EXO Labs Headlines

The AI Realist

The article discusses how EXO Labs, an open-source distributed inference project, achieved a 2.8x speedup by combining DGX Sparks and a Mac Studio M3 Ultra for AI inference workloads.

Companies:EXO Labs

Pawel Dubiel (Personal Blog)

This blog post details Exo Cluster, an open-source software framework from Exo Labs, that enables the creation of distributed AI superclusters using everyday devices for local LLM inference.

Companies:EXO Labs

3DVF

EXO Labs, founded by Oxford University researchers, successfully ran a LLama2.c-based language model on a 1998 Pentium II processor with 128 MB of RAM, achieving 39.31 tokens per second. This experiment challenges the notion that only cutting-edge hardware can handle AI, suggesting a future of affordable and energy-efficient intelligence for older devices. The article highlights the potential for democratizing AI by lowering hardware barriers and enabling local inference for privacy and resilience.

Companies:EXO Labs

Tom's Hardware

EXO Labs demonstrated a significant AI benchmark gain by combining NVIDIA DGX Spark systems with Apple M3 Ultra Mac Studio for disaggregated AI inference. Their open-source EXO framework efficiently distributes LLM workloads across mixed hardware, showcasing a 2.8x speedup for Meta's Llama-3.1 8B model.

Companies:EXO Labs