Mistral AI Expands: New Models and Features!
Mistral AI has unveiled exciting updates, introducing the 8x22B model and a fresh Instruct 8x22B variant with function calling capability on Hugging Face. Accompanying this release is a comprehensive blog post and evaluation, providing insights into the latest advancements. Key highlights of the new Insigns of the 8x22B models include: Multilingual proficiency in English, French, Italian, German, and Spanish. Released under the permissive Apache 2.0 license, fostering accessibility and collaboration. Boasting a 141B parameter count, with 39B actively engaged during generation. Instruct model equipped with native function calling support, enhancing its versatility. Context window expanded to 64K tokens for enriched contextual understanding. Demonstrating stellar performance on open Large Language Model Leaderboard metrics, with over 75% on Multimodal Language Understanding (MMLU) and 45% on Human Evaluation. While evaluations for the Instruct model are pending, Mistral AI's latest offerings signify a significant stride in advancing open AI capabilities, promising enhanced linguistic prowess and functionality for diverse applications.