Emil_Shanaty/open-webui-ai4me
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

refac

Browse Source
This commit is contained in:
Timothy Jaeryang Baek
2026-03-21 20:25:52 -05:00
parent 2ae47cf200
commit f8b3a32caf
1 changed files with 14 additions and 20 deletions
Download Patch File Download Diff File Expand all files Collapse all files
backend/open_webui/utils/middleware.py
+14 -20
View File
@@ -3365,6 +3365,7 @@ async def streaming_chat_response_handler(response, ctx):
output = [] output = []
usage = None usage = None
prior_output = []
reasoning_tags_param = metadata.get('params', {}).get('reasoning_tags') reasoning_tags_param = metadata.get('params', {}).get('reasoning_tags')
DETECT_REASONING_TAGS = reasoning_tags_param is not False DETECT_REASONING_TAGS = reasoning_tags_param is not False
@@ -3399,14 +3400,10 @@ async def streaming_chat_response_handler(response, ctx):
nonlocal content nonlocal content
nonlocal usage nonlocal usage
nonlocal output nonlocal output
nonlocal prior_output
response_tool_calls = [] response_tool_calls = []
# Responses API output_index values are relative to the
# current response (0, 1, ...). During re-invocations,
# output has accumulated history, so we offset indices.
responses_output_offset = len(output)
delta_count = 0 delta_count = 0
delta_chunk_size = max( delta_chunk_size = max(
CHAT_RESPONSE_STREAM_DELTA_CHUNK_SIZE, CHAT_RESPONSE_STREAM_DELTA_CHUNK_SIZE,
@@ -3475,15 +3472,12 @@ async def streaming_chat_response_handler(response, ctx):
) )
# Check for Responses API events (type field starts with "response.") # Check for Responses API events (type field starts with "response.")
elif data.get('type', '').startswith('response.'): elif data.get('type', '').startswith('response.'):
# Offset output_index for re-invocations
if responses_output_offset and 'output_index' in data:
data = {**data, 'output_index': data['output_index'] + responses_output_offset}
output, response_metadata = handle_responses_streaming_event(data, output) output, response_metadata = handle_responses_streaming_event(data, output)
processed_data = { processed_data = {
'output': output, 'output': prior_output + output,
'content': serialize_output(output), 'content': serialize_output(prior_output + output),
} }
# print(data) # print(data)
@@ -4233,17 +4227,17 @@ async def streaming_chat_response_handler(response, ctx):
) )
if isinstance(res, StreamingResponse): if isinstance(res, StreamingResponse):
# Save output before re-invocation. Responses API # Save accumulated output and start fresh.
# response.completed replaces the entire output # Responses API output_index values are relative
# with only the new response's items, losing tool # to the current response — a clean output list
# call history. Restore previous items afterward. # keeps indices aligned. The display prefix
output_prefix = list(output) # ensures the UI shows tool history during
# streaming.
prior_output = list(output)
output = []
await stream_body_handler(res, new_form_data) await stream_body_handler(res, new_form_data)
if output_prefix and ( output[:0] = prior_output
not output prior_output = []
or output[0].get('id') != output_prefix[0].get('id')
):
output[:0] = output_prefix
else: else:
break break
except Exception as e: except Exception as e: