model : add EXAONE MoE (#18543)

* Add EXAONE MoE implementations

Co-authored-by: Junwon Hwang <nuclear1221@gmail.com>

* Address PR feedback

* Address PR feedback

* [WIP] Add MTP for EXAONE-MoE

* Address PR feedback

* Address PR feedback

* Address PR feedback

* Address PR feedback

* Address PR feedback

* Address PR feedback

* Address PR feedback

---------

Co-authored-by: LG-AI-EXAONE <exaonemodels@lgresearch.ai>
This commit is contained in:
Junwon Hwang 2026-01-14 07:28:38 +09:00 committed by GitHub
parent e4832e3ae4
commit 60591f01d4
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
17 changed files with 656 additions and 2 deletions

View file

@ -57,6 +57,7 @@ static const std::map<std::string, llm_chat_template> LLM_CHAT_TEMPLATES = {
{ "minicpm", LLM_CHAT_TEMPLATE_MINICPM },
{ "exaone3", LLM_CHAT_TEMPLATE_EXAONE_3 },
{ "exaone4", LLM_CHAT_TEMPLATE_EXAONE_4 },
{ "exaone-moe", LLM_CHAT_TEMPLATE_EXAONE_MOE },
{ "rwkv-world", LLM_CHAT_TEMPLATE_RWKV_WORLD },
{ "granite", LLM_CHAT_TEMPLATE_GRANITE },
{ "gigachat", LLM_CHAT_TEMPLATE_GIGACHAT },
@ -137,6 +138,9 @@ llm_chat_template llm_chat_detect_template(const std::string & tmpl) {
} else if (tmpl_contains("[gMASK]<sop>")) {
return LLM_CHAT_TEMPLATE_CHATGLM_4;
} else if (tmpl_contains("<|assistant|>") && tmpl_contains("<|user|>")) {
if (tmpl_contains("<|tool_declare|>")) {
return LLM_CHAT_TEMPLATE_EXAONE_MOE;
}
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;
@ -576,6 +580,22 @@ int32_t llm_chat_apply_template(
if (add_ass) {
ss << "[|assistant|]";
}
} else if (tmpl == LLM_CHAT_TEMPLATE_EXAONE_MOE) {
for (auto message : chat) {
std::string role(message->role);
if (role == "system") {
ss << "<|system|>\n" << trim(message->content) << "<|endofturn|>\n";
} else if (role == "user") {
ss << "<|user|>\n" << trim(message->content) << "<|endofturn|>\n";
} else if (role == "assistant") {
ss << "<|assistant|>\n" << trim(message->content) << "<|endofturn|>\n";
} else if (role == "tool") {
ss << "<|tool|>\n" << trim(message->content) << "<|endofturn|>\n";
}
}
if (add_ass) {
ss << "<|assistant|>\n";
}
} else if (tmpl == LLM_CHAT_TEMPLATE_RWKV_WORLD) {
// this template requires the model to have "\n\n" as EOT token
for (size_t i = 0; i < chat.size(); i++) {