Modders Push AI Boundaries: Llama 2 Runs on Windows 98 PC and Xbox 360

Llama 2 Meets Retro Tech: AI on Windows 98 and Xbox 360

In a remarkable display of technological ingenuity, developers have successfully adapted Meta's Llama 2 AI model to run on outdated hardware, pushing the boundaries of AI accessibility and demonstrating the potential for widespread AI integration across diverse platforms.

Windows 98 Revival: Exo Labs' Groundbreaking Project

Exo Labs, an organization dedicated to democratizing AI access, has achieved a significant milestone by running a modified version of Meta's Llama 2 on a Windows 98 Pentium II machine 1

1

. This feat is particularly noteworthy as it operates without relying on power-hungry data centers, addressing environmental concerns associated with AI infrastructure.

The project faced numerous challenges, including:

1. Sourcing compatible PS/2 peripherals
2. Transferring files to the decades-old machine using FTP over Ethernet
3. Compiling modern code for pre-Pentium Pro architecture using Borland C++ 5.02
4. Adapting to older C programming conventions due to C++ compatibility issues

Despite the Pentium II's limited 128 MB of RAM, Exo Labs created a streamlined version of Llama 2 with 1 billion parameters, now available on GitHub for public experimentation 1

1

.

Xbox 360 Adaptation: Pushing Console Boundaries

Building on Exo Labs' success, developer Andrei David took on the challenge of porting llama2.c, a lightweight version of the Llama 2 model, to the Xbox 360 console 2

2

. This endeavor presented its own set of obstacles:

1. Adapting to the Xbox 360's big-endian PowerPC CPU architecture
2. Extensive endianness conversion for model configuration and weights
3. Optimizing code for the console's unified memory architecture

David successfully ran the llama2 model on the Xbox 360, demonstrating the potential for AI integration on gaming consoles 2

2

.

Implications and Future Prospects

These projects highlight several key points:

1. AI Accessibility: The ability to run AI models on older hardware opens up possibilities for wider AI adoption and experimentation.
2. Resource Efficiency: Working within hardware constraints forces developers to prioritize efficient memory usage and optimization.
3. Cross-Platform Potential: The success on both PC and console platforms suggests a broad range of devices could potentially support AI functionality.

As the experiments continue, we may see further adaptations of small LLM implementations, such as Hugging Face's smolLM, on various legacy platforms 2

2

. These developments could pave the way for more inclusive and diverse AI applications across a spectrum of devices, both old and new.

1

1

2

2