Hey all! Is it possible or even optimal to fine tune a model which is intended for both QA, classification and entity abstraction? I’ll provide the example to provide more clarity.
I want to create a bot which discusses sports. Ideally this bot can have its own personality or voice, but be aware of specific current information. I find current vector models are bad at pulling a name in vector form and finding specific information pertaining to this player in both structure and unstructured data (some people have unique names which provide errors in my experience). Therefore I would need to train a bot to perform the new task of “sports analyst”, for example, but also be excellent at entity abstraction so I can pull names from information, and find exact information pertaining to a player from both APIs and vector databases. Is this possible (or even optimal) to do under one fine tuned model?
If I’m on the wrong track, let me know. I’ve built a variety of these separately, just trying to now combine all three.
Models cannot “pull data” from anywhere. Models are self-contained and have no external network access and cannot call other APIs, like embeddings.
Maybe I am misunderstanding your question because you are using a vocabulary for terms like “entity abstraction” but this is not a term which is used in the OpenAI documentation.
I might be good to reformulate your question @pbergin11 in terms of vocabulary and concepts used in the OpenAI guidelines and API docs, etc.
HTH
PS: Yes, we know what “entity extraction” means in the content of AI in general; but since we are working with OpenAI’s models, it makes things easier to discuss if we use the same vocabulary as in the OpenAI docs.
Thanks for the reply @ruby_coder. Sorry for the mix-up, I did mean entity extraction. The concept it to extract the entities to be able to pull external data. These functions would be separate from openAI but need entity extraction to ensure proper API calls externally. I understand fine tuned models can perform better through proper training for entity extraction, classification and summarization, just was curious as to someone experience trying to fine tune one model to perform better at all tasks.
Yes, I know you used “entity extraction” and also you meant to use it; but if you are looking to get a solution for this using OpenAPI, I suggested you change your vocabulary to meet the OpenAI vocabulary and OpenAI does not use this term, as you can see from my reply above.
Or never mind… I’m simply trying to get you to use the same vocabulary and concepts as in the OpenAI docs. If you want to use general concepts from AI that are not part of the OpenAI vocabulary, that’s up to you of course.
