Hey everyone, I've only seen Fine tuning models using the prompt & completion variation. I haven't seen any example of training the model based on a separate DB with complex data. Is it possible to do it? Thanks!
#Fine tune Model with a complex DB as data source?
buoyant merlin
drifting zinc
Hey everyone, I've only seen Fine tuning models using the prompt & completion va...
Hey, it is only possible to fine tune with prompt and completion pairs
buoyant merlin
Oh okay, Thanks! If I were to train a model based on an entire DB worth of data, I'd have to create a custom model based on certain architecture like GPT-3 and then use Deep Learning to feed it?
drifting zinc
Hmm, you can maybe try to convert/transform that DB data into prompt and completion pairs right?
buoyant merlin
Yeah, that could be an option but the data is too complex to be tied down to just Prompt & Completion pairs. I'll have to find a way to dissect it somehow into those pairs. Thanks a lot for your help, btw! 🙂
drifting zinc
No worries! Yeah, I can see how it may be problematic
drifting zinc
Oh okay, Thanks! If I were to train a model based on an entire DB worth of data,...
But yes you can train your own model to achieve this
There are some models online that are open source whose training formats are defined that you may be able to use, such as GPT2 I think maybe?