llama-chat : fix wrong template in GLM4-0414 (#13140)

* fix wrong template in GLM4-0414

* fix spaces

* no bos token since it is already in the template

* moved the chatgml4 check to higher priority

* restored template for old GLM models

* moved the GLM4 template check in the correct place with correct check

diff --git a/convert_hf_to_gguf.py b/convert_hf_to_gguf.py
index ea3a951b93753d2372f9b15ef22af2eb2074beda..d4fec408dd2020c2f05f769a284806e0809cc7e9 100755 (executable)
--- a/convert_hf_to_gguf.py
+++ b/convert_hf_to_gguf.py
@@ -5154,7 +5154,7 @@ class Glm4Model(TextModel):
         special_vocab._set_special_token("eos", tokenizer.get_added_vocab()["<|endoftext|>"])
         special_vocab._set_special_token("eot", tokenizer.get_added_vocab()["<|user|>"])
         special_vocab._set_special_token("unk", tokenizer.get_added_vocab()["<|endoftext|>"])
-        special_vocab._set_special_token("bos", tokenizer.get_added_vocab()["[gMASK]"])
+        special_vocab._set_special_token("bos", tokenizer.get_added_vocab()["<|endoftext|>"])
         special_vocab.add_to_gguf(self.gguf_writer)
 
     def set_gguf_parameters(self):

diff --git a/src/llama-chat.cpp b/src/llama-chat.cpp
index 41f89e3a9d3bde801ade0117250c843493db97fa..698c30ce49710486308dc14f96716b99bed72c69 100644 (file)
--- a/src/llama-chat.cpp
+++ b/src/llama-chat.cpp
@@ -122,6 +122,8 @@ 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("[gMASK]<sop>")) {
+        return LLM_CHAT_TEMPLATE_CHATGML_4;
     } else if (tmpl_contains("<|assistant|>") && tmpl_contains("<|user|>")) {
         return tmpl_contains("</s>") ? LLM_CHAT_TEMPLATE_FALCON_3 : LLM_CHAT_TEMPLATE_GLMEDGE;
     } else if (tmpl_contains("<|{{ item['role'] }}|>") && tmpl_contains("<|begin_of_image|>")) {
@@ -155,8 +157,6 @@ llm_chat_template llm_chat_detect_template(const std::string & tmpl) {
     } else if (tmpl_contains("[gMASK]sop")) {
         // chatglm3-6b
         return LLM_CHAT_TEMPLATE_CHATGML_3;
-    } else if (tmpl_contains("[gMASK]<sop>")) {
-        return LLM_CHAT_TEMPLATE_CHATGML_4;
     } else if (tmpl_contains(LU8("<用户>"))) {
         // MiniCPM-3B-OpenHermes-2.5-v2-GGUF
         return LLM_CHAT_TEMPLATE_MINICPM;