James Ding
February 26, 2025 15:38
Microsoft announces a new PHI SLM, including multimodal PHI-4, which has been trained in the NVIDIA GPU, to improve AI function by using efficient resources.
Microsoft announced the latest additions to the new Phi-4-Multimodal and the PHI-4-Mini model, which has been trained using the NVIDIA GPU. According to NVIDIA, this development is an important stage in the evolution of language models that focus on efficiency and diversity.
Development of small language models
SLM has emerged as a practical solution for the problem raised by LLMS (Lange Language Models), which requires considerable computational resources despite its ability. SLM is designed to work efficiently within a limited environment, suitable for deployment to a device with limited memory and calculation power.
The new Phi-4-Multimodal model of Microsoft is particularly noteworthy as the ability to process multiple types of data, including text, audio and images. This feature opens up new possibilities of applications such as automated voice recognition, translation and visual reasoning. The training of this model includes 512 NVIDIA A100-80GB GPUs for 21 days, emphasizing the intensive calculation efforts necessary to achieve functionality.
PHI-4-Multimodal and Phi-4-Mini
The PHI-4-Multimodal model boasts 5.6 billion parameters, showing excellent performance in automated voice recognition and ranking 6.14%of Word error in WorgingFace OpenASR leader board. This achievement emphasizes the potential of a model that improves voice recognition technology.
With PHI-4-Multimodal, Microsoft introduced Phi-4-Mini, a text-only model optimized for chat applications. PHI-4-Mini is designed to effectively handle long formats with 3.8 billion parameters, providing a 128K token context window. Training included 1024 NVIDIA A100 80GB GPU for 14 days, reflecting the focus of models on high -quality education data and code.
Distribution and accessibility
Both models can be used in Microsoft’s Azure AI Foundry and provide a platform for AI application design, customization and management. The user also allows users to explore these models through the NVIDIA API catalog and provide a sandbox environment for testing and integrating this model in various applications.
NVIDIA’s cooperation with Microsoft expands beyond training these models. Partnerships include software and model optimization such as PHI to promote AI transparency and support open source projects. This collaboration aims to develop AI technology throughout the industry, from medical to life science.
For more information, visit the NVIDIA blog.
Image Source: Shutter Stock