Open-Assistant
Open-Assistant copied to clipboard
Improve inference export script
inference/server/export.py
- Add ability to filter by conditions
- Include a filter for only chats which have at least one assistant message with score != 0
- Include a filter for specific date ranges
- Consider performance improvements such as use of lazy loading from the database
I'd like to work on this
I'd like to work on this
Thank you, have assigned it to you.
I've added the simple filters - date range, user id and chat id. Will need a bit more time to figure out how to set up the dev env to actually test it.
I also have a few things to clarify:
- should user_id and chat_id filters be for single user/chat or should they allow for multiple values?
- there's a todo comment in code that lists a couple of filters, including "... reported, score. You mentioned the score filter above - at least one assistant message with score != 0. What about reported? Should it only include chats with at least one reported message?
- should user_id and chat_id filters be for single user/chat or should they allow for multiple values?
Single value is fine for this
- there's a todo comment in code that lists a couple of filters, including "... reported, score. You mentioned the score filter above - at least one assistant message with score != 0. What about reported? Should it only include chats with at least one reported message?
Yes, a filter for chats with at least one reported message would be a great addition :)