jobs/training_2.slurm

#!/bin/bash
#SBATCH --partition=gpu               # Specify the partition
#SBATCH --account=gpu_gres            # Specify the account
#SBATCH --mem=25000                   # Request 10GB of memory
#SBATCH --time=20:30:00
#SBATCH --job-name=SVJtr2  # Name the job
#SBATCH --output=jobs/training_2_output.log       # Redirect stdout to a log file
#SBATCH --error=jobs/training_2_error.log         # Redirect stderr to a log file
#SBATCH --gres=gpu:1
source env.sh
export APPTAINER_TMPDIR=/work/gkrzmanc/singularity_tmp
export APPTAINER_CACHEDIR=/work/gkrzmanc/singularity_cache
nvidia-smi
#singularity exec -B /t3home/gkrzmanc/ -B /work/gkrzmanc/ --nv docker://dologarcia/gatr:v0 python -m src.train -train scouting_PFNano_signals2/SVJ_hadronic_std3/s-channel_mMed-900_mDark-20_rinv-0.3_alpha-peak_13TeV-pythia8_n-2000 -val scouting_PFNano_signals2/SVJ_hadronic_std2/s-channel_mMed-900_mDark-20_rinv-0.3 -net src/models/GATr/Gatr.py -bs 64 --gpus 0 --run-name Eval03_OC_betaPt_BC --val-dataset-size 1000 --num-epochs 1 --train-dataset-size 1 --attr-loss-weight 0.1 --coord-loss-weight 0.1 --beta-type pt+bc --load-model-weights train/OC_betaPt_BC_train_2024_12_27_18_00_40/step_10000_epoch_8.ckpt
#singularity exec -B /t3home/gkrzmanc/ -B /work/gkrzmanc/ --nv docker://gkrz/lgatr:v3 python -m src.train -train scouting_PFNano_signals2/SVJ_hadronic_std3/s-channel_mMed-900_mDark-20_rinv-0.3_alpha-peak_13TeV-pythia8_n-2000 -val scouting_PFNano_signals2/SVJ_hadronic_std2/s-channel_mMed-900_mDark-20_rinv-0.3 -net src/models/LGATr/lgatr.py -bs 64 --gpus 0 --run-name Train_LGATr --val-dataset-size 100 --num-epochs 1000  --attr-loss-weight 0.1 --coord-loss-weight 0.1 --beta-type pt+bc

#singularity exec -B /t3home/gkrzmanc/ -B /work/gkrzmanc/ --nv docker://gkrz/lgatr:v3 python -m src.train -train scouting_PFNano_signals2/SVJ_hadronic_std3/s-channel_mMed-900_mDark-20_rinv-0.3_alpha-peak_13TeV-pythia8_n-2000 -val scouting_PFNano_signals2/SVJ_hadronic_std2/s-channel_mMed-900_mDark-20_rinv-0.3 -net src/models/LGATr/lgatr.py -bs 64 --gpus 0 --run-name Train_LGATr_SB --val-dataset-size 100 --num-epochs 1000 --attr-loss-weight 0.1 --coord-loss-weight 0.1 --beta-type pt+bc --load-model-weights train/Train_LGATr_2025_01_06_12_48_47/step_10000_epoch_8.ckpt
singularity exec -B /t3home/gkrzmanc/ -B /work/gkrzmanc/ --nv docker://gkrz/lgatr:v3 python -m src.train -train scouting_PFNano_signals2/SVJ_hadronic_std3/s-channel_mMed-900_mDark-20_rinv-0.3_alpha-peak_13TeV-pythia8_n-2000 -val scouting_PFNano_signals2/SVJ_hadronic_std2/s-channel_mMed-900_mDark-20_rinv-0.3 -net src/models/LGATr/lgatr.py -bs 64 --gpus 0 --run-name Train_LGATr_SB --val-dataset-size 100 --num-epochs 1000 --attr-loss-weight 0.1 --coord-loss-weight 0.1 --beta-type pt+bc
# VEGA: python -m src.train -train scouting_PFNano_signals2/SVJ_hadronic_std3/s-channel_mMed-900_mDark-20_rinv-0.3_alpha-peak_13TeV-pythia8_n-2000 -val scouting_PFNano_signals2/SVJ_hadronic_std2/s-channel_mMed-900_mDark-20_rinv-0.3 -net src/models/LGATr/lgatr.py -bs 256 --gpus 0 --run-name Train_LGATr_SB --val-dataset-size 256 --num-epochs 1000 --attr-loss-weight 0.1 --coord-loss-weight 0.1 --beta-type pt+bc



# Overfit test

#singularity exec -B /t3home/gkrzmanc/ -B /work/gkrzmanc/ --nv docker://gkrz/lgatr:v3 python -m src.train -train scouting_PFNano_signals2/SVJ_hadronic_std3/s-channel_mMed-900_mDark-20_rinv-0.3_alpha-peak_13TeV-pythia8_n-2000 -val scouting_PFNano_signals2/SVJ_hadronic_std3/s-channel_mMed-900_mDark-20_rinv-0.3_alpha-peak_13TeV-pythia8_n-2000 -net src/models/LGATr/lgatr.py -bs 16 --gpus 0 --run-name OverfitTest_LGATr --val-dataset-size 16  --train-dataset-size 16 --num-epochs 8000  --attr-loss-weight 0.01 --coord-loss-weight 0.1 --beta-type pt+bc
# Eval overfit
#singularity exec -B /t3home/gkrzmanc/ -B /work/gkrzmanc/ --nv docker://gkrz/lgatr:v3 python -m src.train -train scouting_PFNano_signals2/SVJ_hadronic_std3/s-channel_mMed-900_mDark-20_rinv-0.3_alpha-peak_13TeV-pythia8_n-2000 -val scouting_PFNano_signals2/SVJ_hadronic_std3/s-channel_mMed-900_mDark-20_rinv-0.3_alpha-peak_13TeV-pythia8_n-2000 -net src/models/LGATr/lgatr.py -bs 16 --gpus 0 --run-name OverfitTest_LGATr --val-dataset-size 16  --train-dataset-size 16 --num-epochs 1  --attr-loss-weight 0.01 --coord-loss-weight 0.1 --beta-type pt+bc --load-model-weights  train/OverfitTest_LGATr_2025_01_06_12_58_07/step_8000_epoch_8000.ckpt
# Run the job:
# sbatch jobs/training_2.slurm