Commit 5a9a34d88 for llama.cpp

commit 5a9a34d88847d0875bd21eab0d7c04da4bcfd2af
Author: TANGBUDU <79953480+TANGBUDU@users.noreply.github.com>
Date:   Wed Oct 21 01:32:12 2026 +0900

    server : avoid repeated mmproj memory margin addition on sleep/resume (#28233)

    Co-authored-by: TANGBUDU <TANGBUDU@users.noreply.github.com>

diff --git a/tools/server/server-context.cpp b/tools/server/server-context.cpp
index b2116b174..74812189c 100644
--- a/tools/server/server-context.cpp
+++ b/tools/server/server-context.cpp
@@ -1180,7 +1180,8 @@ private:
         }

         // optionally fit mmproj memory usage
-        if (has_mmproj && params_base.fit_params) {
+        // on resume params_base already holds the margin from the first load
+        if (has_mmproj && params_base.fit_params && !is_resume) {
             const auto & mmproj_mem = mmproj_usage.backend_mem_usage;
             const int64_t t_elapsed = mmproj_usage_t_us;
             if (!mmproj_mem.empty()) {