fix: add greptile fixes

litellm/responses/compaction.py

@@ -33,6 +33,10 @@ def _get_compact_threshold(context_management: List[Dict[str, Any]]) -> Optional
             threshold = entry.get("compact_threshold")
             if threshold is not None:
                 return max(int(threshold), MIN_COMPACT_THRESHOLD)
+            verbose_logger.warning(
+                "compaction: context_management has type='compaction' but no "
+                "compact_threshold — compaction will not trigger"
+            )
             return None
     return None
 

litellm/responses/litellm_completion_transformation/handler.py

@@ -62,12 +62,12 @@ class LiteLLMCompletionTransformationHandler:
         completion_args = {}
         completion_args.update(kwargs)
         completion_args.update(litellm_completion_request)
+        completion_args.pop("context_management", None)
 
         litellm_completion_response: Union[
             ModelResponse, litellm.CustomStreamWrapper
         ] = litellm.completion(
-            **litellm_completion_request,
-            **kwargs,
+            **completion_args,
         )
 
         if isinstance(litellm_completion_response, ModelResponse):
@@ -153,6 +153,7 @@ class LiteLLMCompletionTransformationHandler:
                     "custom_llm_provider"
                 ),
                 litellm_metadata=kwargs.get("litellm_metadata", {}),
+                compaction_summary_text=summary_text,
             )
         raise ValueError(
             f"Unexpected response type: {type(litellm_completion_response)}"

litellm/responses/litellm_completion_transformation/streaming_iterator.py

@@ -57,8 +57,10 @@ class LiteLLMCompletionStreamingIterator(ResponsesAPIStreamingIterator):
         responses_api_request: ResponsesAPIOptionalRequestParams,
         custom_llm_provider: Optional[str] = None,
         litellm_metadata: Optional[dict] = None,
+        compaction_summary_text: Optional[str] = None,
     ):
         self.model: str = model
+        self.compaction_summary_text: Optional[str] = compaction_summary_text
         self.litellm_custom_stream_wrapper: litellm.CustomStreamWrapper = (
             litellm_custom_stream_wrapper
         )
@@ -1176,6 +1178,17 @@ class LiteLLMCompletionStreamingIterator(ResponsesAPIStreamingIterator):
                 responses_api_request=self.responses_api_request,
             )
 
+            # Append compaction output item if compaction was triggered
+            if self.compaction_summary_text is not None:
+                from litellm.responses.litellm_completion_transformation.handler import (
+                    _append_compaction_output,
+                )
+
+                responses_api_response.output = _append_compaction_output(
+                    self.compaction_summary_text,
+                    responses_api_response.output,
+                )
+
             # Use the cached response ID to ensure consistency across all events
             if self._cached_response_id:
                 responses_api_response.id = self._cached_response_id