Various fixes based on messages

docs/source/tasks/nlp/token_classification.rst

@@ -35,7 +35,7 @@ Training
         revision="master",
         tokenizer=tokenizer,
     )
-    model = TokenClassificationTransformer(pretrained_model_name_or_path="bert-base-uncased", labels=dm.labels)
+    model = TokenClassificationTransformer(pretrained_model_name_or_path="bert-base-uncased", labels=dm.num_classes)
     trainer = pl.Trainer(accelerator="auto", devices="auto", max_epochs=1)
 
     trainer.fit(model, dm)

examples/token_classification.py

@@ -17,7 +17,7 @@
         revision="master",
         tokenizer=tokenizer,
     )
-    model = TokenClassificationTransformer(pretrained_model_name_or_path="bert-base-uncased", labels=dm.labels)
+    model = TokenClassificationTransformer(pretrained_model_name_or_path="bert-base-uncased", labels=dm.num_classes)
     trainer = pl.Trainer(accelerator="auto", devices="auto", max_epochs=1)
 
     trainer.fit(model, dm)

examples/translation_wmt.py

@@ -4,7 +4,7 @@
 from lightning_transformers.task.nlp.translation import TranslationTransformer, WMT16TranslationDataModule
 
 if __name__ == "__main__":
-    tokenizer = AutoTokenizer.from_pretrained(pretrained_model_name_or_path="t5-base")
+    tokenizer = AutoTokenizer.from_pretrained(pretrained_model_name_or_path="t5-base", model_max_length=512)
     model = TranslationTransformer(
         pretrained_model_name_or_path="t5-base",
         n_gram=4,
@@ -20,6 +20,7 @@
         target_language="ro",
         max_source_length=128,
         max_target_length=128,
+        padding="max_length",
         tokenizer=tokenizer,
     )
     trainer = pl.Trainer(accelerator="auto", devices="auto", max_epochs=1)

lightning_transformers/core/data.py

@@ -212,12 +212,3 @@ def predict_dataloader(self) -> Optional[DataLoader]:
     @property
     def collate_fn(self) -> Optional[Callable]:
         return None
-
-    @property
-    def model_data_kwargs(self) -> Dict:
-        """Override to provide the model with additional kwargs.
-
-        This is useful to provide the number of classes/pixels to the model or any other data specific args
-        Returns: Dict of args
-        """
-        return {}

lightning_transformers/core/model.py

@@ -42,7 +42,6 @@ class TaskTransformer(pl.LightningModule):
         pretrained_model_name_or_path: Huggingface model to use if backbone config not passed.
         tokenizer: The pre-trained tokenizer.
         pipeline_kwargs: Arguments required for the HuggingFace inference pipeline class.
-        **model_data_kwargs: Arguments passed from the data module to the class.
     """
 
     def __init__(

lightning_transformers/core/seq2seq/data.py

@@ -18,10 +18,9 @@ class Seq2SeqDataModule(TransformerDataModule):
     def __init__(
         self, *args, max_target_length: int = 128, max_source_length: int = 1024, padding: str = "longest", **kwargs
     ) -> None:
-        super().__init__(*args, **kwargs)
+        super().__init__(*args, padding=padding, **kwargs)
         self.max_target_length = max_target_length
         self.max_source_length = max_source_length
-        self.padding = padding
 
     def process_data(self, dataset: Dataset, stage: Optional[str] = None) -> Dataset:
         src_text_column_name, tgt_text_column_name = self.source_target_column_names
@@ -60,14 +59,16 @@ def convert_to_features(
         src_text_column_name: str,
         tgt_text_column_name: str,
     ):
-        encoded_results = tokenizer.prepare_seq2seq_batch(
-            src_texts=examples[src_text_column_name],
-            tgt_texts=examples[tgt_text_column_name],
-            max_length=max_source_length,
-            max_target_length=max_target_length,
-            padding=padding,
-        )
-        return encoded_results
+        inputs = examples[src_text_column_name]
+        targets = examples[tgt_text_column_name]
+        model_inputs = tokenizer(inputs, max_length=max_source_length, padding=padding, truncation=True)
+
+        # Setup the tokenizer for targets
+        with tokenizer.as_target_tokenizer():
+            labels = tokenizer(targets, max_length=max_target_length, padding=padding, truncation=True)
+
+        model_inputs["labels"] = labels["input_ids"]
+        return model_inputs
 
