BenchmarkTransferLearning
BenchmarkTransferLearning copied to clipboard
Question about performance metrics
Hello, congratulations on the excellent paper!
I see the default number of epochs used is 1000. Can you share the number of gpus used or the average gpu-days for each experiment?
Thanks.
Hi,
we use the early stop technique based on the validation loss. That being said, each run takes much fewer epochs than 1000. Please see section E of the paper's appendix for the average number of epochs for each method. Also, we use one DGX-1 V100 32G GPU to run the target tasks.
Thanks for the information. Much appreciated.