Summary Microsoft announces new supercomputer, lays out vision for future AI work - Source news.microsoft.com
2,565 words - html page - View html page
One Line
Microsoft has released the world's largest publicly available AI language model and advancements in distributed training that make large-scale AI models more resource efficient and cost-effective, making the performance and cost of the entire Azure cloud better.
Key Points
- Microsoft has developed one of the top five publicly disclosed supercomputers in the world, which has enabled advances in Office features such as smart find, key insights and suggested replies.
- Microsoft has released the Turing model for natural language generation with 17 billion parameters, the largest publicly available language AI model to date.
- Azure AI provides access to powerful compute with AI accelerators and networks that power supercomputers.
- Microsoft has unveiled a new supercomputer in Azure to train large AI models, part of a partnership with OpenAI.
- Microsoft is making its large AI models and training optimization tools available through Azure AI services and GitHub.
- Microsoft has announced a new supercomputer and outlined its vision for future artificial intelligence work across the UK, India, Latin America, Asia, and Europe.
Summaries
197 word summary
Microsoft has announced a new supercomputer in Azure to train large AI models in partnership with OpenAI. This supercomputer has 285,000 CPU cores, 10,000 GPUs and 400 gigabits per second of network connectivity for each GPU server and ranks in the top five of supercomputers in the world. Azure AI provides access to powerful compute with AI accelerators and networks that power these supercomputers. Microsoft is making its large AI models and training optimization tools available through Azure AI services and GitHub.
The largest publicly available AI language model in the world, the Microsoft Turing model for natural language generation, has been released to researchers. It has a 17 billion parameter capacity for self-supervised learning. Microsoft has released DeepSpeed PyTorch and ONNX Runtime for distributed training, enabling advances in Office features such as smart find, key insights and suggested replies. Bing search accuracy has improved by up to 125%.
Microsoft's leading-edge infrastructure for training large AI models makes the performance and cost of the entire Azure cloud better. Researchers and engineers have achieved efficiencies in distributed training that make large-scale AI models more resource efficient and cost-effective. These advancements are slowly finding their way into everyday life.
529 word summary
Microsoft has developed one of the top five publicly disclosed supercomputers in the world. This leading-edge infrastructure for training large AI models makes the performance and cost of the entire Azure cloud better. Microsoft researchers and engineers have achieved efficiencies in distributed training that make using large-scale AI models more resource efficient and cost-effective. The same advancements used in Formula 1 race cars eventually find their way into everyday cars. Microsoft has developed its own suite of optimization tools, such as DeepSpeed PyTorch and ONNX Runtime, to distribute large AI models across different machines (model parallelism) and split the training data into batches (data parallelism). This allows for more computing power and has enabled advances in Office features such as smart find, key insights and suggested replies. The Microsoft Turing model has been used to improve Bing search accuracy by up to 125%. It also allows for fine tuning of models with smaller datasets and resources. This results in a richer understanding of grammar, concepts and contextual relationships. AI models are advancing rapidly, becoming more sophisticated and larger. In self-supervised learning, AI models can learn from large amounts of unlabeled data by predicting missing words and sentences. Microsoft has released the Turing model for natural language generation with 17 billion parameters, the largest publicly available language AI model to date. Microsoft has also added support for distributed training to the ONNX Runtime and released a new version of DeepSpeed for PyTorch. Finally, Microsoft will soon open source its Microsoft Turing models, giving developers access to the same powerful language models. Azure AI provides access to powerful compute with AI accelerators and networks that power supercomputers. OpenAI and Microsoft have developed a supercomputer designed to accelerate the cycle of developing powerful AI technologies. The supercomputer has 285,000 CPU cores, 10,000 GPUs and 400 gigabits per second of network connectivity for each GPU server and ranks in the top five of supercomputers in the world. Microsoft is making its large AI models and training optimization tools available through Azure AI services and GitHub so developers, data scientists and business customers can easily leverage the power of AI at Scale. This includes the Microsoft Turing models which are used to improve language understanding tasks across Bing, Office, Dynamics and other products. The largest publicly available AI language model in the world, the Microsoft Turing model for natural language generation, has also been released to researchers. Microsoft has unveiled a new supercomputer in Azure to train large AI models, part of a partnership with OpenAI. This new class of models can perform multiple tasks such as summarizing speeches, moderating content, finding relevant passages and generating code. These models excel at learning language, grammar, knowledge, concepts and context. The potential benefits extend beyond narrow advances in one type of AI model to new applications that are hard to imagine. Microsoft is laying out a vision for future AI work. Microsoft has announced a new supercomputer and outlined its vision for future artificial intelligence work. This vision will be implemented across the UK, India, Latin America, Asia, and Europe. The news was released on the Microsoft Official Blog, Company News, and Our Company Home pages.