IONOS Provider (European Cloud)

The IONOS provider implements the LLM as Judge methodology using IONOS Cloud’s European-hosted models. It offers cost-effective, GDPR-compliant evaluation services with excellent performance for European deployments.

Available Models & Their Advantages

IONOS Custom Models

Optimized for evaluation tasks with fine-tuned prompts
European data centers ensuring GDPR compliance and low latency
Cost-effective pricing compared to major cloud providers
Flexible model sizes from efficient to high-performance

Model Selection Guide

Model Type	Use Case	Advantages
Standard Models	General evaluations	Good balance of speed and accuracy
Efficiency Models	High-volume processing	Fast response times, lower costs
Performance Models	Complex evaluations	Enhanced reasoning capabilities

Key Provider Advantages

🇪🇺 European Hosting: Data sovereignty and GDPR compliance
💰 Cost Efficiency: Competitive pricing for high-volume usage
⚡ Low Latency: Optimized for European users
🔒 Privacy: No data retention policies
📈 Scalability: Better rate limits for production workloads

Configuration Setup

Environment Variables:

IONOS_ENDPOINT="https://inference.de-txl.ionos.com/models"
IONOS_API_KEY="your_jwt_token_here"

Programmatic Setup:

from level_core.evaluators.ionos import IonosEvaluator
from level_core.evaluators.schemas import EvaluationConfig
from logging import Logger
 
# Configure IONOS evaluator
config = EvaluationConfig(
    api_url="https://inference.de-txl.ionos.com/models",
    api_key="your_ionos_jwt_token",
    model_id="YOUR-IONOS-MODEL-ID"
)
 
evaluator = IonosEvaluator(config, Logger("IONOS"))

API Endpoint and Authentication

IONOS uses JWT-based authentication:

Endpoint: https://inference.de-txl.ionos.com/models
Authentication: Bearer token (JWT)
Content-Type: application/json

Model-Specific Parameters

llm_config = {
    "top-k": 5,          # Limit to top K tokens
    "top-p": 0.9,        # Nucleus sampling threshold
    "temperature": 0.0,   # Deterministic output
    "max_tokens": 150    # Maximum response length
}

Example Usage

import asyncio
from level_core.evaluators.ionos import IonosEvaluator
from level_core.evaluators.schemas import EvaluationConfig
from logging import Logger
 
async def evaluate_with_ionos():
    # Setup
    config = EvaluationConfig(
        api_url="https://inference.de-txl.ionos.com/models",
        api_key="your_ionos_jwt_token",
        model_id="0b6c4a15-bb8d-4092-82b0-f357b77c59fd"
    )
    
    evaluator = IonosEvaluator(config, Logger("IONOS"))
    
    # Evaluate
    result = await evaluator.evaluate(
        generated_text="The capital of France is Paris.",
        expected_text="Paris is the capital city of France."
    )
    
    print(f"Score: {result.match_level}/5")
    print(f"Reasoning: {result.justification}")
    print(f"Tokens used: {result.metadata.get('inputTokens', 0)} + {result.metadata.get('outputTokens', 0)}")
 
# Run evaluation
asyncio.run(evaluate_with_ionos())

IONOS-Specific Features

European data centers for GDPR compliance
Competitive pricing for high-volume usage
Custom model selection with specific model IDs
Token usage tracking in metadata

OpenAI Provider (GPT Models)Custom Judge Implementations