diff --git a/src/llama.cpp b/src/llama.cpp
index 36c5e681f..8a53df14a 100644
--- a/src/llama.cpp
+++ b/src/llama.cpp
@@ -5103,6 +5103,7 @@ static void llm_load_vocab(
             } else if (
                 tokenizer_pre == "chatglm-bpe") {
                 vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_CHATGLM4;
+                vocab.special_bos_id  = -1;
             } else if (
                 tokenizer_pre == "viking") {
                 vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_VIKING;
@@ -19828,7 +19829,7 @@ static int32_t llama_chat_apply_template_internal(
         if (add_ass) {
             ss << "<|start_header_id|>assistant<|end_header_id|>\n\n";
         }
-    } else if (tmpl == "chatglm3" || tmpl.find("[gMASK]sop") != std::string::npos) {
+    } else if (tmpl.find("chatglm3") != std::string::npos || tmpl.find("[gMASK]sop") != std::string::npos) {
         // chatglm3-6b
         ss << "[gMASK]" << "sop";
         for (auto message : chat) {
@@ -19838,7 +19839,7 @@ static int32_t llama_chat_apply_template_internal(
         if (add_ass) {
             ss << "<|assistant|>";
         }
-    } else if (tmpl == "chatglm4" || tmpl.find("[gMASK]<sop>") != std::string::npos) {
+    } else if (tmpl.find("chatglm4") != std::string::npos || tmpl.find("[gMASK]<sop>") != std::string::npos) {
         ss << "[gMASK]" << "<sop>";
         for (auto message : chat) {
             std::string role(message->role);