Fix sequence_length float coercion in ByteTokenizer

Author: maitry63

keras_hub/src/tokenizers/byte_tokenizer.py

@@ -179,11 +179,21 @@ def __init__(
                     "`sequence_length` must be an int, got bool: "
                     f"{sequence_length}"
                 )
-            if not isinstance(sequence_length, numbers.Integral):
+
+            if isinstance(sequence_length, float):
+                if not sequence_length.is_integer():
+                    raise ValueError(
+                        "`sequence_length` must be a whole number. "
+                        f"Received: {sequence_length}"
+                    )
+                sequence_length = int(sequence_length)
+
+            elif not isinstance(sequence_length, numbers.Integral):
                 raise ValueError(
                     "`sequence_length` must be an int or None. "
-                    f"{sequence_length}"
+                    f"Received: {sequence_length}"
                 )
+
             if sequence_length <= 0:
                 raise ValueError(
                     "`sequence_length` must be > 0. "

keras_hub/src/tokenizers/byte_tokenizer_test.py

@@ -222,8 +222,8 @@ def test_sequence_length_valid_int(self):
         self.assertEqual(tokenizer.sequence_length, 5)
 
     def test_sequence_length_valid_float(self):
-        with self.assertRaises(ValueError):
-            ByteTokenizer(sequence_length=5.0)
+        tokenizer = ByteTokenizer(sequence_length=5.0)
+        self.assertEqual(tokenizer.sequence_length, 5)
 
     def test_sequence_length_invalid_float(self):
         with self.assertRaises(ValueError):