jetclustering / jobs /eval3_quark_dist_loss.slurm
gregorkrzmanc's picture
.
e75a247
#!/bin/bash
#SBATCH --partition=gpu # Specify the partition
#SBATCH --account=gpu_gres # Specify the account
#SBATCH --mem=25000 # Request 10GB of memory
#SBATCH --time=25:30:00
#SBATCH --job-name=SVJtr3 # Name the job
#SBATCH --output=jobs/train3_output1.log # Redirect stdout to a log file
#SBATCH --error=jobs/eval3error1.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
srun singularity exec -B /t3home/gkrzmanc/ -B /work/gkrzmanc/ --nv docker://gkrz/lgatr:v3 python -m src.train -test scouting_PFNano_signals2/SVJ_hadronic_std2/s-channel_mMed-1000_mDark-20_rinv-0.3 scouting_PFNano_signals2/SVJ_hadronic_std2/s-channel_mMed-1500_mDark-20_rinv-0.5 scouting_PFNano_signals2/SVJ_hadronic_std2/s-channel_mMed-700_mDark-20_rinv-0.7 scouting_PFNano_signals2/SVJ_hadronic_std2/s-channel_mMed-900_mDark-20_rinv-0.3 scouting_PFNano_signals2/SVJ_hadronic_std2/s-channel_mMed-1000_mDark-20_rinv-0.5 scouting_PFNano_signals2/SVJ_hadronic_std2/s-channel_mMed-1500_mDark-20_rinv-0.7 scouting_PFNano_signals2/SVJ_hadronic_std2/s-channel_mMed-800_mDark-20_rinv-0.3 scouting_PFNano_signals2/SVJ_hadronic_std2/s-channel_mMed-900_mDark-20_rinv-0.5 scouting_PFNano_signals2/SVJ_hadronic_std2/s-channel_mMed-1000_mDark-20_rinv-0.7 scouting_PFNano_signals2/SVJ_hadronic_std2/s-channel_mMed-700_mDark-20_rinv-0.3 scouting_PFNano_signals2/SVJ_hadronic_std2/s-channel_mMed-800_mDark-20_rinv-0.5 scouting_PFNano_signals2/SVJ_hadronic_std2/s-channel_mMed-900_mDark-20_rinv-0.7 scouting_PFNano_signals2/SVJ_hadronic_std2/s-channel_mMed-1500_mDark-20_rinv-0.3 scouting_PFNano_signals2/SVJ_hadronic_std2/s-channel_mMed-700_mDark-20_rinv-0.5 scouting_PFNano_signals2/SVJ_hadronic_std2/s-channel_mMed-800_mDark-20_rinv-0.7 -net src/models/GATr/Gatr.py -bs 64 --gpus 0 --run-name Eval_Quark_dist_loss_VALSET_STD2 --num-epochs 1000 --attr-loss-weight 0.1 --coord-loss-weight 0.1 --beta-type pt+bc --load-model-weights train/debug_2025_01_15_14_57_46/step_19000_epoch_14.ckpt --num-workers 0