Model Card for mlpf-clic-clusters-v2.2.0

This model reconstructs particles in a detector, based on the tracks and calorimeter clusters recorded by the detector.

Model Details

The performance is measured with respect to generator-level jets and MET computed from Pythia particles, i.e. the truth-level jets and MET. The primary difference with respect to v2.1.0 is the inclusion of the sqrt(pt) weight in the pT and energy loss term.

Jet performance

MET performance

Model Description

Developed by: Joosep Pata, Eric Wulff, Farouk Mokhtar, Mengke Zhang, David Southwick, Maria Girone, David Southwick, Javier Duarte, Michael Kagan
Model type: transformer
License: Apache License

Model Sources

Repository: https://github.com/jpata/particleflow/releases/tag/v2.2.0

Uses

Direct Use

This model may be used to study the physics and computational performance on ML-based reconstruction in simulation.

Out-of-Scope Use

This model is not intended for physics measurements on real data.

Bias, Risks, and Limitations

The model has only been trained on simulation data and has not been validated against real data. The model has not been peer reviewed or published in a peer-reviewed journal.

How to Get Started with the Model

Use the code below to get started with the model.

#get the code
git clone https://github.com/jpata/particleflow
cd particleflow
git checkout v2.2.0

#get the models
git clone https://huggingface.co/jpata/particleflow models

Training Details

Trained on 1x A100 for 5 epochs over ~6 days. The training was continued from a checkpoint due to a runtime limit.

Training Data

The following datasets were used:

4.7G    /scratch/persistent/joosep/tensorflow_datasets/clic_edm_qq_pf/1/2.5.0
4.8G    /scratch/persistent/joosep/tensorflow_datasets/clic_edm_qq_pf/2/2.5.0
4.7G    /scratch/persistent/joosep/tensorflow_datasets/clic_edm_qq_pf/3/2.5.0
4.7G    /scratch/persistent/joosep/tensorflow_datasets/clic_edm_qq_pf/4/2.5.0
4.7G    /scratch/persistent/joosep/tensorflow_datasets/clic_edm_qq_pf/5/2.5.0
4.7G    /scratch/persistent/joosep/tensorflow_datasets/clic_edm_qq_pf/6/2.5.0
4.7G    /scratch/persistent/joosep/tensorflow_datasets/clic_edm_qq_pf/7/2.5.0
4.7G    /scratch/persistent/joosep/tensorflow_datasets/clic_edm_qq_pf/8/2.5.0
4.7G    /scratch/persistent/joosep/tensorflow_datasets/clic_edm_qq_pf/9/2.5.0
4.8G    /scratch/persistent/joosep/tensorflow_datasets/clic_edm_qq_pf/10/2.5.0
9.3G    /scratch/persistent/joosep/tensorflow_datasets/clic_edm_ttbar_pf/1/2.5.0
9.3G    /scratch/persistent/joosep/tensorflow_datasets/clic_edm_ttbar_pf/2/2.5.0
9.3G    /scratch/persistent/joosep/tensorflow_datasets/clic_edm_ttbar_pf/3/2.5.0
9.3G    /scratch/persistent/joosep/tensorflow_datasets/clic_edm_ttbar_pf/4/2.5.0
9.3G    /scratch/persistent/joosep/tensorflow_datasets/clic_edm_ttbar_pf/5/2.5.0
9.3G    /scratch/persistent/joosep/tensorflow_datasets/clic_edm_ttbar_pf/6/2.5.0
9.3G    /scratch/persistent/joosep/tensorflow_datasets/clic_edm_ttbar_pf/7/2.5.0
9.3G    /scratch/persistent/joosep/tensorflow_datasets/clic_edm_ttbar_pf/8/2.5.0
9.3G    /scratch/persistent/joosep/tensorflow_datasets/clic_edm_ttbar_pf/9/2.5.0
9.3G    /scratch/persistent/joosep/tensorflow_datasets/clic_edm_ttbar_pf/10/2.5.0
7.4G    /scratch/persistent/joosep/tensorflow_datasets/clic_edm_ww_fullhad_pf/1/2.5.0
7.4G    /scratch/persistent/joosep/tensorflow_datasets/clic_edm_ww_fullhad_pf/2/2.5.0
7.4G    /scratch/persistent/joosep/tensorflow_datasets/clic_edm_ww_fullhad_pf/3/2.5.0
7.4G    /scratch/persistent/joosep/tensorflow_datasets/clic_edm_ww_fullhad_pf/4/2.5.0
7.4G    /scratch/persistent/joosep/tensorflow_datasets/clic_edm_ww_fullhad_pf/5/2.5.0
7.4G    /scratch/persistent/joosep/tensorflow_datasets/clic_edm_ww_fullhad_pf/6/2.5.0
7.4G    /scratch/persistent/joosep/tensorflow_datasets/clic_edm_ww_fullhad_pf/7/2.5.0
7.4G    /scratch/persistent/joosep/tensorflow_datasets/clic_edm_ww_fullhad_pf/8/2.5.0
7.4G    /scratch/persistent/joosep/tensorflow_datasets/clic_edm_ww_fullhad_pf/9/2.5.0
7.4G    /scratch/persistent/joosep/tensorflow_datasets/clic_edm_ww_fullhad_pf/10/2.5.0

The datasets were generated using Key4HEP with the following scripts:

Training Procedure

#!/bin/bash
#SBATCH --partition gpu
#SBATCH --gres gpu:a100:1
#SBATCH --mem-per-gpu 250G
#SBATCH -o logs/slurm-%x-%j-%N.out

IMG=/home/software/singularity/pytorch.simg:2024-12-03
cd ~/particleflow

ulimit -n 100000
singularity exec -B /scratch/persistent --nv \
    --env PYTHONPATH=`pwd` \
    --env KERAS_BACKEND=torch \
    $IMG python3 mlpf/pipeline.py --gpus 1 \
    --data-dir /scratch/persistent/joosep/tensorflow_datasets --config parameters/pytorch/pyg-clic.yaml \
    --train --conv-type attention \
    --gpu-batch-multiplier 256 --checkpoint-freq 1 --num-workers 8 --prefetch-factor 100 --comet --ntest 2000 --test-datasets clic_edm_qq_pf

Evaluation

#!/bin/bash
#SBATCH --partition gpu
#SBATCH --gres gpu:a100-mig:1
#SBATCH --mem-per-gpu 100G
#SBATCH -o logs/slurm-%x-%j-%N.out

IMG=/home/software/singularity/pytorch.simg:2024-12-03
cd ~/particleflow

WEIGHTS=experiments/pyg-clic_20250106_193536_269746/checkpoints/checkpoint-05-1.995116.pth
singularity exec -B /scratch/persistent --nv \
     --env PYTHONPATH=`pwd` \
     --env KERAS_BACKEND=torch \
     $IMG  python3 mlpf/pipeline.py --gpus 1 \
     --data-dir /scratch/persistent/joosep/tensorflow_datasets --config parameters/pytorch/pyg-clic.yaml \
     --test --make-plots --gpu-batch-multiplier 100 --load $WEIGHTS --dtype bfloat16 --num-workers 0 --ntest 50000

jpata
/

particleflow