Apple Empowers Open Research Community with On-Device OpenELM Models
In a groundbreaking move, Apple has recently released a series of open source large language models (LLMs) that run on-device rather than through traditional cloud servers. Named OpenELM (Open-source Efficient Language Models), these models are now available on the Hugging Face Hub for researchers, developers, and companies alike to explore.
Comprised of eight total models, four pre-trained using the CoreNet library and four instruction-tuned models, Apple has employed a layer-wise scaling strategy to enhance both accuracy and efficiency. By providing access to code, training logs, and multiple versions, Apple aims to drive faster progress and ensure more reliable results within the natural language AI field.
OpenELM boasts a state-of-the-art open language model that leverages this layer-wise scaling strategy for unprecedented accuracy. Apple’s decision to release these models signifies a commitment to empowering and enriching the open research community with cutting-edge language models.
Moreover, this open sharing of information allows for thorough investigation of risks, data, and model biases by a wide range of stakeholders. Not only does it facilitate collaboration and transparency within the industry, but it also serves as a strategic tool for Apple to attract top talent in the form of engineers, scientists, and experts.
Rumors have also emerged suggesting that Apple plans to incorporate new AI features, potentially running these large language models on-device in the upcoming iOS 18 update. This move is said to prioritize user privacy by keeping the processing local rather than relying on external servers.
By introducing OpenELM models, Apple is not only revolutionizing the field of natural language processing but also setting a precedent for open collaboration and innovation within the tech industry.
“Prone to fits of apathy. Devoted music geek. Troublemaker. Typical analyst. Alcohol practitioner. Food junkie. Passionate tv fan. Web expert.”