how much memory was used

alibaba / lightweight-neural-architecture-search

This is a collection of our zero-cost NAS and efficient vision applications.

Apache License 2.0

383 stars 50 forks source link

how much memory was used #19

Open YiChuangDai opened 1 year ago

YiChuangDai commented 1 year ago

in paper said that we can use cpu with small memory source ,but when i run sh tools/dist_search.sh configs/classification/deepmad_29M_224.py 32G memory was used in 10 sec and the job has been aborted.

xiuyu-sxy commented 1 year ago

To ensure unified implementation of different methods, our repo first transforms the model description into a PyTorch model before computing relevant scores. As a result, more content is cached during the search process, which is unnecessary for DeepMAD. To reduce the memory usage, the simple way is to change the "nproc=64" to "nproc=32/16/8" in the dist_search.sh . It will cost more search time.