alpa icon indicating copy to clipboard operation
alpa copied to clipboard

when i check installation by running python3 -m alpa.test_install,AssertionError happend

Open Ghiad opened this issue 1 year ago • 6 comments

Please describe the bug

Please describe the expected behavior

System information and environment

  • OS Platform and Distribution (e.g., Linux Ubuntu 16.04, docker):
  • Python version:3.95
  • CUDA version:11.3
  • NCCL version:
  • cupy version:cupy-cuda11x 12.2.0
  • GPU model and memory:A100
  • Alpa version:0.2.3
  • JAX version:0.3.22

To Reproduce Steps to reproduce the behavior: 1.run python3 -m alpa.test_install 4. See error

Screenshots If applicable, add screenshots to help explain your problem. image image

Code snippet to reproduce the problem

Additional information Add any other context about the problem here or include any logs that would be helpful to diagnose the problem.

Ghiad avatar Oct 17 '23 10:10 Ghiad

it says we compiled for one device but ran on another.CUDA_VISIBLE_DEVICES=0 fixes this problem for me.However,when i run this command CUDA_VISIBLE_DEVICES=0,1,2 python3 -m alpa.test_install, another error occured , there was a AssertionError,how can i fix that? image

Ghiad avatar Oct 18 '23 03:10 Ghiad

你有几张卡,num_stages默认是2,你的num_devices不能整除num_satges就会出现这个问题,最好保证num_devices为偶数

caixiiaoyang avatar Oct 18 '23 06:10 caixiiaoyang

你有几张卡,num_stages默认是2,你的num_devices不能整除num_satges就会出现这个问题,最好保证num_devices为偶数

我的服务器上有5张卡,如下图 image

我应该怎么修改num_devices的数量,当我尝试输入CUDA_VISIBLE_DEVICES=0,1 python3 -m alpa.test_install,还是会有相同的问题

Ghiad avatar Oct 18 '23 06:10 Ghiad

你有几张卡,num_stages默认是2,你的num_devices不能整除num_satges就会出现这个问题,最好保证num_devices为偶数

我的服务器上有5张卡,如下图 image

我应该怎么修改num_devices的数量,当我尝试输入CUDA_VISIBLE_DEVICES=0,1 python3 -m alpa.test_install,还是会有相同的问题

我不知道怎么修改卡的数目,你有五张卡,并且卡的型号也不一样,你在a100上编译的,在3090上运行可能会出问题

caixiiaoyang avatar Oct 18 '23 07:10 caixiiaoyang

你有几张卡,num_stages默认是2,你的num_devices不能整除num_satges就会出现这个问题,最好保证num_devices为偶数

我的服务器上有5张卡,如下图 image 我应该怎么修改num_devices的数量,当我尝试输入CUDA_VISIBLE_DEVICES=0,1 python3 -m alpa.test_install,还是会有相同的问题

我不知道怎么修改卡的数目,你有五张卡,并且卡的型号也不一样,你在a100上编译的,在3090上运行可能会出问题

我已经解决了这个error,通过在启动ray的时候限制gpu数量ray start --head --num-gpus=2,然后当我继续运行CUDA_VISIBLE_DEVICES=0,1 python3 -m alpa.test_install 另一个error报错了,另外通过CUDA_VISIBLE_DEVICES=0,1可以限定运行的设备,让代码在同一种设备上运行 image

image

Ghiad avatar Oct 18 '23 07:10 Ghiad

一个愚蠢的错误,一张显卡被其他用户占用了,换另外的显卡就能顺利执行,十分感谢

Ghiad avatar Oct 18 '23 08:10 Ghiad