[to #42322933] Redo an unmerged CR:https://code.alibaba-inc.com/Ali-MaaS/MaaS-lib/codereview/9427959

1. Redo a CR in current code
2. Refactor sbert's model configs
        Link: https://code.alibaba-inc.com/Ali-MaaS/MaaS-lib/codereview/9643861

modelscope/models/nlp/structbert/modeling_sbert.py

@@ -53,7 +53,7 @@ from .configuration_sbert import SbertConfig
 
 logger = get_logger(__name__)
 
-_CHECKPOINT_FOR_DOC = 'chinese_sbert-large-std-512'
+_CHECKPOINT_FOR_DOC = 'nlp_structbert_backbone_base_std'
 _CONFIG_FOR_DOC = 'SbertConfig'
 _TOKENIZER_FOR_DOC = 'SbertTokenizer'
 

modelscope/models/nlp/structbert/tokenization_sbert.py

@@ -32,8 +32,10 @@ VOCAB_FILES_NAMES = {'vocab_file': 'vocab.txt'}
 PRETRAINED_VOCAB_FILES_MAP = {'vocab_file': {}}
 
 PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES = {
-    'chinese_sbert-large-std-512': 512,
-    'english_sbert-large-std-512': 512,
+    'nlp_structbert_backbone_large_std': 512,
+    'nlp_structbert_backbone_base_std': 512,
+    'nlp_structbert_backbone_lite_std': 512,
+    'nlp_structbert_backbone_tiny_std': 512,
 }
 
 PRETRAINED_INIT_CONFIGURATION = {

modelscope/models/nlp/structbert/tokenization_sbert_fast.py

@@ -38,8 +38,10 @@ PRETRAINED_VOCAB_FILES_MAP = {
 }
 
 PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES = {
-    'chinese_sbert-large-std-512': 512,
-    'english_sbert-large-std-512': 512,
+    'nlp_structbert_backbone_large_std': 512,
+    'nlp_structbert_backbone_base_std': 512,
+    'nlp_structbert_backbone_lite_std': 512,
+    'nlp_structbert_backbone_tiny_std': 512,
 }
 
 PRETRAINED_INIT_CONFIGURATION = {

modelscope/pipelines/nlp/fill_mask_pipeline.py

@@ -37,7 +37,8 @@ class FillMaskPipeline(Pipeline):
             preprocessor = FillMaskPreprocessor(
                 fill_mask_model.model_dir,
                 first_sequence=first_sequence,
-                second_sequence=None)
+                second_sequence=None,
+                sequence_length=kwargs.pop('sequence_length', 128))
         fill_mask_model.eval()
         super().__init__(
             model=fill_mask_model, preprocessor=preprocessor, **kwargs)

modelscope/pipelines/nlp/named_entity_recognition_pipeline.py

@@ -26,7 +26,9 @@ class NamedEntityRecognitionPipeline(Pipeline):
         model = model if isinstance(model,
                                     Model) else Model.from_pretrained(model)
         if preprocessor is None:
-            preprocessor = NERPreprocessor(model.model_dir)
+            preprocessor = NERPreprocessor(
+                model.model_dir,
+                sequence_length=kwargs.pop('sequence_length', 512))
         model.eval()
         super().__init__(model=model, preprocessor=preprocessor, **kwargs)
         self.tokenizer = preprocessor.tokenizer

modelscope/pipelines/nlp/pair_sentence_classification_pipeline.py

@@ -33,5 +33,6 @@ class PairSentenceClassificationPipeline(SequenceClassificationPipelineBase):
             preprocessor = PairSentenceClassificationPreprocessor(
                 model.model_dir if isinstance(model, Model) else model,
                 first_sequence=first_sequence,
-                second_sequence=second_sequence)
+                second_sequence=second_sequence,
+                sequence_length=kwargs.pop('sequence_length', 512))
         super().__init__(model=model, preprocessor=preprocessor, **kwargs)

modelscope/pipelines/nlp/sequence_classification_pipeline.py

@@ -37,7 +37,8 @@ class SequenceClassificationPipeline(Pipeline):
             preprocessor = SequenceClassificationPreprocessor(
                 sc_model.model_dir,
                 first_sequence='sentence',
-                second_sequence=None)
+                second_sequence=None,
+                sequence_length=kwargs.pop('sequence_length', 512))
         super().__init__(model=sc_model, preprocessor=preprocessor, **kwargs)
 
         assert hasattr(self.model, 'id2label'), \

modelscope/pipelines/nlp/single_sentence_classification_pipeline.py

@@ -31,5 +31,6 @@ class SingleSentenceClassificationPipeline(SequenceClassificationPipelineBase):
         if preprocessor is None:
             preprocessor = SingleSentenceClassificationPreprocessor(
                 model.model_dir if isinstance(model, Model) else model,
-                first_sequence=first_sequence)
+                first_sequence=first_sequence,
+                sequence_length=kwargs.pop('sequence_length', 512))
         super().__init__(model=model, preprocessor=preprocessor, **kwargs)

