logo
30
5
WeChat Login

Deploying DeepSeek AI Applications to HAI

Tencent Cloud High-performance Application Service (HAI) is a GPU application service product for AI and scientific computing.

This tutorial explains how to deploy deepseek to HAI using CNB and Docker.

1. Build Docker Image

Build and publish a Docker image based on deepseek. Alternatively, you can choose pre-built deepseek images from the cnb deepseek repository, such as docker.cnb.cool/examples/ecosystem/deepseek/1.5b:latest.

2. Create HAI Instance

Log in to the HAI Console and click Create New.

Under Application Selection > Basic Environment, choose Ubuntu 20.04.

Fill in other form fields as needed and create the instance.

After successful creation, Tencent Cloud will send a notification containing the default username and password for SSH login.

Open Ports

In the instance details page, open ports 11434 and 8080 in the port configuration.

11434 is the service port for ollama's deepseek service, and 8080 is the web access port for open-webui.

3. Deploy deepseek

3.1 Manual Deployment

Using the default username and password recorded earlier, log in to the HAI instance via SSH client.

Copy the startup script deploy.sh to the HAI instance (e.g., /app/deepseek/deploy.sh) and execute the following commands:

sudo su
cd /app/deepseek/
chmod +x ./deploy.sh
./deploy.sh

After starting the deepseek container, you can access http://{Instance Public IP}:11434 to view the ollama service.

After starting the open-webui container, wait about 1-2 minutes until the container status becomes healthy, then access http://{Instance Public IP}:8080 to view the page and start conversations.

3.2 Deployment via Cloud Native Build

Alternatively, you can use CNB pipeline with the ssh plugin to execute deployment commands:

Add a yaml file in private or key repository to record the instance's public IP, username and password, for example:

# secret.yml
HAI_IP: xxx
HAI_USERNAME: xxx
HAI_PASSWORD: xxx

# allow_* controls the accessibility scope of this file
allow_slug: xxx
# .cnb.yml
.host-info: &host-info
  host: $HAI_IP
  username: $HAI_USERNAME
  password: $HAI_PASSWORD
  port: 22

main:
  push:
    # Import the secret.yml mentioned above as environment variables
    - imports: https://xxx/secret.yml
      stages:
        # Upload script via plugin (can also be pre-uploaded)
        - name: trans shell
          image: tencentcom/rsync
          settings:
            host:
              - ${HAI_IP}
            user: ${HAI_USERNAME}
            key: ${HAI_KEY}
            target: /home/ubuntu/deepseek
            source: ./deploy.sh
        - name: exec shell
          image: tencentcom/ssh
          settings:
            <<: *host-info
            script: |
              sudo chmod +x /home/ubuntu/deepseek/deploy.sh
              sudo bash /home/ubuntu/deepseek/deploy.sh

After starting the deepseek container, you can access http://{Instance Public IP}:11434 to view the ollama service.

After starting the open-webui container, wait about 1-2 minutes until the container status becomes healthy, then access http://{Instance Public IP}:8080 to view the page.

About

部署 deepseek 到 腾讯云高性能应用服务 HAI

example
Language
Markdown94.6%
Shell5.4%