fix bugs

log

-------------------------- main configuration -------------------------
-target_tag               	1, 0
-vocab_path               	pretrain_model/ernie/vocab.txt
-do_lower_case            	True
-learning_rate            	3e-05
-backbone                 	ernie
-batch_size               	4
-max_seq_len              	512
-backbone_config_path     	pretrain_model/ernie/ernie_config.json
-mix_ratio                	0.5, 0.5
-warmup_proportion        	0.1
-save_path                	output_model/secondrun
-num_epochs               	0.5
-print_every_n_steps      	1
-weight_decay             	0.1
-optimizer                	adam
-task_instance            	mrqa, match4mrqa
-
-------------------------- mrqa configuration -------------------------
-pred_file                	data/mrqa/dev.json
-verbose                  	False
-max_query_len            	64
-pred_output_path         	mrqa_output
-train_file               	data/mrqa/train.json
-null_score_diff_threshold	0.0
-reader                   	mrc
-doc_stride               	128
-max_answer_len           	30
-n_best_size              	20
-paradigm                 	mrc
-
----------------------- match4mrqa configuration ----------------------
-train_file               	data/match4mrqa/train.tsv
-paradigm                 	match
-reader                   	match
-
-mrqa: set as target task.
-match4mrqa: set as aux task.
-mrqa: mix_ratio is set to 0.5
-match4mrqa: mix_ratio is set to 0.5
------------------------ backbone configuration -----------------------
-vocab_size               	30522
-target_tag               	1, 0
-num_hidden_layers        	24
-sent_type_vocab_size     	4
-warmup_proportion        	0.1
-num_epochs               	0.5
-print_every_n_steps      	1
-weight_decay             	0.1
-optimizer                	adam
-hidden_dropout_prob      	0.1
-do_lower_case            	True
-max_position_embeddings  	512
-max_seq_len              	512
-mix_ratio                	0.5, 0.5
-save_path                	output_model/secondrun
-attention_probs_dropout_prob	0.1
-hidden_size              	1024
-num_attention_heads      	16
-vocab_path               	pretrain_model/ernie/vocab.txt
-learning_rate            	3e-05
-backbone                 	ernie
-batch_size               	4
-backbone_config_path     	pretrain_model/ernie/ernie_config.json
-task_instance            	mrqa, match4mrqa
-task_type_vocab_size     	16
-initializer_range        	0.02
-hidden_act               	gelu
-
-initialing for training...
-mrqa: set as main task
-mrqa: preparing data...