Hello! I need to use a particular loss function that looks like a cosine loss. I have a few questions. First, am I correct to assume that no cosine loss is coded? Second, is this something I could contribute to the code? If so, am I correct to modify losses.pytorch.ipynb?

Yes, everything you said is correct! Make sure to inherit BasePointLoss and refer to the contributing guidelines, so we can review your PR. Looking forward to it! 🙂

Sounds good. Thank you!