How to run DeepSeek's AI model on PC for free
What's the story
DeepSeek R1, an innovative open-source AI model from a Chinese start-up, is making waves in the tech world. The company's app has just hit top spot on Apple's App Store in the US.
This groundbreaking model has been developed through "large-scale reinforcement learning (RL) without supervised fine-tuning (SFT) as a preliminary step," as revealed in a research paper by Chinese AI lab DeepSeek.
DeepSeek R1's training method consists of two key phases.
Training method
Training and application process
In the first SFT phase, humans give the model desired outputs to teach it what is good or bad.
This lays the groundwork for the next phase, RL, where the model produces different outputs and humans rank them by quality.
This continues until the model learns to produce satisfactory results consistently.
Usage scope
DeepSeek R1: A tool for research and beyond
Currently, DeepSeek R1 is mainly designed for research. However, being open-source and licensed by the Massachusetts Institute of Technology (MIT), it can be used for personal as well as professional purposes.
The model runs on Ollama, an open-source platform that provides a space on your computer for AI models to run without cloud computing. This way, your data stays secure with you.
Model features
Capabilities and cost-effectiveness
DeepSeek R1 is a powerful model with 671 billion parameters, capable of performing everything from complex math problems to real-time decision-making.
It comes with a range of models from 1.5 billion to 70 billion parameters, each requiring different RAM configurations for optimal performance.
As per DeepSeek's research paper, running DeepSeek R1 costs about ₹684 per million tokens, much cheaper than OpenAI's o1 model which costs around ₹1,281 for input and a whopping ₹5,130 for output per million tokens.
Setup process
Using DeepSeek R1 on your computer
Setting up DeepSeek R1 on your computer is pretty easy.
Windows users can download the installer from the official Ollama website, while Mac users can download it directly from Olla or use Homebrew, a software package manager.
To run DeepSeek for the first time, just open your terminal and type: "ollama run deepseek-r1:8b" (without the double quotes).
The model downloads and launches on your machine automatically.