L1TSC82ProngJetModel by henripetrow · Pull Request #150 · p2l1pfp/cmssw

henripetrow · 2026-02-17T11:48:33Z

PR description:

This PR adds the v0.0.0 Level-1 Trigger SeededCone 8 2-Prong Jet Model to CMSSW. This model tags seededcone 8 jets with 2-prong score.

The model is documented here.

This PR also includes changes to the jet data format tag scores by adding a nprong score.

The cms-hls4ml model file tagged v0.0.0 is found here

This PR relies on an addition to the cms-dist externals, with the open PR here .

PR validation:

The testing and the performance of the model is presented here.

cerminar · 2026-02-17T16:11:27Z

Hello @henripetrow, it looks like this PR needs to be rebased...

cerminar

Thanks for rebasing!

cerminar · 2026-02-23T09:04:59Z

+#include "PhysicsTools/TensorFlow/interface/TensorFlow.h"
+#include "DataFormats/L1TParticleFlow/interface/PFCandidate.h"
+#include "DataFormats/L1TParticleFlow/interface/PFJet.h"
+#include "DataFormats/L1TParticleFlow/interface/datatypes.h"
+#include "L1Trigger/Phase2L1ParticleFlow/interface/jetmet/L1SeedConePFJetEmulator.h"


Are all of these necessary? I believe they need some cleanup.

Cleaned up redundant includes.

cerminar · 2026-02-23T09:05:17Z

    addGenObjects(process)
    addPh2L1Objects(process)
-    addPh2GTObjects(process)
+    #addPh2GTObjects(process)


Is this intended?

It wasn't intended. Removed the comment sign.

cerminar · 2026-02-23T09:08:54Z

The file name should rather be L1TSC82ProngJetID.cc

Changed the filename.

thesps

A first round of review from running this branch

thesps · 2026-03-09T13:07:09Z


  struct Jet {
-    static const unsigned NTagFields = 8;
+    static const unsigned NTagFields = 9;


This should be kept at 8 since it refers to the number of fields of the narrow-cone / regular jets

Changed it back to 8 fields.

thesps · 2026-03-09T13:26:28Z

We need a from L1Trigger.Phase2L1ParticleFlow.l1tSC82ProngJetProducer_cfi import l1tSC82ProngJetProducer in the imports section

Added the new entry to the imports section.

…cms-merge-topic

thesps

I've left a few more comments. Some minor changes requested, and a few due-diligence checks.

thesps · 2026-04-08T08:46:46Z

+  desc.add<double>("minPt", 10);
+  desc.add<double>("maxEta", 2.4);


These should be double checked against the model training. As far as I know there's no minPt cut on the training of this model, and I'm not sure about any maxEta

It seems the training didn't use these cuts. Removed them from the producer.

thesps · 2026-04-08T08:46:57Z

    addGenObjects(process)
    addPh2L1Objects(process)
-    addPh2GTObjects(process)
+   # addPh2GTObjects(process)


Could you revert this change?

thesps · 2026-04-08T08:47:08Z

    sc8JetTable,
    sc4ExtJetTable, 
-    sc4NGJetTable,
+    ##sc4NGJetTable,


Could you revert this change?

thesps · 2026-04-08T08:47:17Z


 sc8JetTable = pfJetTable.clone(
-    src = 'l1tSC8PFL1PuppiCorrectedEmulator',
+    #src = 'l1tSC8PFL1PuppiCorrectedEmulator',


Remove the commented out line

thesps · 2026-04-08T08:48:43Z

+    fPt_.get()[i0] = iParts[i0]->pt();
+    fZ0_.get()[i0] = iParts[i0]->z0();
+    fPt_rel_.get()[i0] = iParts[i0]->pt() / jet_pt_;
+    fPt_log_.get()[i0] = std::log(iParts[i0]->pt());
+    fDEta_.get()[i0] = jet_eta_ - float(iParts[i0]->eta());


Are we sure that the model training used these physical values and not the hardware ones?

Yes, the model training used physical values.

cerminar · 2026-04-16T11:39:02Z

+private:
+  std::vector<inputtype> NNvectorVar_;
+  int fNParticles_;
+  unique_ptr<float[]> fPt_;


Shoud these be std::unique_ptr

Copilot

Pull request overview

Adds a new Phase-2 L1 ParticleFlow jet tagger producer based on an hls4ml model (“SeededCone 0.8 2-prong”) and extends the PFJet tag-class plumbing so the new nprong tag score can be stored and exposed (e.g., in Phase-2 L1 NanoAOD).

Changes:

Introduce L1TSC82ProngJetID inference helper and L1TSC82ProngJetProducer EDProducer to compute and attach an nprong tag score to SC8 PFJets.
Extend l1ct::JetTagClass/defaults to add the nprong tag label.
Wire the new producer into L1TPFJetsTask / L1TPFJetsEmulationTask and add nprong to the SC8 NanoAOD table.

Reviewed changes

Copilot reviewed 7 out of 7 changed files in this pull request and generated 8 comments.

