Fix the gemini adapter

app/services/providers/google_adapter.py

@@ -301,6 +301,7 @@ async def process_completion(
     async def _stream_google_response(
         self, api_key: str, model: str, google_payload: dict[str, Any]
     ) -> AsyncGenerator[bytes, None]:
+        # https://ai.google.dev/api/generate-content#method:-models.streamgeneratecontent
         model_path = model if model.startswith("models/") else f"models/{model}"
         url = f"{self._base_url}/{model_path}:streamGenerateContent"
 
@@ -325,9 +326,6 @@ async def _stream_google_response(
                     error=ValueError(error_text)
                 )
             headers = {"Content-Type": "application/json", "Accept": "application/json"}
-            logger.debug(
-                f"Google API request - URL: {url}, Payload sample: {str(google_payload)[:200]}..."
-            )
 
             async with (
                 aiohttp.ClientSession() as session,
@@ -345,8 +343,10 @@ async def _stream_google_response(
                     )
 
                 # Process response in chunks
-                # https://ai.google.dev/api/generate-content#v1beta.GenerateContentResponse
                 buffer = ""
+                # According to https://ai.google.dev/api/generate-content#UsageMetadata
+                # thoughtsTokenCount is output only. We should only record it once
+                logged_thoughts_tokens = False
                 async for chunk in response.content.iter_chunks():
                     if not chunk[0]:  # Empty chunk
                         continue
@@ -379,6 +379,11 @@ async def _stream_google_response(
                                 usage_data = self.format_google_usage(
                                     json_obj["usageMetadata"]
                                 )
+                                if logged_thoughts_tokens:
+                                    del usage_data['completion_tokens_details']
+                                elif usage_data.get('completion_tokens_details', {}).get('reasoning_tokens'):
+                                    logged_thoughts_tokens = True
+
 
                             if "candidates" in json_obj:
                                 choices = []
@@ -431,6 +436,7 @@ async def _stream_google_response(
 
     @staticmethod
     def format_google_usage(metadata: dict) -> dict:
+        # https://ai.google.dev/api/generate-content#UsageMetadata
         """Format Google usage metadata to OpenAI format"""
         if not metadata:
             return None
@@ -457,7 +463,7 @@ async def convert_openai_completion_payload_to_google(
                 "stopSequences": payload.get("stop", []),
                 "temperature": payload.get("temperature", 0.7),
                 "topP": payload.get("top_p", 0.95),
-                "maxOutputTokens": payload.get("max_completion_tokens", payload.get("max_tokens", 2048)),
+                "maxOutputTokens": payload.get("max_completion_tokens", payload.get("max_tokens")),
             },
         }
 

tests/unit_tests/test_google_provider.py

@@ -87,7 +87,7 @@ async def test_chat_completion(self):
                 "generationConfig": {
                     "temperature": 0.7,
                     "topP": 0.95,
-                    "maxOutputTokens": 2048,
+                    "maxOutputTokens": None,
                     "stopSequences": [],
                 },
                 "contents": [
@@ -127,7 +127,7 @@ async def test_chat_completion_streaming(self):
                 "generationConfig": {
                     "temperature": 0.7,
                     "topP": 0.95,
-                    "maxOutputTokens": 2048,
+                    "maxOutputTokens": None,
                     "stopSequences": [],
                 },
                 "contents": [