IBM Granite Models

10/22/2024

This post announces the release and availability of IBM's Granite family of LLMs, detailing their different models (Granite 3.0, Granite for Code, Granite for Time Series, Granite Guardian) and their respective capabilities. It highlights their open-source nature, performance, and trustworthiness, and provides links to Hugging Face, Ollama, and Replicate for access. It also introduces developer resources like Cookbooks, Granite Kitchen, Granite Playground, and a Granite Workshop to facilitate model usage for various AI applications, including RAG, time-series forecasting, and code generation. Integration points with watsonx.ai are also mentioned.

6/27/2023

This post details IBM's collaboration with Hugging Face, highlighting contributions to open-source projects like TGIS and peft, and the creation of the Caikit AI toolkit. Caikit provides containerization and consistent APIs for serving Hugging Face models on any cloud. The post also mentions the Granite 3.2 models with new reasoning and vision capabilities, and their integration into the watsonx platform.

10/13/2022

This post details how to run AI inferencing on IBM Power10 processors by leveraging the new Matrix Math Accelerator (MMA) units. It explains the developer setup using Anaconda and Conda packages from the RocketCE channel, and presents a container-based demo framework with six real-world scenarios showcasing performance improvements (up to 3x) with MMA-optimized libraries compared to non-optimized ones. The demos cover image detection, action recognition, credit card fraud detection, and speech-to-text.