fix wrong template in GLM4-0414

Author: matteoserva

src/llama-chat.cpp

@@ -123,6 +123,9 @@ llm_chat_template llm_chat_detect_template(const std::string & tmpl) {
     } else if (tmpl_contains("<|assistant|>") && tmpl_contains("<|end|>")) {
         return LLM_CHAT_TEMPLATE_PHI_3;
     } else if (tmpl_contains("<|assistant|>") && tmpl_contains("<|user|>")) {
+        if (tmpl_contains("[gMASK]<sop>")) { /* new GLM4 0414 models */
+			return LLM_CHAT_TEMPLATE_CHATGML_4;
+		}
         return tmpl_contains("</s>") ? LLM_CHAT_TEMPLATE_FALCON_3 : LLM_CHAT_TEMPLATE_GLMEDGE;
     } else if (tmpl_contains("<|{{ item['role'] }}|>") && tmpl_contains("<|begin_of_image|>")) {
         return LLM_CHAT_TEMPLATE_GLMEDGE;