Skip to main content
Customize PAS2 behavior by configuring models, progress tracking, and other advanced options.

​
Using custom models

Specify which models to use for different tasks: 
from pas2 import PAS2
import os

# Initialize with custom model selection
pas2 = PAS2(
    mistral_api_key=os.environ.get("MISTRAL_API_KEY"),
    openai_api_key=os.environ.get("OPENAI_API_KEY")
)

# The default models are:
# - mistral_model: "mistral-large-latest"
# - openai_model: "o3-mini"

print(f"Using Mistral model: {pas2.mistral_model}")
print(f"Using OpenAI model: {pas2.openai_model}")
Expected output: 
Using Mistral model: mistral-large-latest
Using OpenAI model: o3-mini

​
Progress tracking with callbacks

Monitor the detection process with progress callbacks: 
from pas2 import PAS2
import os

def progress_callback(stage, **kwargs):
    """Custom progress tracking callback"""
    if stage == "starting":
        print(f"Starting analysis for: {kwargs.get('query')}")
    elif stage == "generating_paraphrases":
        print("Generating paraphrases...")
    elif stage == "paraphrases_complete":
        print(f"Generated {kwargs.get('count')} paraphrases")
    elif stage == "responses_progress":
        completed = kwargs.get('completed_responses', 0)
        total = kwargs.get('total_responses', 0)
        print(f"Getting responses: {completed}/{total}")
    elif stage == "judging":
        print("Analyzing for hallucinations...")
    elif stage == "complete":
        print("Analysis complete!")

# Initialize with progress callback
pas2 = PAS2(
    mistral_api_key=os.environ.get("MISTRAL_API_KEY"),
    openai_api_key=os.environ.get("OPENAI_API_KEY"),
    progress_callback=progress_callback
)

results = pas2.detect_hallucination(
    "Who wrote the novel 1984?",
    n_paraphrases=2
)
Expected output (with callback): 
Starting analysis for: Who wrote the novel 1984?
Generating paraphrases...
Generated 3 paraphrases
Getting responses: 0/3
Getting responses: 1/3
Getting responses: 2/3
Getting responses: 3/3
Analyzing for hallucinations...
Analysis complete!

​
Configuring number of paraphrases

Adjust the number of paraphrases to balance accuracy and speed: 
from pas2 import PAS2
import os

pas2 = PAS2(
    mistral_api_key=os.environ.get("MISTRAL_API_KEY"),
    openai_api_key=os.environ.get("OPENAI_API_KEY")
)

# Faster, less thorough
results = pas2.detect_hallucination(
    "What was the first computer?",
    n_paraphrases=2
)
Performance comparison:
ParaphrasesProcessing timeAccuracy
2~10-15 secondsGood
3~15-20 secondsBetter
5~25-35 secondsBest

​
Using the hallucination judgment model

Direct access to the judgment functionality: 
from pas2 import PAS2
import os

pas2 = PAS2(
    mistral_api_key=os.environ.get("MISTRAL_API_KEY"),
    openai_api_key=os.environ.get("OPENAI_API_KEY")
)

# Prepare your queries and responses
original_query = "What is the capital of France?"
original_response = "The capital of France is Paris."
paraphrased_queries = [
    "Can you tell me France's capital city?",
    "Which city is the capital of France?"
]
paraphrased_responses = [
    "France's capital city is Paris.",
    "The capital city of France is Paris."
]

# Judge for hallucinations
judgment = pas2.judge_hallucination(
    original_query=original_query,
    original_response=original_response,
    paraphrased_queries=paraphrased_queries,
    paraphrased_responses=paraphrased_responses
)

print(f"Hallucination detected: {judgment.hallucination_detected}")
print(f"Confidence: {judgment.confidence_score}")
print(f"Summary: {judgment.summary}")
Expected output: 
Hallucination detected: False
Confidence: 0.95
Summary: All responses consistently and accurately identify Paris as the capital of France with no conflicting information.

​
Logging configuration

Customize logging output for debugging: 
import logging
from pas2 import PAS2
import os

# Configure logging level
logging.basicConfig(
    level=logging.DEBUG,  # Show detailed debug information
    format='%(asctime)s [%(levelname)s] %(message)s',
    handlers=[logging.StreamHandler()]
)

# Or set specific logger level
logger = logging.getLogger('pas2')
logger.setLevel(logging.INFO)

pas2 = PAS2(
    mistral_api_key=os.environ.get("MISTRAL_API_KEY"),
    openai_api_key=os.environ.get("OPENAI_API_KEY")
)

results = pas2.detect_hallucination(
    "What is the speed of light?",
    n_paraphrases=2
)
Expected output: 
2026-03-03 10:15:23 [INFO] PAS2 initialized with Mistral model: mistral-large-latest and OpenAI model: o3-mini
2026-03-03 10:15:23 [INFO] Starting hallucination detection for query: What is the speed of light?
2026-03-03 10:15:23 [INFO] Generating 2 paraphrases for query: What is the speed of light?
2026-03-03 10:15:25 [INFO] Generated 2 paraphrases in 2.15 seconds
2026-03-03 10:15:25 [INFO] Getting responses for 3 queries in parallel
...

Build docs developers (and LLMs) love

Get started for freeTalk to us