[Bug] Qwen3-235B-A22B-Instruct-W8A8 is very easy npu OOM #12016
Description
Checklist
- 1. I have searched related issues but cannot get the expected help.
- 2. The bug has not been fixed in the latest version.
- 3. Please note that if the bug-related issue you submitted lacks corresponding environment info and a minimal reproducible demo, it will be challenging for us to reproduce and resolve the issue, reducing the likelihood of receiving feedback.
- 4. If the issue you raised is not a bug but a question, please raise a discussion at https://github.com/sgl-project/sglang/discussions/new/choose Otherwise, it will be closed.
- 5. Please use English, otherwise it will be closed.
Describe the bug
no any other complex configugres, mem-fraction-statio < 0.9 almost can not start service. even context_len very small
for example：
for 800I A2 single node，if context_len=10K，mem-fraction-statio must > 0.92, is not correct；
for 800I A2，1P(8 npus)1D(8 npus) disaggregation , context_len=4K，mem-fraction-statio must > 0.95, even 0.98
Reproduction
Environment
800I A2 64G, 2 nodes, stand-alone computer(using 8 fiber to connect the npus)