sij/khoj
1
0
Fork 0
mirror of https://github.com/khoj-ai/khoj.git synced 2024-11-27 17:35:07 +01:00
Code Issues Projects Releases Packages Wiki Activity Actions

Truncate last message if still over max supported prompt size by model

Browse source
This commit is contained in:
Debanjum Singh Solanky 2023-06-01 08:05:18 +05:30
parent ed4d0f9076
commit 1cd9ecd449
1 changed files with 10 additions and 1 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

11
src/khoj/processor/conversation/utils.py
View file

@ -97,10 +97,19 @@ def generate_chatml_messages_with_context(
# Truncate oldest messages from conversation history until under max supported prompt size by model
encoder = tiktoken.encoding_for_model(model_name)
tokens = sum([len(encoder.encode(content)) for message in messages for content in message.content])
while tokens > max_prompt_size[model_name]:
while tokens > max_prompt_size[model_name] and len(messages) > 1:
messages.pop()
tokens = sum([len(encoder.encode(content)) for message in messages for content in message.content])
# Truncate last message if still over max supported prompt size by model
if tokens > max_prompt_size[model_name]:
last_message = messages[-1]
truncated_message = encoder.decode(encoder.encode(last_message.content))
logger.debug(
f"Truncate last message to fit within max prompt size of {max_prompt_size[model_name]} supported by {model_name} model:\n {truncated_message}"
)
messages = [ChatMessage(content=[truncated_message], role=last_message.role)]
# Return message in chronological order
return messages[::-1]