Fix CI for metrics

Author: albertvillanova

setup.py

@@ -197,7 +197,7 @@
     "jiwer",
     "langdetect",
     "mauve-text",
-    "nltk",
+    "nltk<3.8.2",
     "rouge_score",
     "sacrebleu",
     "sacremoses",

tests/test_inspect.py

@@ -29,7 +29,7 @@ def test_inspect_dataset(path, tmp_path):
 @pytest.mark.filterwarnings("ignore:metric_module_factory is deprecated:FutureWarning")
 @pytest.mark.parametrize("path", ["accuracy"])
 def test_inspect_metric(path, tmp_path):
-    inspect_metric(path, tmp_path, trust_remote_code=True)
+    inspect_metric(path, tmp_path, trust_remote_code=True, revision="2.21")
     script_name = path + ".py"
     assert script_name in os.listdir(tmp_path)
     assert "__pycache__" not in os.listdir(tmp_path)

tests/test_load.py

@@ -452,6 +452,7 @@ def test_GithubMetricModuleFactory_with_internal_import(self):
         # "squad_v2" requires additional imports (internal)
         factory = GithubMetricModuleFactory(
             "squad_v2",
+            revision="2.21",
             download_config=self.download_config,
             dynamic_modules_path=self.dynamic_modules_path,
             trust_remote_code=True,
@@ -464,6 +465,7 @@ def test_GithubMetricModuleFactory_with_external_import(self):
         # "bleu" requires additional imports (external from github)
         factory = GithubMetricModuleFactory(
             "bleu",
+            revision="2.21",
             download_config=self.download_config,
             dynamic_modules_path=self.dynamic_modules_path,
             trust_remote_code=True,

tests/test_metric.py

@@ -10,7 +10,7 @@
 from datasets.features import Features, Sequence, Value
 from datasets.metric import Metric, MetricInfo
 
-from .utils import require_tf, require_torch
+from .utils import require_numpy1_on_windows, require_tf, require_torch
 
 
 class DummyMetric(Metric):
@@ -433,6 +433,7 @@ def test_input_numpy(self):
         self.assertDictEqual(expected_results, metric.compute())
         del metric
 
+    @require_numpy1_on_windows
     @require_torch
     def test_input_torch(self):
         import torch