oobabooga/text-generation-webui
1
0
Fork 0
mirror of https://github.com/oobabooga/text-generation-webui.git synced 2025-12-06 07:12:10 +01:00
Code Issues Actions Packages Projects Releases Wiki Activity

chat.py improvements

Browse source
This commit is contained in:
oobabooga 2025-08-25 17:03:37 -07:00
parent 3bc48014a5
commit d08800c359
1 changed files with 7 additions and 9 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

16
modules/chat.py
View file

@ -336,10 +336,6 @@ def generate_chat_prompt(user_input, state, **kwargs):
prompt = renderer(messages=messages) prompt = renderer(messages=messages)
if state['mode'] == 'chat-instruct': if state['mode'] == 'chat-instruct':
outer_messages = []
if state['custom_system_message'].strip() != '':
outer_messages.append({"role": "system", "content": state['custom_system_message']})
command = state['chat-instruct_command'] command = state['chat-instruct_command']
command = command.replace('<|character|>', state['name2'] if not impersonate else state['name1']) command = command.replace('<|character|>', state['name2'] if not impersonate else state['name1'])
command = command.replace('<|prompt|>', prompt) command = command.replace('<|prompt|>', prompt)
@ -353,29 +349,31 @@ def generate_chat_prompt(user_input, state, **kwargs):
if not impersonate: if not impersonate:
prefix = apply_extensions('bot_prefix', prefix, state) prefix = apply_extensions('bot_prefix', prefix, state)
suffix = get_generation_prompt(instruct_renderer, impersonate=False)[1]
outer_messages = []
if state['custom_system_message'].strip() != '':
outer_messages.append({"role": "system", "content": state['custom_system_message']})
outer_messages.append({"role": "user", "content": command}) outer_messages.append({"role": "user", "content": command})
outer_messages.append({"role": "assistant", "content": prefix}) outer_messages.append({"role": "assistant", "content": prefix})
prompt = instruct_renderer(messages=outer_messages) prompt = instruct_renderer(messages=outer_messages)
suffix = get_generation_prompt(instruct_renderer, impersonate=False)[1]
if len(suffix) > 0: if len(suffix) > 0:
prompt = prompt[:-len(suffix)] prompt = prompt[:-len(suffix)]
else: else:
# Handle GPT-OSS as a special case when continuing # Handle GPT-OSS as a special case when continuing
# (otherwise the thinking block gets removed...)
if _continue and '<|channel|>final<|message|>' in state['instruction_template_str']: if _continue and '<|channel|>final<|message|>' in state['instruction_template_str']:
last_message_to_continue = messages[-1] last_message_to_continue = messages[-1]
prompt = renderer(messages=messages[:-1]) prompt = renderer(messages=messages[:-1])
# Start the assistant turn wrapper
assistant_reply_so_far = "<|start|>assistant" assistant_reply_so_far = "<|start|>assistant"
if 'thinking' in last_message_to_continue: if 'thinking' in last_message_to_continue:
assistant_reply_so_far += f"<|channel|>analysis<|message|>{last_message_to_continue['thinking']}<|end|>" assistant_reply_so_far += f"<|channel|>analysis<|message|>{last_message_to_continue['thinking']}<|end|>"
assistant_reply_so_far += f"<|channel|>final<|message|>{last_message_to_continue.get('content', '')}" assistant_reply_so_far += f"<|channel|>final<|message|>{last_message_to_continue.get('content', '')}"
prompt += assistant_reply_so_far prompt += assistant_reply_so_far
else: else:
prompt = renderer(messages=messages) prompt = renderer(messages=messages)
if _continue: if _continue: