Fix bug of get_embedding interface

Fix bugs

hub_module/modules/text/semantic_model/bert_cased_L_12_H_768_A_12/README.md

@@ -96,7 +96,7 @@ embedding_result = module.get_embedding(texts=[["Sample1_text_a"],["Sample2_text
 
 # Use "get_params_layer" to get params layer and used to ULMFiTStrategy.
 params_layer = module.get_params_layer()
-strategy = hub.finetune.strategy.ULMFiTStrategy(params_layer=params_layer)
+strategy = hub.finetune.strategy.ULMFiTStrategy(frz_params_layer=params_layer, dis_params_layer=params_layer)
 ```
 
 

hub_module/modules/text/semantic_model/bert_cased_L_24_H_1024_A_16/README.md

@@ -96,7 +96,7 @@ embedding_result = module.get_embedding(texts=[["Sample1_text_a"],["Sample2_text
 
 # Use "get_params_layer" to get params layer and used to ULMFiTStrategy.
 params_layer = module.get_params_layer()
-strategy = hub.finetune.strategy.ULMFiTStrategy(params_layer=params_layer)
+strategy = hub.finetune.strategy.ULMFiTStrategy(frz_params_layer=params_layer, dis_params_layer=params_layer)
 ```
 
 ##   查看代码

hub_module/modules/text/semantic_model/bert_chinese_L_12_H_768_A_12/README.md

@@ -96,7 +96,7 @@ embedding_result = module.get_embedding(texts=[["Sample1_text_a"],["Sample2_text
 
 # Use "get_params_layer" to get params layer and used to ULMFiTStrategy.
 params_layer = module.get_params_layer()
-strategy = hub.finetune.strategy.ULMFiTStrategy(params_layer=params_layer)
+strategy = hub.finetune.strategy.ULMFiTStrategy(frz_params_layer=params_layer, dis_params_layer=params_layer)
 ```
 
 ##   查看代码

hub_module/modules/text/semantic_model/bert_multi_cased_L_12_H_768_A_12/README.md

@@ -96,7 +96,7 @@ embedding_result = module.get_embedding(texts=[["Sample1_text_a"],["Sample2_text
 
 # Use "get_params_layer" to get params layer and used to ULMFiTStrategy.
 params_layer = module.get_params_layer()
-strategy = hub.finetune.strategy.ULMFiTStrategy(params_layer=params_layer)
+strategy = hub.finetune.strategy.ULMFiTStrategy(frz_params_layer=params_layer, dis_params_layer=params_layer)
 ```
 
 ##   查看代码

hub_module/modules/text/semantic_model/bert_multi_uncased_L_12_H_768_A_12/README.md

@@ -96,7 +96,7 @@ embedding_result = module.get_embedding(texts=[["Sample1_text_a"],["Sample2_text
 
 # Use "get_params_layer" to get params layer and used to ULMFiTStrategy.
 params_layer = module.get_params_layer()
-strategy = hub.finetune.strategy.ULMFiTStrategy(params_layer=params_layer)
+strategy = hub.finetune.strategy.ULMFiTStrategy(frz_params_layer=params_layer, dis_params_layer=params_layer)
 ```
 
 ##   查看代码

hub_module/modules/text/semantic_model/bert_uncased_L_12_H_768_A_12/README.md

@@ -96,7 +96,7 @@ embedding_result = module.get_embedding(texts=[["Sample1_text_a"],["Sample2_text
 
 # Use "get_params_layer" to get params layer and used to ULMFiTStrategy.
 params_layer = module.get_params_layer()
-strategy = hub.finetune.strategy.ULMFiTStrategy(params_layer=params_layer)
+strategy = hub.finetune.strategy.ULMFiTStrategy(frz_params_layer=params_layer, dis_params_layer=params_layer)
 ```
 
 ##   查看代码

hub_module/modules/text/semantic_model/bert_uncased_L_24_H_1024_A_16/README.md

@@ -96,7 +96,7 @@ embedding_result = module.get_embedding(texts=[["Sample1_text_a"],["Sample2_text
 
 # Use "get_params_layer" to get params layer and used to ULMFiTStrategy.
 params_layer = module.get_params_layer()
-strategy = hub.finetune.strategy.ULMFiTStrategy(params_layer=params_layer)
+strategy = hub.finetune.strategy.ULMFiTStrategy(frz_params_layer=params_layer, dis_params_layer=params_layer)
 ```
 
 ##   查看代码

hub_module/modules/text/semantic_model/chinese_bert_wwm/README.md

@@ -114,7 +114,7 @@ embedding_result = module.get_embedding(texts=[["Sample1_text_a"],["Sample2_text
 
 # Use "get_params_layer" to get params layer and used to ULMFiTStrategy.
 params_layer = module.get_params_layer()
-strategy = hub.finetune.strategy.ULMFiTStrategy(params_layer=params_layer)
+strategy = hub.finetune.strategy.ULMFiTStrategy(frz_params_layer=params_layer, dis_params_layer=params_layer)
 ```
 
