config_diff() lists the values in module$config next to baseline values,
changed rows first. Use it to see which settings an optimizer changed.
Value
A tibble with columns parameter, before and after (values
formatted as text) and changed (logical).
Details
The default baseline assumes provider defaults: temperature = 1,
top_p = 1, frequency_penalty = 0 and presence_penalty = 0. A value
the module never set is shown as "<default>" and counts as changed, as do
internal fields such as .module_kind, so pass a baseline that matches
your starting configuration for a meaningful comparison.
See also
Other optimization results:
apply_best_config(),
best_demos(),
best_params(),
export_module_code(),
optimization_result(),
optimization_summary(),
top_trials()
Examples
mod <- module(signature("text -> sentiment"), config = list(temperature = 0))
config_diff(mod, baseline = list(temperature = 0.7))
#> # A tibble: 6 × 4
#> parameter before after changed
#> <chr> <chr> <chr> <lgl>
#> 1 temperature 0.70 0 TRUE
#> 2 top_p 1 <default> TRUE
#> 3 frequency_penalty 0 <default> TRUE
#> 4 presence_penalty 0 <default> TRUE
#> 5 params <default> 0 TRUE
#> 6 .module_kind <default> predict TRUE