Roboticker Posted April 26 Share Posted April 26 Apple has released eight open-source AI language models, collectively named OpenELM. These models are designed to operate directly on devices, marking a significant shift from the traditional cloud-based AI systems. The OpenELM models are divided into two categories. Four of them are “pretrained,” essentially raw, next-token versions of the model. The remaining four are instruction-tuned, fine-tuned for instruction following, making them ideal for developing AI assistants and chatbots. These models are not just proof-of-concept research models. They could potentially form the foundation of future on-device AI offerings from Apple. The models range from 270 million to 3 billion parameters, making them much smaller than many existing AI models. Apple’s innovative approach with OpenELM includes a “layer-wise scaling strategy” that reportedly allocates parameters more efficiently across each layer. This not only saves computational resources but also enhances the model’s performance while being trained on fewer tokens. The models are currently available on the Hugging Face Hub under an Apple Sample Code License. While there are some restrictions in the license, the source code for OpenELM is readily available. This move by Apple mirrors efforts by other tech giants to create useful small AI language models that can run locally. It’s a significant step towards making AI more accessible and efficient, and it opens up new possibilities for on-device AI applications. What are your thoughts on this development? Do you think on-device AI is the future of artificial intelligence? Share your views in the comments below. Read more: https://huggingface.co/apple/OpenELM https://applemagazine.com/apple-introduces-openelm-open-source-ai-models-for-on-device-use/65059 https://www.macrumors.com/2024/04/24/apple-ai-open-source-models/ https://www.theinformation.com/briefings/apple-releases-small-open-source-ai-models-for-on-device-applications Image: Rimidolove | Dreamstime.com Quote Link to comment Share on other sites More sharing options...
Recommended Posts