 ## 查看代码
@@ -218,7 +218,7 @@ embedding_result = module.get_embedding(texts=[["Sample1_text_a"],["Sample2_text
 
 # Use "get_params_layer" to get params layer and used to ULMFiTStrategy.
 params_layer = module.get_params_layer()
-strategy = hub.finetune.strategy.ULMFiTStrategy(params_layer=params_layer)
+strategy = hub.finetune.strategy.ULMFiTStrategy(frz_params_layer=params_layer, dis_params_layer=params_layer)
 ```
 
 ## 查看代码

hub_module/modules/text/semantic_model/chinese_bert_wwm_ext/README.md

@@ -96,7 +96,7 @@ embedding_result = module.get_embedding(texts=[["Sample1_text_a"],["Sample2_text
 
 # Use "get_params_layer" to get params layer and used to ULMFiTStrategy.
 params_layer = module.get_params_layer()
-strategy = hub.finetune.strategy.ULMFiTStrategy(params_layer=params_layer)
+strategy = hub.finetune.strategy.ULMFiTStrategy(frz_params_layer=params_layer, dis_params_layer=params_layer)
 ```
 
 ## 查看代码

hub_module/modules/text/semantic_model/chinese_electra_base/README.md

@@ -96,7 +96,7 @@ embedding_result = module.get_embedding(texts=[["Sample1_text_a"],["Sample2_text
 
 # Use "get_params_layer" to get params layer and used to ULMFiTStrategy.
 params_layer = module.get_params_layer()
-strategy = hub.finetune.strategy.ULMFiTStrategy(params_layer=params_layer)
+strategy = hub.finetune.strategy.ULMFiTStrategy(frz_params_layer=params_layer, dis_params_layer=params_layer)
 ```
 
 

hub_module/modules/text/semantic_model/chinese_electra_small/README.md

@@ -96,7 +96,7 @@ embedding_result = module.get_embedding(texts=[["Sample1_text_a"],["Sample2_text
 
 # Use "get_params_layer" to get params layer and used to ULMFiTStrategy.
 params_layer = module.get_params_layer()
-strategy = hub.finetune.strategy.ULMFiTStrategy(params_layer=params_layer)
+strategy = hub.finetune.strategy.ULMFiTStrategy(frz_params_layer=params_layer, dis_params_layer=params_layer)
 ```
 
 

hub_module/modules/text/semantic_model/chinese_roberta_wwm_ext/README.md

@@ -96,7 +96,7 @@ embedding_result = module.get_embedding(texts=[["Sample1_text_a"],["Sample2_text
 
 # Use "get_params_layer" to get params layer and used to ULMFiTStrategy.
 params_layer = module.get_params_layer()
-strategy = hub.finetune.strategy.ULMFiTStrategy(params_layer=params_layer)
+strategy = hub.finetune.strategy.ULMFiTStrategy(frz_params_layer=params_layer, dis_params_layer=params_layer)
 ```
 
 ## 查看代码

hub_module/modules/text/semantic_model/chinese_roberta_wwm_ext_large/README.md

@@ -96,7 +96,7 @@ embedding_result = module.get_embedding(texts=[["Sample1_text_a"],["Sample2_text
 
 # Use "get_params_layer" to get params layer and used to ULMFiTStrategy.
 params_layer = module.get_params_layer()
-strategy = hub.finetune.strategy.ULMFiTStrategy(params_layer=params_layer)
+strategy = hub.finetune.strategy.ULMFiTStrategy(frz_params_layer=params_layer, dis_params_layer=params_layer)
 ```
 
 ## 查看代码

hub_module/modules/text/semantic_model/ernie/README.md

@@ -107,7 +107,7 @@ embedding_result = module.get_embedding(texts=[["Sample1_text_a"],["Sample2_text
 
