### What this PR does / why we need it?
There was an accuracy issue with **Qwen3-Next-80B-A3B-Instruct-W8A8**
model in the old version of **Triton-Ascend**, so, we are now adding one
nightly test to maintain it.
### Does this PR introduce _any_ user-facing change?
N/A
### How was this patch tested?
- vLLM version: v0.13.0
- vLLM main:
7157596103
Signed-off-by: IncSec <1790766300@qq.com>