upgrade_dgu_api (#4413)

PaddleNLP/dialogue_system/dialogue_general_understanding/README.md

@@ -145,7 +145,7 @@ batch_size: 一个batch内输入的样本个数
 do_lower_case: 是否进行大小写转换
 random_seed: 随机种子设置
 use_cuda: 是否使用cuda, 如果是gpu训练时，设置成true
-in_tokens: 是否采用in_tokens模式来计算batch_siz数量, 如果in_tokens为false, 则batch_size等于真实设置的batch_size大小, 如果in_tokens为true, 则batch_size=batch_size*max_seq_len，即按照token计数
+in_tokens: false
 do_save_inference_model: 是否保存inference model
 encable_ce: 是否开启ce
 ```
@@ -213,9 +213,8 @@ python -u main.py \
        --task_name=${TASK_NAME} \
        --use_cuda=${use_cuda} \
        --do_train=true \
-       --in_tokens=true \
        --epoch=20 \
-       --batch_size=4096 \
+       --batch_size=32 \
        --do_lower_case=true \
        --data_dir="./data/input/data/atis/${TASK_NAME}" \
        --bert_config_path="${BERT_BASE_PATH}/bert_config.json" \
@@ -236,7 +235,7 @@ python -u main.py \
 
 #### windows环境下
 ```
-python -u main.py --task_name=atis_intent --use_cuda=false --do_train=true --in_tokens=true --epoch=20 --batch_size=4096 --do_lower_case=true --data_dir=data\input\data\atis\atis_intent --bert_config_path=data\pretrain_model\uncased_L-12_H-768_A-12\bert_config.json --vocab_path=data\pretrain_model\uncased_L-12_H-768_A-12\vocab.txt --init_from_pretrain_model=data\pretrain_model\uncased_L-12_H-768_A-12\params --save_model_path=data\saved_models\atis_intent --save_param=params --save_steps=100 --learning_rate=2e-5 --weight_decay=0.01 --max_seq_len=128 --print_steps=10
+python -u main.py --task_name=atis_intent --use_cuda=false --do_train=true --epoch=20 --batch_size=32 --do_lower_case=true --data_dir=data\input\data\atis\atis_intent --bert_config_path=data\pretrain_model\uncased_L-12_H-768_A-12\bert_config.json --vocab_path=data\pretrain_model\uncased_L-12_H-768_A-12\vocab.txt --init_from_pretrain_model=data\pretrain_model\uncased_L-12_H-768_A-12\params --save_model_path=data\saved_models\atis_intent --save_param=params --save_steps=100 --learning_rate=2e-5 --weight_decay=0.01 --max_seq_len=128 --print_steps=10
 ```
 
 ### 模型预测
@@ -292,8 +291,7 @@ python -u main.py \
        --task_name=${TASK_NAME} \
        --use_cuda=${use_cuda} \
        --do_predict=true \
-       --in_tokens=true \
-       --batch_size=4096 \
+       --batch_size=32 \
        --do_lower_case=true \
        --data_dir="./data/input/data/atis/${TASK_NAME}" \
        --init_from_params="./data/saved_models/trained_models/${TASK_NAME}/params" \
@@ -307,7 +305,7 @@ python -u main.py \
 
 #### windows环境下
 ```
-python -u main.py --task_name=atis_intent --use_cuda=false --do_predict=true --in_tokens=true --batch_size=4096 --do_lower_case=true --data_dir=data\input\data\atis\atis_intent --init_from_params=data\saved_models\trained_models\atis_intent\params --bert_config_path=data\pretrain_model\uncased_L-12_H-768_A-12\bert_config.json --vocab_path=data\pretrain_model\uncased_L-12_H-768_A-12\vocab.txt --output_prediction_file=data\output\pred_atis_intent --max_seq_len=128
+python -u main.py --task_name=atis_intent --use_cuda=false --do_predict=true --batch_size=32 --do_lower_case=true --data_dir=data\input\data\atis\atis_intent --init_from_params=data\saved_models\trained_models\atis_intent\params --bert_config_path=data\pretrain_model\uncased_L-12_H-768_A-12\bert_config.json --vocab_path=data\pretrain_model\uncased_L-12_H-768_A-12\vocab.txt --output_prediction_file=data\output\pred_atis_intent --max_seq_len=128
 ```
 
 ### 模型评估

PaddleNLP/dialogue_system/dialogue_general_understanding/predict.py

@@ -71,7 +71,7 @@ def do_predict(args):
                 name='sent_ids', shape=[-1, args.max_seq_len], dtype='int64')
             input_mask = fluid.data(
                 name='input_mask',
-                shape=[-1, args.max_seq_len],
+                shape=[-1, args.max_seq_len, 1],
                 dtype='float32')
             if args.task_name == 'atis_slot':
                 labels = fluid.data(

PaddleNLP/dialogue_system/dialogue_general_understanding/run.sh

@@ -3,7 +3,7 @@
 export FLAGS_sync_nccl_allreduce=0
 export FLAGS_eager_delete_tensor_gb=1
 
-export CUDA_VISIBLE_DEVICES=0
+export CUDA_VISIBLE_DEVICES=1
 if  [ ! "$CUDA_VISIBLE_DEVICES" ]
 then
     export CPU_NUM=1
@@ -21,7 +21,7 @@ SAVE_MODEL_PATH="./data/saved_models/${TASK_NAME}"