Show a summary per file

File	Description
`L1Trigger/Phase2L1ParticleFlow/src/L1TSC82ProngJetID.cc`	Implements feature building and model evaluation for the 2-prong score.
`L1Trigger/Phase2L1ParticleFlow/interface/L1TSC82ProngJetID.h`	Declares the ID helper class used by the producer.
`L1Trigger/Phase2L1ParticleFlow/plugins/L1TSC82ProngJetModelProducer.cc`	New EDProducer that runs inference and attaches `nprong` tag scores to output PFJets.
`L1Trigger/Phase2L1ParticleFlow/python/l1pfJetMet_cff.py`	Adds the new producer to standard PF jet tasks.
`DataFormats/L1TParticleFlow/interface/jets.h`	Extends tag class enum and changes WideJet nprong-score propagation.
`DataFormats/L1TParticleFlow/src/jets.cpp`	Adds `nprong` string label and updates default tag-class list.
`DPGAnalysis/Phase2L1TNanoAOD/python/l1tPh2Nanotables_cff.py`	Switches SC8 source to the new producer and stores `nprong` in NanoAOD.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2026-04-16T11:48:07Z

      j.v3.eta = CTtoGT_eta(hwEta);
      j.z0(l1ct::z0_t::width - 1, 0) = hwZ0(l1ct::z0_t::width - 1, 0);
-      j.hwNProngScore = 0;
+      j.hwNProngScore = hwTagScores[0]; // currently only one WideJet tagger


toGTWide() now fills hwNProngScore from hwTagScores[0], but index 0 corresponds to the first jet tag score field ("b" in the default ordering). This will propagate the b-tag score into the WideJet n-prong field unless the hwTagScores packing has been repurposed accordingly. Please either source hwNProngScore from the actual n-prong score field, or document/update the hwTagScores layout so index 0 is guaranteed to be n-prong for WideJets.

Suggested change

j.hwNProngScore = hwTagScores[0]; // currently only one WideJet tagger

j.hwNProngScore = hwTagScores[static_cast<unsigned>(JetTagClass::nprong)];

I think this suggestion is wrong.

Copilot · 2026-04-16T11:48:08Z

+void L1TSC82ProngJetProducer::fillDescriptions(edm::ConfigurationDescriptions& descriptions) {
+  edm::ParameterSetDescription desc;
+  desc.add<edm::InputTag>("jets", edm::InputTag("l1tSC8PFL1PuppiEmulator"));
+  desc.add<std::string>("l1tSC82ProngJetModelPath", std::string("L1TSC82ProngJetModel_v0"));


fillDescriptions() does not declare the minPt and maxEta parameters, but the constructor unconditionally reads them. With parameter validation enabled, configs that set minPt/maxEta will be rejected as having unknown parameters (and omitting them will throw at runtime). Add desc.add<double>("minPt", ...) and desc.add<double>("maxEta", ...) (and consider using an unsigned type for maxJets).

Suggested change

desc.add<std::string>("l1tSC82ProngJetModelPath", std::string("L1TSC82ProngJetModel_v0"));

desc.add<std::string>("l1tSC82ProngJetModelPath", std::string("L1TSC82ProngJetModel_v0"));

desc.add<double>("minPt", 0.);

desc.add<double>("maxEta", 999.);

This one makes sense, just set the maxEta value to 5.

Added minPt and maxEta.

Copilot · 2026-04-16T11:48:08Z

+class L1TSC82ProngJetID {
+public:
+  L1TSC82ProngJetID(const std::shared_ptr<hls4mlEmulator::Model> model, int iNParticles);
+
+  typedef ap_fixed<24, 12, AP_RND, AP_SAT, 0> inputtype;
+  typedef ap_ufixed<20, 10, AP_RND, AP_SAT, 0> prong_score;
+
+  void setNNVectorVar();
+  std::vector<float> EvaluateNNFixed();
+  std::vector<float> computeFixed(const l1t::PFJet &iJet);
+
+private:
+  std::vector<inputtype> NNvectorVar_;
+  int fNParticles_;
+  unique_ptr<float[]> fPt_;
+  unique_ptr<float[]> fPt_rel_;
+  unique_ptr<float[]> fDEta_;
+  unique_ptr<float[]> fDPhi_;
+  unique_ptr<float[]> fPt_log_;
+  unique_ptr<float[]> fMass_;
+  unique_ptr<float[]> fZ0_;
+  unique_ptr<float[]> fDxy_;
+  unique_ptr<int[]> fIs_filled_;
+  unique_ptr<float[]> fPuppi_weight_;
+  unique_ptr<int[]> fEmID_;
+  unique_ptr<float[]> fQuality_;
+
+  unique_ptr<int[]> fCharge_;
+  unique_ptr<int[]> fId_;
+  std::shared_ptr<hls4mlEmulator::Model> modelRef_;


This header uses std::shared_ptr, std::vector, and unique_ptr but doesn't include the standard headers that define them, and the member declarations use unqualified unique_ptr (no std::). As written, this is likely to fail to compile depending on include order. Please include <memory> / <vector> explicitly and use std::unique_ptr<...> for the member types.

