Add val_loss and test_loss calculation and logging for QnA task (#832)

flash/text/question_answering/data.py

@@ -87,8 +87,16 @@ def _tokenize_fn(self, samples: Any) -> Callable:
 
         if stage == RunningStage.TRAINING:
             # Preprocess function for training
-            tokenized_samples = self._prepare_train_features(samples, tokenized_samples)
+            tokenized_samples, _, _ = self._prepare_train_features(samples, tokenized_samples)
         elif self._running_stage.evaluating or stage == RunningStage.PREDICTING:
+            if self._running_stage.evaluating:
+                tokenized_samples, _sample_mapping, _offset_mapping = self._prepare_train_features(
+                    samples, tokenized_samples
+                )
+
+                tokenized_samples["overflow_to_sample_mapping"] = _sample_mapping
+                tokenized_samples["offset_mapping"] = _offset_mapping
+
             # Preprocess function for eval or predict
             tokenized_samples = self._prepare_val_features(samples, tokenized_samples)
 
@@ -169,7 +177,7 @@ def _prepare_train_features(self, samples: Any, tokenized_samples: Any):
                         token_end_index -= 1
                     tokenized_samples["end_positions"].append(token_end_index + 1)
 
-        return tokenized_samples
+        return tokenized_samples, sample_mapping, offset_mapping
 
     def _prepare_val_features(self, samples: Any, tokenized_samples: Any):
         # Since one example might give us several features if it has a long context, we need a map from a feature to

flash/text/question_answering/model.py

@@ -251,12 +251,13 @@ def _generate_answers(self, pred_start_logits, pred_end_logits, examples):
     def forward(self, batch: Any) -> Any:
         metadata = batch.pop(DefaultDataKeys.METADATA)
         outputs = self.model(**batch)
+        loss = outputs.loss
         start_logits = outputs.start_logits
         end_logits = outputs.end_logits
 
         generated_answers = self._generate_answers(start_logits, end_logits, metadata)
         batch[DefaultDataKeys.METADATA] = metadata
-        return generated_answers
+        return loss, generated_answers
 
     def training_step(self, batch: Any, batch_idx: int) -> Tensor:
         outputs = self.model(**batch)
@@ -265,9 +266,10 @@ def training_step(self, batch: Any, batch_idx: int) -> Tensor:
         return loss
 
     def common_step(self, prefix: str, batch: Any) -> torch.Tensor:
-        generated_answers = self(batch)
+        loss, generated_answers = self(batch)
         result = self.compute_metrics(generated_answers, batch[DefaultDataKeys.METADATA])
-        self.log_dict(result, on_step=False, on_epoch=True, prog_bar=True)
+        self.log(f"{prefix}_loss", loss, on_step=False, on_epoch=True, prog_bar=True)
+        self.log_dict(result, on_step=False, on_epoch=True, prog_bar=False)
 
     def compute_metrics(self, generated_tokens, batch):
         for example in batch:
@@ -285,7 +287,7 @@ def test_step(self, batch: Any, batch_idx: int, dataloader_idx: int = 0):
         self.common_step("test", batch)
 
     def predict_step(self, batch: Any, batch_idx: int, dataloader_idx: int = 0) -> Any:
-        generated_answers = self(batch)
+        _, generated_answers = self(batch)
         return generated_answers
 
     @property

tests/text/question_answering/test_data.py

@@ -136,6 +136,8 @@ def test_from_files(tmpdir):
     batch = next(iter(dm.val_dataloader()))
     assert "input_ids" in batch
     assert "attention_mask" in batch
+    assert "start_positions" in batch
+    assert "end_positions" in batch
     assert DefaultDataKeys.METADATA in batch
     assert "context" in batch[DefaultDataKeys.METADATA][0]
     assert "answer" in batch[DefaultDataKeys.METADATA][0]
@@ -145,6 +147,8 @@ def test_from_files(tmpdir):
     batch = next(iter(dm.test_dataloader()))
     assert "input_ids" in batch
     assert "attention_mask" in batch
+    assert "start_positions" in batch
+    assert "end_positions" in batch
     assert DefaultDataKeys.METADATA in batch
     assert "context" in batch[DefaultDataKeys.METADATA][0]
     assert "answer" in batch[DefaultDataKeys.METADATA][0]