Skip to contents

Extract outputs, metadata and costs from results

Source: R/accessors.R
accessors.Rd

These functions read the parts of a result so you do not need to know its structure. They work on results of run() with .return_format = "structured", for single inputs and batches, and on evaluate() results.

  • get_output() returns the outputs: the named list of output fields for a single result, a list of them for a batch, and the predictions of an evaluation.

  • get_metadata() returns the call metadata: a list for a single result and one list per row for a batch or an evaluation.

  • get_tokens() returns input_tokens, output_tokens and total_tokens.

  • get_cost() returns the estimated cost in US dollars.

For run_dataset() results, use the result and .metadata columns instead.

Usage

get_output(x, ...)

get_metadata(x, ...)

get_tokens(x, ...)

get_cost(x, ...)

Arguments

x

A result: a dsprrr_result or dsprrr_batch_result from run() with .return_format = "structured", or a dsprrr_evaluation from evaluate(). For other objects, get_output() returns x itself and the other functions return empty or missing values.

...

Not used.

Value

  • get_output(): the outputs, as described above.

  • get_metadata(): a list, or a list of lists; an empty list when x has no metadata.

  • get_tokens(): a list of the three counts for a single result, or a tibble with columns index, input_tokens, output_tokens and total_tokens for a batch or an evaluation. Unknown counts are NA.

  • get_cost(): a number (NA when unknown) for a single result. For a batch or an evaluation, a list of class dsprrr_cost_summary with costs (a tibble of index and cost), total and n_missing; missing costs give a warning and make total NA instead of counting as free.

See also

Other inspection: clear_prompt_history(), clear_traces(), export_traces(), get_last_prompt(), inspect_history(), session_cost(), summarize_traces()

Examples

shout <- module_fn("text -> reply", function(text) toupper(text))
result <- run(shout, text = "hello", .return_format = "structured")

get_output(result)
#> $reply
#> [1] "HELLO"
#> 
get_metadata(result)$latency_ms
#> [1] 0.0629425
# Function-backed modules make no model calls, so these are NA
get_tokens(result)
#> $input_tokens
#> [1] NA
#> 
#> $output_tokens
#> [1] NA
#> 
#> $total_tokens
#> [1] NA
#> 
get_cost(result)
#> [1] NA

if (FALSE) { # \dontrun{
qa <- module(signature("question -> answer"))
batch <- run(
  qa,
  question = c("What is 2 + 2?", "What is 3 + 3?"),
  .llm = ellmer::chat_openai(model = "gpt-6-luna"),
  .return_format = "structured"
)
get_tokens(batch)
get_cost(batch)$total
} # }