According to the company, Z-code models use “Mixture of Experts” learning techniques, which Microsoft says means they are more efficient. That’s because only parts of the model are necessary to complete tasks. Microsoft says the set of Z-code medals will enhance translations driven by machine learning. Furthermore, Translator and Azure AI will support more languages, including those that have less data available for training. Because of their efficiency, the models are significantly more scalable than previous resources, able to handle more parameters without increasing compute load.
Details
Microsoft has already used the Z-code models to boost common language understanding for tasks like text summarization and name entity recognition across Azure AI services. With this week’s announcement, this is the “the first time a company has publicly demonstrated that it can use this new class of Mixture of Experts models to power machine translation products.” The next phase is putting the models into production. To achieve this, Microsoft is using NVIDIA GPUs. Speaking of which, the company also announced this week the ability to support trusted execution environments (TEEs) for confidential computing on GPUs in Azure, also with the help of NVIDIA. One of the problems with TEEs in their current form is that they are built in CPUs. Microsoft wants to extend their capabilities to GPUs and to make that happen is collaborating with NVIDIA. This means data can become more secure at a hardware level for more powerful machines that handle intensive workloads. Tip of the day: Did you know that your data and privacy might be at risk if you run Windows without encryption? A bootable USB with a live-linux distribution is often just enough to gain access to all of your files. If you want to change that, check out our detailed BitLocker guide where we show you how to turn on encryption for your system disk or any other drive you might be using in your computer.
