Deprecate text metrics

flash/core/utilities/imports.py

@@ -92,9 +92,9 @@ def _compare_version(package: str, op, version) -> bool:
 _TORCH_SPARSE_AVAILABLE = _module_available("torch_sparse")
 _TORCH_GEOMETRIC_AVAILABLE = _module_available("torch_geometric")
 _TORCHAUDIO_AVAILABLE = _module_available("torchaudio")
-_ROUGE_SCORE_AVAILABLE = _module_available("rouge_score")
 _SENTENCEPIECE_AVAILABLE = _module_available("sentencepiece")
 _DATASETS_AVAILABLE = _module_available("datasets")
+_TM_TEXT_AVAILABLE: bool = _module_available("torchmetrics.text")
 _ICEVISION_AVAILABLE = _module_available("icevision")
 
 if Version:
@@ -103,9 +103,9 @@ def _compare_version(package: str, op, version) -> bool:
 _TEXT_AVAILABLE = all(
     [
         _TRANSFORMERS_AVAILABLE,
-        _ROUGE_SCORE_AVAILABLE,
         _SENTENCEPIECE_AVAILABLE,
         _DATASETS_AVAILABLE,
+        _TM_TEXT_AVAILABLE,
     ]
 )
 _TABULAR_AVAILABLE = _TABNET_AVAILABLE and _PANDAS_AVAILABLE

flash/text/seq2seq/core/metrics.py

@@ -18,6 +18,7 @@
 # Link: https://pytorch.org/text/_modules/torchtext/data/metrics.html#bleu_score
 from collections import Counter
 from typing import Dict, List, Tuple
+from warnings import warn
 
 import numpy as np
 import torch
@@ -64,6 +65,9 @@ class BLEUScore(Metric):
         >>> metric = BLEUScore()
         >>> metric(translate_corpus, reference_corpus)
         tensor(0.7598)
+
+    .. deprecated:: v0.5
+        Use :func:`torchmetrics.text.BLEUScore` instead. Will be removed in v0.6.
     """
 
     def __init__(self, n_gram: int = 4, smooth: bool = False):
@@ -72,6 +76,12 @@ def __init__(self, n_gram: int = 4, smooth: bool = False):
             n_gram: Gram value ranged from 1 to 4 (Default 4)
             smooth: Whether or not to apply smoothing – Lin et al. 2004
         """
+        warn(
+            "Metric `text.seq2seq.core.metrics.BLEUScore` is deprecated in v0.5 and will be removed in v0.6."
+            " Use `torchmetrics.text.BLEUScore` instead.",
+            DeprecationWarning,
+        )
+
         super().__init__()
         self.n_gram = n_gram
         self.smooth = smooth
@@ -151,6 +161,9 @@ class RougeMetric(Metric):
          'rougeLsum_fmeasure': 0.25,
          'rougeLsum_precision': 0.25,
          'rougeLsum_recall': 0.25}
+
+    .. deprecated:: v0.5
+        Use :func:`torchmetrics.text.ROUGEScore` instead. Will be removed in v0.6.
     """
 
     @requires_extras("text")
@@ -160,6 +173,11 @@ def __init__(
         use_stemmer: bool = False,
         rouge_keys: Tuple[str] = ("rouge1", "rouge2", "rougeL", "rougeLsum"),
     ):
+        warn(
+            "Metric `text.seq2seq.core.metrics.RougeScore` is deprecated in v0.5 and will be removed in v0.6."
+            " Use `torchmetrics.text.ROUGEScore` instead.",
+            DeprecationWarning,
+        )
         super().__init__()
 
         self.rouge_newline_sep = rouge_newline_sep

flash/text/seq2seq/summarization/model.py

@@ -14,9 +14,8 @@
 from typing import Any, Callable, Dict, List, Mapping, Optional, Sequence, Type, Union
 
 import torch
-from torchmetrics import Metric
+from torchmetrics import Metric, ROUGEScore
 
-from flash.text.seq2seq.core.metrics import RougeMetric
 from flash.text.seq2seq.core.model import Seq2SeqTask
 
 
@@ -66,8 +65,8 @@ def __init__(
             val_target_max_length=val_target_max_length,
             num_beams=num_beams,
         )
-        self.rouge = RougeMetric(
-            rouge_newline_sep=rouge_newline_sep,
+        self.rouge = ROUGEScore(
+            newline_sep=rouge_newline_sep,
             use_stemmer=use_stemmer,
         )
 

flash/text/seq2seq/translation/model.py

@@ -14,9 +14,8 @@
 from typing import Any, Callable, Dict, List, Mapping, Optional, Sequence, Type, Union
 
 import torch
-from torchmetrics import Metric
+from torchmetrics import BLEUScore, Metric
 
-from flash.text.seq2seq.core.metrics import BLEUScore
 from flash.text.seq2seq.core.model import Seq2SeqTask
 
 

requirements.txt

@@ -1,6 +1,6 @@
 packaging
 torch
-torchmetrics
+torchmetrics>=0.5.0
 pytorch-lightning>=1.4.0
 pyDeprecate
 pandas<1.3.0

requirements/datatype_text.txt

@@ -1,5 +1,5 @@
-rouge-score>=0.0.4
 sentencepiece>=0.1.95
 filelock
 transformers>=4.5
 datasets>=1.2, <1.3
+torchmetrics[text]