This Practical AI podcast episode interviews Kate Soule, Director of Technical Product Management at Granite for IBM, about IBM's family of large language models (LLMs). The discussion covers IBM's decision to open-source Granite under the Apache 2 license, the rationale behind developing different sized models (1B to 8B parameters), and the incorporation of "mixture of experts" architecture for efficient inference. Soule also details Granite's multimodal capabilities, including vision and time series models, and the role of Granite Guardian for ensuring responsible AI. Listeners gain insights into the practical considerations of building and deploying LLMs, including the trade-offs between model size, performance, and cost, and the importance of safety and security features.
