![\"\"](\"https:\/\/i.gzn.jp\/img\/2025\/07\/06\/exo-ai-cluster\/00.jpg\")
<\/a><\/p>\nInference of large-scale language models requires a computer with powerful computing resources, but high-performance computers are expensive and difficult to obtain. ‘ exo<\/b> ‘ is a system that can connect multiple devices to a network to create a cluster for AI processing, and is promoted as a system that can gather familiar devices and use them for AI processing.<\/p>\n
EXO Labs – Enterprise On-Premise AI<\/b>
https:\/\/exolabs.net\/<\/a><\/p>\n GitHub – exo-explore\/exo: Run your own AI cluster at home with everyday devices \ud83d\udcf1\ud83d\udcbb \ud83d\udda5\ufe0f\u231a<\/b> Below is an image showing an example of exo in action. A single AI cluster is built by combining a MacBook Pro with a processing power of 10.6 TFLOPS and a PC equipped with a GeForce RTX 4060 Ti with a processing power of 1544.0 TFLOPS. According to the exo project page, it supports devices such as ‘iPhone’, ‘iPad’, ‘Android devices’, ‘Mac’, ‘NVIDIA GPU’, and ‘Raspberry Pi’. In other words, by using exo, you can effectively use ‘smartphones that have been replaced and are no longer of use’ or ‘Raspberry Pis that were bought for electronic work but are not used’ for AI processing.<\/p>\n exo supports AI model series such as ‘Llama’, ‘Mistral’, ‘LLaVA’, ‘Qwen’, and ‘Deepseek’. For example, in the case of ‘llama 3.1 8B (fp16)’, which requires 16GB of memory to run, it can be run by combining multiple devices such as ‘two MacBook Airs with M3 chips with 8GB of memory’ or ‘two Raspberry Pi 400s with 4GB of memory and one Mac Mini with 8GB of memory’.<\/p>\n At the time of writing, exo is in the early experimental stages of development and is looking for developers. In addition, a development contribution program with a bounty is also being developed. The bounty varies depending on the task, and you can check the task content and bounty amount at the following link.<\/p>\n
https:\/\/github.com\/exo-explore\/exo
<\/a><\/b><\/p>\n<\/p>\n<\/a><\/p>\n