Copilot · 2026-04-16T11:48:09Z

+std::vector<float> L1TSC82ProngJetID::EvaluateNNFixed() {
+  const int NInputs = 160;
+  prong_score prong_scores;
+  inputtype fillzero = 0.0;
+
+  inputtype modelInput[NInputs] = {};  // Do something
+  std::fill(modelInput, modelInput + NInputs, fillzero);
+
+  for (unsigned int i = 0; i < NNvectorVar_.size(); i++) {
+    modelInput[i] = NNvectorVar_[i];
+  }


NInputs is hard-coded to 160, but the size of NNvectorVar_ depends on fNParticles_ (20 features × fNParticles_). If nParticles is configured to anything other than 8, the loop can write past the end of modelInput (or leave inputs unfilled). Please either (a) enforce fNParticles_ == 8 with a clear exception, or (b) compute NInputs from fNParticles_ and cap the copy at NInputs.

Added a check for the size of fNParticles.

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

…ducer.cc Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

…ripetrow/cmssw into L1TSC82ProngJetModel_p2l1pfp

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

cerminar · 2026-04-23T15:07:00Z

Last comment I promise:
https://gitlab.cern.ch/cms-cactus/phase2/firmware/correlator-common/-/pipelines/14604119

There seem to be a code-format change needed!

henripetrow · 2026-04-29T09:53:44Z

Last comment I promise: https://gitlab.cern.ch/cms-cactus/phase2/firmware/correlator-common/-/pipelines/14604119

There seem to be a code-format change needed!

Applied code-format to remove the diffs.

henripetrow force-pushed the L1TSC82ProngJetModel_p2l1pfp branch from ba2254a to ab20985 Compare February 20, 2026 21:25

henripetrow added 4 commits February 20, 2026 22:39

Added nprong score to DataFormats.

d5c7fbf

Added L1TSC82ProngJet to DPGAnalysis.

4bf12cb

Added L1TSC82ProngJetModel to L1Trigger/Phase2L1ParticleFlow

9765ee3

Fix for nprong tag.

64ead43

henripetrow force-pushed the L1TSC82ProngJetModel_p2l1pfp branch from ab20985 to 64ead43 Compare February 20, 2026 21:43

cerminar requested changes Feb 23, 2026

View reviewed changes

henripetrow added 3 commits March 3, 2026 10:58

Cleaned includes.

4146781

File renamed.

a20196a

Removed comment from addPh2GTObjects(process)

bb02d43

thesps requested changes Mar 9, 2026

View reviewed changes

henripetrow added 2 commits March 10, 2026 09:53

Merged L1TSC82ProngJetModel_p2l1pfp from repository henripetrow with …

da47c35

…cms-merge-topic

fixes.

415e52f

thesps requested changes Apr 8, 2026

View reviewed changes

henripetrow added 2 commits April 15, 2026 13:17

Reverted changes.

2f9df92

Removed cuts..

912dfe4

thesps approved these changes Apr 16, 2026

View reviewed changes

cerminar reviewed Apr 16, 2026

View reviewed changes

cerminar requested a review from Copilot April 16, 2026 11:39

Copilot started reviewing on behalf of cerminar April 16, 2026 11:40 View session

Copilot AI reviewed Apr 16, 2026

View reviewed changes

henripetrow and others added 8 commits April 17, 2026 08:17

Update L1Trigger/Phase2L1ParticleFlow/src/L1TSC82ProngJetID.cc

c4fbe83

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

Update L1Trigger/Phase2L1ParticleFlow/plugins/L1TSC82ProngJetModelPro…

5fe0d52

…ducer.cc Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

Update DataFormats/L1TParticleFlow/src/jets.cpp

039521f

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

added values for minPt and maxEta.

5c3a721

Merge branch 'L1TSC82ProngJetModel_p2l1pfp' of https://github.com/hen…

85744a2

…ripetrow/cmssw into L1TSC82ProngJetModel_p2l1pfp

Added std headers.

c0e2949

Added a check for fNParticles.

fd5052d

Update L1Trigger/Phase2L1ParticleFlow/python/l1pfJetMet_cff.py

063a582

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

Apply CMSSW code-format

db3dc45

cerminar merged commit 40a6aff into p2l1pfp:L1PF_15_1_X May 4, 2026

cerminar mentioned this pull request May 29, 2026

Phase2 L1T: Correlator development for AR2026 cms-sw/cmssw#51093

Closed

This was referenced Jun 19, 2026

Phase2 L1T: Correlator development for AR2026 cms-sw/cmssw#51270

Open

Phase2 L1T: Correlator development for AR2026 (backport) cms-sw/cmssw#51293

Open

		desc.add<double>("minPt", 10);
		desc.add<double>("maxEta", 2.4);

	j.hwNProngScore = hwTagScores[0]; // currently only one WideJet tagger
	j.hwNProngScore = hwTagScores[static_cast<unsigned>(JetTagClass::nprong)];

Uh oh!

Conversation

henripetrow commented Feb 17, 2026

PR description:

PR validation:

Uh oh!

cerminar commented Feb 17, 2026

Uh oh!

cerminar left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

thesps left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

thesps left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

Uh oh!

Copilot AI Apr 16, 2026

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Copilot AI Apr 16, 2026

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!