Mistral Provider

The Mistral provider enables access to Mistral AI's language models through the AiCore system, offering both chat and embedding capabilities.

Features

• Multiple Model Support: Access all Mistral AI models through a unified interface
• Streaming Support: Real-time token streaming for chat completions
• Token Counting: Automatic token usage tracking
• Cost Estimation: Built-in cost calculation based on model pricing
• Async/Sync APIs: Both synchronous and asynchronous operation modes

Supported Models

Model Name	Description	Context Window
mistral-tiny	Mistral 7B model	8k tokens
mistral-small	Mixtral 8x7B model	32k tokens
mistral-medium	Latest medium-sized model	32k tokens
mistral-large	Latest large model (best performance)	32k tokens

Configuration

YAML Configuration

llm:
  provider: mistral
  api_key: "your_mistral_api_key"
  model: "mistral-large"
  temperature: 0.7
  max_tokens: 1000

Python Configuration

from aicore.llm.config import LlmConfig

config = LlmConfig(
    provider="mistral",
    api_key="your_mistral_api_key",
    model="mistral-large",
    temperature=0.7,
    max_tokens=1000
)

Usage Examples

Basic Completion

from aicore.llm import Llm

llm = Llm(config=config)
response = llm.complete("Explain quantum computing in simple terms")
print(response)

Async Usage

import asyncio
from aicore.llm import AsyncLlm

async def main():
    llm = AsyncLlm(config=config)
    response = await llm.acomplete("What is the capital of France?")
    print(response)

asyncio.run(main())

Observability Integration

All Mistral provider operations are automatically tracked by the AiCore observability system:

from aicore.observability.collector import LlmOperationCollector

# View metrics for Mistral operations
mistral_metrics = LlmOperationCollector.polars_from_db(provider="mistral")
print(mistral_metrics.select(["model", "latency_ms", "total_tokens"]))

Troubleshooting

Common Issues:

• 401 Unauthorized: Verify your API key is correct
• Model not found: Check the model name spelling and your account access
• Rate limit exceeded: Implement retry logic or reduce request frequency

For additional help, refer to the Mistral API documentation.