luzai

About model parallel for weight matrix in the head

2

Thank you very much for your high performance repo! By splitting large matrix(refer to [here](https://github.com/ZhaoJ9014/face.evoLVe.PyTorch/blob/f8b8a982b5fe92f8d91d111f43a77071c115c66f/head/metrics.py#L107)), the memory consumption is more balanced, however the training process seems not speed up. I...

pred will contain nan if k_at_hop is large

4

Thank you very much for your inspiring work! As suggested in the paper, "In the testing phase, it is not necessary to keep the same configuration with the training phase....

May I ask which transfer accuracy used in LogME paper?

2

Hi, In LEEP paper, there is two transfer accuracy depending on which transfer learning method is used, re-train head or fine-tune (whole model). Is LogME correlated well with the transfer...

Hi, kalviny. Thank you very much for this repo! Would you share the pretrained 4-scale MSDNet w.o./w GE/IMTA on Imagenet? Training on imagenet takes a long time. In the paper,...

Thank you for your great work and elegant idea! Just wondering what if Int8 BMM overflows? Will it do wrapping or saturation?

May I ask the way to split SUN397 dataset? I saw this file `data/SUN397/Training_01.txt` is needed in sun397.py. Thank you!

About scheduler

2

Thank you for your great work! May I ask about some details on the scheduler? 1. In paper, it is mentioned that "To minimize latency penalty, we limit the prefill...