]> git.djapps.eu Git - pkg/ggml/sources/llama.cpp/commitdiff
server : adjust prompt similarity thold + add logs (#15913)
authorGeorgi Gerganov <redacted>
Fri, 12 Sep 2025 14:02:55 +0000 (17:02 +0300)
committerGitHub <redacted>
Fri, 12 Sep 2025 14:02:55 +0000 (17:02 +0300)
ggml-ci

common/common.h
tools/server/server.cpp

index 85b3b879d45362962d8f22b0161d85a3c0f97b7b..740168743aec94027741a44759a53895714fcb0f 100644 (file)
@@ -452,7 +452,7 @@ struct common_params {
 
     std::string slot_save_path;
 
-    float slot_prompt_similarity = 0.5f;
+    float slot_prompt_similarity = 0.1f;
 
     // batched-bench params
     bool is_pp_shared = false;
index 39ef439e94f97071492cab708cbb3f37d27f1294..160b97cf7d2c9804ae7963f8d022973939eb6e65 100644 (file)
@@ -2372,7 +2372,7 @@ struct server_context {
             }
 
             if (ret != nullptr) {
-                SLT_DBG(*ret, "selected slot by lcs similarity, lcs_len = %d, similarity = %f\n", lcs_len, similarity);
+                SLT_INF(*ret, "selected slot by lcs similarity, lcs_len = %d, similarity = %.3f (> %.3f thold)\n", lcs_len, similarity, slot_prompt_similarity);
             }
         }
 
@@ -2394,7 +2394,7 @@ struct server_context {
             }
 
             if (ret != nullptr) {
-                SLT_DBG(*ret, "selected slot by lru, t_last = %" PRId64 "\n", t_last);
+                SLT_INF(*ret, "selected slot by LRU, t_last = %" PRId64 "\n", t_last);
             }
         }