Shujun-He / RibonanzaNet

MIT License
10 stars 5 forks source link

Update pairwise.yaml to match public pretrained checkpoint #1

Closed chaitjo closed 9 months ago

chaitjo commented 9 months ago

It seems that the publicly available pretrained RibonanzaNet checkpoint does not match the config provided in the Github repo: https://www.kaggle.com/code/shujun717/ribonanzanet-inference/notebook

The k parameter is set to 9 here, whereas the model weights use k = 5.

I've updated k to 5 in the config file here and everything seems to work for me now.

I'd also suggest updating the public Kaggle notebook, in case what I did is correct.