Paris-based artificial intelligence startup, Mistral AI, has rolled out its inaugural multimodal model named Pixtral 12B. The model is designed to process both text and images, marking a noteworthy advancement for the company.
Overview of Pixtral 12B
Pixtral 12B builds upon Mistral's prior text model, Nemo 12B, by adding a 400 million-parameter vision adapter. Using the enhancement enables image processing via URLs or base64 encoding within texts. With 12 billion parameters and a size of about 24GB, it can perform tasks such as image captioning and object identification.
In terms of functionality, Pixtral 12B is positioned alongside other multimodal models like Anthropic's Claude family and OpenAI's GPT-4. It aims to excel in image analysis tasks, including generating captions, identifying objects, and answering image-related questions.
Access and Licensing
Developers can obtain Pixtral 12B through a torrent link on GitHub and on the Hugging Face platform. It is distributed under an Apache 2.0 license, allowing for unrestricted usage and customization. The open-source strategy is intended to foster widespread adoption and modification.
Although Mistral has not specified the exact datasets used to train Pixtral 12B, it's common for generative AI models to leverage extensive publicly available data. Mistral's approach has sparked legal debates, particularly around the use of copyrighted material. Some assert that public data scraping falls under “fair use,” a view that remains contested and has led to legal disputes involving prominent AI firms.
Mistral recently secured $645 million in funding, led by General Catalyst, and is now valued at $6 billion. Partly owned by Microsoft, the company aims to become a European alternative to OpenAI. Mistral's strategy includes offering open models for free while monetizing managed versions and consulting services.
This collaboration marks a potentially important step for Microsoft as it seeks to expand its AI offerings, allowing Azure AI customers to access large language models developed by Mistral AI. The partnership commenced with the integration of Mistral's Large Language Model (LLM) into Azure AI services, and now, the Mistral Small LLM has been made accessible to customers as well.
