Apple keeps advancing the open-source Artificial market with more innovative products.

Apple’s ability create AI models that are incredibly compact and powerful is unmatched in the industry.
Apple Intelligence’s research team has released two new language models that are small, but perform well. They can be used to train AI generators. Apple’s Machine Learning team is participating in an open-source DataComp for Language Models Project with others in the industry. Apple’s two new models have been shown to be on par with or even better than other leading training models such as Llama 3 or Gemma. These language models are used to train AI engines like ChatGPT by providing a standardised framework. This includes an architectural framework, parameters, as well as filtering datasets for higher-quality data that AI engines can draw from. Apple’s submission for the project includes two models, a larger model with seven billion parameter and a smaller model with 1.4 billion parameter. Apple’s team stated that the larger model outperformed MAP-Neo by 6.6 per cent in benchmarks. Apple’s DataComp LM model is 40 percent more efficient than the MAP-Neo in achieving these benchmarks. It was the most efficient model amongst those who used open datasets and comparable to those using private datasets. Apple has made all of its models open – the dataset, weight models and training code is available for researchers to use. Both the smaller and larger models scored high enough in the Massive Multi-task Language Understanding (MMLU) benchmarks to be competitive with commercial models. Apple’s benchmarks for its larger dataset are competitive with other models. By debuting Apple Intelligence at its WWDC conference, in June, and Private Cloud Compute in July, the company silenced critics that claimed Apple was lagging behind in artificial intelligence applications on its devices. Research papers published by the Machine Learning team before and after the event proved that Apple is an AI industry leader. Apple’s models are not intended to be used in future Apple products. These are community research projects that aim to improve the efficiency of curating small or big datasets for AI models. Apple’s Machine Learning Team has shared its research with the AI community in the past. HuggingFace.co is a platform that aims to expand the AI community. It contains datasets, research papers, and other assets.