DeepSeek-r1 has caused a heated discussion in the global community due to its outstanding performance. However, as the number of users surges, and with cyberattacks in some unknown areas, official services are often overwhelmed. Today, I will teach you how to build a dedicated, never-down AI assistant at zero cost.
Why choose private deployment? What are the benefits of private deployment? In addition to service stability, private deployment also gives us complete control over data security, and allows us to freely adjust parameters and expand functions as needed. The free GPU resources currently provided by Tencent Cloud (equivalent to RTX 4080 level, 10,000 minutes per month, 16G graphics memory cloud service) are more than enough, allowing us to enjoy a smooth AI experience without having to invest in hardware. The detailed deployment steps are as follows:
Step 1: Preparatory work
Visit Tencent Cloud Studio
to complete registration and real-name verification. This step is crucial, as only real-name users can receive free GPU resources.
You can also choose resources provided by other cloud service providers. This article uses Tencent Cloud as an example. Free GPU resources are now available
Step 2: Create a workspace
Search for “Ollama” in the workspace template selection interface and select the basic type (free version) to create a workspace. Wait 2-5 minutes for initialization to complete.
Step 3: Model deployment
Create a new terminal in the workspace and execute the following commands in sequence:
Note: It takes about 10-15 minutes to download the model, depending on the network condition. After the download is complete, you can directly talk to AI in the terminal.
Step 4: Configure Intranet Penetration
In order to allow other devices to access your AI assistant, you need to configure intranet penetration:
- Check the Ollama running port
- Start intranet penetration
Note: 6399 is an example port number, please replace it according to the result of the previous step
After execution, you will get a public network access address in the format:
Step 5: Set automatic restart (optional but recommended)
In order to maintain stable service operation, it is recommended to add an automatic restart script:
Step 6: Configuration
CherryStudio is used as an example. For a detailed introduction,
- Download and install CherryStudio
- Add a new model configuration in the settings:
- Model name: deepseek-r1:14b
- Server address: fill in your gradio public network address
- API key: leave blank
Performance optimization tips
Want your AI assistant to respond faster? Try these optimization parameters:
Frequently asked questions
If you encounter a port being occupied:
Model download interrupted:
Insufficient memory
Tips
- Free GPU instances will automatically go to sleep at 2am every morning. It is recommended to avoid this time period
- The intranet penetration address will expire after 72 hours. Remember to check it regularly
- You can control the length of the response by adjusting the –max-tokens parameter to balance speed and completeness
- If you find that the response is slowing down, you can try restarting the service
At this point, your own AI assistant is ready to go. It is not only stable and reliable, but also completely under your control. Whether it is daily writing, programming, or data analysis, you will have a capable assistant to support you.
Remember to keep your access address safe, and if you have any questions about the deployment process, please feel free to discuss them in the comments section. I wish you all a new year in which you can reach new heights with the help of AI!
