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

Add Jinja template support (#11016)

Browse source 
* Copy minja from 58f0ca6dd7

* Add --jinja and --chat-template-file flags

* Add missing <optional> include

* Avoid print in get_hf_chat_template.py

* No designated initializers yet

* Try and work around msvc++ non-macro max resolution quirk

* Update test_chat_completion.py

* Wire LLM_KV_TOKENIZER_CHAT_TEMPLATE_N in llama_model_chat_template

* Refactor test-chat-template

* Test templates w/ minja

* Fix deprecation

* Add --jinja to llama-run

* Update common_chat_format_example to use minja template wrapper

* Test chat_template in e2e test

* Update utils.py

* Update test_chat_completion.py

* Update run.cpp

* Update arg.cpp

* Refactor common_chat_* functions to accept minja template + use_jinja option

* Attempt to fix linkage of LLAMA_CHATML_TEMPLATE

* Revert LLAMA_CHATML_TEMPLATE refactor

* Normalize newlines in test-chat-templates for windows tests

* Forward decl minja::chat_template to avoid eager json dep

* Flush stdout in chat template before potential crash

* Fix copy elision warning

* Rm unused optional include

* Add missing optional include to server.cpp

* Disable jinja test that has a cryptic windows failure

* minja: fix vigogne (https://github.com/google/minja/pull/22)

* Apply suggestions from code review

Co-authored-by: Xuan Son Nguyen <thichthat@gmail.com>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>

* Finish suggested renamings

* Move chat_templates inside server_context + remove mutex

* Update --chat-template-file w/ recent change to --chat-template

* Refactor chat template validation

* Guard against missing eos/bos tokens (null token otherwise throws in llama_vocab::impl::token_get_attr)

* Warn against missing eos / bos tokens when jinja template references them

* rename: common_chat_template[s]

* reinstate assert on chat_templates.template_default

* Update minja to b8437df626

* Update minja to https://github.com/google/minja/pull/25

* Update minja from https://github.com/google/minja/pull/27

* rm unused optional header

---------

Co-authored-by: Xuan Son Nguyen <thichthat@gmail.com>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
This commit is contained in:
Olivier Chafik 2025-01-21 13:18:51 +00:00 • committed by GitHub
parent e28245f35f
commit 6171c9d258
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
22 changed files with 3563 additions and 133 deletions
Download patch file Download diff file Expand all files Collapse all files

51
examples/server/server.cpp
View file

@ -1688,6 +1688,8 @@ struct server_context {
// Necessary similarity of prompt for slot selection
float slot_prompt_similarity = 0.0f;
common_chat_templates chat_templates;
~server_context() {
// Clear any sampling context
for (server_slot & slot : slots) {
@ -1767,14 +1769,39 @@ struct server_context {
cparams_dft.type_v = GGML_TYPE_F16;
}
chat_templates = common_chat_templates_from_model(model, params_base.chat_template);
GGML_ASSERT(chat_templates.template_default.get() != nullptr);
return true;
}
bool validate_builtin_chat_template() const {
bool validate_builtin_chat_template(bool use_jinja) const {
llama_chat_message chat[] = {{"user", "test"}};
const char * tmpl = llama_model_chat_template(model);
const int32_t chat_res = llama_chat_apply_template(tmpl, chat, 1, true, nullptr, 0);
return chat_res > 0;
if (use_jinja) {
auto templates = common_chat_templates_from_model(model, "");
GGML_ASSERT(templates.template_default);
try {
templates.template_default->apply({{
{"role", "user"},
{"content", "test"},
}}, json(), true);
if (templates.template_tool_use) {
templates.template_tool_use->apply({{
{"role", "user"},
{"content", "test"},
}}, json(), true);
}
return true;
} catch (const std::exception & e) {
SRV_ERR("failed to apply template: %s\n", e.what());
return false;
}
} else {
const char * tmpl = llama_model_chat_template(model, /* name */ nullptr);
const int32_t chat_res = llama_chat_apply_template(tmpl, chat, 1, true, nullptr, 0);
return chat_res > 0;
}
}
void init() {
@ -3659,9 +3686,12 @@ int main(int argc, char ** argv) {
{ "default_generation_settings", ctx_server.default_generation_settings_for_props },
{ "total_slots", ctx_server.params_base.n_parallel },
{ "model_path", ctx_server.params_base.model },
{ "chat_template", common_get_builtin_chat_template(ctx_server.model) },
{ "chat_template", ctx_server.chat_templates.template_default->source() },
{ "build_info", build_info },
};
if (ctx_server.params_base.use_jinja && ctx_server.chat_templates.template_tool_use) {
data["chat_template_tool_use"] = ctx_server.chat_templates.template_tool_use->source();
}
res_ok(res, data);
};
@ -3889,7 +3919,10 @@ int main(int argc, char ** argv) {
return;
}
json data = oaicompat_chat_completion_params_parse(ctx_server.model, json::parse(req.body), params.chat_template);
auto body = json::parse(req.body);
const auto & chat_template = body.contains("tools") && ctx_server.chat_templates.template_tool_use ? *ctx_server.chat_templates.template_tool_use : *ctx_server.chat_templates.template_default;
json data = oaicompat_completion_params_parse(body, chat_template, params.use_jinja);
return handle_completions_impl(
SERVER_TASK_TYPE_COMPLETION,
data,
@ -4299,7 +4332,7 @@ int main(int argc, char ** argv) {
// if a custom chat template is not supplied, we will use the one that comes with the model (if any)
if (params.chat_template.empty()) {
if (!ctx_server.validate_builtin_chat_template()) {
if (!ctx_server.validate_builtin_chat_template(params.use_jinja)) {
LOG_WRN("%s: The chat template that comes with this model is not yet supported, falling back to chatml. This may cause the model to output suboptimal responses\n", __func__);
params.chat_template = "chatml";
}
@ -4307,8 +4340,8 @@ int main(int argc, char ** argv) {
// print sample chat example to make it clear which template is used
LOG_INF("%s: chat template, chat_template: %s, example_format: '%s'\n", __func__,
params.chat_template.empty() ? "(built-in)" : params.chat_template.c_str(),
common_chat_format_example(ctx_server.model, params.chat_template).c_str());
ctx_server.chat_templates.template_default->source().c_str(),
common_chat_format_example(*ctx_server.chat_templates.template_default, ctx_server.params_base.use_jinja).c_str());
ctx_server.queue_tasks.on_new_task(std::bind(
&server_context::process_single_task, &ctx_server, std::placeholders::_1));