PickleGPT

Inspiration

Me and Ryan wanted to learn more about machine learning since we were both intrested in the subject beforehand and decided on a chatbot using minGPT as the frame

What it does

The program will answer the user's question's on pickles

How we built it

we used minGPT as a frame and used a dataset of pickle-facts, then configured the temperature and layer's to decrease the training loss further every iteration

Challenges we ran into

re-learning python and HTML there was a learning curve in understanding what exactly mingpt was doing

Accomplishments that we're proud of

we got the trainer to run perfectly within 3 hours of trial and error we made a decent looking user-interface we were able to create a method which understand's general facts about pickles

What we learned

machine learning takes a long time to train re-learning python re-learning HTML and CSS learning PHP on the fly Machine learning algorithms how to convert a python file onto a PHP using pyscript how to use GitHub importing datafiles containing the method to use for our user interface how minGPT's default constructor's affect the training model, we figured out on 4/15 that if we let the default constructor run instead of the overloaded constructor, we can implement a layer,head, and embed

What's next for PickleGPT

better user-interface, more general chatbot, reduced training loss through training and increasing efficiencies on the trainer, removing redundancies

Built With

• css
• html
• mingpt
• php
• python
• pytorch