Tengxun cloud account purchase: GPU server one-click deployment of Stable Diffusion and DeepSeek practical operation guide

cloud 2026-07-24 阅读 6

For many friends who want to play AI painting and local language model, the biggest headache is

Environment Configuration

. When matching the environment, problems such as CUDA version conflict, PyTorch dependency error reporting, model download interruption, etc. are enough to dissuade 90% of the novices.

So, is there a way for us to avoid these underlying tedious details and achieve "out of the box"?

The answer is yes. Using tengxunyun's GPU cloud server (HAI or GN series) and matching the preset application image or Docker container, you can build your own in a few minutes at the earliest.

Stable Diffusion(AI drawing)

DeepSeek (large language model)

Exclusive private services.

This article will take you hand in hand to complete the whole process of deployment, to ensure that novices can follow suit.

1. Preparation: Server Selection and Infrastructure Configuration

Before you start, you need to prepare the infrastructure. Deploying the AI model requires hard computing resources (especially video memory):

Stable Diffusion(SD WebUI / ComfyUI): $\ge 8\text{GB}$(for example, NVIDIA T4 or V100/A10) and $16\text{GB}$or more are recommended.

DeepSeek Distilled Edition (7B/14B): $\ge 16\text{GB }$$ is recommended. If you want to run the large-parameter version, you need to configure a higher GPU instance.

1. Account and permission settings

If you don't have a cloud environment, you need to complete account registration and real name authentication first. For enterprise teams or enterprise users who need to purchase in bulk, it is recommended to do so through formal channels.

Purchase a Tencent Cloud account

and open for better architecture consulting support and exclusive discounts.

2. Create a GPU instance

Select a region (we recommend that you select the node closest to you, such as Guangzhou, Shanghai, and Beijing).

Key configuration: operating system/image: directly select the basic image of the built-in Ubuntu 20.04/22.04 + NVIDIA CUDA 11.8/12.x, or directly select the application image pre-installed with the AI environment. Network and security group: Open the required ports (such as SD port 7860, Ollama/DeepSeek 11434, or 8080 port).

2. scenario 1: one-click deployment of stable Diffusion(AI drawing)

If you want to have a AI studio that can be called at any time, you can follow these steps:

Method 1: Use a preinstalled image

(The fastest, suitable for pure novice)

When creating HAI instances in Tengxun Cloud, check directly in the application template

Stable Diffusion WebUI

After the deployment is completed, the console will automatically generate an access link, click to enter the WebUI interface, even the command line does not need to type.

Method 2: Deployment based on Docker containers (recommended, flexible and controllable)

If you choose a standard Linux GPU instance, it is recommended to use Docker to pull up the environment at speed:

#1. Mount the graphics driver and run the SD container

docker run -d --gpus all \

-p 7860:7860 \

-v /data/sd-models:/app/models \

--name sd-webui \

yanmitsu/sd-webui:latest

Tips: Model files are usually large. It is recommended to purchase a separate cloud hard disk and mount it to the/data directory to facilitate storage of various Checkpoint (such as SD 1.5, SDXL, and ControlNet) and Lora models to prevent the system disk from bursting.

After setting up, in the browser access

http:// your server IP:7860

You can start creating.

3. scenario 2: one-click deployment of DeepSeek private large models

DeepSeek is the best of the big open source models right now. Deploying DeepSeek on Tengxun cloud GPU server can ensure data privacy and access your own business system through API.

Here we recommend using

Ollama Open-WebUI

to create a ChatGPT-like interface.

Step 1: Install Ollama and pull the DeepSeek

After logging in to the server, execute the official one-click installation script:

The system automatically downloads and loads the model. At this time, the terminal can already have a dialogue with the DeepSeek!

Step 2: Deploy the Open-WebUI visualization interface

The terminal dialogue is not intuitive enough, we use Docker to pull up a beautiful Web interface:

After the deployment is completed, open the browser to visit http:// your server IP:8080, register the administrator account, and select the just downloaded deepseek-r1:7b to enjoy a fast and completely private AI question-and-answer experience!

4. common potholes and performance optimization tips

Display memory overflow (OOM)SD optimization: startup parameters plus-medvram or-lowvram can be

A small amount of performance loss in exchange for a significant reduction in the memory footprint. help.aliyun.com DeepSeek optimization: Select the quantized version (such as Q4_K_M) based on the GPU memory, and do not blindly pursue the high-parameter version. Alibaba Cloud Documentation

Bandwidth limit causes slow model download. Under the tengxun cloud volume charging mode, it is recommended to pull the peak bandwidth of the public network to more than 100Mbps. Because it is settled by traffic, large bandwidth allows you to download tens of GB of models in a few minutes, and then turn them down or shut them down after use, which is extremely cost-effective.

For data persistence and security, remember to set an IP access whitelist for the port in the security group, or configure Basic Auth password authentication to prevent your private GPU from being scanned and stolen by others.

Summary

Taking advantage of the computing power of Tengxun cloud GPU cloud server, with Docker and ready-made mirrors, whether it is to generate pictures with one click or to build a local large model knowledge base, it has become within reach.

For enterprises that have batch deployment requirements and need to flexibly open up the private cloud architecture, it is recommended to make resource planning in advance and identify compliant ones.

Purchase a Tencent Cloud account

and service support channels to ensure operational efficiency while reducing computing costs. Open the console now and set up your first AI server!