[Performance]: When deployed DeepSeek-V3 on 8*H20(96GB), maximum model length only reaches 6500 using vllm, but with sglang can achieve 163840.

[zhaotyer]

Proposal to improve performance

No response

Report of performance regression

No response

Misc discussion on performance

vllm command python3 -m vllm.entrypoints.openai.api_server --model ${model_path} --port 8108 --max-model-len 6500 --gpu-memory-utilization 0.98 --tensor-parallel-size 8 --trust-remote-code --quantization fp8 sglang command python3 -m sglang.launch_server --model-path ${model_path} --tp 8 --trust-remote-code --mem-fraction-static 0.9 --host 0.0.0.0 --port 50050 --served-model-name atom

Your current environment (if you think it is necessary)

The output of `python collect_env.py`

Before submitting a new issue...

• [x] Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the documentation page, which can answer lots of frequently asked questions.