In a landscape where major language models are often behind paywalls or limited licenses, the emergence of easier options is a breath of fresh air. Mistral, a French startup, has launched its first speaker model, the Mistral 7B, with the promise of high efficiency and flexibility.
Mistral is a France-based AI startup that recently raised a large round of seed funding. Its purpose is to become a pillar of the open generative AI community.
The Mistral 7B model
she Mistral 7B a 13.4 gigabyte language model that the company says outperforms other models of its size in terms of performance. It stands for lower computational cost compared to models like Llama 2.
Apache License 2.0
The model is offered under Apache License 2.0, which allows unlimited use as long as the corresponding credit is given. This means that any entity, from a hobbyist to a multi-million dollar corporation, can use it.
Community support
Mistral has built a repository of GitHub and a channel of Discord to promote collaboration and solve problems, showing a commitment to the development of the community.
Mistral’s business model
Although the model is free, Mistral plans to offer a commercial product that will include weights and code, aimed at business implementations.
Is it really open?
Despite the permissive license, the model was developed with private funds and the data sets and weights are also private. This raises questions about how “open” the model really is.
Previous team experience
The founders of Mistral have previous experience in well-known technology companies such as Meta y Google DeepMind which probably influenced the rapid development of the model.
Challenges and the future
The main challenge for Mistral is maintaining the balance between accessibility and financial sustainability, especially in an increasingly competitive market.
The launch of Mistral 7B is not only a milestone for the company, but may also mark a change in how the AI community approaches the issue of accessibility and openness.
