added run script

apps/pretrained_compound/ChemRL/GEM/tasks/conf_gen/train_vae.sh

@@ -0,0 +1,120 @@
+#!/bin/bash
+
+# cd ..
+
+source /fs1/software/modules/bashrc
+module load loginnode
+# module load CUDA/11.2
+# module load cudnn/8.4.1-cuda11.x
+module load CUDA/10.2
+module load cudnn/8.0.5-cuda10.2
+export PYTHONUNBUFFERED=1
+
+root_path="../../../../../.."
+export PYTHONPATH="$(pwd)/$root_path/":$PYTHONPATH
+export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
+# export LD_LIBRARY_PATH='/fs1/home/wangll1/miniconda3/envs/paddlehelix-cu112-gt/lib'
+
+BATCH_SIZE=256
+NUM_WORKERS=1
+EPOCHS=100
+NUM_LAYERS=3
+
+LR=1e-4
+
+DATASET="qm9_bond_angle_diheral_gt_no_position"
+LOG_PATH="./log/"
+
+TRAIN_DATA_PATH="./data/raw/train"
+VALID_DATA_PATH="./data/raw/valid"
+
+TRAIN_DATA_PATH_DEBUG="./data/raw/test"
+VALID_DATA_PATH_DEBUG="./data/raw/test"
+
+PRIOR_CONFIG="./data/configs/prior.json"
+ENCODER_CONFIG="./data/configs/encoder.json"
+DECODER_CONFIG="./data/configs/decoder.json"
+HEAD_CONFIG="./data/configs/head.json"
+
+INIT_MODEL=''
+
+LEARNING_RATE=1e-3
+WEIGHT_DECAY=5e-4
+DROPOUT_RATE=0.2
+
+N_NOISE_MOL=1
+VAE_BETA=1.0
+
+((BATCH_SIZE=$BATCH_SIZE/$N_NOISE_MOL))
+
+DEBUG=$1
+DISTRIBUTED=$2
+
+if [ $DEBUG -eq 1 ]; then
+  echo "Run as debug mode："
+  echo "=================="
+  python train_vae.py \
+    --batch_size $BATCH_SIZE \
+    --num_workers 1 \
+    --epochs 10 \
+    --num_layers $NUM_LAYERS \
+    --dataset 'debug' \
+    --train_data_path $TRAIN_DATA_PATH \
+    --valid_data_path $VALID_DATA_PATH_DEBUG \
+    --log_path $LOG_PATH \
+    --prior_config $PRIOR_CONFIG \
+    --encoder_config $ENCODER_CONFIG \
+    --decoder_config $DECODER_CONFIG \
+    --head_config $HEAD_CONFIG \
+    --learning_rate $LEARNING_RATE \
+    --weight_decay $WEIGHT_DECAY \
+    --dropout_rate $DROPOUT_RATE \
+    --n_noise_mol $N_NOISE_MOL \
+    --vae_beta $VAE_BETA \
+    --isomorphism \
+    --debug
+elif [ $DEBUG -eq 0 ] && [ $DISTRIBUTED -eq 0 ]; then
+  echo "Run as single gpu mode："
+  echo "=================="
+  python train_vae.py \
+    --batch_size $BATCH_SIZE \
+    --num_workers $NUM_WORKERS \
+    --epochs $EPOCHS \
+    --num_layers $NUM_LAYERS \
+    --dataset $DATASET \
+    --train_data_path $TRAIN_DATA_PATH \
+    --valid_data_path $VALID_DATA_PATH \
+    --log_path $LOG_PATH \
+    --prior_config $PRIOR_CONFIG \
+    --encoder_config $ENCODER_CONFIG \
+    --decoder_config $DECODER_CONFIG \
+    --head_config $HEAD_CONFIG \
+    --learning_rate $LEARNING_RATE \
+    --weight_decay $WEIGHT_DECAY \
+    --dropout_rate $DROPOUT_RATE \
+    --n_noise_mol $N_NOISE_MOL \
+    --vae_beta $VAE_BETA \
+    --isomorphism
+elif [ $DEBUG -eq 0 ] && [ $DISTRIBUTED -eq 1 ]; then
+  echo "Run as multi gpu mode："
+  echo "=================="
+  python -m paddle.distributed.launch train_vae.py \
+    --batch_size $BATCH_SIZE \
+    --num_workers $NUM_WORKERS \
+    --epochs $EPOCHS \
+    --num_layers $NUM_LAYERS \
+    --dataset $DATASET \
+    --train_data_path $TRAIN_DATA_PATH \
+    --valid_data_path $VALID_DATA_PATH \
+    --log_path $LOG_PATH \
+    --prior_config $PRIOR_CONFIG \
+    --encoder_config $ENCODER_CONFIG \
+    --decoder_config $DECODER_CONFIG \
+    --head_config $HEAD_CONFIG \
+    --learning_rate $LEARNING_RATE \
+    --weight_decay $WEIGHT_DECAY \
+    --dropout_rate $DROPOUT_RATE \
+    --n_noise_mol $N_NOISE_MOL \
+    --vae_beta $VAE_BETA \
+    --distributed
+fi