Add Phase-2 Level-1 LLP trigger emulator: TOoLLiP [13_3_0_pre3]

[ddiaz006]

PR description:

• Add ML-based LLP Jet tagger to Phase-2 Level-1 trigger emulator; codenamed TOoLLiP
• Requires TOoLLiP external integrated here: Add TOoLLiP extra cms-sw/cmsdist#8906
• Compiled object file added under L1Trigger/Phase2L1ParticleFlow/test until above PR is approved
• Presented at L1 Tau-Jets-MET meetings, e.g.
  • https://indico.cern.ch/event/1288096/#9-llp-jet-tagging-status-and-p
  • https://indico.cern.ch/event/1350143/contributions/5683985/
• This model represents a v1 baseline with post-training quantization. We expect to update it with a more realistic firmware implementation.
• Model and expected performance plots shown here: https://github.com/cms-hls4ml/TOoLLiP/blob/main/README.md
• Central CMSSW PR: TOoLLiP integration cms-sw/cmssw#43638

PR validation:

In progress

[triggerDoctor]

Hello, I'm triggerDoctor. @aloeliger is testing this script for L1T offline software validation.

Attempts to compile this PR succeeded!

Info	Value
return code	0
command	eval scramv1 runtime -sh && scram b -j 8

[triggerDoctor]

Hello, I'm triggerDoctor. @aloeliger is testing this script for L1T offline software validation.

I found no issues with the code checks!

Info	Value
return code	0
command	eval scramv1 runtime -sh && scram b -k -j 8 code-checks && scram b -k -j 8 code-checks

I found no issues with the headers!

Info	Value
return code	0
command	eval scramv1 runtime -sh && scram b -k -j 8 check-headers

[triggerDoctor]

Hello, I'm triggerDoctor. @aloeliger is testing this script for L1T offline software validation.

I found 1 files that did not meet formatting requirements:

• L1Trigger/L1TTrackMatch/plugins/L1TrackJetClustering.h

Please run scram b code-format to auto-apply code formatting

Info	Value
return code	0
command	eval scramv1 runtime -sh && scram b -k -j 8 code-format-all

[EmyrClement]

@ddiaz006 thanks for this PR! For my understanding, you are reusing/generalizing the jet ID class that was originally written for the b tag ID so that it can calculate an ID from any NN (assuming the same/similar input features) by either running tensorflow directly or the "HLS4ML emulation"? I think this makes sense, though I would like to see a quick validation that the b tag scores are the same before/after this PR to make sure they are unaffected.

[aloeliger]

@ddiaz006 Just a gentle ping on the review comments.

[aloeliger]

Hi @ddiaz006, I'm going around starting to figure out how to update all PRs to our new IB

For this PR I would recommend:

1. First, make a backup of this branch and it's commits pushed to github under a different name
2. checkout CMSSW_14_0_0_pre3 via cmsrel CMSSW_14_0_0_pre3 && cd CMSSW_14_0_0_pre3/src/ && cmsenv && git cms-init
3. Get a copy of this branch locally via git cms-checkout-topic -u ddiaz006:TOoLLiP_1330pre3_2
4. Perform an interactive rebase via git rebase from-CMSSW_14_0_0_pre3 --interactive
5. In the interactive rebase, please drop all commits that were not originally part of this pull request. In my tests, there seemed to be some earlier commits from track trigger commits that get caught up in this somehow?
6. You can either then push the resulting branch to the current PR branch, or make a new one and make a PR from that branch to cms-l1t-offline:phase2-l1t-integration-14_0_0_pre3

Alternatively, because you have relatively few commits, you could just cherry-pick commits by doing something like:

1. First, make a backup of this branch and it's commits pushed to github under a different name
2. checkout CMSSW_14_0_0_pre3 via cmsrel CMSSW_14_0_0_pre3 && cd CMSSW_14_0_0_pre3/src/ && cmsenv && git cms-init
3. Fetch your changes git fetch my-cmssw
4. Cherry pick the 4 necessary commits via git cherry-pick, the 4 commands needed (in order) I think are: git cherry-pick 9eaa43f096c1d45995bc647ef04e8017de403b9d git cherry-pick 1e4a6d6040201bc79be960fc81d2816b2d04987a git cherry-pick 666efd5f97c5b85e2e5340d26f3e7a0a1374613f and git cherry-pick fc0fa3d7f251aab0249caf6ddb1d503283022322
5. You can either then push the resulting branch to the current PR branch, or make a new one and make a PR from that branch to cms-l1t-offline:phase2-l1t-integration-14_0_0_pre3

[EmyrClement]

@ddiaz006 In addition to the rebase above, can you also add python configuration for running your tagger, as is done for the b jets here? You can then add the tagger to the L1 sequence (see example here for the b tagger) and to the event content (see here).

[EmyrClement]

I tried running the tagger, but found the code crashes in this section:

cmssw/L1Trigger/Phase2L1ParticleFlow/src/JetId.cc

Lines 83 to 90 in fc0fa3d

	ap_fixed<16, 6> modelResult = -1;
	//for (int i = 0; i < 1; i++) {
	// modelResult[i] = -1;
	//}
	modelRef_->prepare_input(modelInput);
	modelRef_->predict();
	modelRef_->read_result(modelResult);

The problem is modelResults is an ap_fixed<> but modelRef_->read_result(modelResult); requires modelResults to be an array of ap_fixed<> with size 1.

I guess this wasn't caught in the tests by trigger doctor (or in master) as the producer isn't being run in the L1 sequence.

[aloeliger]

@ddiaz006 Any updates on rebasing this PR?

[ddiaz006]

Thank you for your instructions @aloeliger. I have rebased to CMSSW_14_0_0_pre3 and created a new PR: #1213. This one can be closed and I will update the other.