Skip to content

GitLab

Closed
Created by gaoxuan@gaoxuan

qwen2.5 7B跑8卡报错

qwen2.5 7B跑8卡报错Total number of attention heads (28) must be divisible by tensor parallel size (8). image