Introduction

In a jaw-dropping revelation, [EXO Labs] has successfully adapted the Llama 2 language model to run on the nearly ancient operating system, Windows 98. This bold move raises intriguing questions about software control and personalised computing. After all, if a program can function on Windows 98, it implies a degree of freedom from the clutches of major tech corporations that seek to regulate user experiences.

The Experience of Running LLM on Outdated Hardware

So, what’s the experience like running a large language model (LLM) on such outdated hardware? Despite facing hurdles like compatibility issues with peripherals (hello, PS/2 connectors!) and the challenge of file transfers (thank goodness for FTP over Ethernet!), the experimental setup surprisingly delivers decent results.

Performance Metrics

In fact, a Windows 98 machine equipped with a Pentium II processor and a mere 128 MB of RAM manages to generate approximately 39.31 tokens per second using a 260K parameter version of the Llama 2 model. Upgrading to the more extensive 15M parameter model yields a slower but workable 1.03 tokens per second. While these figures may seem painfully slow by modern standards, the crucial takeaway is that it actually works — proving that older hardware still has the potential to engage with cutting-edge tech.

Coherent Outputs from Smaller Models

It's essential to highlight that although contemporary LLMs boast billions of parameters, the smaller models being run on a vintage platform can still deliver coherent outputs. Likely fueled by focused training and specific applications, these miniature models can perform surprisingly well with the right conditions and tools, many of which are conveniently available on GitHub for enthusiasts and developers alike.

Broader Initiative by EXO Labs

This remarkable project is not just a one-off feat but part of a broader initiative by EXO Labs, which encompasses twelve unique projects. Their goal? To ensure that the future of AI is accessible to everyone, running independently of the powerful tech giants that currently dominate the market.

Conclusion

In a world where software freedom is continuously threatened, this impressive demonstration serves as a reminder that innovation can thrive even on the oldest machines. Are we witnessing the dawn of a new era of computing freedom? Only time will tell!