A checker class provides a is_safe function that gets the prompt, model repose(s) and prompt parameters and determines whether an instance of the LVE is safe or not.
The amiable checkers can be found in Core Checkers.
Checkers can specify a postprocess_response function that can alter the saved instance (either a response string or dictionary of variables) before saving it to a file. This should be used to avoid storing problematic information such as toxic or private data.
We recommend to use or adapt the existing checkers where possible. However, if a custom checkers is required, there are two possibilities:
lve-tools. Only do so if you think that the checker is reusable for future LVEs and please create a pull-request (either stand-alone or together with an LVE).A single-file checker can be defined inside checker/checker.py inside an LVE. For an example see responsibility/copyright/songs/translation.
Each checker must extend BaseChecker (or MultiRunBaseChecker, respectively) and implement the function is_safe and optionally postprocess_response. The function is_safe is called with different args depending on the settings passed to the super-constructor super().__init__(...).
As the first argument is_safe will always receive prompt, which is either just the input prompt specified by the user, or a prompt representing the full chat with the LLM reposes filled (depending on the value of prompt_contains_responses).
Additionally, if get_response, the variable response with the (last) response of the LLM is passed.
Lastly if get_variables is set, the variable variables is passed, which is a dictionary of all variables in the prompt (see Multi-Variable Prompts).
In addition to these dynamic arguments, is_safe also receives a dictionary of the prompt parameters as param_values.
As a summary, arguments of the constructor with their respective default values are:
prompt_contains_responses=False, # wether to fill in the model responses in the prompt
get_variables=False, # wether to pass a dictionary of all variables to the checker
get_response=True # wether to pass the (last) model response to the checker
Based on the default values the default signature of is_safe is:
def is_safe(self, prompt: list[Message], response: str, param_values:dict[str, any] = None) -> bool
For MultiRunBaseChecker the arguments to the constructor are the same.
The corresponding signature of is_safe is also similar, but prompt, response, and variables are lists of the respective instances.
In the case of Multi-Variable Prompts multi-variable aware checkers must be used. Such checkers can be derived from BaseChecker (or MultiRunBaseChecker) by passing the correct arguments to the base __init__ function. See Extending BaseCheckers for details.
Checkers that operate on multiple runs for a single instance must extend MultiRunBaseChecker. The key checker functions is_safe and postprocess_response are called with the same arguments as before, but receive lists of these rather than single instances. Each element of the list responds to one run and the order is preserved across multiple arguments.