Mirrors/text-generation-webui
1
0
Fork 0
mirror of https://github.com/oobabooga/text-generation-webui.git synced 2024-11-25 17:29:22 +01:00
Code Issues Packages Projects Releases Wiki Activity

Add user_input in impersonate reply for consistency

Browse Source
This commit is contained in:
Yiximail 2024-08-23 09:46:17 +08:00
parent 38b3daad55
commit 9fdfd5ecb0
1 changed files with 7 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

9
extensions/openai/completions.py
View File

@ -245,7 +245,7 @@ def chat_completions_common(body: dict, is_legacy: bool = False, stream=False, p
continue_ = body['continue_'] continue_ = body['continue_']
impersonate = body['impersonate'] impersonate = body['impersonate']
if impersonate: if impersonate:
continue_ = False continue_ = False # While impersonate, continue_ should be False. References impersonate_wrapper in chat.py
# Instruction template # Instruction template
if body['instruction_template_str']: if body['instruction_template_str']:
@ -342,7 +342,12 @@ def chat_completions_common(body: dict, is_legacy: bool = False, stream=False, p
seen_content = '' seen_content = ''
for a in generator: for a in generator:
answer = a if impersonate else a['internal'][-1][1] if impersonate:
# The generate_chat_reply returns the entire message, but generate_reply will only start from new content.
# So we need to add the user_input to keep output consistent.
answer = user_input + a
else:
answer = a['internal'][-1][1]
if stream: if stream:
len_seen = len(seen_content) len_seen = len(seen_content)
new_content = answer[len_seen:] new_content = answer[len_seen:]