Closed mrmachine closed 1 year ago
Currently this is not supported, but i might add a Q&A mode in the near future, i can't do it now because i'm not at home and i don't have C++ dev env setup in my laptop. For now you need to make a customized automatic embedding tool and match the result with my chat history json format for it to work, or just wait until i add this function to this program.
Please, if I can help with this, it would be awesome for a similar project I'm working on right now.
Please, if I can help with this, it would be awesome for a similar project I'm working on right now.
hmmm i found out theres a chatgpt plugin made by OpenAI for Q&A purposes already, now my motivation is destroyed😢 but maybe i'll still make it🙂
Can you preload it with a bunch of embeddings?
Can you preload it with a bunch of embeddings?
if you are talking about the chatgpt plugin, idk because its not public currently. if you are talking about this repo, you kinda can but its not supported currently so you need to do it by yourself, probably need to modify some code too.
is this possible yet? I definitely need this feature.
is this possible yet? I definitely need this feature.
For the time of this comment(2023/4/19), no, it's currently not possible.
I'm currently focusing on improving other things, but I will work on this asap!
Edit: Feature added in v1.2.0.
Progress update: I'm now working on it, planning to release it in the next 3 days, but no guarantees!
It's mostly done, releasing it in 2 hours, closed as completed.
I have a 4 hour long transcript which is about 5,000 lines and 172,000 characters. It was created by Whisper API. I would like to preload this transcript to generate embeddings, and then ask ChatGPT questions and have it automatically find relevant sections of the transcript via the embeddings and include those sections in the prompt so that it can answer my questions about the transcript.
I tried to just paste a small section of the document with a prompt like:
But the bot immediately responded to the first line (only) as if it was the entire prompt:
How can I provide a multi-line prompt to the bot, to progressively load in data to then be saved as embeddings?
Or, how can I preload a document to generate embeddings before I start the chat session?