GLM-130B icon indicating copy to clipboard operation
GLM-130B copied to clipboard

6 cards inference

Open wangheqi987 opened this issue 1 year ago • 1 comments

When I tried to inference with 6 cards, I got AssertionError: 32768 is not divisible by 6.

I found 32768 sourced from inner-hidden-size.

Why is the card number should be divisible by 32768?

What shoud I do to make it run with 6 cards?

wangheqi987 avatar Jul 05 '23 02:07 wangheqi987

你这个6张卡是集群部署吗?

ChenBinfighting1 avatar Jul 26 '23 09:07 ChenBinfighting1