happyz
/
llama.cpp
mirror of https://github.com/ggerganov/llama.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

spec : simplify time measurement using common_time_meas (#19262)

This commit is contained in:
Georgi Gerganov 2026-02-03 08:20:15 +02:00 committed by GitHub
parent 91ea44e89b
commit aeb827a3cc
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 10 additions and 19 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

29
common/speculative.cpp
View File

@ -951,12 +951,8 @@ void common_speculative_begin(common_speculative * spec, const llama_tokens & pr
} }
for (auto & impl : spec->impls) { for (auto & impl : spec->impls) {
const int64_t t_start_us = impl->gen_perf ? ggml_time_us() : 0; common_time_meas tm(impl->t_begin_us, !impl->gen_perf);
impl->begin(prompt); impl->begin(prompt);
const int64_t t_now_us = impl->gen_perf ? ggml_time_us() : 0;
impl->t_begin_us += t_now_us - t_start_us; // accumulate duration for this refresh
} }
} }
@ -971,14 +967,9 @@ llama_tokens common_speculative_draft(
for (auto & impl : spec->impls) { for (auto & impl : spec->impls) {
{ {
const int64_t t_start_us = impl->gen_perf ? ggml_time_us() : 0; common_time_meas tm(impl->t_draft_us, !impl->gen_perf);
impl->draft(params, prompt_tgt, id_last, result); impl->draft(params, prompt_tgt, id_last, result);
const int64_t t_now_us = impl->gen_perf ? ggml_time_us() : 0;
impl->drafts_call_count++; impl->drafts_call_count++;
impl->t_draft_us += t_now_us - t_start_us; // accumulate duration for this implementation
} }
if (!result.empty()) { if (!result.empty()) {
@ -1006,15 +997,15 @@ void common_speculative_accept(common_speculative * spec, uint16_t n_accepted) {
GGML_ASSERT(impl); GGML_ASSERT(impl);
const int64_t t_start_us = impl->gen_perf ? ggml_time_us() : 0; {
if (n_accepted > 0) { common_time_meas tm(impl->t_accept_us, !impl->gen_perf);
impl->drafts_accepted_count++; if (n_accepted > 0) {
impl->drafts_accepted_tokens += n_accepted; impl->drafts_accepted_count++;
} impl->drafts_accepted_tokens += n_accepted;
}
impl->accept(n_accepted); impl->accept(n_accepted);
const int64_t t_now_us = impl->gen_perf ? ggml_time_us() : 0; }
impl->t_accept_us += t_now_us - t_start_us; // accumulate duration for this acculumulation
} }
void common_speculative_print_stats(const common_speculative * spec) { void common_speculative_print_stats(const common_speculative * spec) {