Mistral AI's 8x22B Models Unveiled
Exciting news from Mistral AI as they release the highly anticipated 8x22B model series, accompanied by a wealth of new insights and functionalities. 🚀 The latest additions to their repertoire include the 8x22B model, proficient in five languages—English, French, Italian, German, and Spanish—catering to a diverse global audience. 🌎 Under the permissive Apache 2.0 license, these models offer accessibility and adaptability to users and developers alike. 📜 With a staggering 141 billion parameters, of which 39 billion are active during generation, Mistral AI sets a new standard for model complexity and performance. 🏎️ Notably, the instruct model within this series introduces native function calling support, empowering users with enhanced control and versatility in their interactions. 🧑🏻💻 Sporting a spacious 64K token context window, these models exhibit remarkable contextual understanding and generation capabilities. 📚 Evaluation results confirm their prowess, with performance metrics exceeding 75% on MMLU and 45% on Human Evaluation, reinforcing Mistral AI's position on the open LLM Leaderboard. 📊 While evaluation data for the instruct model is not available at present, its innovative features promise exciting avenues for exploration and application. Stay tuned as Mistral AI continues to push the boundaries of natural language understanding and generation. 🔥