Add detailed error logging to trace API failures.

Will show the exact exception type and response on HuggingFace.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

providers/openai_compat.py

@@ -210,7 +210,14 @@ class OpenAIChatTransport(BaseProvider):
 
     async def _create_stream(self, body: dict) -> tuple[Any, dict]:
         """Create a streaming chat completion, optionally retrying once."""
+        from loguru import logger
+
         try:
+            logger.info(
+                "{}_CREATE_STREAM: calling API with model={}",
+                self._provider_name,
+                body.get("model"),
+            )
             stream = await self._global_rate_limiter.execute_with_retry(
                 self._client.chat.completions.create,
                 **body,
@@ -219,6 +226,13 @@ class OpenAIChatTransport(BaseProvider):
             )
             return stream, body
         except Exception as error:
+            logger.error(
+                "{}_CREATE_STREAM_ERROR: {} - {} - response={}",
+                self._provider_name,
+                type(error).__name__,
+                str(error),
+                getattr(error, "response", None),
+            )
             retry_body = self._get_retry_request_body(error, body)
             if retry_body is None:
                 raise