metrics_flush_idle

This commit is contained in:
Xuan Son Nguyen
2026-08-12 13:47:09 +02:00
parent 6849901f4a
commit a55abe44aa
+13
View File
@@ -2703,6 +2703,9 @@ private:
if (all_idle) {
SRV_TRC("%s", "all slots are idle\n");
metrics_flush_idle();
return; // skip further processing
} else {
@@ -3922,6 +3925,16 @@ private:
}
}
// flush any queued prompt metrics if all slots are now idle
void metrics_flush_idle() {
if (metrics.n_prompt_queued == 0) {
return;
}
llama_synchronize(ctx_tgt);
metrics.flush_prompt();
}
void metrics_on_prediction(const server_slot & slot) {
const uint64_t t_us = slot.stats.t_gen_us();
const uint64_t n = slot.stats.n_gen;