LLM methods

This module provides basic methods to support LLM conditions. These methods return bool values based on LLM result.

class BaseMethod(**data)

Bases: BaseModel, ABC

Base class to evaluate models response as condition.

abstract async __call__(ctx, model_result)

Determine if result of an LLM invocation satisfies the condition of this method.

Parameters:

• ctx (Context) – Current dialog context.

• model_result (LLMResult) – Result of langchain model’s invoke.

Return type:

bool

model_result_to_text(model_result)

Extract text from raw model result.

Return type:

str

class Contains(**data)

Bases: BaseMethod

Simple method to check if a string contains a pattern.

pattern: str

Pattern that will be searched in model_result.

async __call__(ctx, model_result)

Return type:

bool

Returns:

True if pattern is contained in model_result.

class LogProb(**data)

Bases: BaseMethod

Method to check whether a target token’s log probability is higher than a threshold.

target_token: str

Token to check (e.g. “TRUE”)

threshold: float

Threshold to bypass. by default -0.5

async __call__(ctx, model_result)

Return type:

bool

Returns:

True if logprob of the token is higher than threshold.