scriptoriumadmin/open-webui
1
0
Fork 0
mirror of https://github.com/open-webui/open-webui.git synced 2026-10-09 03:18:18 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 3

fix: earlier tool calls are sent twice after answering the model's question or pressing Continue (#32029)

Browse source 
When a reply was paused and then picked up again, either by answering a question from the built-in Ask User tool or by pressing Continue, every tool round from the second one on sent everything from before the pause to the model twice. Providers that reject repeated tool calls, like DeepSeek, then fail with "Duplicate 'call_id'" and the chat stops, while others quietly see the earlier tool calls and text twice. Everything from before the pause is now sent exactly once. Tested against a mock provider that records every request: three tool rounds after an answered question and after Continue send everything once, and replies that were never paused send exactly what they sent before.

Fixes #31991
This commit is contained in:
Classic298 2026-10-07 13:04:52 +02:00 • committed by GitHub
parent f694210fdc
commit 728f397ff4
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
1 changed files with 7 additions and 3 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

10
backend/open_webui/utils/middleware.py
View file

@ -6183,6 +6183,8 @@ async def streaming_chat_response_handler(response, ctx):
# keeps indices aligned. The display prefix
# ensures the UI shows tool history during
# streaming.
continued_output = prior_output
round_output = output
prior_output = list(full_output())
# Trim the trailing empty placeholder message
# so it doesn't persist as a ghost item once
@ -6195,11 +6197,13 @@ async def streaming_chat_response_handler(response, ctx):
msg_parts = prior_output[-1].get('content', [])
if not msg_parts or (len(msg_parts) == 1 and not msg_parts[0].get('text', '').strip()):
prior_output.pop()
round_output = round_output[:-1]
output = []
output_start = len(prior_output)
await stream_body_handler(res, new_form_data)
output = full_output()
prior_output = []
# A continued reply's earlier items are already in form_data['messages']
output = [*round_output, *output]
prior_output = continued_output
elif getattr(res, 'status_code', 200) >= 400:
await emit_message_error(get_message_error_content(get_response_error_detail(res)))
break
@ -6411,7 +6415,7 @@ async def streaming_chat_response_handler(response, ctx):
break
# Mark all in-progress items as completed
for item in output:
for item in [*prior_output, *output]:
if item.get('status') == 'in_progress':
item['status'] = 'completed'