TchaikovskyBear

Results 3 comments of TchaikovskyBear

error: identifier "AT_CHECK" is undefined

I have solved this problem after switching pytorch to version 1.3. Can the program only run in version 1.3?

Multi-GPU inference Error: RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:5 and cuda:6!

可以修改一下device_map, 某一层的权重被分配到不同显卡上了, 比如像这样: ![image](https://github.com/THUDM/CogVLM2/assets/29391214/287ddc49-f15b-4789-9a08-67997735dfaf) 这里在vision.transformer.layers.8下就会出现tensor计算不在同一个device上, 像我举得例子里你可以把layer.8整个改在同一个设备上

[Bug] 真的难用，每次配个环境都像手搓原子弹一样，改来改去的，这不适配那用不了

只能说配环境是DL的一部分了, 说实话这个还不算我经历过的最糟糕的, 好几年前华为的NNIE那套用CAFFE的东西才让我煎熬, 还有早期版本的Ascend