XianchiMeng

icon indicating an email [email protected]

icon company 天津科技大学智能驾驶与大数据实验室 icon location 天津科技大学滨海校区 icon location Undergraduate at Tianjin University of Science & Technology. Computer Vision：Object Detection（YOLO）、Multimodal（BEiTv3）&AIGC（Stable Diffusion） Baidu Practical

Results 2 issues of XianchiMeng

In Windows, multiple Gpus train my VOC datasets to report NCCL problems

3 comment

Hello, the model can be trained normally when I use 1 GPU. But the following problem occurred when I tried to convert the model to 4 Gpus for training ![image-20220622161317189](https://user-images.githubusercontent.com/69078348/174979768-568c8a98-53ee-459d-9bf7-64ae03b136e9.png)...

请问如何进行多卡训练呢？

您好，感谢您的elegant工作， 想询问下我在调整分辨率到512后训练batch_size=1（A100）需要52gb的显存，2.5小时每epoch 在训练batch_zize=2时显存爆掉 然后希望使用多卡训练，该如何使用呢？