运行PLATO模型，训练时进程被Killed

您好，我在PaddlePaddle 1.6.0，8核8G内存的Linux机器上运行PLATO模型时，报如下错误，请帮忙看看，谢谢！

$ bash scripts/DailyDialog/train.sh

SAVE_DIR=outputs/DailyDialog
VOCAB_PATH=model/Bert/vocab.txt
DATA_DIR=data/DailyDialog
INIT_CHECKPOINT=model/PLATO
DATA_TYPE=multi
USE_VISUALDL=false
export CUDA_VISIBLE_DEVICES=
CUDA_VISIBLE_DEVICES=
export FLAGS_fraction_of_gpu_memory_to_use=0.1
FLAGS_fraction_of_gpu_memory_to_use=0.1
export FLAGS_eager_delete_scope=True
FLAGS_eager_delete_scope=True
export FLAGS_eager_delete_tensor_gb=0.0
FLAGS_eager_delete_tensor_gb=0.0
python -u ./preprocess.py --vocab_path model/Bert/vocab.txt --data_dir data/DailyDialog --data_type multi
[[ false = true ]]
python -u ./run.py --do_train true --vocab_path model/Bert/vocab.txt --data_dir data/DailyDialog --data_type multi --batch_size 6 --valid_steps 2000 --num_type_embeddings 2 --use_discriminator true --num_epoch 20 --lr 1e-5 --save_checkpoint false --save_summary false --init_checkpoint model/PLATO --save_dir outputs/DailyDialog { "do_train": true, "do_test": false, "do_infer": false, "num_infer_batches": null, "hparams_file": null, "BPETextField": { "vocab_path": "model/Bert/vocab.txt", "filtered": false, "max_len": 256, "min_utt_len": 1, "max_utt_len": 50, "min_ctx_turn": 1, "max_ctx_turn": 16, "max_knowledge_num": 16, "max_knowledge_len": 16, "tokenizer_type": "Bert" }, "Dataset": { "data_dir": "data/DailyDialog", "data_type": "multi" }, "Trainer": { "use_data_distributed": false, "valid_metric_name": "-loss", "num_epochs": 20, "save_dir": "outputs/DailyDialog", "batch_size": 6, "log_steps": 100, "valid_steps": 2000, "save_checkpoint": false, "save_summary": false, "shuffle": true, "sort_pool_size": 0 }, "Model": { "init_checkpoint": "model/PLATO", "model": "UnifiedTransformer", "num_token_embeddings": -1, "num_pos_embeddings": 512, "num_type_embeddings": 2, "num_turn_embeddings": 16, "num_latent": 20, "tau": 0.67, "with_bow": true, "hidden_dim": 768, "num_heads": 12, "num_layers": 12, "padding_idx": 0, "dropout": 0.1, "embed_dropout": 0.0, "attn_dropout": 0.1, "ff_dropout": 0.1, "use_discriminator": true, "dis_ratio": 1.0, "weight_sharing": true, "pos_trainable": true, "two_layer_predictor": false, "bidirectional_context": true, "label_smooth": 0.0, "initializer_range": 0.02, "lr": 1e-05, "weight_decay": 0.0, "max_grad_norm": null }, "Generator": { "generator": "BeamSearch", "min_gen_len": 1, "max_gen_len": 30, "beam_size": 5, "length_average": false, "length_penalty": -1.0, "ignore_unk": true } } Loading parameters from model/PLATO Loaded parameters from model/PLATO scripts/DailyDialog/train.sh: line 45: 2041 Killed python -u ./run.py --do_train true --vocab_path $VOCAB_PATH --data_dir $DATA_DIR --data_type $DATA_TYPE --batch_size 6 --valid_steps 2000 --num_type_embeddings 2 --use_discriminator true --num_epoch 20 --lr 1e-5 --save_checkpoint false --save_summary $USE_VISUALDL --init_checkpoint $INIT_CHECKPOINT --save_dir $SAVE_DIR
[[ false = true ]]

PaddlePaddle / Research

运行PLATO模型，训练时进程被Killed #56