server : final touches

ggml-ci

Author: ggerganov

examples/server/server.cpp

@@ -2296,7 +2296,7 @@ struct server_context {
                 }
 
                 // check if the slot supports speculative decoding
-                if (!slot.ctx_dft || slot.params.speculative.n_max <= 0) {
+                if (!slot.ctx_dft || slot.params.speculative.n_max <= 0 || !slot.params.cache_prompt) {
                     continue;
                 }
 
@@ -2348,6 +2348,8 @@ struct server_context {
                         break;
                     }
                 }
+
+                SRV_DBG("accepted %d/%d draft tokens\n", (int) ids.size() - 1, (int) draft.size());
             }
         }