AdityaNG/kan-gpt

We need comparation results.

Closed this issue · 5 comments

yuedajiong commented
We need comparation results.
aiva00 commented

It is in the author's todo list. Any ideas on the comparison options?

PriNova commented

Maybe some perplexity metrics against the base GTP-2 model?

AdityaNG commented

Hi all, we have added results in the README comparing the KAN-GPT to the MLP-GPT. These are preliminary results and we are running large hyperparameter sweeps. For now, we observe that the KAN-GPT performs slightly better than the MLP-GPT as shown below:

results

README Results: https://github.com/AdityaNG/kan-gpt/?tab=readme-ov-file#results

We are currently running an extensive hyperparameter sweep as shown below:

image

  • 👍2
AdityaNG commented

We have added the metrics of cross entropy, perplexity along with the loss curves to the docs and the README

  • 👍2
yuedajiong commented

Greeeeeeeeaaaaaat !!!

  • ❤️1