Fix TrainingArguments.torch_empty_cache_steps post_init check (#36734)

Mistaken use of De Morgan's law. Fixed "not (X or Y)"
to correct "not (X and Y)" check to raise a ValueError.

Added corresponding test to check "positive int or None" condition.

Co-authored-by: Yih-Dar <2521628+ydshieh@users.noreply.github.com>

src/transformers/training_args.py

@@ -1641,7 +1641,7 @@ class TrainingArguments:
             self.do_eval = True
 
         if self.torch_empty_cache_steps is not None:
-            if not (isinstance(self.torch_empty_cache_steps, int) or self.torch_empty_cache_steps > 0):
+            if not (isinstance(self.torch_empty_cache_steps, int) and self.torch_empty_cache_steps > 0):
                 raise ValueError(
                     f"`torch_empty_cache_steps` must be an integer bigger than 0, got {self.torch_empty_cache_steps}."
                 )

tests/test_training_args.py

@@ -40,3 +40,28 @@ class TestTrainingArguments(unittest.TestCase):
             self.assertFalse(os.path.exists(output_dir))  # Still shouldn't exist
 
             # Directory should be created when actually needed (e.g. in Trainer)
+
+    def test_torch_empty_cache_steps_requirements(self):
+        """Test that torch_empty_cache_steps is a positive integer or None."""
+
+        # None is acceptable (feature is disabled):
+        args = TrainingArguments(torch_empty_cache_steps=None)
+        self.assertIsNone(args.torch_empty_cache_steps)
+
+        # non-int is unacceptable:
+        with self.assertRaises(ValueError):
+            TrainingArguments(torch_empty_cache_steps=1.0)
+        with self.assertRaises(ValueError):
+            TrainingArguments(torch_empty_cache_steps="none")
+
+        # negative int is unacceptable:
+        with self.assertRaises(ValueError):
+            TrainingArguments(torch_empty_cache_steps=-1)
+
+        # zero is unacceptable:
+        with self.assertRaises(ValueError):
+            TrainingArguments(torch_empty_cache_steps=0)
+
+        # positive int is acceptable:
+        args = TrainingArguments(torch_empty_cache_steps=1)
+        self.assertEqual(args.torch_empty_cache_steps, 1)