Skip to content

server : add last-5-seconds generation speed display #24291

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
ngxson merged 2 commits into from
Jun 18, 2026
Merged

server : add last-5-seconds generation speed display #24291

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
bdb1420
server : add last-5-seconds generation speed display
akx Jun 8, 2026
db10354
cont : clean-up
ggerganov Jun 18, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
27 changes: 16 additions & 11 deletions tools/server/server-context.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -193,9 +193,10 @@ struct server_slot {
// stats
size_t n_sent_text = 0; // number of sent text character

int64_t t_print_last = 0;
int64_t t_start_process_prompt;
int64_t t_start_generation;
int64_t t_print_last = 0;
int32_t n_decoded_last = 0;

double t_prompt_processing = 0.0; // ms
double t_token_generation = 0.0; // ms
Expand Down Expand Up @@ -467,11 +468,13 @@ struct server_slot {
return;
}

t_print_last = t_now;
const double n_gen_second = 1e3 / (t_token_generation) * (n_decoded);
const double n_gen_second_win = 1e6 / (t_now - t_print_last) * (n_decoded - n_decoded_last);

const double n_gen_second = 1e3 / t_token_generation * n_decoded;
t_print_last = t_now;
n_decoded_last = n_decoded;

SLT_INF(*this, "n_decoded = %6d, tg = %6.2f t/s\n", n_decoded, n_gen_second);
SLT_INF(*this, "n_decoded = %6d, tg = %6.2f t/s, tg_3s = %6.2f t/s\n", n_decoded, n_gen_second, n_gen_second_win);
}

void print_timings_pp() const {
Expand Down Expand Up @@ -2937,8 +2940,8 @@ struct server_context_impl {
}
}

const int64_t t_current = ggml_time_us();
slot.t_prompt_processing = (t_current - slot.t_start_process_prompt) / 1e3;
const int64_t t_now = ggml_time_us();
slot.t_prompt_processing = (t_now - slot.t_start_process_prompt) / 1e3;
slot.print_timings_pp();

// truncate any tokens that are beyond n_past for this slot
Expand Down Expand Up @@ -3391,17 +3394,19 @@ struct server_context_impl {
common_sampler_accept(slot.smpl.get(), id, true);

// here we have synchronized the llama_context (due to the sampling above), so we can do time measurement
const int64_t t_current = ggml_time_us();
const int64_t t_now = ggml_time_us();

slot.n_decoded += 1;

if (slot.n_decoded == 1) {
slot.t_start_generation = t_current;
slot.t_start_generation = t_now;
slot.t_print_last = t_now;
slot.n_decoded_last = 0;
slot.t_prompt_processing = (slot.t_start_generation - slot.t_start_process_prompt) / 1e3;
metrics.on_prompt_eval(slot);
}

slot.t_token_generation = std::max<int64_t>(1, t_current - slot.t_start_generation) / 1e3;
slot.t_token_generation = std::max<int64_t>(1, t_now - slot.t_start_generation) / 1e3;

completion_token_output result;
result.tok = id;
Expand Down Expand Up @@ -3495,11 +3500,11 @@ struct server_context_impl {
slot.spec_draft = std::move(accepted);
}

const int64_t t_current = ggml_time_us();
const int64_t t_now = ggml_time_us();

const auto ids = std::move(slot.spec_draft);

slot.t_token_generation = std::max<int64_t>(1, t_current - slot.t_start_generation) / 1e3;
slot.t_token_generation = std::max<int64_t>(1, t_now - slot.t_start_generation) / 1e3;

// update how many tokens out of those tested were accepted
slot.n_draft_accepted += ids.size() - 1;
Expand Down