Replace as_target context managers by direct calls (#18325)

* Preliminary work on tokenizers

* Quality + fix tests

* Treat processors

* Fix pad

* Remove all uses of  in tests, docs and examples

* Replace all as_target_tokenizer

* Fix tests

* Fix quality

* Update examples/flax/image-captioning/run_image_captioning_flax.py

Co-authored-by: amyeroberts <amy@huggingface.co>

* Style

Co-authored-by: amyeroberts <amy@huggingface.co>

examples/research_projects/robust-speech-event/run_speech_recognition_ctc_bnb.py

@@ -304,13 +304,12 @@ class DataCollatorCTCWithPadding:
             return_tensors="pt",
         )
 
-        with self.processor.as_target_processor():
-            labels_batch = self.processor.pad(
-                label_features,
-                padding=self.padding,
-                pad_to_multiple_of=self.pad_to_multiple_of_labels,
-                return_tensors="pt",
-            )
+        labels_batch = self.processor.pad(
+            labels=label_features,
+            padding=self.padding,
+            pad_to_multiple_of=self.pad_to_multiple_of_labels,
+            return_tensors="pt",
+        )
 
         # replace padding with -100 to ignore loss correctly
         labels = labels_batch["input_ids"].masked_fill(labels_batch.attention_mask.ne(1), -100)

examples/research_projects/robust-speech-event/run_speech_recognition_ctc_streaming.py

@@ -301,13 +301,12 @@ class DataCollatorCTCWithPadding:
             return_tensors="pt",
         )
 
-        with self.processor.as_target_processor():
-            labels_batch = self.processor.pad(
-                label_features,
-                padding=self.padding,
-                pad_to_multiple_of=self.pad_to_multiple_of_labels,
-                return_tensors="pt",
-            )
+        labels_batch = self.processor.pad(
+            labels=label_features,
+            padding=self.padding,
+            pad_to_multiple_of=self.pad_to_multiple_of_labels,
+            return_tensors="pt",
+        )
 
         # replace padding with -100 to ignore loss correctly
         labels = labels_batch["input_ids"].masked_fill(labels_batch.attention_mask.ne(1), -100)

examples/research_projects/tapex/run_wikisql_with_tapex.py

@@ -437,13 +437,12 @@ def main():
                 table=tables, query=questions, max_length=data_args.max_source_length, padding=padding, truncation=True
             )
 
-        with tokenizer.as_target_tokenizer():
-            labels = tokenizer(
-                answer=[", ".join(answer) for answer in answers],
-                max_length=max_target_length,
-                padding=padding,
-                truncation=True,
-            )
+        labels = tokenizer(
+            answer=[", ".join(answer) for answer in answers],
+            max_length=max_target_length,
+            padding=padding,
+            truncation=True,
+        )
 
         # If we are padding here, replace all tokenizer.pad_token_id in the labels by -100 when we want to ignore
         # padding in the loss.

examples/research_projects/tapex/run_wikitablequestions_with_tapex.py

@@ -413,13 +413,12 @@ def main():
                 table=tables, query=questions, max_length=data_args.max_source_length, padding=padding, truncation=True
             )
 
-        with tokenizer.as_target_tokenizer():
-            labels = tokenizer(
-                answer=[", ".join(answer) for answer in answers],
-                max_length=max_target_length,
-                padding=padding,
-                truncation=True,
-            )
+        labels = tokenizer(
+            answer=[", ".join(answer) for answer in answers],
+            max_length=max_target_length,
+            padding=padding,
+            truncation=True,
+        )
 
         # If we are padding here, replace all tokenizer.pad_token_id in the labels by -100 when we want to ignore
         # padding in the loss.