 TRAIN_MODEL_PATH="./data/saved_models/trained_models"
 OUTPUT_PATH="./data/output"
 INFERENCE_MODEL="data/inference_models"
-PYTHON_PATH="python"
+PYTHON_PATH="python3"
 
 if [ -f ${SAVE_MODEL_PATH} ]; then
     rm ${SAVE_MODEL_PATH}
@@ -37,8 +37,7 @@ then
   save_steps=1000
   max_seq_len=210
   print_steps=1000
-  batch_size=6720
-  in_tokens=true
+  batch_size=32
   epoch=2
   learning_rate=2e-5
 elif [ "${TASK_NAME}" = "swda" ]
@@ -46,8 +45,7 @@ then
   save_steps=500
   max_seq_len=128
   print_steps=200
-  batch_size=6720
-  in_tokens=true
+  batch_size=32
   epoch=3
   learning_rate=2e-5
 elif [ "${TASK_NAME}" = "mrda" ]
@@ -55,8 +53,7 @@ then
   save_steps=500
   max_seq_len=128
   print_steps=200
-  batch_size=4096
-  in_tokens=true
+  batch_size=32
   epoch=7
   learning_rate=2e-5
 elif [ "${TASK_NAME}" = "atis_intent" ]
@@ -64,8 +61,7 @@ then
   save_steps=100
   max_seq_len=128
   print_steps=10
-  batch_size=4096
-  in_tokens=true
+  batch_size=32
   epoch=20
   learning_rate=2e-5
   INPUT_PATH="./data/input/data/atis/${TASK_NAME}"
@@ -75,7 +71,6 @@ then
   max_seq_len=128
   print_steps=10
   batch_size=32
-  in_tokens=False
   epoch=50
   learning_rate=2e-5
   INPUT_PATH="./data/input/data/atis/${TASK_NAME}"
@@ -83,22 +78,23 @@ elif [ "${TASK_NAME}" = "dstc2" ]
 then
   save_steps=400
   print_steps=20
-  batch_size=8192
-  in_tokens=true
   epoch=40
   learning_rate=5e-5
   INPUT_PATH="./data/input/data/dstc2/${TASK_NAME}"
   if [ "${TASK_TYPE}" = "train" ]
   then
     max_seq_len=256
+    batch_size=32
   else
     max_seq_len=512
+    batch_size=16
   fi
 else
   echo "not support ${TASK_NAME} dataset.."
   exit 255
 fi
 
+
 #training
 function train()
 {
@@ -106,7 +102,6 @@ function train()
        --task_name=${TASK_NAME} \
        --use_cuda=$1 \
        --do_train=true \
-       --in_tokens=${in_tokens} \
        --epoch=${epoch} \
        --batch_size=${batch_size} \
        --do_lower_case=true \
@@ -130,7 +125,6 @@ function predict()
        --task_name=${TASK_NAME} \
        --use_cuda=$1 \
        --do_predict=true \
-       --in_tokens=${in_tokens} \
        --batch_size=${batch_size} \
        --data_dir=${INPUT_PATH} \
        --do_lower_case=true \

PaddleNLP/dialogue_system/dialogue_general_understanding/train.py

@@ -67,7 +67,7 @@ def do_train(args):
                 name='sent_ids', shape=[-1, args.max_seq_len], dtype='int64')
             input_mask = fluid.data(
                 name='input_mask',
-                shape=[-1, args.max_seq_len],
+                shape=[-1, args.max_seq_len, 1],
                 dtype='float32')
             if args.task_name == 'atis_slot':
                 labels = fluid.data(
@@ -80,8 +80,9 @@ def do_train(args):
 
             input_inst = [src_ids, pos_ids, sent_ids, input_mask, labels]
             input_field = InputField(input_inst)
-            data_reader = fluid.io.PyReader(
-                feed_list=input_inst, capacity=4, iterable=False)
+            
+            data_reader = fluid.io.DataLoader.from_generator(feed_list=input_inst, capacity=4, iterable=False)
+
             processor = processors[task_name](data_dir=args.data_dir,
                                               vocab_path=args.vocab_path,
                                               max_seq_len=args.max_seq_len,
@@ -108,10 +109,8 @@ def do_train(args):
                 accuracy.persistable = True
             num_seqs.persistable = True
 
-            if args.use_cuda:
-                dev_count = fluid.core.get_cuda_device_count()
-            else:
-                dev_count = int(os.environ.get('CPU_NUM', 1))
+            places = fluid.cuda_places() if args.use_cuda else fluid.cpu_places()
+            dev_count = len(places)
 
             batch_generator = processor.data_generator(
                 batch_size=args.batch_size, phase='train', shuffle=True)
@@ -140,7 +139,7 @@ def do_train(args):
                 use_fp16=False,
                 loss_scaling=args.loss_scaling)
 
-    data_reader.decorate_batch_generator(batch_generator)
+    data_reader.set_batch_generator(batch_generator, places=places)
 
     if args.use_cuda:
         place = fluid.CUDAPlace(int(os.getenv('FLAGS_selected_gpus', '0')))