guardrail.py

from weave.scorers import WeaveToxicityScorerV1

toxicity_scorer = WeaveToxicityScorerV1()


@weave.op
def call_llm(prompt: str) -> str:
    """Generate text using an LLM."""
    # Your LLM generation logic here
    return prompt.upper()

# Call our guardrailed function
async def generate_safe_response(prompt: str) -> str:
    # Call the function and return call object (from the weave.op'd function)
    result, call = call_llm.call(prompt)

    # Check Toxicity
    safety = await call.apply_scorer(toxicity_scorer)
    if not safety.result.passed:
        return f"Sorry but I cannot respond. Guardrail triggered: \n{safety.result.metadata}"

    return result
    
    
response = await generate_safe_response("Hey, how is it going?")
print(response)