Prerequisites

Before installing Voxtral, ensure your system meets these requirements:

• Python 3.8 or higher
• GPU with CUDA support (recommended)
• 10GB+ available storage
• Internet connection for initial model download

Installation

The easiest way to get started with Voxtral is using the UV package manager:

# Install UV package manager curl -LsSf https://astral.sh/uv/install.sh | sh # Install vLLM with UV uv pip install vllm

First Run

Start the Voxtral model server locally:

# Serve Voxtral 3B model (for local/edge deployment) vllm serve mistralai/voxtral-3b # Or serve Voxtral 24B model (for production) vllm serve mistralai/voxtral-24b

Basic Usage

Once the server is running, you can interact with Voxtral through the API:

from mistral_common import ChatCompletionRequest
import requests

# Initialize client
client_url = "http://localhost:8000"

# Process audio file
audio_file = "path/to/your/audio.wav"
with open(audio_file, "rb") as f:
    response = requests.post(f"{client_url}/transcribe", files={"file": f})

print(response.json())

Understanding Output

Voxtral provides comprehensive speech understanding capabilities:

• Transcription: Accurate speech-to-text conversion
• Question Answering: Ask questions about audio content
• Summarization: Automatic content summaries
• Language Detection: Automatic multilingual processing

Performance Tips

• Use GPU acceleration for faster processing
• Choose the 3B model for local deployment and lower resource usage
• Use the 24B model for production environments requiring maximum accuracy
• Process audio files up to 40 minutes for optimal results

Next Steps

Now that you have Voxtral running, explore these advanced features:

• Function calling capabilities
• Multilingual speech processing
• Custom deployment configurations
• Integration with existing applications

Continue reading our Features & Capabilities guide to learn more about Voxtral's advanced functionality.