Apple released OpenELM (Open-source Efficient Language Models), a suite of Open source big language models (LLM) designed to run directly on devices rather than relying on cloud servers.
These new language models are available on the Cuddly facea community where people can share AI codes.
According to a white paper published by Apple, OpenELM includes eight distinct models, four of which were pre-trained using the CoreNet library, while the other four are instruction-optimized variants. Apple’s approach notably uses a layered scaling strategy aimed at improving both accuracy and efficiency, setting a new standard for on-device AI performance.
Unlike previous versions that only provided model weights and inference code, Apple has gone a step further by offering comprehensive training frameworks, evaluation tools, and multiple model versions.
The impact of OpenELM is highlighted by its demonstrated performance gains. With a parameter budget of around a billion parameters, OpenELM shows a remarkable 2.36% improvement in accuracy over existing models like OLMo, while requiring half the pre-training tokens.
What is processing on the device?
On-device processing when it comes to AI or large language models means that the entire language model runs on the device and is powered by the chipset itself. Usually, AI processing requires a transfer to the cloud where the command is sent to the cloud server for processing and then the response is sent back to the user. On-device processing improves privacy and security. Along with this, it also makes things much faster compared to cloud processing. And it also costs companies less because they don’t need to power large servers to process all of the AI commands and tasks.
Apple to bring AI-based features in iOS 18
Apple is expected to bring a host of new AI-powered features to iPhones, iPads and other devices with future software updates. The AI features at the forefront will be iOS 18 and iPadOS 18. Leaks and rumors suggest that Apple is exploring the idea of running AI entirely on the device to improve privacy and security of device and user data.
These new language models are available on the Cuddly facea community where people can share AI codes.
According to a white paper published by Apple, OpenELM includes eight distinct models, four of which were pre-trained using the CoreNet library, while the other four are instruction-optimized variants. Apple’s approach notably uses a layered scaling strategy aimed at improving both accuracy and efficiency, setting a new standard for on-device AI performance.
Unlike previous versions that only provided model weights and inference code, Apple has gone a step further by offering comprehensive training frameworks, evaluation tools, and multiple model versions.
The impact of OpenELM is highlighted by its demonstrated performance gains. With a parameter budget of around a billion parameters, OpenELM shows a remarkable 2.36% improvement in accuracy over existing models like OLMo, while requiring half the pre-training tokens.
What is processing on the device?
On-device processing when it comes to AI or large language models means that the entire language model runs on the device and is powered by the chipset itself. Usually, AI processing requires a transfer to the cloud where the command is sent to the cloud server for processing and then the response is sent back to the user. On-device processing improves privacy and security. Along with this, it also makes things much faster compared to cloud processing. And it also costs companies less because they don’t need to power large servers to process all of the AI commands and tasks.
Apple to bring AI-based features in iOS 18
Apple is expected to bring a host of new AI-powered features to iPhones, iPads and other devices with future software updates. The AI features at the forefront will be iOS 18 and iPadOS 18. Leaks and rumors suggest that Apple is exploring the idea of running AI entirely on the device to improve privacy and security of device and user data.
