/vlm-pgd

Primary LanguagePython

Reminders:

  • Make sure none of your modeling code is wrapped in a torch.no_grad(). For example, OpenFlamingo + LLaVA do this by default around the vision encoders