Deployment Models for LLMs
The chart below shows four basic deployment models, with examples, advantages, and disadvantages for each. The purple background models are better for privacy.
Brand Name SaaS
Examples:
- Website: OpenAI.com
- Website: X.com/grok
- Website: Perplexity.ai
- Website: Deepseek.com
Advantages: Available on any connected device, usually have free tier.
Disadvantages: User data may be at risk. User not in control of inputs and outputs. Monthly fee for desired features, models.
User Device
Examples:
- Software: Brad’s LLMs
- Software: Ollama
- Hardware: Laptop or Desktop PC
Advantages: No data leaves the device, free or fixed software cost.
Disadvantages: Large download, large (16+ GB) RAM requirement, not wicked fast without GPU.
Note: Software and model run on the device with no communication to other servers.
Private Local Server
Examples:
- Software: Brad’s LLMs
- Software: Ollama
- Hardware: M4 Mac Mini, Raspberry Pi 5, PC
Advantages: Available on any connected device on your local network or with access to your local network.
Disadvantages: Requires a dedicated device with adequate resources (16+ GB RAM), maintenance, security.
Note: Software and model run on the local server with no communication to other servers.
Private Hosted Server
Examples:
- Website: KevinFerron.com
Advantages: Available on any connected device, data sent to and from server not retained (by contract).
Disadvantages: Monthly fee for private server.
Note: Software and model run on the private server with no communication to other servers.