Un Ministral, des Ministraux | Mistral AI

What this post added

Introduction of Ministral 3B and Ministral 8B models, specifically designed for edge and on-device inference. Key technical features include support for 128k context length and an interleaved sliding-window attention pattern in Ministral 8B for efficient inference. The post presents benchmark results demonstrating superior performance compared to existing models in the sub-10B parameter class.