NVIDIA Unveils New AI Models: Phi-3 and Granite Code

Generative AI continues to reshape industries with advanced foundation models that enhance content creation and data interpretation. According to the NVIDIA Technical Blog, the tech giant has introduced two new model families under its NVIDIA AI Foundation: Phi-3 and Granite Code.

Phi-3 Language Models

The Phi-3 series, developed in collaboration with Microsoft, encompasses small language models (SLMs) optimized for high performance and computational efficiency. These models excel in tasks such as content generation, summarization, question-answering, and sentiment analysis. Their robust reasoning capabilities make them ideal for various applications requiring logical deductions and accurate responses.

Phi-3 Vision Model

The Phi-3 family also features the Phi-3 Vision model, a 4.2 billion parameter multimodal model designed to process and interpret both text and visual data. With support for 128K tokens, this model can analyze complex visual elements within images, such as charts, graphs, and tables, making it highly suitable for data-intensive tasks.

Granite Code

IBM has contributed the Granite Code models, which are open programming models designed to assist with various coding tasks. These models, trained on 116 programming languages, can generate code examples, identify and fix errors, and provide explanations for code segments. Their performance on coding benchmarks is state-of-the-art, and they are trained on license-permissible data, making them suitable for enterprise use.

Optimized for Performance

Both Phi-3 and Granite Code models are optimized for latency and throughput using NVIDIA TensorRT-LLM. These models join over three dozen popular AI models supported by NVIDIA NIM, a microservice designed to simplify the deployment of performance-optimized models at scale. NVIDIA NIM enables a significant increase in the number of enterprise application developers who can contribute to AI transformations.

NVIDIA continues to collaborate with leading model builders to support their models on a fully accelerated stack, ensuring optimal performance and ease of deployment.

Get Started

To experience, customize, and deploy these models in enterprise applications, visit the API catalog. With free NVIDIA cloud credits, developers can start testing the models at scale and build a proof of concept by connecting their applications to the NVIDIA-hosted API endpoint running on a fully accelerated stack.