如果之前装过先删掉之前的执行文件（如果没有，先跳过这步）
rm -f /usr/local/bin/ix-ctk
rm -f /usr/local/bin/ix-container-runtime
rm -rf /etc/ix-container-runtime
rm -rf /usr/local/ix-container-toolkit
下载新的ix-container-toolkit包
密码请咨询天数售后
sftp -P 29880 iluvatar_mr@iftp.iluvatar.com.cn
get /SDK_4.4.0/x86/cloudnative/ix-container-toolkit-1.1.0_x86_64.run /home (x86下载这个）
get /SDK_4.4.0/arm/cloudnative/ix-container-toolkit-1.1.0_aarch64.run /home (arm下载这个）
安装ix-container-toolkit
apt update
apt install -y jq
chmod a+x /home/ix-container-toolkit-1.1.0_x86_64.run
bash /home/ix-container-toolkit-1.1.0_x86_64.run --install
ix-ctk runtime configure --runtime docker --ix-set-as-default
systemctl daemon-reload
systemctl restart docker
通过runtime指定卡并且启动模型
docker run -it --rm \
--runtime iluvatar \
-e IX_VISIBLE_DEVICES=0,1,2,3 \
--network=host \
-v /data:/data \
registry.iluvatar.com.cn:10443/customer/sz/vllm0.11.2-4.4.0-x86:v9 \
/bin/bash -c "
ixsmi
export VLLM_ENFORCE_CUDA_GRAPH=1
export VLLM_W8A8_MOE_USE_W4A8=1
export VLLM_KV_DISABLE_CROSS_GROUP_SHARE=1
vllm serve /data/model/Qwen3.5-122B-A10B-w4a8 \
--trust-remote-code \
--tensor-parallel-size 4 \
--pipeline-parallel-size 1 \
--max-num-seqs 64 \
--enable-chunked-prefill \
--max-model-len 16384 \
--host 0.0.0.0 \
--port 1234 \
--gpu-memory-utilization 0.90 \
--compilation-config \"{\\\"cudagraph_mode\\\":\\\"FULL_DECODE_ONLY\\\",\\\"level\\\":0}\"
"
