Enquiry About Using Guardrails Validators for LLM Output Validation

[gstomar]

Dear Guardrails Team,

I hope this message finds you well. My name is Gaurav and I am currently working on a AIML project. We are currently exploring the integration of Large Language Models (LLM) in our project.

We are interested in using Guardrails to validate the output of our LLMs, and I have reviewed your documentation on validators and output schema. However, I have some specific questions about how we can effectively implement this in our project, particularly around LLM outputs validation

Could you please provide some insights on the following if it's possible?
We want to keep the control of LLM execution and just use the LLM output for validation.

1. How can I use the Validators apis directly like below mentioned in the example?

For e.g.

Define the input schema

input_schema = {
"text": str,
"parameters": {
"name": str,
"age": int,
},
}

Define the output schema

output_schema = {
"response": str,
}

LLM Output

llm_result = "Microsoft's earnings per share for 2022 was $9.65, a 20% increase from the previous year."

Create a Guardrails validator

validator = guardrails.Validator(input_schema, output_schema)

try:
validator.validate(input, output)
print("Validation successful")
except guardrails.ValidationError as e:
print("Validation failed:", e)

2. Can I extend and create my own validators based on Guardrails validators ?

Thank you for considering my request. I am looking forward to your response.

Best regards,
Gaurav Tomar

[gstomar]

Hello Team,

Can you please reply if possible ?
Eagerly waiting for your valuable input.

Many Thanks
Gaurav Tomar

[gstomar]

Hello Team,
Can you please update on the above question ?
Many Thanks

[irgolic]

Hi, sorry for the late reply. You can use guard.parse to process your LLM result. See here for an example: https://docs.guardrailsai.com/concepts/guard/#parse

Input validation is on the way, coming this week!

I'm closing this discussion for now, but feel free to reopen it if you have any more questions.

[gstomar]

Thanks @irgolic
Much appreciated.
Will test the parse api.
When are we releasing the input validation one ?
Cheers

[irgolic]

Hi, we're planning to release it with 0.3.0. You can track our progress here: #490