#i'm trying to make a transformers model for text generation to help me generate diet plans and so on
10 messages · Page 1 of 1 (latest)
you can use the assistants or "GPTs"/"myGPT" from openai, there you can define the task you want to solve by the model and upload additional files, like pdfs were you store additional informations for your/any diet plan. Give it a try, they work quite well
So I would caution against using a text generator for coming up with diet plans. This is primarily because you cannot confirm if a plan is edible or even a structured plan for a diet (ie does the food match the diet constraints, does it take into account macros, etc),. I think you’d do better with looking at the problem like search or graph problem. Set your restrictions based on diet and allergies and then create an algorithm that will search through a recipe database to construct your diet.
Remember, the text generator is going to generate text that sounds human. It doesn’t actually “know” enough to be useful outside of that.
do you think you have a working model i can work on? i tried some articles but i didn't really understand much, if i'm able to see a code it'd help much better qwq
the thing is it's mostly similar to "roleplaying" so it doesn't need to be real since i kinda wanna link it to a dataset to train it over it
Aye, fair enough
yep, i've tried to search for a way to learn it so i'm kinda trying to find a model that i can try and read it's code and learn it. :(
If youre using openai / chatGPT, you dont need any coding, you can do it in the webbrowser:
https://chat.openai.com/gpts/mine there you can click on configure and add your files, its verry simple.
If you want to code it by yourself, i would say take a look into the #📣・announcements channel, there are some videos about RAG, you could try with the last one: #📣・announcements message not sure if its a good point to start, havent watched it myself 😄 but on the youtube channel you should be able to find what youre looking for
If you know nothing about AI I'd strongly recommend you understand some classical machine learning and then deep learning, before you go for a very applied domain with more domain specific stuff (NLP in this case) like tokenization