modelscope/pipelines/nlp/text_generation_pipeline.py

@@ -32,7 +32,8 @@ class TextGenerationPipeline(Pipeline):
             preprocessor = TextGenerationPreprocessor(
                 model.model_dir,
                 first_sequence='sentence',
-                second_sequence=None)
+                second_sequence=None,
+                sequence_length=kwargs.pop('sequence_length', 128))
         model.eval()
         super().__init__(model=model, preprocessor=preprocessor, **kwargs)
 

modelscope/pipelines/nlp/word_segmentation_pipeline.py

@@ -31,7 +31,9 @@ class WordSegmentationPipeline(Pipeline):
         model = model if isinstance(model,
                                     Model) else Model.from_pretrained(model)
         if preprocessor is None:
-            preprocessor = TokenClassificationPreprocessor(model.model_dir)
+            preprocessor = TokenClassificationPreprocessor(
+                model.model_dir,
+                sequence_length=kwargs.pop('sequence_length', 128))
         model.eval()
         super().__init__(model=model, preprocessor=preprocessor, **kwargs)
         self.id2label = kwargs.get('id2label')

modelscope/pipelines/nlp/zero_shot_classification_pipeline.py

@@ -36,7 +36,9 @@ class ZeroShotClassificationPipeline(Pipeline):
         self.entailment_id = 0
         self.contradiction_id = 2
         if preprocessor is None:
-            preprocessor = ZeroShotClassificationPreprocessor(model.model_dir)
+            preprocessor = ZeroShotClassificationPreprocessor(
+                model.model_dir,
+                sequence_length=kwargs.pop('sequence_length', 512))
         model.eval()
         super().__init__(model=model, preprocessor=preprocessor, **kwargs)
 

modelscope/preprocessors/nlp.py

@@ -216,7 +216,7 @@ class PairSentenceClassificationPreprocessor(NLPTokenizerPreprocessorBase):
     def __init__(self, model_dir: str, mode=ModeKeys.INFERENCE, **kwargs):
         kwargs['truncation'] = kwargs.get('truncation', True)
         kwargs['padding'] = kwargs.get(
-            'padding', False if mode == 'inference' else 'max_length')
+            'padding', False if mode == ModeKeys.INFERENCE else 'max_length')
         kwargs['max_length'] = kwargs.pop('sequence_length', 128)
         super().__init__(model_dir, pair=True, mode=mode, **kwargs)
 
@@ -228,7 +228,7 @@ class SingleSentenceClassificationPreprocessor(NLPTokenizerPreprocessorBase):
     def __init__(self, model_dir: str, mode=ModeKeys.INFERENCE, **kwargs):
         kwargs['truncation'] = kwargs.get('truncation', True)
         kwargs['padding'] = kwargs.get(
-            'padding', False if mode == 'inference' else 'max_length')
+            'padding', False if mode == ModeKeys.INFERENCE else 'max_length')
         kwargs['max_length'] = kwargs.pop('sequence_length', 128)
         super().__init__(model_dir, pair=False, mode=mode, **kwargs)
 
@@ -309,7 +309,7 @@ class TextGenerationPreprocessor(NLPTokenizerPreprocessorBase):
         return super().build_tokenizer(model_dir)
 
     def __call__(self, data: Union[Dict, str]) -> Dict[str, Any]:
-        if self._mode == 'inference':
+        if self._mode == ModeKeys.INFERENCE:
             return super().__call__(data)
         src_txt = data['src_txt']
         tgt_txt = data['tgt_txt']
@@ -420,6 +420,7 @@ class TokenClassificationPreprocessor(NLPTokenizerPreprocessorBase):
         elif isinstance(data, dict):
             text_a = data.get(self.first_sequence)
             labels_list = data.get(self.label)
+        text_a = text_a.replace(' ', '').strip()
         tokenized_inputs = self.tokenizer(
             text_a,
             return_tensors='pt' if self._mode == ModeKeys.INFERENCE else None,

tests/pipelines/test_sentiment_classification.py

@@ -12,7 +12,7 @@ from modelscope.utils.test_utils import test_level
 
 
 class SentimentClassificationTest(unittest.TestCase):
-    model_id = 'damo/nlp_structbert_sentiment-classification_chinese-base'
+    model_id = 'damo/nlp_structbert_sentiment-classification_chinese-tiny'
     sentence1 = '启动的时候很大声音，然后就会听到1.2秒的卡察的声音，类似齿轮摩擦的声音'
 
     @unittest.skipUnless(test_level() >= 2, 'skip test in current test level')