scriptoriumadmin/litellm
1
0
Fork 0
mirror of https://github.com/BerriAI/litellm.git synced 2026-09-12 23:01:41 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 88

fix(gemini realtime): preserve sibling toolCall when serverContent has only transcription

Browse source 
Previously, when a Gemini frame contained both a transcription-only
serverContent and a sibling toolCall, the transcription handler would
early-return and silently drop the toolCall. Instead, mark serverContent
as handled and fall through so the main loop still processes siblings
like toolCall, while preserving the prior no-op behavior for empty/
transcription-only frames.

Co-authored-by: Yassin Kortam <yassin@berri.ai>
This commit is contained in:
Cursor Agent 2026-05-23 00:15:54 +00:00
parent 84791fc3f5
commit f1b76d99e8
No known key found for this signature in database
1 changed files with 24 additions and 13 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

37
litellm/llms/gemini/realtime/transformation.py
View file

@ -1259,24 +1259,21 @@ class GeminiRealtimeConfig(BaseRealtimeConfig):
)
# If serverContent only contained transcription(s) and no model
# content, return early — the main loop would fail on unknown keys.
# content, mark it as already handled so the main loop skips it
# (map_openai_event would raise on an unknown serverContent
# subkey). Fall through so sibling top-level keys such as
# ``toolCall`` are still processed in the main loop.
_model_content_keys = {
"modelTurn",
"turnComplete",
"interrupted",
"generationComplete",
}
if not any(k in server_content for k in _model_content_keys):
return {
"response": returned_message,
"current_output_item_id": current_output_item_id,
"current_response_id": current_response_id,
"current_delta_chunks": current_delta_chunks,
"current_conversation_id": current_conversation_id,
"current_item_chunks": current_item_chunks,
"current_delta_type": current_delta_type,
"session_configuration_request": session_configuration_request,
}
server_content_handled = not any(
k in server_content for k in _model_content_keys
)
else:
server_content_handled = False
for key, value in json_message.items():
# Skip sibling metadata keys (e.g. ``usageMetadata``) that can
@ -1285,6 +1282,11 @@ class GeminiRealtimeConfig(BaseRealtimeConfig):
# would otherwise terminate the WebSocket session.
if key not in _KNOWN_GEMINI_TOP_LEVEL_KEYS:
continue
# serverContent was a transcription-only payload already emitted
# above; skip it here so map_openai_event doesn't raise on the
# missing model-content subkeys.
if key == "serverContent" and server_content_handled:
continue
# Check if this key or any nested key matches our mapping
openai_event = self.map_openai_event(
key=key,
@ -1534,7 +1536,16 @@ class GeminiRealtimeConfig(BaseRealtimeConfig):
# to forward to the OpenAI-shaped client. Returning the
# unchanged state keeps the WebSocket alive; raising would
# terminate the session for a benign no-op frame.
if not any(key in _KNOWN_GEMINI_TOP_LEVEL_KEYS for key in json_message):
# serverContent already consumed by the transcription handler is
# a benign no-op for downstream — treat it like a metadata-only
# key when deciding whether to raise.
unhandled_known_keys = [
key
for key in json_message
if key in _KNOWN_GEMINI_TOP_LEVEL_KEYS
and not (key == "serverContent" and server_content_handled)
]
if not unhandled_known_keys:
return {
"response": returned_message,
"current_output_item_id": current_output_item_id,