Pick the highest reward-model score among N samples. best_index is
0-BASED.
Usage
morie_kamath_best_of_n_sampling(
samples,
rewards = NULL,
reward_fn = NULL,
x = NULL
)
Arguments
- samples
Candidate responses.
- rewards
Optional scores.
- reward_fn
Optional function (x, y).
- x
Prompt.
Value
List with estimate, best, best_index, reward_spread.
Examples
morie_kamath_best_of_n_sampling(list("a", "b", "c"), rewards = c(0.1,
0.9, 0.4))
#> $estimate
#> [1] 0.9
#>
#> $best
#> [1] "b"
#>
#> $best_index
#> [1] 1
#>
#> $rewards
#> [1] 0.1 0.9 0.4
#>
#> $reward_spread
#> [1] 0.8
#>
#> $n
#> [1] 3
#>
#> $method
#> [1] "best-of-N sampling (Kamath Ch 5)"
#>
