Fixed bytes encoding for pandas codec (#2117)

Author: RobertSamoilescu

mlserver/codecs/pandas.py

@@ -81,11 +81,11 @@ def _process_bytes(
     content_type: Optional[str] = StringCodec.ContentType
     for elem in data:
         converted = elem
-        if not isinstance(elem, str):
+        if not isinstance(elem, (str, bytes)):
             # There was a non-string element, so we can't determine a content
             # type
             content_type = None
-        elif use_bytes:
+        elif isinstance(elem, str) and use_bytes:
             converted = encode_str(elem)
 
         processed.append(converted)

tests/codecs/test_pandas.py

@@ -32,6 +32,28 @@ def test_can_encode(payload: Any, expected: bool):
 @pytest.mark.parametrize(
     "series, use_bytes, expected",
     [
+        (
+            pd.Series(data=[b"hey", b"abc"], name="foo"),
+            True,
+            ResponseOutput(
+                name="foo",
+                shape=[2, 1],
+                data=[b"hey", b"abc"],
+                datatype="BYTES",
+                parameters=Parameters(content_type=StringCodec.ContentType),
+            ),
+        ),
+        (
+            pd.Series(data=[b"hey", b"abc"], name="foo"),
+            False,
+            ResponseOutput(
+                name="foo",
+                shape=[2, 1],
+                data=[b"hey", b"abc"],
+                datatype="BYTES",
+                parameters=Parameters(content_type=StringCodec.ContentType),
+            ),
+        ),
         (
             pd.Series(data=["hey", "abc"], name="foo"),
             True,