报错RuntimeError: Error building extension 'fused_adam'

[zbxx-423]

这是cuda版本算力问题吗

[1049451037]

这不是算力的问题，是nvcc压根没安装……或者没加到环境变量里

[zbxx-423]

这不是算力的问题，是nvcc压根没安装……或者没加到环境变量里

nvcc放到他指定的目录里解决掉之后 还是报RuntimeError: Error building extension 'fused_adam' 这个错误 我用的是两张3090 报错之前看显卡使用都是100%

[1049451037]

参考这里：https://github.com/THUDM/VisualGLM-6B/issues/125#issuecomment-1630407747

[LijunRio]

我将cuda的版本改为11.7之后能解决这个问题，主要是deepspeed的问题

[chenchen333-dev]

我将cuda的版本改为11.7之后能解决这个问题，主要是deepspeed的问题

是deepspeed版本的问题吗，什么版本呢

[a1stupid]

怎么解决呢？

[zzyzeyuan]

Same issue, how to fix it?

[Gae-Zhang]

@zzyzeyuan Have you fixed it?

[zzyzeyuan]

@zzyzeyuan Have you fixed it?

Yes, see this page.

The error may be caused by the wrong installation of deepspeed, and you just need to clone the repo of deepspeed and install it manually like:

git clone https://github.com/microsoft/DeepSpeed.git
cd DeepSpeed
DS_BUILD_FUSED_ADAM=1 pip3 install .

[Gae-Zhang]

@zzyzeyuan Have you fixed it?

Yes, see this page.

The error may be caused by the wrong installation of deepspeed, and you just need to clone the repo of deepspeed and install it manually like:

git clone https://github.com/microsoft/DeepSpeed.git
cd DeepSpeed
DS_BUILD_FUSED_ADAM=1 pip3 install .

Thank you so much!! 祝您科研顺利！