4 ms·
to my understanding, fine tuning is slow and would be quite bad to update. embeddings seems to be the way to go. i don't understand it well enough, but it seems
by underlines 4y ago
to my understanding, fine tuning is slow and would be quite bad to update. embeddings seems to be the way to go. i don't understand it well enough, but it seems with the langchain framework you can create an embedding of your own data and submit it to the GPT API and i believe emeddings should be a similar principle in llama. at least i did it with diffusers in stablediffusion.