Mistral launches 8X22B mix of expert AI models in open source, benchmarks reveal major upgrade

Mistral released its latest. Artificial intelligence (AI) model, 8X22B, on Wednesday. The Mixture of Expert AI model follows the 8X7B model released in December 2023 and offers a larger parameter size. The company known for its fully open-source AI models released the latest model unconventionally, with no announcement posts or blog posts. Although no benchmarks have been provided by the AI ​​firm itself, users in the Hugging Face community have tested it and posted benchmark scores for the 8X22B model. Results appear to bridge the gap with closed models. Open AI And Google.

Mistral’s official X (formerly known as Twitter) account Continued The 8X22B AI model via torrent magnet link continues its unorthodox approach of releasing AI models for people to download without announcement and directly. Mistral is also one of the only truly open source platforms that not only opens up the weights, but also the entire architecture. However, it should be noted that most devices are not equipped to run powerful AI models on-device, and this may cause device damage. Total file size is 262 GB.

Mistral’s 8X22B is an automated AI model. These are usually different from AI models’ instructions or chat variations. Open AI Chat GPT And Google’s Gemini AI Cheats are models, while Meta’s code Llama 7B and 13B are instruction models. Chat models are those that can understand natural language and context to provide accurate answers to questions. Instruct models are primarily used by developers who ask an AI model to perform a specific task. In contrast, an automatic model completes the sentence that is provided in the prompt.

Preliminary benchmark scores based on the hugging face Community Post Show that the latest Mistral model offers significant upgrades over its predecessors and closes the gap with other major models. In the Hellaswag benchmark, the Mistral 8X22B scored 88.9, followed by the GPT-4 at 95.3, the Claude 3 Opus at 95.4, and the Gemini 1.5 Pro at 92.5. However, it outscores the GPT-3.5 at 85.5 and the Gemini 1.0 Ultra at 87.8.

Comparing benchmarks with Mistral’s own models, the 8X22B falls between the Mistral-medium and Mistral-large models. With a later release of an instructor or chat variant, it may also be equivalent to the Mistral-large.