Fix intolerance for new field in TGI stream response: 'index' (#2006)

* Fixing intolerance for new TGI field 'index'

* make style

* fix test python38

---------

Co-authored-by: Lucain Pouget <lucainp@gmail.com>

src/huggingface_hub/inference/_text_generation.py

@@ -451,6 +451,8 @@ class TextGenerationStreamResponse:
     Args:
         token (`Token`):
             The generated token.
+        index (`Optional[int]`, *optional*):
+            The token index within the stream. Optional to support older clients that omit it.
         generated_text (`Optional[str]`, *optional*):
             The complete generated text. Only available when the generation is finished.
         details (`Optional[StreamDetails]`, *optional*):
@@ -459,6 +461,9 @@ class TextGenerationStreamResponse:
 
     # Generated token
     token: Token
+    # The token index within the stream
+    # Optional to support  older clients that omit it.
+    index: Optional[int] = None
     # Complete generated text
     # Only available when the generation is finished
     generated_text: Optional[str] = None

tests/test_inference_text_generation.py

@@ -20,6 +20,7 @@
     OverloadedError,
     TextGenerationParameters,
     TextGenerationRequest,
+    TextGenerationStreamResponse,
     raise_text_generation_error,
 )
 from huggingface_hub.inference._text_generation import (
@@ -106,6 +107,35 @@ def test_request_validation(self):
                 inputs="test", parameters=TextGenerationParameters(best_of=2, do_sample=True), stream=True
             )
 
+    def test_streaming_response_validation(self):
+        """
+        Regression test for #2005.
+
+        See https://github.com/huggingface/huggingface_hub/issues/2005
+        """
+        json_payload_latest = {
+            "index": 97,
+            "token": {"id": 264, "text": " a", "logprob": -0.0003259182, "special": False},
+            "generated_text": None,
+            "details": None,
+        }
+        TextGenerationStreamResponse(**json_payload_latest)
+
+        json_payload_older_had_no_index = {
+            "token": {"id": 264, "text": " a", "logprob": -0.0003259182, "special": False},
+            "generated_text": None,
+            "details": None,
+        }
+        TextGenerationStreamResponse(**json_payload_older_had_no_index)
+
+        json_payload_without_required_field_token = {
+            "index": 97,
+            "generated_text": None,
+            "details": None,
+        }
+        with self.assertRaises((ValidationError, TypeError)):
+            TextGenerationStreamResponse(**json_payload_without_required_field_token)
+
 
 class TestTextGenerationErrors(unittest.TestCase):
     def test_generation_error(self):