ver4a/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

mtmd : fix glm-edge redundant token count (#13139)

Browse source 
* mtmd : fix glm-edge redundant token count

* fix chat template

* temporary disable GLMEdge test chat tmpl
This commit is contained in:
Xuan-Son Nguyen 2025-04-28 16:12:56 +02:00 • committed by GitHub
parent fb0471d175
commit 4e87962e34
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
3 changed files with 11 additions and 26 deletions
Download patch file Download diff file Expand all files Collapse all files

10
src/llama-chat.cpp
View file

@ -447,7 +447,7 @@ int32_t llm_chat_apply_template(
if (add_ass) {
ss << "<|assistant|>";
}
} else if (tmpl == LLM_CHAT_TEMPLATE_CHATGLM_4) {
} else if (tmpl == LLM_CHAT_TEMPLATE_CHATGLM_4 || tmpl == LLM_CHAT_TEMPLATE_GLMEDGE) {
ss << "[gMASK]" << "<sop>";
for (auto message : chat) {
std::string role(message->role);
@ -456,14 +456,6 @@ int32_t llm_chat_apply_template(
if (add_ass) {
ss << "<|assistant|>";
}
} else if (tmpl == LLM_CHAT_TEMPLATE_GLMEDGE) {
for (auto message : chat) {
std::string role(message->role);
ss << "<|" << role << "|>" << "\n" << message->content;
}
if (add_ass) {
ss << "<|assistant|>";
}
} else if (tmpl == LLM_CHAT_TEMPLATE_MINICPM) {
// MiniCPM-3B-OpenHermes-2.5-v2-GGUF
for (auto message : chat) {