vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Min P disabled if set to 1.0 or 0, otherwise Top P

Browse source
This commit is contained in:
kalomaze 2023-10-28 21:08:26 -05:00
parent a235a0d226
commit 838d58dc32
2 changed files with 6 additions and 3 deletions
Download patch file Download diff file Expand all files Collapse all files

2
common/common.cpp
View file

@ -685,7 +685,7 @@ void gpt_print_usage(int /*argc*/, char ** argv, const gpt_params & params) {
printf(" -b N, --batch-size N batch size for prompt processing (default: %d)\n", params.n_batch);
printf(" --top-k N top-k sampling (default: %d, 0 = disabled)\n", sparams.top_k);
printf(" --top-p N top-p sampling (default: %.1f, 1.0 = disabled)\n", (double)sparams.top_p);
printf(" --min-p N min-p sampling (default: %.1f, 0.0 = disabled)\n", (double)sparams.min_p);
printf(" --min-p N min-p sampling (default: %.2f, 1.0 = disabled)\n", (double)sparams.min_p);
printf(" --tfs N tail free sampling, parameter z (default: %.1f, 1.0 = disabled)\n", (double)sparams.tfs_z);
printf(" --typical N locally typical sampling, parameter p (default: %.1f, 1.0 = disabled)\n", (double)sparams.typical_p);
printf(" --repeat-last-n N last n tokens to consider for penalize (default: %d, 0 = disabled, -1 = ctx_size)\n", sparams.penalty_last_n);

7
common/sampling.cpp
View file

@ -190,8 +190,11 @@ llama_token llama_sampling_sample(
llama_sample_top_k (ctx_main, &cur_p, top_k, min_keep);
llama_sample_tail_free(ctx_main, &cur_p, tfs_z, min_keep);
llama_sample_typical (ctx_main, &cur_p, typical_p, min_keep);
llama_sample_top_p (ctx_main, &cur_p, top_p, min_keep);
llama_sample_min_p (ctx_main, &cur_p, min_p, min_keep);
if (min_p != 1.0 && min_p != 0.0) {
llama_sample_min_p(ctx_main, &cur_p, min_p, min_keep);
} else {
llama_sample_top_p(ctx_main, &cur_p, top_p, min_keep);
}
llama_sample_temp (ctx_main, &cur_p, temp);
id = llama_sample_token(ctx_main, &cur_p);