Quick Deployment

This section helps you quickly building-a-deepseek-inference-system on Huawei Cloud.

1. Log in to Huawei Cloud Quick-Start Guides and choose Building a DeepSeek Inference System. Select a region from the Data Center drop-down list and click Deploy.
   Figure 1 Selecting a solution
   

2. On the Select Template page, click Next.
   Figure 2 Selecting a solution
   

3. On the Configure Parameters page, enter a stack name, configure parameters based on Table1 Parameter description, and click Next.
   Figure 3 Configuring parameters
   

   Table 1 Parameter description

   Parameter	Type	Mandatory	Description	Default Value
   vpc_name	string	Yes	Virtual Private Cloud (VPC) name. This template uses a newly created VPC and the VPC name must be unique. It can contain 1 to 54 characters, including only letters, digits, underscores (_), hyphens (-), and periods (.).	building-a-deepseek-Inference-system-demo
   security_group_name	string	Yes	Security group name. This template uses a newly created security group. For details about how to configure security group rules, see (Optional) Modifying Security Group Rules. It can contain 1 to 64 characters, including only letters, digits, underscores (_), hyphens (-), and periods (.).	building-a-deepseek-Inference-system-demo
   ecs_name	string	Yes	Cloud server name, which must be unique. It consists of 1 to 64 characters and can include letters, digits, underscores (_), hyphens (-), and periods (.).	building-a-deepseek-Inference-system-demo
   distilled_model	string	Yes	DeepSeek-R1-Distill model. Supports Qwen-1.5B, Qwen-7B, Llama-8B. Default is Qwen-1.5B.	DeepSeek-R1-Distill-Qwen-1.5B
   dify_enable	string	Yes (GPU-based)	Whether installing Dify along with Ollama. Dify provides the out-of-box web application to interact with the model.	enable
   ecs_flavor	string	Yes	Cloud Server Instance Specifications: For 1.5B model, it is recommended to use x1.4u.4g or higher; for 7B and 8B models, it is recommended to use x1.16u.16g or higher. NOTE: For GPU-based solution, GPU-accelerated type is required. Value can be found from the specification list page of the documentation. (Before executing the plan, please ensure the resource is available in the corresponding region).	DeepSeek-R1-Distill-Qwen-1.5B
   ecs_password	string	Yes	Initial password of the cloud server. The password can include 8 to 26 characters and must contain at least three of the following character types: uppercase letters, lowercase letters, digits, and special characters (!@$%^-_=+[{}]:,./?). The password cannot contain any username or the username spelled backwards. The administrator username is root.	false
   system_disk_size	number	Yes	System disk size of the cloud server. .The default disk type is General Purpose SSD, and the unit is GB. The system disk can only be increased. The default value is 100. Value range: 40-1,024.	40
   charging_mode	string	Yes	Billing mode. By default, expenses are automatically deducted. The value can be postPaid (pay-per-use) or prePaid (yearly/monthly).	postPaid
   charging_unit	string	Yes	Subscription period type. This parameter is valid only when the charging_mode is set to prePaid (yearly/monthly). The value can be month or year.	month
   charge_period	number	Yes	Subscription period. This parameter is valid only when charging_mode is set to prePaid (yearly/monthly). The default value is 1. Value range: 1-9 (charging_unit set to month); 1-3 (charging_unit set to year).	1

4. On the Configure Stack page, select rf_admin_trust from the Agency drop-down list and click Next. This step is optional if you use an account (HUAWEI ID) or use an IAM user in the admin user group.
   Figure 4 Configuring a stack
   

5. On the Confirm Configurations page, confirm the configurations and click Create Execution Plan.
   Figure 5 Confirming the configurations
   

6. In the displayed Create Execution Plan dialog box, enter an execution plan name and click OK.
   Figure 6 Creating an execution plan
   

7. Wait until the status of the execution plan changes to Available and then click Deploy in the Operation column. In the displayed dialog box, click Execute.
   Figure 7 Execution plan page
   
   Figure 8 Confirming the execution plan
   

8. (Optional) If you select the yearly/monthly billing mode and your account balance is insufficient, log in to the Billing Center to manually pay for the order. You can refer to Table2 Resource planning costs (yearly/monthly) to see the total price.
9. Wait until the message "Apply required resource success" is displayed on the Events tab page. This means the deployment is complete. The deployment takes about 10 minutes, which will be delayed by network fluctuations.
   Figure 9 Resources created