pdf. Please note that the producer introduced here does not affect the main DeepMET producer in any way. This is an alternate way to run the standard DeepMET model using inference as a service. As noted in the linked slides, this producer has been validated locally on Tier 2 resources and at scale with cloud computing, but with this PR, we hope to perform more "official" production tests. But again, this producer does not any workflow unless explicitly called. A helper-function macro is slightly modified, as are a few config files. There will be an accompanying PR into https://github.com/cms-data/RecoMET-METPUSubtraction, which adds needed model files.
Tagging @kpedro88 @yongbinfeng @violatingcp @nhanvtran @jmduarte
PR validation:
This PR was tested on LPC using a Triton server running on the GPU-enabled AILab machines at FermiLab. Tests have also been performed using the DeepMET SONIC producer at Purdue T2 and in Google Cloud. When the producer is not called, there is no performance impact, and when it is called, the MET results are the same as for the generic DeepMET producer.