CogVideo icon indicating copy to clipboard operation
CogVideo copied to clipboard

NPU out of memory

Open fallbernana123456 opened this issue 1 year ago • 1 comments

我在昇腾910B上部署完成,python inference/cli_demo.py 时报错: hidden_states = F.scaled_dot_product_attention( RuntimeError: NPU out of memory. Tried to allocate 35.31 GiB (NPU 0; 60.97 GiB total capacity; 26.13 GiB already allocated; 26.13 GiB current active; 33.81 GiB free; 26.61 GiB reserved in total by PyTorch)

那这个需要在多大的NPU上才能运行呢?有什么办法可以降低对NPU的要求吗?

fallbernana123456 avatar Aug 06 '24 10:08 fallbernana123456

36g是GPU的,没有测试过NPU的这是消耗了多少G了

zRzRzRzRzRzRzR avatar Aug 06 '24 10:08 zRzRzRzRzRzRzR

我在昇腾910B上部署完成,python inference/cli_demo.py 时报错: hidden_states = F.scaled_dot_product_attention( RuntimeError: NPU out of memory. Tried to allocate 35.31 GiB (NPU 0; 60.97 GiB total capacity; 26.13 GiB already allocated; 26.13 GiB current active; 33.81 GiB free; 26.61 GiB reserved in total by PyTorch)

那这个需要在多大的NPU上才能运行呢?有什么办法可以降低对NPU的要求吗?

可以分享一下部署相关的资料吗,我在晟腾官网没有找到支持CogVideox的信息。

ql390962 avatar Aug 12 '24 07:08 ql390962

请问这个问题解决了吗,我在昇腾npu上部署也发生了一样的问题,不知道是哪里出问题了

zyang6 avatar Sep 19 '24 02:09 zyang6