MMSRL

Our Multimodal Semantic Role Labeling model for CONSTRAINT 2022
git clone https://esimon.eu/repos/MMSRL.git
Log | Files | Refs | Submodules | README

generic_features_visualbert.sh (2331B)


      1 #!/bin/bash
      2 #SBATCH -A lco@gpu
      3 #SBATCH --job-name=mmsrl             # job name
      4 #SBATCH --ntasks=1                   # number of MP tasks
      5 #SBATCH --ntasks-per-node=1          # number of MPI tasks per node
      6 #SBATCH --gres=gpu:1                 # number of GPUs per node
      7 #SBATCH --cpus-per-task=10           # number of cores per tasks
      8 #SBATCH --hint=nomultithread         # we get physical cores not logical
      9 #SBATCH --distribution=block:block   # we pin the tasks on contiguous cores
     10 #SBATCH --time=10:00:00               # maximum execution time (HH:MM:SS)
     11 #SBATCH --output=/gpfswork/rech/lco/url46ht/outputs/generic_features_visualbert/generic_features_visualbert_%j_%x_%A_%a.out  # output file name
     12 #SBATCH --error=/gpfswork/rech/lco/url46ht/outputs/generic_features_visualbert/generic_features_visualbert_%j_%x_%A_%a.err   # error file name
     13 #SBATCH --array=0-95
     14 
     15 
     16 cd $WORK/mmsrl
     17 
     18 module purge
     19 source ~/.bashrc
     20 conda activate pytorch
     21 export DATA_PATH=$WORK/data
     22 export TRANSFORMERS_CACHE=$HOME/.cache/huggingface/transformers
     23 export TRANSFORMERS_OFFLINE=1
     24 
     25 base_config=("--features_path='all' --use_sbert=True --use_caption=True --use_entities=True"
     26              "--features_path=features_frcnn"
     27              "--features_path=features_b7"
     28              "--features_path=features_vgg"
     29              "--features_path=features_b7 --use_sbert=True  --use_entities=True"
     30              "--features_path=features_b7 --use_caption=True")
     31 
     32 base_name=("all"
     33            "frcnn"
     34            "b7"
     35            "vgg"
     36            "entities"
     37            "caption")
     38 base_pooling=(attention mlp)
     39 base_lr=(5e-3 1e-4 5e-4 1e-5)
     40 
     41 iconfig=$(($SLURM_ARRAY_TASK_ID%6))
     42 ipooling=$(($SLURM_ARRAY_TASK_ID/6%2))
     43 ilr=$(($SLURM_ARRAY_TASK_ID/12%4))
     44 
     45 config=${base_config[$iconfig]}
     46 pooling=${base_pooling[$ipooling]}
     47 lr=${base_lr[$ilr]}
     48 name=${base_name[$iconfig]}_${pooling}_lr${lr}
     49 
     50 python -m mmsrl.train configs/generic.py --use_clip=False --pooling=$pooling $config --patience=26 \
     51         --batch_per_epoch=1000 --use_visualbert=True\
     52         --learning_rate=$lr \
     53         --cyclic_subsample=None --subsample_labels='interpolate_micro_to_macro' \
     54         --output_val=\"$WORK/outputs/generic_features_visualbert/val_${name}_${SLURM_ARRAY_TASK_ID}.pkl\" \
     55         --output_test=\"$WORK/outputs/generic_features_visualbert/test_${name}_${SLURM_ARRAY_TASK_ID}.pkl\"