Joke Maker using Mistral

The notebook key_pharse_extraction:

1. A pretrained Mistral 7b Instruct v0.2 model is used to extract list of topics from each dad joke
2. Mistral is finetuned on the set of topics and the target joke using Autotrain

Joke Maker using GPT2

The notebook fine_tune_joke_maker:

The same procedure as above

At the moment, GPT2 was trained on only 2000 data elements, and Mistral is trained on 18000 elements. Definitely Mistral is performing better.

I will update GPT2 on the same data and then will determine which is good To Do: make an evaluation function to find out which joke is good