Add possibility to control the truncation in bert score

CHANGELOG.md

@@ -24,6 +24,9 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - Added multi-output support for MAE metric ([#2605](https://github.com/Lightning-AI/torchmetrics/pull/2605))
 
 
+- Added `truncation` argument to `BERTScore` ([#2776](https://github.com/Lightning-AI/torchmetrics/pull/2776))
+
+
 ### Changed
 
 - Tracker higher is better integration ([#2649](https://github.com/Lightning-AI/torchmetrics/pull/2649))

src/torchmetrics/functional/text/bert.py

@@ -276,6 +276,7 @@ def bert_score(
     rescale_with_baseline: bool = False,
     baseline_path: Optional[str] = None,
     baseline_url: Optional[str] = None,
+    truncation: bool = False,
 ) -> Dict[str, Union[Tensor, List[float], str]]:
     """`Bert_score Evaluating Text Generation`_ for text similirity matching.
 
@@ -323,6 +324,7 @@ def bert_score(
             of the files from `BERT_score`_
         baseline_path: A path to the user's own local csv/tsv file with the baseline scale.
         baseline_url: A url path to the user's own  csv/tsv file with the baseline scale.
+        truncation: An indication of whether the input sequences should be truncated to the maximum length.
 
     Returns:
         Python dictionary containing the keys ``precision``, ``recall`` and ``f1`` with corresponding values.
@@ -417,13 +419,14 @@ def bert_score(
 
     # We ignore mypy typing below as the proper typing is ensured by conditions above, only mypy cannot infer that.
     if _are_valid_lists:
-        target_dataset = TextDataset(target, tokenizer, max_length, idf=idf)  # type: ignore
+        target_dataset = TextDataset(target, tokenizer, max_length, idf=idf, truncation=truncation)  # type: ignore
         preds_dataset = TextDataset(
             preds,  # type: ignore
             tokenizer,
             max_length,
             idf=idf,
             tokens_idf=target_dataset.tokens_idf,
+            truncation=truncation,
         )
     elif _are_valid_tensors:
         target_dataset = TokenizedDataset(**target, idf=idf)  # type: ignore

src/torchmetrics/functional/text/helper_embedding_metric.py

@@ -199,6 +199,7 @@ def __init__(
         ] = _preprocess_text,
         idf: bool = False,
         tokens_idf: Optional[Dict[int, float]] = None,
+        truncation: bool = False,
     ) -> None:
         """Initialize text dataset class.
 
@@ -209,9 +210,10 @@ def __init__(
             preprocess_text_fn: A function used for processing the input sentences.
             idf: An indication of whether calculate token inverse document frequencies to weight the model embeddings.
             tokens_idf: Inverse document frequencies (these should be calculated on reference sentences).
+            truncation: An indication of whether tokenized sequences should be padded only to the length of the longest
 
         """
-        _text = preprocess_text_fn(text, tokenizer, max_length)
+        _text = preprocess_text_fn(text, tokenizer, max_length, truncation)
         if isinstance(_text, tuple):
             self.text, self.sorting_indices = _text
         else:

src/torchmetrics/text/bert.py

@@ -107,6 +107,7 @@ class BERTScore(Metric):
             of the files from `BERT_score`_.
         baseline_path: A path to the user's own local csv/tsv file with the baseline scale.
         baseline_url: A url path to the user's own  csv/tsv file with the baseline scale.
+        truncation: An indication of whether the input sequences should be truncated to the ``max_length``.
         kwargs: Additional keyword arguments, see :ref:`Metric kwargs` for more info.
 
     Example:
@@ -150,6 +151,7 @@ def __init__(
         rescale_with_baseline: bool = False,
         baseline_path: Optional[str] = None,
         baseline_url: Optional[str] = None,
+        truncation: bool = False,
         **kwargs: Any,
     ) -> None:
         super().__init__(**kwargs)
@@ -169,6 +171,7 @@ def __init__(
         self.rescale_with_baseline = rescale_with_baseline
         self.baseline_path = baseline_path
         self.baseline_url = baseline_url
+        self.truncation = truncation
 
         if user_tokenizer:
             self.tokenizer = user_tokenizer
@@ -210,15 +213,15 @@ def update(self, preds: Union[str, Sequence[str]], target: Union[str, Sequence[s
             preds,
             self.tokenizer,
             self.max_length,
-            truncation=False,
+            truncation=self.truncation,
             sort_according_length=False,
             own_tokenizer=self.user_tokenizer,
         )
         target_dict, _ = _preprocess_text(
             target,
             self.tokenizer,
             self.max_length,
-            truncation=False,
+            truncation=self.truncation,
             sort_according_length=False,
             own_tokenizer=self.user_tokenizer,
         )

tests/unittests/text/test_bertscore.py

@@ -188,3 +188,20 @@ def test_bertscore_sorting(idf: bool):
 
     # First index should be the self-comparison - sorting by length should not shuffle this
     assert score["f1"][0] > score["f1"][1]
+
+
+@skip_on_connection_issues()
+@pytest.mark.skipif(not _TRANSFORMERS_GREATER_EQUAL_4_4, reason="test requires transformers>4.4")
+@pytest.mark.parametrize("truncation", [True, False])
+def test_bertscore_truncation(truncation: bool):
+    """Test that BERTScore truncation works as expected."""
+    pred = ["abc " * 2000]
+    gt = ["def " * 2000]
+    bert_score = BERTScore(truncation=truncation)
+
+    if truncation:
+        res = bert_score(pred, gt)
+        assert res["f1"] > 0.0
+    else:
+        with pytest.raises(RuntimeError, match="The expanded size of the tensor.*must match.*"):
+            bert_score(pred, gt)