fixing find stop when maximum number of tokens

mgonzs13 · Jan 20, 2025 · ea2357a · ea2357a
1 parent 986b539
commit ea2357a
Showing 1 changed file with 2 additions and 2 deletions.
diff --git a/llama_ros/src/llama_ros/llama.cpp b/llama_ros/src/llama_ros/llama.cpp
@@ -824,13 +824,13 @@ Llama::find_stop(std::vector<struct CompletionOutput> completion_result_list,
   }
 
   // respect the maximum number of tokens
-  if (this->n_past > this->params.n_predict && this->params.n_predict != -1) {
+  if (this->n_past >= this->params.n_predict && this->params.n_predict >= 0) {
     LLAMA_LOG_INFO("Maximum number of tokens reached %d",
                    this->params.n_predict);
     return FULL_STOP;
   }
 
-  if (this->n_past > this->get_n_ctx() && this->params.n_predict == -2) {
+  if (this->n_past >= this->get_n_ctx() && this->params.n_predict == -2) {
     LLAMA_LOG_INFO("Maximum number of tokens reached %d", this->get_n_ctx());
     return FULL_STOP;
   }