The Mistral Large 2 could deliver similar performance to the Meta Llama 3.1 405B
Mistral on Wednesday released the next generation of its flagship open-source artificial intelligence (AI) model, Mistral Large 2. The company claims that the AI model offers significantly improved capabilities in code generation, mathematics and reasoning. It also gets support for several new languages and advanced function calling capabilities. It is also said that despite being a third of the size of the recently released Meta Llama 3.1 405B AI model, Mistral’s flagship large language model (LLM) offers comparable performance. Notably, Mistral Large 2 is only available for research and non-commercial use.
Mistral Large 2 Features
The company announced the AI model in a newsroom after. The Mistral Large 2 comes with a context window of 1,28,000 tokens, which is comparable to Meta’s latest AI offering. Furthermore, the flagship Mistral AI model supports several new languages, including Arabic, Chinese, French, German, Hindi, Italian, Japanese, Korean, Portuguese, Russian, and Spanish. In addition, it can also generate code in over 80 programming languages.
Mistral’s new AI model has a size of 123 billion parameters and can be executed on a single node. The company said there were three main areas of focus to improve the Large 2 model. First, there was code generation, with the LLM being trained on a large volume of coding data. Second, to improve reasoning ability and minimize instances of hallucinations, the AI company refined the model to respond more cautiously. Finally, the AI model was trained to “recognize when it cannot find solutions or does not have enough information to give a confident answer.”
Despite being a third of the size of Llama 3.1 405B, the company claims its LLM outperforms it. Based on its internal benchmark tests, Mistral said its AI model outperformed it in code generation and mathematical performance. It also claimed to outperform GPT-4o in Java code generation.
Furthermore, the company claims that the Mistral Large 2 has improved function calling and retrieval, allowing it to drive complex business applications. Function calling is a capability of AI models to interact with external tools or functions. This allows them to obtain data from different sources and provide more accurate, informative and efficient answers.
The company has partnered with Google Cloud Platform to bring the Large 2 AI model to Vertex AI via a managed application programming interface (API). It is also available in the cloud via Azure AI Studio, Amazon Bedrock, and IBM Watsonx. Since it is an open-source AI model, interested parties can also access the LLM via the website under the name mistral-large-2407.
To download the instruction model, users can refer to the HuggingFace mentionIn particular, it is available under the Mistral Research License, which permits use and modification for research and non-commercial use only.
