Hi，Dr Wang, nice work!
I found that the IO speed is quite slow in the result that utility of GPU is relative low(~30). The GPU utilization rate will even become 0 every few seconds. The training time on ycbv is more than 26 hours. I whether is there any optimized strategy on this dataloader to speed up this process.
Thank you!
Hi，Dr Wang, nice work! I found that the IO speed is quite slow in the result that utility of GPU is relative low(~30). The GPU utilization rate will even become 0 every few seconds. The training time on ycbv is more than 26 hours. I whether is there any optimized strategy on this dataloader to speed up this process. Thank you!