From: Georgi Gerganov Date: Sun, 8 Feb 2026 07:40:04 +0000 (+0200) Subject: server : improve context checkpoint logic (#19408) X-Git-Tag: upstream/0.0.8067~97 X-Git-Url: https://git.djapps.eu/?a=commitdiff_plain;h=eb449cdfa4319d8fd9066e4633b49f4c867dd11c;p=pkg%2Fggml%2Fsources%2Fllama.cpp server : improve context checkpoint logic (#19408) --- diff --git a/tools/server/server-context.cpp b/tools/server/server-context.cpp index b71d496ee..8ec845133 100644 --- a/tools/server/server-context.cpp +++ b/tools/server/server-context.cpp @@ -2507,7 +2507,8 @@ private: slot.n_prompt_tokens_processed++; // process the last few tokens of the prompt separately in order to allow for a checkpoint to be created. - if (do_checkpoint && slot.task->n_tokens() - slot.prompt.n_tokens() == 64) { + const int n_last = std::min(n_batch, 512); + if (do_checkpoint && slot.task->n_tokens() == slot.prompt.n_tokens() + n_last) { break; } }