run包装环境

省流版本
host端：
TopsRider*.run -y，一键装驱动

容器内：
TopsRider*.run -y，一键装默认topsplatform的包
TopsRider*.run -C vllm -y，一键安装vllm

软件栈安装参考文档
https://support.enflame-tech.com/onlinedoc_dev_3.3/2-install/sw_install/content/source/installation.html#id12

run包网址
http://10.9.130.163:8080/qodftp_builds/tops_daily_packages/master/release_package/internal/x86_64/run/i3x/

daily包地址
http://10.9.130.163:8080/qodftp_builds/tops_daily_packages/master/test_package/x86_64/gcc7_with_paddle/

./TopsRider_i3x_3.5.20250808_deb_amd64.run --uninstall

一、环境安装

1、下载run包

sudo wget http://10.9.130.163:8080/qodftp_builds/tops_daily_packages/master/release_package/internal/x86_64/run/i3x/TopsRider_i3x_3.5.20250816_deb_amd64.run

2、静默安装
host里面
sudo chmod +x TopsRider_i3x_3.5.20250816_deb_amd64.run
sudo ./TopsRider_i3x_3.5.20250816_deb_amd64.run -y

3、起容器
sudo docker run -it --name eccl0801 -v /home:/home -v /sys/:/sys/ --network host --privileged --ipc=host artifact.enflame.cn/enflame_docker_images/ubuntu/qic_ubuntu_2204_gcc11:1.4.0
sudo docker run -it --name eccl0802 -v /home:/home -v /sys/:/sys/ --network host --privileged --ipc=host artifact.enflame.cn/enflame_docker_images/ubuntu/qic_ubuntu_2204_gcc11:1.4.0

4、容器里面

./TopsRider_i3x_3.5.20250816_deb_amd64.run -y 一键装默认topsplatform的包
./TopsRider_i3x_3.5.20250816_deb_amd64.run -C vllm -y
./TopsRider_i3x_3.5.20250816_deb_amd64.run -C vllm-gcu -y
./TopsRider_i3x_3.5.20250816_deb_amd64.run -C topscompressor -y
./TopsRider_i3x_3.5.20250816_deb_amd64.run -C fast-diffusers -y
./TopsRider_i3x_3.5.20250816_deb_amd64.run -C TopsInference -y
./TopsRider_i3x_3.5.20250816_deb_amd64.run -C xfuser -y

5、软件栈安装状态检查
topsinfo

二、验证环境是否OK
1、下个模型
sudo wget -nH --no-parent --recursive http://10.12.110.200:8080/sse_ard_data/inference/pretrained_models/DeepSeek-R1-Distill-Qwen-14B/

2、执行一下命令看一下能不能出结果
nohup vllm serve /home/cssg/yhy/model/sse_ard_data/inference/pretrained_models/DeepSeek-R1-Distill-Qwen-14B/
–max-model-len 32768
–block-size=64
–disable-log-requests &> server.log &

python3 -m vllm_utils.benchmark_serving
–backend vllm
–dataset-name random
–model /home/cssg/yhy/model/sse_ard_data/inference/pretrained_models/DeepSeek-R1-Distill-Qwen-14B/
–num-prompts 1
–random-input-len 1024
–random-output-len 1024
–trust-remote-code
–ignore_eos
–strict-in-out-len
–keep-special-tokens