From a55abe44aa5a7e2b0f350ea897b1fd481eba6ffe Mon Sep 17 00:00:00 2001 From: Xuan Son Nguyen Date: Wed, 12 Aug 2026 13:47:09 +0200 Subject: [PATCH] metrics_flush_idle --- tools/server/server-context.cpp | 13 +++++++++++++ 1 file changed, 13 insertions(+) diff --git a/tools/server/server-context.cpp b/tools/server/server-context.cpp index 6b0eade736..c3afe3d901 100644 --- a/tools/server/server-context.cpp +++ b/tools/server/server-context.cpp @@ -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;