search

google-research / l2p

Learning to Prompt (L2P) for Continual Learning @ CVPR22 and DualPrompt: Complementary Prompting for Rehearsal-free Continual Learning @ ECCV22
https://arxiv.org/pdf/2112.08654.pdf
Apache License 2.0
408 stars 41 forks source link

Loss become NaN. Results mismatch between different convolution algorithms. #24

Closed GengDavid closed 2 years ago

GengDavid commented 2 years ago

When running on GPUs, loss becomes NaN. You can find the training log here.

GengDavid commented 2 years ago

solved by re-installing cudnn