CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-10-31 08:51:55 +00:00
Code Issues Packages Projects Releases Wiki Activity

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

Browse Source 
* 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
This commit is contained in:
matteo
2025-04-27 21:57:32 +02:00
committed by GitHub
parent e291450b76
commit ced44be342
2 changed files with 3 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
convert_hf_to_gguf.py
View File

@@ -5154,7 +5154,7 @@ class Glm4Model(TextModel):
special_vocab._set_special_token("eos", tokenizer.get_added_vocab()["<|endoftext|>"]) 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("eot", tokenizer.get_added_vocab()["<|user|>"])
special_vocab._set_special_token("unk", tokenizer.get_added_vocab()["<|endoftext|>"]) 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) special_vocab.add_to_gguf(self.gguf_writer)
def set_gguf_parameters(self): def set_gguf_parameters(self):

4
src/llama-chat.cpp
View File

@@ -122,6 +122,8 @@ llm_chat_template llm_chat_detect_template(const std::string & tmpl) {
} }
} else if (tmpl_contains("<|assistant|>") && tmpl_contains("<|end|>")) { } else if (tmpl_contains("<|assistant|>") && tmpl_contains("<|end|>")) {
return LLM_CHAT_TEMPLATE_PHI_3; 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|>")) { } else if (tmpl_contains("<|assistant|>") && tmpl_contains("<|user|>")) {
return tmpl_contains("</s>") ? LLM_CHAT_TEMPLATE_FALCON_3 : LLM_CHAT_TEMPLATE_GLMEDGE; return tmpl_contains("</s>") ? LLM_CHAT_TEMPLATE_FALCON_3 : LLM_CHAT_TEMPLATE_GLMEDGE;
} else if (tmpl_contains("<|{{ item['role'] }}|>") && tmpl_contains("<|begin_of_image|>")) { } 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")) { } else if (tmpl_contains("[gMASK]sop")) {
// chatglm3-6b // chatglm3-6b
return LLM_CHAT_TEMPLATE_CHATGML_3; 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>"))) { } else if (tmpl_contains(LU8("<用户>"))) {
// MiniCPM-3B-OpenHermes-2.5-v2-GGUF // MiniCPM-3B-OpenHermes-2.5-v2-GGUF
return LLM_CHAT_TEMPLATE_MINICPM; return LLM_CHAT_TEMPLATE_MINICPM;