library(parsnip)
data("hpc_data", package = "modeldata", envir = rlang::current_env())

mr_spec <- multinom_reg(penalty = 0.1) %>% set_engine("glmnet")
f_fit <- fit(mr_spec, class ~ protocol + log(compounds) + input_fields,
             data = hpc_data)
f_pred_class <- multi_predict(f_fit, hpc_data, penalty = c(0.01, 0.1), 
                              type = "class")
f_pred_class$.pred[[1]]
#> # A tibble: 2 × 2
#>   penalty .pred_class
#>     <dbl> <fct>      
#> 1    0.01 VF         
#> 2    0.1  VF

f_pred_prob <- multi_predict(f_fit, hpc_data, penalty = c(0.01, 0.1), 
                             type = "prob")
f_pred_prob$.pred[[1]]
#> # A tibble: 2 × 5
#>   penalty .pred_VF .pred_F .pred_M .pred_L
#>     <dbl>    <dbl>   <dbl>   <dbl>   <dbl>
#> 1    0.01    0.464   0.287   0.174  0.0758
#> 2    0.1     0.488   0.325   0.124  0.0625