 # Use "get_params_layer" to get params layer and used to ULMFiTStrategy.
 params_layer = module.get_params_layer()
-strategy = hub.finetune.strategy.ULMFiTStrategy(params_layer=params_layer)
+strategy = hub.finetune.strategy.ULMFiTStrategy(frz_params_layer=params_layer, dis_params_layer=params_layer)
 ```
 利用该PaddleHub Module Fine-tune示例，可参考[文本分类](https://github.com/PaddlePaddle/PaddleHub/tree/release/v1.2/demo/text-classification)、[序列标注](https://github.com/PaddlePaddle/PaddleHub/tree/release/v1.2/demo/sequence-labeling)。
 

hub_module/modules/text/semantic_model/ernie_tiny/README.md

@@ -94,7 +94,7 @@ embedding_result = module.get_embedding(texts=[["Sample1_text_a"],["Sample2_text
 
 # Use "get_params_layer" to get params layer and used to ULMFiTStrategy.
 params_layer = module.get_params_layer()
-strategy = hub.finetune.strategy.ULMFiTStrategy(params_layer=params_layer)
+strategy = hub.finetune.strategy.ULMFiTStrategy(frz_params_layer=params_layer, dis_params_layer=params_layer)
 ```
 利用该PaddleHub Module Fine-tune示例，可参考[文本分类](https://github.com/PaddlePaddle/PaddleHub/tree/release/v1.4.0/demo/text-classification)。
 

hub_module/modules/text/semantic_model/ernie_v2_eng_base/README.md

@@ -100,7 +100,7 @@ embedding_result = module.get_embedding(texts=[["Sample1_text_a"],["Sample2_text
 
 # Use "get_params_layer" to get params layer and used to ULMFiTStrategy.
 params_layer = module.get_params_layer()
-strategy = hub.finetune.strategy.ULMFiTStrategy(params_layer=params_layer)
+strategy = hub.finetune.strategy.ULMFiTStrategy(frz_params_layer=params_layer, dis_params_layer=params_layer)
 ```
 利用该PaddleHub Module Fine-tune示例，可参考[文本分类](https://github.com/PaddlePaddle/PaddleHub/tree/release/v1.4.0/demo/text-classification)。
 

hub_module/modules/text/semantic_model/ernie_v2_eng_large/README.md

@@ -103,7 +103,7 @@ embedding_result = module.get_embedding(texts=[["Sample1_text_a"],["Sample2_text
 
 # Use "get_params_layer" to get params layer and used to ULMFiTStrategy.
 params_layer = module.get_params_layer()
-strategy = hub.finetune.strategy.ULMFiTStrategy(params_layer=params_layer)
+strategy = hub.finetune.strategy.ULMFiTStrategy(frz_params_layer=params_layer, dis_params_layer=params_layer)
 ```
 利用该PaddleHub Module Fine-tune示例，可参考[文本分类](https://github.com/PaddlePaddle/PaddleHub/tree/release/v1.2/demo/text-classification)。
 

hub_module/modules/text/semantic_model/rbt3/README.md

@@ -96,7 +96,7 @@ embedding_result = module.get_embedding(texts=[["Sample1_text_a"],["Sample2_text
 
 # Use "get_params_layer" to get params layer and used to ULMFiTStrategy.
 params_layer = module.get_params_layer()
-strategy = hub.finetune.strategy.ULMFiTStrategy(params_layer=params_layer)
+strategy = hub.finetune.strategy.ULMFiTStrategy(frz_params_layer=params_layer, dis_params_layer=params_layer)
 ```
 
 ## 查看代码

hub_module/modules/text/semantic_model/rbtl3/README.md

@@ -96,7 +96,7 @@ embedding_result = module.get_embedding(texts=[["Sample1_text_a"],["Sample2_text
 
 # Use "get_params_layer" to get params layer and used to ULMFiTStrategy.
 params_layer = module.get_params_layer()
-strategy = hub.finetune.strategy.ULMFiTStrategy(params_layer=params_layer)
+strategy = hub.finetune.strategy.ULMFiTStrategy(frz_params_layer=params_layer, dis_params_layer=params_layer)
 ```
 
 ## 查看代码

paddlehub/finetune/strategy.py

@@ -511,10 +511,6 @@ class CombinedStrategy(DefaultStrategy):
                     unfreeze_depths=self.
                     sorted_depth[:self.max_depth * self.epoch //
                                  self.scheduler["gradual_unfreeze"]["blocks"]])
-            else:
-                logger.warning(
-                    "The max op-depth in the network is %s. That results in that can't use the gradual unfreeze finetune strategy."
-                    % (self.max_depth))
         elif self.scheduler["gradual_unfreeze"]["params_layer"]:
             max_layer = max(
                 self.scheduler["gradual_unfreeze"]["params_layer"].values())
@@ -631,8 +627,9 @@ class ULMFiTStrategy(CombinedStrategy):
                  ratio=32,
                  dis_blocks=3,
                  factor=2.6,
+                 dis_params_layer=None,
                  frz_blocks=3,
-                 params_layer=None):
+                 frz_params_layer=None):
 
