bhetherman/open-webui
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 committed 2026-06-29 12:16:58 -05:00
1 parent c89fd237b8
commit 75db531c12
2 files changed
+68 -19
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

backend/open_webui/utils/middleware.py
+49 -16
View File
@@ -3623,7 +3623,31 @@ async def non_streaming_chat_response_handler(response, ctx):
# otherwise generate from response content # otherwise generate from response content
response_output = response_data.get('output') response_output = response_data.get('output')
if not response_output: if not response_output:
response_output = [ choice_message = choices[0].get('message', {})
reasoning_content = choice_message.get('reasoning_content') or choice_message.get('reasoning')
reasoning_details = choice_message.get('reasoning_details')
response_output = []
if reasoning_content or reasoning_details:
reasoning_item = {
'type': 'reasoning',
'id': output_id('r'),
'status': 'completed',
'start_tag': '<think>',
'end_tag': '</think>',
'attributes': {'type': 'reasoning_content'},
'content': (
[{'type': 'output_text', 'text': reasoning_content}]
if reasoning_content
else []
),
'summary': None,
}
if reasoning_details:
reasoning_item['reasoning_details'] = (
reasoning_details if isinstance(reasoning_details, list) else [reasoning_details]
)
response_output.append(reasoning_item)
response_output.append(
{ {
'type': 'message', 'type': 'message',
'id': output_id('msg'), 'id': output_id('msg'),
@@ -3631,7 +3655,7 @@ async def non_streaming_chat_response_handler(response, ctx):
'role': 'assistant', 'role': 'assistant',
'content': [{'type': 'output_text', 'text': content}], 'content': [{'type': 'output_text', 'text': content}],
} }
] )
await event_emitter( await event_emitter(
{ {
@@ -4398,7 +4422,8 @@ async def streaming_chat_response_handler(response, ctx):
or delta.get('reasoning') or delta.get('reasoning')
or delta.get('thinking') or delta.get('thinking')
) )
if reasoning_content: reasoning_details = delta.get('reasoning_details')
if reasoning_content or reasoning_details:
if not output or output[-1].get('type') != 'reasoning': if not output or output[-1].get('type') != 'reasoning':
reasoning_item = { reasoning_item = {
'type': 'reasoning', 'type': 'reasoning',
@@ -4415,20 +4440,28 @@ async def streaming_chat_response_handler(response, ctx):
else: else:
reasoning_item = output[-1] reasoning_item = output[-1]
# Append to reasoning content if reasoning_content:
parts = reasoning_item.get('content', []) # Append to reasoning content
if parts and parts[-1].get('type') == 'output_text': parts = reasoning_item.get('content', [])
parts[-1]['text'] += reasoning_content if parts and parts[-1].get('type') == 'output_text':
else: parts[-1]['text'] += reasoning_content
reasoning_item['content'] = [ else:
{ reasoning_item['content'] = [
'type': 'output_text', {
'text': reasoning_content, 'type': 'output_text',
} 'text': reasoning_content,
] }
]
data = {'content': serialize_output(full_output())} data = {'content': serialize_output(full_output())}
delta_type = 'content' delta_type = 'content'
if reasoning_details:
reasoning_item.setdefault('reasoning_details', []).extend(
reasoning_details
if isinstance(reasoning_details, list)
else [reasoning_details]
)
if value: if value:
if ( if (
backend/open_webui/utils/misc.py
+19 -3
View File
@@ -250,10 +250,16 @@ def convert_output_to_messages(
pending_tool_calls = [] pending_tool_calls = []
pending_content = [] pending_content = []
pending_reasoning = [] # Only populated when reasoning_format == 'reasoning_content' pending_reasoning = [] # Only populated when reasoning_format == 'reasoning_content'
pending_reasoning_details = []
def flush_pending(): def flush_pending():
nonlocal pending_content, pending_tool_calls, pending_reasoning nonlocal pending_content, pending_tool_calls, pending_reasoning, pending_reasoning_details
if not pending_content and not pending_tool_calls and not pending_reasoning: if (
not pending_content
and not pending_tool_calls
and not pending_reasoning
and not pending_reasoning_details
):
return return
message = { message = {
@@ -265,10 +271,14 @@ def convert_output_to_messages(
if pending_reasoning: if pending_reasoning:
message['reasoning_content'] = '\n'.join(pending_reasoning) message['reasoning_content'] = '\n'.join(pending_reasoning)
if pending_reasoning_details:
message['reasoning_details'] = pending_reasoning_details
messages.append(message) messages.append(message)
pending_content = [] pending_content = []
pending_tool_calls = [] pending_tool_calls = []
pending_reasoning = [] pending_reasoning = []
pending_reasoning_details = []
for item in output: for item in output:
item_type = item.get('type', '') item_type = item.get('type', '')
@@ -339,7 +349,8 @@ def convert_output_to_messages(
) )
elif item_type == 'reasoning': elif item_type == 'reasoning':
if not reasoning_format: reasoning_details = item.get('reasoning_details') if raw else None
if not reasoning_format and not reasoning_details:
continue continue
reasoning_text = '' reasoning_text = ''
@@ -360,6 +371,11 @@ def convert_output_to_messages(
# llama.cpp: collect for reasoning_content field # llama.cpp: collect for reasoning_content field
pending_reasoning.append(reasoning_text) pending_reasoning.append(reasoning_text)
if reasoning_details:
pending_reasoning_details.extend(
reasoning_details if isinstance(reasoning_details, list) else [reasoning_details]
)
elif item_type == 'open_webui:code_interpreter': elif item_type == 'open_webui:code_interpreter':
# Always include code interpreter content so the LLM knows # Always include code interpreter content so the LLM knows
# the code was already executed and doesn't retry. # the code was already executed and doesn't retry.