scriptoriumadmin/litellm
1
0
Fork 0
mirror of https://github.com/BerriAI/litellm.git synced 2026-10-08 03:08:45 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 114

Fix Gemini parallel tool call ID mismatch in Responses API translation

Browse source 
When litellm converts Responses API function_call input items to Chat
Completion messages, each function_call item becomes a separate assistant
message with a single tool_calls entry. For providers like Gemini that
issue parallel tool calls in one model turn, this produces multiple
assistant messages instead of one with multiple tool_calls.

The Gemini message converter later walks the history looking for the
'last assistant message with tool calls' to match a tool result. With
split messages, it only finds the last call's assistant message and
fails to match earlier calls' tool results, producing:

  Missing corresponding tool call for tool response message.
  Received - message={'role': 'tool', 'content': '...', 'tool_call_id': 'call_xxx'}

Fix: merge consecutive function_call-derived assistant messages into a
single assistant message with multiple tool_calls entries, re-indexing
as needed. This preserves the one-turn-one-message invariant that
downstream converters expect.

Affects: Gemini 3 models with parallel tool calls via the Responses API
(litellm.aresponses / /v1/responses endpoint).
This commit is contained in:
Guilherme Pires 2026-02-12 22:26:49 -08:00
parent a06113ec82
commit d3e695f055
1 changed files with 45 additions and 1 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

46
litellm/responses/litellm_completion_transformation/transformation.py
View file

@ -434,7 +434,51 @@ class LiteLLMCompletionResponsesConfig:
messages.extend(deduped_in_place)
continue
messages.extend(chat_completion_messages)
###########################################################
# Merge consecutive function_call items into a single
# assistant message. Providers like Gemini require that
# parallel tool calls that were issued in one model turn
# appear as *one* assistant message with multiple
# tool_calls, not as separate assistant messages.
###########################################################
for m in chat_completion_messages:
role = m.get("role") if isinstance(m, dict) else getattr(m, "role", "")
if role != "assistant":
messages.append(m)
continue
new_tcs: Any = (
m.get("tool_calls")
if isinstance(m, dict)
else getattr(m, "tool_calls", None)
) or []
# Try to merge into the last message if it is an
# assistant message that already carries tool_calls.
merged = False
if messages:
last = messages[-1]
last_role = last.get("role") if isinstance(last, dict) else getattr(last, "role", "")
if last_role == "assistant":
existing_tcs: Any = (
last.get("tool_calls")
if isinstance(last, dict)
else getattr(last, "tool_calls", None)
)
if existing_tcs is not None:
# Re-index and append
for tc in new_tcs:
idx = len(existing_tcs)
if isinstance(tc, dict):
tc["index"] = idx
elif hasattr(tc, "index"):
tc.index = idx
existing_tcs.append(tc)
merged = True
if not merged:
messages.append(m)
return messages
@staticmethod