         scheduler = {
             "slanted_triangle": {
@@ -641,12 +638,12 @@ class ULMFiTStrategy(CombinedStrategy):
             },
             "gradual_unfreeze": {
                 "blocks": frz_blocks,
-                "params_layer": params_layer
+                "params_layer": frz_params_layer
             },
             "discriminative": {
                 "blocks": dis_blocks,
                 "factor": factor,
-                "params_layer": params_layer
+                "params_layer": dis_params_layer
             }
         }
         regularization = {}

paddlehub/finetune/task/base_task.py

@@ -36,7 +36,7 @@ from visualdl import LogWriter
 
 import paddlehub as hub
 from paddlehub.common.paddle_helper import dtype_map, clone_program
-from paddlehub.common.utils import mkdir, version_compare
+from paddlehub.common.utils import mkdir
 from paddlehub.common.dir import tmp_dir
 from paddlehub.common.logger import logger
 from paddlehub.finetune.checkpoint import load_checkpoint, save_checkpoint
@@ -951,12 +951,6 @@ class BaseTask(object):
         Returns:
             RunState: the running result of predict phase
         """
-
-        if isinstance(self._base_data_reader, hub.reader.LACClassifyReader):
-            raise Exception(
-                "LACClassifyReader does not support predictor, please close accelerate_mode"
-            )
-
         global_run_states = []
         period_run_states = []
 
@@ -998,6 +992,12 @@ class BaseTask(object):
         Returns:
             RunState: the running result of predict phase
         """
+        if accelerate_mode and isinstance(self._base_data_reader,
+                                          hub.reader.LACClassifyReader):
+            logger.warning(
+                "LACClassifyReader does not support predictor, the accelerate_mode is closed now."
+            )
+            accelerate_mode = False
         self.accelerate_mode = accelerate_mode
 
         with self.phase_guard(phase="predict"):

paddlehub/finetune/task/reading_comprehension_task.py

@@ -205,7 +205,7 @@ def get_predictions(all_examples, all_features, all_results, n_best_size,
         for (feature_index, feature) in enumerate(features):
             if feature.unique_id not in unique_id_to_result:
                 logger.info(
-                    "As using pyreader, the last one batch is so small that the feature %s in the last batch is discarded "
+                    "As using multidevice, the last one batch is so small that the feature %s in the last batch is discarded "
                     % feature.unique_id)
                 continue
             result = unique_id_to_result[feature.unique_id]

paddlehub/module/nlp_module.py

@@ -397,7 +397,8 @@ class TransformerModule(NLPBaseModule):
 
         return inputs, outputs, module_program
 
-    def get_embedding(self, texts, use_gpu=False, batch_size=1):
+    def get_embedding(self, texts, max_seq_len=512, use_gpu=False,
+                      batch_size=1):
         """
         get pooled_output and sequence_output for input texts.
         Warnings: this method depends on Paddle Inference Library, it may not work properly in PaddlePaddle <= 1.6.2.
@@ -405,6 +406,7 @@ class TransformerModule(NLPBaseModule):
         Args:
             texts (list): each element is a text sample, each sample include text_a and text_b where text_b can be omitted.
                           for example: [[sample0_text_a, sample0_text_b], [sample1_text_a, sample1_text_b], ...]
+            max_seq_len (int): the max sequence length.
             use_gpu (bool): use gpu or not, default False.
             batch_size (int): the data batch size, default 1.
 
@@ -417,12 +419,12 @@ class TransformerModule(NLPBaseModule):
         ) or self.emb_job["batch_size"] != batch_size or self.emb_job[
                 "use_gpu"] != use_gpu:
             inputs, outputs, program = self.context(
-                trainable=True, max_seq_len=self.MAX_SEQ_LEN)
+                trainable=True, max_seq_len=max_seq_len)
 
             reader = hub.reader.ClassifyReader(
                 dataset=None,
                 vocab_path=self.get_vocab_path(),
-                max_seq_len=self.MAX_SEQ_LEN,
+                max_seq_len=max_seq_len,
                 sp_model_path=self.get_spm_path() if hasattr(
                     self, "get_spm_path") else None,
                 word_dict_path=self.get_word_dict_path() if hasattr(

paddlehub/reader/nlp_reader.py

@@ -1113,7 +1113,7 @@ class LACClassifyReader(BaseReader):
 
             return processed
 
-        if not self.has_processed[phase]:
+        if not self.has_processed[phase] or phase == "predict":
             logger.info(
                 "processing %s data now... this may take a few minutes" % phase)
             for i in range(len(data)):