examples/research_projects/wav2vec2/run_asr.py

@@ -266,14 +266,13 @@ class DataCollatorCTCWithPadding:
             pad_to_multiple_of=self.pad_to_multiple_of,
             return_tensors="pt",
         )
-        with self.processor.as_target_processor():
-            labels_batch = self.processor.pad(
-                label_features,
-                padding=self.padding,
-                max_length=self.max_length_labels,
-                pad_to_multiple_of=self.pad_to_multiple_of_labels,
-                return_tensors="pt",
-            )
+        labels_batch = self.processor.pad(
+            labels=label_features,
+            padding=self.padding,
+            max_length=self.max_length_labels,
+            pad_to_multiple_of=self.pad_to_multiple_of_labels,
+            return_tensors="pt",
+        )
 
         # replace padding with -100 to ignore loss correctly
         labels = labels_batch["input_ids"].masked_fill(labels_batch.attention_mask.ne(1), -100)
@@ -419,9 +418,10 @@ def main():
             len(set(batch["sampling_rate"])) == 1
         ), f"Make sure all inputs have the same sampling rate of {processor.feature_extractor.sampling_rate}."
 
-        batch["input_values"] = processor(batch["speech"], sampling_rate=batch["sampling_rate"][0]).input_values
-        with processor.as_target_processor():
-            batch["labels"] = processor(batch[data_args.target_text_column]).input_ids
+        processed_batch = processor(
+            audio=batch["speech"], text=batch[data_args.target_text_column], sampling_rate=batch["sampling_rate"][0]
+        )
+        batch.update(processed_batch)
         return batch
 
     train_dataset = train_dataset.map(

examples/research_projects/wav2vec2/run_common_voice.py

@@ -185,14 +185,13 @@ class DataCollatorCTCWithPadding:
             pad_to_multiple_of=self.pad_to_multiple_of,
             return_tensors="pt",
         )
-        with self.processor.as_target_processor():
-            labels_batch = self.processor.pad(
-                label_features,
-                padding=self.padding,
-                max_length=self.max_length_labels,
-                pad_to_multiple_of=self.pad_to_multiple_of_labels,
-                return_tensors="pt",
-            )
+        labels_batch = self.processor.pad(
+            labels=label_features,
+            padding=self.padding,
+            max_length=self.max_length_labels,
+            pad_to_multiple_of=self.pad_to_multiple_of_labels,
+            return_tensors="pt",
+        )
 
         # replace padding with -100 to ignore loss correctly
         labels = labels_batch["input_ids"].masked_fill(labels_batch.attention_mask.ne(1), -100)
@@ -414,10 +413,11 @@ def main():
         assert (
             len(set(batch["sampling_rate"])) == 1
         ), f"Make sure all inputs have the same sampling rate of {processor.feature_extractor.sampling_rate}."
-        batch["input_values"] = processor(batch["speech"], sampling_rate=batch["sampling_rate"][0]).input_values
-        # Setup the processor for targets
-        with processor.as_target_processor():
-            batch["labels"] = processor(batch["target_text"]).input_ids
+
+        processed_batch = processor(
+            audio=batch["speech"], text=batch["target_text"], sampling_rate=batch["sampling_rate"][0]
+        )
+        batch.update(processed_batch)
         return batch
 
     train_dataset = train_dataset.map(

examples/research_projects/xtreme-s/run_xtreme_s.py

@@ -349,13 +349,12 @@ class SpeechDataCollatorWithPadding:
 
         if self.pad_labels:
             label_features = [{"input_ids": feature["labels"]} for feature in features]
-            with self.processor.as_target_processor():
-                labels_batch = self.processor.pad(
-                    label_features,
-                    padding=self.padding,
-                    pad_to_multiple_of=self.pad_to_multiple_of_labels,
-                    return_tensors="pt",
-                )
+            labels_batch = self.processor.pad(
+                labels=label_features,
+                padding=self.padding,
+                pad_to_multiple_of=self.pad_to_multiple_of_labels,
+                return_tensors="pt",
+            )
 
             # replace padding with -100 to ignore loss correctly
             labels = labels_batch["input_ids"].masked_fill(labels_batch.attention_mask.ne(1), -100)