In the original implementation of DARTS, it is input_search, target_search = next(iter(valid_queue), which slows down the training when using PyTorch 0.4 and above.

chenxin061 / pdarts

Codes for our paper "Progressive Differentiable Architecture Search:Bridging the Depth Gap between Search and Evaluation"

Other

359 stars 83 forks source link

Closed BangguWu closed 4 years ago

BangguWu commented 4 years ago

Regarding this issue, when I run the program using pytorch 1.1, I will still be prompted with a warning.

OMP: Warning #190: Forking a process while a parallel region is active is potentially unsafe.

Is there a better solution?

chenxin061 commented 4 years ago

I have modified those lines to fit PyTorch 0.4 and above in my code. If you have a better solution, please kindly share it with us.