Voxtral models from Mistral AI combine text and audio processing in a unified framework, available in 3B and 24B parameter variants. This guide demonstrates deploying Voxtral on Amazon SageMaker using vLLM and the Bring Your Own Container approach. The solution includes custom Docker container configuration with audio processing libraries, model deployment through SageMaker notebooks, and support for multiple use cases including transcription, multimodal understanding, and function calling from voice input. The architecture uses Amazon ECR for container storage, S3 for configuration files, and enables switching between model variants through simple configuration updates without container rebuilds.
Table of contents
Solution overviewPrerequisitesDeploy Voxtral modelsModel use casesClean upConclusion362 Impressions