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

llama : add enum for built-in chat templates (#10623)

Browse source 
* llama : add enum for supported chat templates

* use "built-in" instead of "supported"

* arg: print list of built-in templates

* fix test

* update server README
This commit is contained in:
Xuan Son Nguyen 2024-12-02 22:10:19 +01:00 • committed by GitHub
parent 8648c52101
commit 642330ac7c
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
5 changed files with 310 additions and 107 deletions
Download patch file Download diff file Expand all files Collapse all files

20
tests/test-chat-template.cpp
View file

@ -82,9 +82,9 @@ int main(void) {
// mistralai/Mistral-7B-Instruct-v0.2 (NOTE: Old pre-v1 without a system prompt)
"[INST] You are a helpful assistant\nHello [/INST]Hi there</s>[INST] Who are you [/INST] I am an assistant </s>[INST] Another question [/INST]",
// TheBloke/FusionNet_34Bx2_MoE-AWQ
"[INST] <<SYS>>\nYou are a helpful assistant\n<</SYS>>\n\nHello [/INST] Hi there </s><s>[INST] Who are you [/INST] I am an assistant </s><s>[INST] Another question [/INST]",
"[INST] <<SYS>>\nYou are a helpful assistant\n<</SYS>>\n\nHello [/INST]Hi there</s><s>[INST] Who are you [/INST] I am an assistant </s><s>[INST] Another question [/INST]",
// bofenghuang/vigogne-2-70b-chat
"[INST] <<SYS>>\nYou are a helpful assistant\n<</SYS>>\n\nHello [/INST] Hi there </s>[INST] Who are you [/INST] I am an assistant </s>[INST] Another question [/INST]",
"[INST] <<SYS>>\nYou are a helpful assistant\n<</SYS>>\n\nHello [/INST]Hi there</s>[INST] Who are you [/INST]I am an assistant</s>[INST] Another question [/INST]",
// mlabonne/AlphaMonarch-7B
"system\nYou are a helpful assistant</s>\n<s>user\nHello</s>\n<s>assistant\nHi there</s>\n<s>user\nWho are you</s>\n<s>assistant\n I am an assistant </s>\n<s>user\nAnother question</s>\n<s>assistant\n",
// google/gemma-7b-it
@ -133,6 +133,17 @@ int main(void) {
std::vector<char> formatted_chat(1024);
int32_t res;
// list all supported templates
std::vector<const char *> supported_tmpl;
res = llama_chat_builtin_templates(nullptr, 0);
assert(res > 0);
supported_tmpl.resize(res);
res = llama_chat_builtin_templates(supported_tmpl.data(), supported_tmpl.size());
printf("Built-in chat templates:\n");
for (auto tmpl : supported_tmpl) {
printf(" %s\n", tmpl);
}
// test invalid chat template
res = llama_chat_apply_template(nullptr, "INVALID TEMPLATE", conversation, message_count, true, formatted_chat.data(), formatted_chat.size());
assert(res < 0);
@ -174,7 +185,8 @@ int main(void) {
assert(fmt_sys("mistral-v3") == "[INST] You are a helpful assistant\n\n");
assert(fmt_sys("mistral-v3-tekken") == "[INST]You are a helpful assistant\n\n");
assert(fmt_sys("mistral-v7") == "[SYSTEM_PROMPT] You are a helpful assistant[/SYSTEM_PROMPT]");
assert(fmt_sys("llama2") == "[INST] <<SYS>>\nYou are a helpful assistant\n<</SYS>>\n\n");
assert(fmt_sys("llama2") == "[INST] You are a helpful assistant\n");
assert(fmt_sys("llama2-sys") == "[INST] <<SYS>>\nYou are a helpful assistant\n<</SYS>>\n\n");
assert(fmt_sys("mistral") == "[INST] You are a helpful assistant\n"); // for old pre-v1 templates
assert(fmt_sys("gemma") == ""); // for gemma, system message is merged with user message
assert(fmt_sys("llama3") == "<|start_header_id|>system<|end_header_id|>\n\nYou are a helpful assistant<|eot_id|>");
@ -203,5 +215,7 @@ int main(void) {
assert(fmt_single("gemma") == "\n<start_of_turn>user\nHow are you<end_of_turn>\n<start_of_turn>model\n");
assert(fmt_single("llama3") == "<|start_header_id|>user<|end_header_id|>\n\nHow are you<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n");
printf("Test chat templates: OK\n");
return 0;
}