Support other types of parameters

[liwii]

Updated the get_standard_metric_inputs function so that it can any types of inputs flexibly (system prompts, conversation histories, stringified numeric scores... etc) Hopefully with this update LangCheck can help verification of wider variety of LLM-based apps!!

Making use of the infra, I

• Enabled flexible inputs for custom_evaluator (only for non-pairwise inputs)
• Added prompt_leakage functions which take system_prompts as an input arg.

You can run prompt_leakage by

result_ja = langcheck.metrics.prompt_leakage(
    generated_outputs=generated_outputs,
    system_prompts=system_prompts,
    eval_model=eval_client,
)

You can do the same thing with custom_evaluator by

result_ja = langcheck.metrics.custom_evaluator(
    generated_outputs=generated_outputs,
    prompts=None,
    sources=None,
    reference_outputs=None,
    eval_model=eval_client,
    metric_name="prompt_leakage",
    score_map=score_map,
    template_path="path/to/prompt_leakage.j2",
    language="en",
    additional_params={"system_prompts": system_prompts},
    additional_params_to_prompt_var_mapping={"system_prompts": "system_prompt"},
)