     @property
     def collate_fn(self) -> Callable:

lightning_transformers/task/nlp/multiple_choice/data.py

@@ -11,7 +11,6 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-from typing import Dict
 
 from transformers import default_data_collator
 
@@ -35,7 +34,3 @@ def collate_fn(self) -> callable:
     @property
     def num_classes(self) -> int:
         raise NotImplementedError
-
-    @property
-    def model_data_kwargs(self) -> Dict[str, int]:
-        return {"num_labels": self.num_classes}

lightning_transformers/task/nlp/text_classification/data.py

@@ -11,7 +11,7 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-from typing import Any, Dict, List, Optional
+from typing import Any, List, Optional
 
 from datasets import ClassLabel, Dataset
 from pytorch_lightning.utilities import rank_zero_warn
@@ -53,10 +53,6 @@ def num_classes(self) -> int:
             self.setup("fit")
         return self.labels.num_classes
 
-    @property
-    def model_data_kwargs(self) -> Dict[str, int]:
-        return {"num_labels": self.num_classes}
-
     @staticmethod
     def convert_to_features(
         example_batch: Any, _, tokenizer: PreTrainedTokenizerBase, input_feature_fields: List[str], **tokenizer_kwargs

lightning_transformers/task/nlp/token_classification/data.py

@@ -12,7 +12,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 from functools import partial
-from typing import Any, Callable, Dict, Optional
+from typing import Any, Callable, Optional
 
 from datasets import ClassLabel, Dataset
 from pytorch_lightning.utilities import rank_zero_warn
@@ -90,10 +90,6 @@ def num_classes(self) -> int:
             self.setup("fit")
         return len(self.labels)
 
-    @property
-    def model_data_kwargs(self) -> Dict[str, Any]:
-        return {"labels": self.labels}
-
     @staticmethod
     def convert_to_features(
         examples: Any,

lightning_transformers/task/nlp/translation/datasets/wmt16.py

@@ -32,16 +32,13 @@ def convert_to_features(
         src_text_column_name: str,
         tgt_text_column_name: str,
     ):
-        translations = examples["translation"]  # Extract translations from dict
+        inputs = [ex[src_text_column_name] for ex in examples["translation"]]
+        targets = [ex[tgt_text_column_name] for ex in examples["translation"]]
+        model_inputs = tokenizer(inputs, max_length=max_source_length, padding=padding, truncation=True)
 
-        def extract_text(lang):
-            return [text[lang] for text in translations]
+        # Setup the tokenizer for targets
+        with tokenizer.as_target_tokenizer():
+            labels = tokenizer(targets, max_length=max_target_length, padding=padding, truncation=True)
 
-        encoded_results = tokenizer.prepare_seq2seq_batch(
-            src_texts=extract_text(src_text_column_name),
-            tgt_texts=extract_text(tgt_text_column_name),
-            max_length=max_source_length,
-            max_target_length=max_target_length,
-            padding=padding,
-        )
-        return encoded_results
+        model_inputs["labels"] = labels["input_ids"]
+        return model_inputs

lightning_transformers/task/nlp/translation/model.py

@@ -53,7 +53,7 @@ def compute_generate_metrics(self, batch, prefix):
         tgt_lns = self.tokenize_labels(batch["labels"])
         pred_lns = self.generate(batch["input_ids"], batch["attention_mask"])
         # wrap targets in list as score expects a list of potential references
-        result = self.bleu(pred_lns, tgt_lns)
+        result = self.bleu(preds=pred_lns, target=tgt_lns)
         self.log(f"{prefix}_bleu_score", result, on_step=False, on_epoch=True, prog_bar=True)
 
     def configure_metrics(self, stage: str):

lightning_transformers/task/vision/image_classification/data.py

@@ -11,7 +11,7 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-from typing import Any, Dict, Optional
+from typing import Any, Optional
 
 from datasets import ClassLabel, Dataset
 from pytorch_lightning.utilities import rank_zero_warn
@@ -52,7 +52,3 @@ def num_classes(self) -> int:
             rank_zero_warn("Labels has not been set, calling `setup('fit')`.")
             self.setup("fit")
         return self.labels.num_classes
-
-    @property
-    def model_data_kwargs(self) -> Dict[str, int]:
-        return {"num_labels": self.num_classes}