#!/bin/bash export FLAGS_fraction_of_gpu_memory_to_use=0.02 export FLAGS_eager_delete_tensor_gb=0.0 export FLAGS_fast_eager_deletion_mode=1 export CUDA_VISIBLE_DEVICES=0,1,2,3 python -m paddle.distributed.launch --selected_gpus=0,1,2,3 train.py \ --train_data ./data/train.tsv \ --test_data ./data/test.tsv \ --model_save_dir ./padding_models \ --validation_steps 1000 \ --save_steps 10000 \ --print_steps 200 \ --batch_size 400 \ --epoch 10 \ --traindata_shuffle_buffer 20000 \ --word_emb_dim 128 \ --grnn_hidden_dim 128 \ --bigru_num 2 \ --base_learning_rate 1e-3 \ --emb_learning_rate 2 \ --crf_learning_rate 0.2 \ --word_dict_path ./conf/word.dic \ --label_dict_path ./conf/tag.dic \ --word_rep_dict_path ./conf/q2b.dic \ --enable_ce false \ --use_cuda true \ --cpu_num 1 \ --use_data_paralle True