mirrors/cosmopolitan
1
0
Fork 0
mirror of https://github.com/jart/cosmopolitan.git synced 2025-03-16 05:46:25 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

Fix load time measurement

Browse source
This commit is contained in:
Justine Tunney 2023-05-10 07:54:21 -07:00
parent 290a49952e
commit bb3ebedfce
No known key found for this signature in database
GPG key ID: BE714B4575D6E328
1 changed files with 7 additions and 2 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

9
third_party/ggml/llama.cc vendored
View file

@ -2690,13 +2690,18 @@ int llama_eval(
int n_tokens, int n_tokens,
int n_past, int n_past,
int n_threads) { int n_threads) {
int64_t start_eval = 0;
if (!ctx->has_evaluated_once) {
start_eval = ggml_time_us();
}
if (!llama_eval_internal(*ctx, tokens, n_tokens, n_past, n_threads)) { if (!llama_eval_internal(*ctx, tokens, n_tokens, n_past, n_threads)) {
fprintf(stderr, "%s: failed to eval\n", __func__); fprintf(stderr, "%s: failed to eval\n", __func__);
return 1; return 1;
} }
// get a more accurate load time, upon first eval // get a more accurate load time by measuring the first eval
// this will take into consideration any page fault slowdown
if (!ctx->has_evaluated_once) { if (!ctx->has_evaluated_once) {
ctx->t_load_us = ggml_time_us() - ctx->t_start_us; ctx->t_load_us += ggml_time_us() - start_eval;
ctx->has_evaluated_once = true; ctx->has_evaluated_once = true;
} }
return 0; return 0;