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

perplexity only uses pp_threads

Browse source
This commit is contained in:
netrunnereve 2023-08-08 21:30:12 -04:00
parent 5624a29c1f
commit d854348992
1 changed files with 3 additions and 3 deletions
Download patch file Download diff file Expand all files Collapse all files

6
examples/perplexity/perplexity.cpp
View file

@ -66,7 +66,7 @@ void perplexity(llama_context * ctx, const gpt_params & params) {
tokens[batch_start] = llama_token_bos();
}
if (llama_eval(ctx, tokens.data() + batch_start, batch_size, j * n_batch, params.n_threads, params.n_threads)) {
if (llama_eval(ctx, tokens.data() + batch_start, batch_size, j * n_batch, params.n_threads, params.pp_threads)) {
fprintf(stderr, "%s : failed to eval\n", __func__);
return;
}
@ -337,8 +337,8 @@ int main(int argc, char ** argv) {
// print system information
{
fprintf(stderr, "\n");
fprintf(stderr, "system_info: n_threads = %d / %d | %s\n",
params.n_threads, std::thread::hardware_concurrency(), llama_print_system_info());
fprintf(stderr, "system_info: pp_threads = %d / %d | %s\n",
params.pp_threads, std::thread::hardware_concurrency(), llama_print_system_info());
}
if (params.hellaswag) {