Add chat template for vicuna

llama.cpp

@@ -15713,6 +15713,23 @@ static int32_t llama_chat_apply_template_internal(
         if (add_ass) {
             ss << "GPT4 Correct Assistant:";
         }
+    } else if (tmpl == "vicuna" || tmpl.find("USER: ") != std::string::npos) {
+        for (auto message : chat) {
+            std::string role(message->role);
+            if (role == "user") {
+                ss << "USER: ";
+            } else if (role == "assistant") {
+                ss << "ASSISTANT: ";
+            }
+            ss << message->content << "\n";
+            if (role == "system") {
+                // Extra newline after system message
+                ss << "\n";
+            }
+        }
+        if (add_ass) {
+            ss << "ASSISTANT:";
+        }
     } else {
         // template not supported
         return -1;