[SIRNN] How to perfrom the benchmark of the Extension/Truncation in SIRNN?

mpc-msri / EzPC

MIT License

401 stars 122 forks source link

[SIRNN] How to perfrom the benchmark of the Extension/Truncation in SIRNN? #132

Open llCurious opened 2 years ago

llCurious commented 2 years ago

Hey, I want to better understand the performance of the mixed-bitwidth computatation setting in SiRNN and try to benchmark the extension/ truncation operations which i think is the basis for SIRNN.

However, the README only gives examples for end-to-end training of RNNs.

Could you bother give a description on how i could customly test the efficiency of protocols in SIRNN?
BTW, the corresponding code implementations for SIRNN i think is here. How could i directly test this function alone?

llCurious commented 2 years ago

BTW, if i want to obtain the statsitics for communication size, how could i enable this log or access this data?

llCurious commented 2 years ago

hey @bhatuzdaname. Maybe anyone could give some advice?

zwxandy commented 1 year ago

Hey, I want to better understand the performance of the mixed-bitwidth computatation setting in SiRNN and try to benchmark the extension/ truncation operations which i think is the basis for SIRNN.

However, the README only gives examples for end-to-end training of RNNs.

Could you bother give a description on how i could customly test the efficiency of protocols in SIRNN?

BTW, the corresponding code implementations for SIRNN i think is here. How could i directly test this function alone?

Hi, I am also interested in SiRNN, and I have difficulties in replacing RNN model and Google-30 dataset with CNN and CIFAR-10. I still have no idea of solving this problem. If there‘s a chance, we can talk about this code.

xingpz2008 commented 1 year ago

Hey, I want to better understand the performance of the mixed-bitwidth computatation setting in SiRNN and try to benchmark the extension/ truncation operations which i think is the basis for SIRNN. However, the README only gives examples for end-to-end training of RNNs.

Could you bother give a description on how i could customly test the efficiency of protocols in SIRNN?

BTW, the corresponding code implementations for SIRNN i think is here. How could i directly test this function alone?

Hi, I am also interested in SiRNN, and I have difficulties in replacing RNN model and Google-30 dataset with CNN and CIFAR-10. I still have no idea of solving this problem. If there‘s a chance, we can talk about this code.

Maybe you can open a new issue for this repo. :)

zwxandy commented 1 year ago

Hey, I want to better understand the performance of the mixed-bitwidth computatation setting in SiRNN and try to benchmark the extension/ truncation operations which i think is the basis for SIRNN. However, the README only gives examples for end-to-end training of RNNs.

Could you bother give a description on how i could customly test the efficiency of protocols in SIRNN?

BTW, the corresponding code implementations for SIRNN i think is here. How could i directly test this function alone?

Hi, I am also interested in SiRNN, and I have difficulties in replacing RNN model and Google-30 dataset with CNN and CIFAR-10. I still have no idea of solving this problem. If there‘s a chance, we can talk about this code.

Maybe you can open a new issue for this repo. :)

Thanks for your advice.