Overview

Developed by	Alejandro Esquivel
Date of development	Jul 09, 2024
Validator type	Format
Blog
License	Apache 2
Input/Output	Output

Description

Intended Use

This validator checks that generated text is parseable HTML. It relies on html5lib to ensure that markup returned by an LLM can be parsed without errors before it is rendered or passed downstream.

Requirements

• Dependencies:

  • guardrails-ai>=0.4.0
  • html5lib>=1.1,<2.0

• Foundation model access keys:

  • None

Installation

$ guardrails hub install hub://guardrails/valid_html

Usage Examples

Validating string output via Python

In this example, we apply the validator to a string output generated by an LLM.

# Import Guard and Validator
from guardrails.hub import ValidHtml
from guardrails import Guard

# Setup Guard
guard = Guard().use(
    ValidHtml()
)

guard.validate("<p>hello</p>")  # Validator passes
guard.validate("<phello</p>")  # Validator fails

Validating JSON output via Python

Use the validator to protect HTML-bearing fields of structured outputs.

# Import Guard and Validator
from pydantic import BaseModel, Field
from guardrails.hub import ValidHtml
from guardrails import Guard

# Initialize Validator
html_validator = ValidHtml(strict=True)

# Create Pydantic BaseModel
class WebSection(BaseModel):
		title: str
		body: str = Field(validators=[html_validator])

# Create a Guard to check for valid Pydantic output
guard = Guard.from_pydantic(output_class=WebSection)

# Run LLM output generating JSON through guard
guard.parse("""
{
	"title": "About",
	"body": "<section><h1>About Us</h1></section>"
}
""")

API Reference

__init__(self, *, strict=True, on_fail=None)

Initializes a new instance of the `ValidHtml` validator.

Parameters

• strict (bool): When True, the validator raises on the first HTML parsing error. When False, html5lib attempts to recover from malformed markup.
• on_fail (str, Callable): The policy to enact when a validator fails. If str, must be one of reask, fix, filter, refrain, noop, exception or fix_reask. Otherwise, provide a callable invoked when the validator fails.

validate(self, value, metadata=None) -> ValidationResult

Validates the given `value` using `html5lib`. If the parser cannot consume the markup—even after attempting to prepend a doctype—the validator returns a `FailResult` with the parser's error message.

Parameters

• value (Any): The input value to validate.
• metadata (dict | None): Optional metadata forwarded by Guardrails. It is unused by this validator but retained for API compatibility.