llm_break_soft: Create an 'llm_break_soft' object

In tidyprompt: Prompt Large Language Models and Enhance Their Functionality

Create an llm_break_soft object

Description

This object is used to break a extraction and validation loop defined in a prompt_wrap(), as evaluated by send_prompt(). When an extraction or validation function returns this object, it will prevent any future interactions with the LLM provider for the current prompt. Remaining extraction and validation functions will still be applied and it will still be possible to pass these with the current response from the LLM provider; only, no more new tries will be made if the current response is not satisfactory.

This is useful when, e.g., the token limit for the LLM provider has been reached, but the final response that we got may still be satisfactory. In this case, llm_break() cannot be used, as it would instantly return the current response as the final result, which is not what we want. Instead, llm_break_soft() can be used to prevent any further interactions with the LLM provider, but still allow the remaining extraction and validation functions to be applied (and have those decide the success of the current response).

Usage

llm_break_soft(object_to_return = NULL)

Arguments

object_to_return

The object to return as the response result from send_prompt() when this object is returned from an extraction or validation function

Value

An list of class "llm_break_soft" containing the object to return

Examples

# Quitting when total token count is exceeded (Google Gemini API example)
## Not run: 
  "How are you?" |>
    # Forcing multi-response via initial error, for demonstration purposes
    answer_as_integer(add_instruction_to_prompt = FALSE) |>
    # Validation function to check total token count
    prompt_wrap(validation_fn = function(response, llm_provider, http_list) {
      total_tokens <- purrr::map_dbl(
        http_list$responses,
        ~ .x$body |>
          rawToChar() |>
          jsonlite::fromJSON() |>
          purrr::pluck("usageMetadata", "totalTokenCount")
      ) |> sum()
      if (total_tokens > 50) {
        warning("Token count exceeded; preventing further interactions")
        # Using llm_break_soft() to prevent further interactions
        return(llm_break_soft(response))
      }
    }) |>
    send_prompt(llm_provider_google_gemini(), return_mode = "full")

## End(Not run)

tidyprompt documentation built on Aug. 25, 2025, 5:11 p.m.