oobabooga/text-generation-webui
1
0
Fork 0
mirror of https://github.com/oobabooga/text-generation-webui.git synced 2026-03-18 03:14:39 +01:00
Code Issues Actions Packages Projects Releases Wiki Activity

UI: Prepend thinking tag when template appends it to prompt

Browse source 
Makes Qwen models have a thinking block straight away during streaming.
This commit is contained in:
oobabooga 2026-03-12 14:30:51 -07:00
parent 9a7428b627
commit 3e6bd1a310
1 changed files with 14 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

14
modules/chat.py
View file

@ -28,6 +28,7 @@ from modules.html_generator import (
)
from modules.image_utils import open_image_safely
from modules.logging_colors import logger
from modules.reasoning import THINKING_FORMATS
from modules.text_generation import (
generate_reply,
get_encoded_length,
@ -986,10 +987,23 @@ def chatbot_wrapper(text, state, regenerate=False, _continue=False, loading_mess
# Add timestamp for assistant's response at the start of generation
update_message_metadata(output['metadata'], "assistant", row_idx, timestamp=get_current_timestamp(), model_name=shared.model_name)
# Detect if the template appended a thinking start tag to the prompt
thinking_prefix = None
if not _continue:
stripped_prompt = prompt.rstrip('\n')
for start_tag, end_tag, content_tag in THINKING_FORMATS:
if start_tag is not None and stripped_prompt.endswith(start_tag):
thinking_prefix = start_tag
break
# Generate
reply = None
for j, reply in enumerate(generate_reply(prompt, state, stopping_strings=stopping_strings, is_chat=True, for_ui=for_ui)):
# Prepend thinking tag if the template appended it to the prompt
if thinking_prefix:
reply = thinking_prefix + reply
# Extract the reply
if state['mode'] in ['chat', 'chat-instruct']:
if not _continue: