Mistral Large 2: A New Advanced AI Model for Complex Applications | Festina Lente - Your leading source of AI news | Turtles AI

Mistral Large 2: A New Advanced AI Model for Complex Applications
The Mistral Large 2 AI model, with a 128k context window and support for 80 languages, offers high performance and unprecedented flexibility
DukeRem24 July 2024

Mistral Large 2: New Leading AI Model for Advanced Applications

Highlights:

  • 128k context window and support for over 80 programming languages.
  • High performance on MMLU with 84.0% accuracy.
  • Advanced reasoning capabilities and reduced "hallucinations."
  • Extensive multilingual competence and advanced function calling abilities.

 

Mistral Large 2, Mistral AI’s latest AI model, marks a significant advancement with its 128k context window and support for dozens of languages. The model, boasting 123 billion parameters, is designed for single-node operation, optimizing efficiency and performance. Available for non-commercial and commercial use, Mistral Large 2 excels in code generation and reasoning benchmarks.

Mistral Large 2, recently launched by Mistral AI, stands out as one of the most advanced models currently available on the market. Its 128k context window and support for over 80 programming languages, including Python, Java, and C++, make it particularly suitable for single-node inference applications with extensive contexts. With 123 billion parameters, this model has been optimized to deliver high performance with controlled costs, a crucial aspect for widespread adoption across various sectors.

The model is released under the Mistral Research License, which allows for use and modification for research and non-commercial purposes, while a Mistral Commercial License is required for commercial use. This approach allows for flexible use, potentially spurring further innovations and customized applications.

One of the most notable aspects of Mistral Large 2 is its performance in benchmarks. For instance, it achieved an accuracy of 84.0% on the Multi-task Language Understanding (MMLU), surpassing many competing open-source models in terms of performance/cost ratio. This result highlights the model’s ability to effectively handle a wide range of natural language processing tasks, including the most complex and diverse ones.

The model’s training benefited from previous experiences with the Codestral 22B and Codestral Mamba models, with a particular focus on code generation. Mistral Large 2 not only outperforms its predecessor but also aligns with leading models like GPT-4o and Llama 3 405B, demonstrating advanced reasoning and code generation capabilities. Additionally, a robust fine-tuning process was implemented to reduce the occurrence of "hallucinations," or the generation of plausible but inaccurate information. This improvement is reflected in mathematical benchmarks, where the model demonstrated greater precision and problem-solving capabilities.

Another critical aspect is Mistral Large 2’s ability to follow precise instructions and handle multi-turn conversations more efficiently than previous models. Tests conducted on benchmarks like MT-Bench and Wild Bench confirmed the model’s superiority in terms of alignment and ability to execute complex instructions, which is crucial for business applications where succinctness and efficiency are essential.

Linguistic diversity is another strength of Mistral Large 2. The model was trained on a vast corpus of multilingual data, making it particularly competent not only in English but also in languages like French, German, Spanish, Italian, Portuguese, Russian, Chinese, Japanese, Korean, Arabic, and Hindi. In tests on the Multilingual MMLU benchmark, Mistral Large 2 showed superior performance compared to previous models, indicating a significant improvement in multilingual understanding and generation.

The model is also equipped with advanced function calling and tool usage capabilities, allowing for efficient execution of both parallel and sequential function calls. This makes it a powerful option for complex business applications, where integrating different systems and processes is crucial. Mistral Large 2 is currently available on platforms like Google Cloud Platform, Azure AI Studio, Amazon Bedrock, and IBM watsonx.ai, significantly expanding its accessibility and usability globally.