Reproducing paper on ChatGPT 2 with possible improvements and additional documentation
First problem: I don't get how does he come up with the form of code itself. He just Copy-Paste blocks of code and brifly explanes it. It doesn't seem to be understandable
Important "low resulution" questions:
- Where did they take the data?
- How did they process and input data?
- How did they test results and goals of the trainig?
- How did they tuned and impoved the model?