Apple is set to join the company of developers of artificial intelligence models with its own OpenELM family. This is a lineup of Large Language Models (LLM) with open source code that can run entirely on a single device rather than connecting to cloud servers.
OpenELM stands for Open-source Efficient Language Models. They are aimed at on-device applications. Apple says the model family "aims to expand capabilities and strengthen the open research community, supporting future research efforts." The models were pre-trained on publicly available datasets with 1.8 trillion tokens from Reddit, Wikipedia, arXiv.org, and the like.
Apple is joining the public AI game with 4 new models on the Hugging Face hub! https://t.co/oOefpK37J9
— clem 🤗 (@ClementDelangue) April 24, 2024
OpenELM consists of small models designed for efficient text generation tasks. The corresponding code is published in the artificial intelligence code community Hugging Face. There are a total of 8 OpenELM models: 4 pre-trained and 4 instructionally tuned. They cover various parameter sizes, ranging from 270 million to 3 billion parameters.
OpenELM models can run on ordinary user devices such as laptops and smartphones. Apple notes that tests were conducted on "a workstation with an Intel i9-13900KF processor, 64GB of DDR5-4000 DRAM memory, and an NVIDIA RTX 4090 graphics processor with 24GB of video memory running Ubuntu 22.04," as well as "an Apple MacBook Pro with an integrated M2 Max system and 64GB of memory under macOS 14.4.1."
Apple offers its OpenELM models under a so-called "sample code license," as well as providing various training checkpoints, model performance statistics, and instructions for pre-training, evaluation, instruction tuning, and parameterization.
The sample code license does not prohibit commercial use or modification, but only states that "if you distribute Apple software in its entirety and without modification, you must retain this notice as well as the following text and disclaimers in all such distributions of Apple software."
In addition, the company notes that the models "are provided without any security guarantees. Thus, there is a possibility that these models will produce inaccurate, harmful, biased, or undesirable results in response to user queries."
Source: venturebeat
Comments (0)
There are no comments for now