https://github.com/ggerganov/llama.cpp/pull/11736

Author: Animaxx

ggml/src/ggml-cpu/ggml-cpu.c

@@ -13416,9 +13416,13 @@ static thread_ret_t ggml_graph_compute_thread(void * data) {
             tp->ec    = GGML_STATUS_ABORTED;
         }
 
-        ggml_barrier(state->threadpool);
+        if (node_n + 1 < cgraph->n_nodes) {
+            ggml_barrier(state->threadpool);
+        }
     }
 
+    ggml_barrier(state->threadpool);
+
     return 0;
 }