Preprocessing decorator fixes (#1843)

* Fix handling bytesting input to tokenizers, preprocessing

* Fix no convert scope in multithreaded contexts

keras_nlp/src/utils/tensor_utils.py

@@ -30,20 +30,19 @@
 
 
 NO_CONVERT_COUNTER = threading.local()
-NO_CONVERT_COUNTER.count = 0
 
 
 @contextlib.contextmanager
 def no_convert_scope():
     try:
-        NO_CONVERT_COUNTER.count += 1
+        NO_CONVERT_COUNTER.count = getattr(NO_CONVERT_COUNTER, "count", 0) + 1
         yield
     finally:
-        NO_CONVERT_COUNTER.count -= 1
+        NO_CONVERT_COUNTER.count = getattr(NO_CONVERT_COUNTER, "count", 0) - 1
 
 
 def in_no_convert_scope():
-    return NO_CONVERT_COUNTER.count > 0
+    return getattr(NO_CONVERT_COUNTER, "count", 0) > 0
 
 
 def preprocessing_function(fn):
@@ -119,7 +118,7 @@ def convert_preprocessing_inputs(x):
         return {k: convert_preprocessing_inputs(x[k]) for k, v in x.items()}
     if isinstance(x, tuple):
         return tuple(convert_preprocessing_inputs(v) for v in x)
-    if isinstance(x, str):
+    if isinstance(x, (str, bytes)):
         return tf.constant(x)
     if isinstance(x, list):
         try:
@@ -132,7 +131,7 @@ def convert_preprocessing_inputs(x):
                 # If ragged conversion failed return to the numpy error.
                 raise e
         # If we have a string input, use tf.tensor.
-        if numpy_x.dtype.type is np.str_:
+        if numpy_x.dtype.type is np.str_ or numpy_x.dtype.type is np.bytes_:
             return tf.convert_to_tensor(x)
         # Numpy will default to int64, int32 works with more ops.
         if numpy_x.dtype == np.int64:

keras_nlp/src/utils/tensor_utils_test.py

@@ -49,6 +49,17 @@ def test_strings(self):
         self.assertIsInstance(outputs, list)
         self.assertEqual(outputs, inputs)
 
+    def test_bytestrings(self):
+        inputs = ["one".encode("utf-8"), "two".encode("utf-8")]
+        # Convert to tf.
+        outputs = convert_preprocessing_inputs(inputs)
+        self.assertIsInstance(outputs, tf.Tensor)
+        self.assertAllEqual(outputs, tf.constant(inputs))
+        # Convert from tf.
+        outputs = convert_preprocessing_outputs(outputs)
+        self.assertIsInstance(outputs, list)
+        self.assertEqual(outputs, [x.decode("utf-8") for x in inputs])
+
     def test_ragged(self):
         inputs = [np.ones((1, 3)), np.ones((1, 2))]
         # Convert to tf.