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

add new chat template

Browse source
This commit is contained in:
ngxson 2024-02-21 11:12:58 +01:00
parent 9576df906a
commit b599545e67
2 changed files with 10 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

9
llama.cpp
View file

@ -12580,6 +12580,15 @@ static int32_t llama_chat_apply_template_internal(
if (add_ass) {
ss << "<|assistant|>\n";
}
} else if (tmpl.find("bos_token + message['role']") != std::string::npos) {
// mlabonne/AlphaMonarch-7B template (the <s> is included inside history)
for (auto message : chat) {
std::string bos = (message == chat.front()) ? "" : "<s>"; // skip BOS for first message
ss << bos << message->role << "\n" << message->content << "</s>\n";
}
if (add_ass) {
ss << "<s>assistant\n";
}
} else {
// template not supported
return -1;

1
llama.h
View file

@ -708,6 +708,7 @@ extern "C" {
/// Apply chat template. Inspired by hf apply_chat_template() on python.
/// Both "model" and "custom_template" are optional, but at least one is required. "custom_template" has higher precedence than "model"
/// NOTE: This function only support some known jinja templates. It is not a jinja parser.
/// TODO: Add link to docs for this function so that developers can know the list of supported templates
/// @param tmpl A Jinja template to use for this chat. If this is nullptr, the model’s default chat template will be used instead.
/// @param chat Pointer to a list of multiple llama_chat_message
/// @param n_msg Number of llama_chat_message in this chat