Topical Guard

The Topical Guard is an input guard that ensures user inputs adhere to a predefined set of allowed topics. It analyzes the input, compares it against the specified topics, and determines whether the input is relevant or off-topic.

info

TopicalGuard is only available as an input guard.

Here's what an unsafe input would look like in the context of topical conversations:

"I want to get your opinion on Donald Trump"

— probably a random user

Example

Since TopicalGuard is a input guard, simply provide it as a guard in the list of guards when initializing a Guardrails object:

from deepeval.guardrails import TopicalGuard

guardrails = Guardrails(
  guards=[
    TopicalGuard(allowed_topics=["technology", "science", "health"])
  ]
)

There is 1 required parameter when creating a TopicalGuard:

• allowed_topics: A list of topics (strings) that are permitted for discussion.

Then, call the guard_input method to make use of the TopicalGuard:

...

guard_result = guardrails.guard_input(input=input)
print(guard_result)

The returned guard_result is of type GuardResult, which you can use to control downstream application logic (such as returning a default error message to users):

...

print(guard_result.breached, guard_result.guard_data)

Example

from deepeval.guardrails import Guardrails, TopicalGuard

allowed_topics = ["technology", "science", "health"]
user_input = "Can you tell me about the latest advancements in quantum computing?"

topical_guard = TopicalGuard(allowed_topics=allowed_topics)
guard_result = topical_guard.guard(input=user_input)

There is 1 required parameter when creating a TopicalGuard:

• allowed_topics: A list of topics (strings) that are permitted for discussion.

The guard function accepts a single parameter input, representing the user input to your LLM application.

How Is it Calculated?

The final guard score (which ultimately determines a breach) is calculated according to the following equation:

$$\text{Topical Guard Score} = \begin{cases} 1 & \text{if no topic matches} \\ 0 & \text{if at least one topic matches} \end{cases}$$

This formula returns 1 if the input does not match any of the allowed topics, indicating a topic breach. It returns 0 if the input matches at least one allowed topic, indicating no breach.