Mistral AI has announced today the release of its large-scale base model, Pixtral Large, now generally available on Amazon Bedrock. This innovation enables users to leverage an advanced multimodal model to responsibly develop, experiment, and scale their generative artificial intelligence ideas within the Amazon Web Services (AWS) cloud. In this way, AWS becomes the first major cloud provider to offer Pixtral Large as a fully managed and serverless model.
Pixtral Large is distinguished by its sophisticated architecture, which features a total of 124 billion parameters, integrating a multimodal decoder of 123 billion parameters and a specialized vision encoder of 1 billion parameters. This powerful combination allows it to tackle complex visual and textual tasks, as well as incorporating the outstanding language processing capabilities of its predecessor, Mistral Large 2. One of the most prominent features of this model is its extended context window of 128,000 tokens, allowing it to handle multiple images along with extensive textual data, proving to be especially effective in document analysis and detailed visualizations.
The model offers a range of key capabilities. Among these, text analysis in multiple languages stands out, enabling accurate interpretation of written information found in images and documents. Its ability to interpret graphs and data visualizations is also notable, facilitating the identification of trends and anomalies in complex visual representations. Additionally, Pixtral Large performs overall visual analysis and contextual understanding, expanding its utility in numerous applications, including automated content moderation and contextual image retrieval.
Currently, the model is available in the AWS regions of Northern Europe and Western US, with inter-regional inference enabled for several other areas. Interested parties can request access through the Amazon Bedrock console, where they will have the option to test it in different modes, such as text or chat.
Furthermore, Pixtral Large has the capability to generate responses in JSON format, allowing developers to easily integrate their outputs into various applications and systems. Its compatibility with various Amazon Bedrock APIs enables flexible access to tools that support response generation.
Use cases for Pixtral Large are varied, ranging from generating SQL code from entity relationship diagrams to extracting organizational hierarchies in structured format. It also offers the ability to analyze graphs and reason about them, ensuring a comprehensive approach to understanding and analyzing visual data.
Mistral AI assures that Pixtral Large on Amazon Bedrock will take productivity to new levels in various business applications, spanning sectors such as e-commerce and financial services. The model is already accessible on the Amazon Bedrock console for those wishing to explore its capabilities.
via: MiMub in Spanish
