dgp.sh

#!/bin/bash
#SBATCH --job-name=DGP128
#SBATCH --partition=CLUSTER
#SBATCH -t 7-00:00

### e.g. request 2 nodes with 1 gpu each, totally 2 gpus (WORLD_SIZE==2)
### Note: --gres=gpu:x should equal to ntasks-per-node
#SBATCH --nodelist=compute-0-1
#SBATCH --ntasks-per-node=1
#SBATCH --cpus-per-task=20
#SBATCH -o ./output/%x-%j.out
#SBATCH -e ./output/%x-%j.err

### export MASTER_PORT=12342
### export WORLD_SIZE=1
### echo "NODELIST="${SLURM_NODELIST}
### master_addr=$(scontrol show hostnames "$SLURM_JOB_NODELIST" | head -n 1)
### export MASTER_ADDR=$master_addr
### echo "MASTER_ADDR="$MASTER_ADDR

module load python/anaconda3
module load cuda/cuda-11.4
source activate tf_tc

D='2048'
OBJ='wing'
for d in $D
do
  for obj in $OBJ
  do
    python DGP_untrain.py \
    --class -1 \
    --seed 1314 \
    --random_G \
    --update_embed \
    --lr_ratio 1.0 \
    --select_num 1 \
    --iterations 1000 1000 1000 1000 1000 \
    --G_lrs 1.4e-5 1.4e-5 1.4e-5 0.7e-5 0.35e-5 \
    --z_lrs 7e-4 7e-5 7e-5 7e-6 7e-7  \
    --use_in False False False False False \
    --resolution 256 \
    --weights_root pretrained \
    --load_weights 256 \
    --G_ch 96 \
    --G_shared \
    --hier --dim_z 120 --shared_dim 128 \
    --skip_init --use_ema \
    --dims $d \
    --object $obj
  done
done