Mistral has introduced a sustainability auditing tool to assess the environmental impact of AI models, specifically its Large 2 model. Collaborating with ADEME and Carbone 4, Mistral reported a carbon footprint of 20.4ktCO₂e and a water consumption of 281,000 m³ by January 2025. A significant percentage of these environmental impacts arose from the model’s training and inference phases. The resource-intensive nature of training large language models (LLMs) was highlighted, with every prompt contributing further to emissions. Despite recent sustainability initiatives, Mistral emphasizes the need for transparency to guide users in choosing models wisely. As major tech firms like Google and Microsoft grapple with rising emissions, Mistral aims to construct a data center in France, leveraging low-carbon nuclear energy. The company advocates for the development of international sustainability standards and will regularly update its environmental impact reports, contributing to ADEME’s Base Empreinte database for improved transparency in AI.
Source link
Share
Read more