Expose padding_strategy on squad processor to fix QA pipeline performance regression (#5932)

* Attempt to fix the way squad_convert_examples_to_features pad the elements for the QA pipeline.

Signed-off-by: Morgan Funtowicz <funtowiczmo@gmail.com>

* Quality

Signed-off-by: Morgan Funtowicz <funtowiczmo@gmail.com>

* Make the code easier to read and avoid testing multiple test the same thing.

Signed-off-by: Morgan Funtowicz <funtowiczmo@gmail.com>

* missing enum value on truncation_strategy.

Signed-off-by: Morgan Funtowicz <funtowiczmo@gmail.com>

* Rethinking for the easiest fix: expose the padding strategy on squad_convert_examples_to_features.

Signed-off-by: Morgan Funtowicz <funtowiczmo@gmail.com>

* Remove unused imports.

Signed-off-by: Morgan Funtowicz <funtowiczmo@gmail.com>

src/transformers/data/processors/squad.py

@@ -9,6 +9,7 @@ from tqdm import tqdm
 
 from ...file_utils import is_tf_available, is_torch_available
 from ...tokenization_bert import whitespace_tokenize
+from ...tokenization_utils_base import TruncationStrategy
 from .utils import DataProcessor
 
 
@@ -87,7 +88,9 @@ def _is_whitespace(c):
     return False
 
 
-def squad_convert_example_to_features(example, max_seq_length, doc_stride, max_query_length, is_training):
+def squad_convert_example_to_features(
+    example, max_seq_length, doc_stride, max_query_length, padding_strategy, is_training
+):
     features = []
     if is_training and not example.is_impossible:
         # Get start and end position
@@ -141,11 +144,21 @@ def squad_convert_example_to_features(example, max_seq_length, doc_stride, max_q
     span_doc_tokens = all_doc_tokens
     while len(spans) * doc_stride < len(all_doc_tokens):
 
+        # Define the side we want to truncate / pad and the text/pair sorting
+        if tokenizer.padding_side == "right":
+            texts = truncated_query
+            pairs = span_doc_tokens
+            truncation = TruncationStrategy.ONLY_SECOND.value
+        else:
+            texts = span_doc_tokens
+            pairs = truncated_query
+            truncation = TruncationStrategy.ONLY_FIRST.value
+
         encoded_dict = tokenizer.encode_plus(  # TODO(thom) update this logic
-            truncated_query if tokenizer.padding_side == "right" else span_doc_tokens,
-            span_doc_tokens if tokenizer.padding_side == "right" else truncated_query,
-            truncation="only_second" if tokenizer.padding_side == "right" else "only_first",
-            padding="max_length",
+            texts,
+            pairs,
+            truncation=truncation,
+            padding=padding_strategy,
             max_length=max_seq_length,
             return_overflowing_tokens=True,
             stride=max_seq_length - doc_stride - len(truncated_query) - sequence_pair_added_tokens,
@@ -285,6 +298,7 @@ def squad_convert_examples_to_features(
     doc_stride,
     max_query_length,
     is_training,
+    padding_strategy="max_length",
     return_dataset=False,
     threads=1,
     tqdm_enabled=True,
@@ -300,6 +314,7 @@ def squad_convert_examples_to_features(
         doc_stride: The stride used when the context is too large and is split across several features.
         max_query_length: The maximum length of the query.
         is_training: whether to create features for model evaluation or model training.
+        padding_strategy: Default to "max_length". Which padding strategy to use
         return_dataset: Default False. Either 'pt' or 'tf'.
             if 'pt': returns a torch.data.TensorDataset,
             if 'tf': returns a tf.data.Dataset
@@ -333,6 +348,7 @@ def squad_convert_examples_to_features(
             max_seq_length=max_seq_length,
             doc_stride=doc_stride,
             max_query_length=max_query_length,
+            padding_strategy=padding_strategy,
             is_training=is_training,
         )
         features = list(

src/transformers/pipelines.py

@@ -36,6 +36,7 @@ from .modelcard import ModelCard
 from .tokenization_auto import AutoTokenizer
 from .tokenization_bert import BasicTokenizer
 from .tokenization_utils import PreTrainedTokenizer
+from .tokenization_utils_base import PaddingStrategy
 
 
 if is_tf_available():
@@ -1318,6 +1319,7 @@ class QuestionAnsweringPipeline(Pipeline):
                 max_seq_length=kwargs["max_seq_len"],
                 doc_stride=kwargs["doc_stride"],
                 max_query_length=kwargs["max_question_len"],
+                padding_strategy=PaddingStrategy.DO_NOT_PAD.value,
                 is_training=False,
                 tqdm_enabled=False,
             )