实验条件

[Jamie-Cheung]

请问多少G显存是合适的，或者你们用的是什么显卡

[HolmesShuan]

FYI.

Our model is trained on two NVIDIA Tesla P100 GPUs and two NVIDIA GeForce RTX 3090 GPUs.

FP16 seems to be a good option if you encounter OOM.

[Jamie-Cheung]

font{ line-height: 1.6; } ul,ol{ padding-left: 20px; list-style-position: inside; } Thank you for your time. I want to know whether your code can be trained on 2*3090. I failed to train it on my computer, Ubuntu18 2*3090.  Error:RuntimeError: CUDA out of memory. Tried to allocate 256.00 MiB (GPU 0; 24.00 GiB total capacity; 21.24 GiB already allocated; 210.21 MiB free; 21.30 GiB reserved in total by PyTorch) ***@***.*** 在2022年3月12日 ***@***.***> 写道： Closed #6. —Reply to this email directly, view it on GitHub, or unsubscribe.Triage notifications on the go with GitHub Mobile for iOS or Android. You are receiving this because you authored the thread.Message ID: ***@***.***>