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

126
common/common.cpp
View file

@ -12,6 +12,7 @@
#include "json.hpp"
#include "json-schema-to-grammar.h"
#include "llama.h"
#include "chat-template.hpp"
#include <algorithm>
#include <cinttypes>
@ -1728,67 +1729,75 @@ std::string common_detokenize(const struct llama_vocab * vocab, const std::vecto
// Chat template utils
//
std::string common_get_builtin_chat_template(const struct llama_model * model) {
const char * ptr_tmpl = llama_model_chat_template(model);
return ptr_tmpl == nullptr ? "" : ptr_tmpl;
}
bool common_chat_verify_template(const std::string & tmpl) {
bool common_chat_verify_template(const std::string & tmpl, bool use_jinja) {
if (use_jinja) {
try {
auto chat_template = minja::chat_template(tmpl, "<s>", "</s>");
chat_template.apply({{
{"role", "user"},
{"content", "test"},
}}, json(), true);
return true;
} catch (const std::exception & e) {
LOG_ERR("%s: failed to apply template: %s\n", __func__, e.what());
return false;
}
}
llama_chat_message chat[] = {{"user", "test"}};
const int res = llama_chat_apply_template(tmpl.c_str(), chat, 1, true, nullptr, 0);
return res >= 0;
}
std::string common_chat_apply_template(const struct llama_model * model,
const std::string & tmpl,
std::string common_chat_apply_template(
const common_chat_template & tmpl,
const std::vector<common_chat_msg> & msgs,
bool add_ass) {
bool add_ass,
bool use_jinja) {
if (use_jinja) {
auto messages = json::array();
for (const auto & msg : msgs) {
messages.push_back({{"role", msg.role}, {"content", msg.content}});
}
return tmpl.apply(messages, /* tools= */ json(), add_ass);
}
int alloc_size = 0;
bool fallback = false; // indicate if we must fallback to default chatml
std::vector<llama_chat_message> chat;
for (const auto & msg : msgs) {
chat.push_back({msg.role.c_str(), msg.content.c_str()});
alloc_size += (msg.role.size() + msg.content.size()) * 1.25;
}
const char * ptr_tmpl = tmpl.empty() ? llama_model_chat_template(model) : tmpl.c_str();
std::vector<char> buf(alloc_size);
// run the first time to get the total output length
int32_t res = llama_chat_apply_template(ptr_tmpl, chat.data(), chat.size(), add_ass, buf.data(), buf.size());
int32_t res = llama_chat_apply_template(tmpl.source().c_str(), chat.data(), chat.size(), add_ass, buf.data(), buf.size());
// error: chat template is not supported
if (res < 0) {
if (ptr_tmpl != nullptr) {
// if the custom "tmpl" is not supported, we throw an error
// this is a bit redundant (for good), since we're not sure if user validated the custom template with llama_chat_verify_template()
throw std::runtime_error("this custom template is not supported");
}
// If the built-in template is not supported, we default to chatml
res = llama_chat_apply_template("chatml", chat.data(), chat.size(), add_ass, buf.data(), buf.size());
fallback = true;
// if the custom "tmpl" is not supported, we throw an error
// this is a bit redundant (for good), since we're not sure if user validated the custom template with llama_chat_verify_template()
throw std::runtime_error("this custom template is not supported");
}
// if it turns out that our buffer is too small, we resize it
if ((size_t) res > buf.size()) {
buf.resize(res);
res = llama_chat_apply_template(
fallback ? "chatml" : ptr_tmpl,
chat.data(), chat.size(), add_ass, buf.data(), buf.size());
res = llama_chat_apply_template(tmpl.source().c_str(), chat.data(), chat.size(), add_ass, buf.data(), buf.size());
}
std::string formatted_chat(buf.data(), res);
return formatted_chat;
}
std::string common_chat_format_single(const struct llama_model * model,
const std::string & tmpl,
std::string common_chat_format_single(
const common_chat_template & tmpl,
const std::vector<common_chat_msg> & past_msg,
const common_chat_msg & new_msg,
bool add_ass) {
bool add_ass,
bool use_jinja) {
std::ostringstream ss;
auto fmt_past_msg = past_msg.empty() ? "" : common_chat_apply_template(model, tmpl, past_msg, false);
auto fmt_past_msg = past_msg.empty() ? "" : common_chat_apply_template(tmpl, past_msg, false, use_jinja);
std::vector<common_chat_msg> chat_new(past_msg);
// if the past_msg ends with a newline, we must preserve it in the formatted version
if (add_ass && !fmt_past_msg.empty() && fmt_past_msg.back() == '\n') {
@ -1796,21 +1805,74 @@ std::string common_chat_format_single(const struct llama_model * model,
};
// format chat with new_msg
chat_new.push_back(new_msg);
auto fmt_new_msg = common_chat_apply_template(model, tmpl, chat_new, add_ass);
auto fmt_new_msg = common_chat_apply_template(tmpl, chat_new, add_ass, use_jinja);
// get the diff part
ss << fmt_new_msg.substr(fmt_past_msg.size(), fmt_new_msg.size() - fmt_past_msg.size());
return ss.str();
}
std::string common_chat_format_example(const struct llama_model * model,
const std::string & tmpl) {
std::string common_chat_format_example(const common_chat_template & tmpl, bool use_jinja) {
std::vector<common_chat_msg> msgs = {
{"system", "You are a helpful assistant"},
{"user", "Hello"},
{"assistant", "Hi there"},
{"user", "How are you?"},
};
return common_chat_apply_template(model, tmpl, msgs, true);
return common_chat_apply_template(tmpl, msgs, true, use_jinja);
}
common_chat_templates common_chat_templates_from_model(const struct llama_model * model, const std::string & chat_template_override)
{
auto vocab = llama_model_get_vocab(model);
std::string default_template_src = chat_template_override;
std::string template_tool_use_src = chat_template_override;
bool has_explicit_template = !chat_template_override.empty();
if (chat_template_override.empty()) {
auto str = llama_model_chat_template(model, /* name */ nullptr);
if (str) {
default_template_src = str;
has_explicit_template = true;
}
str = llama_model_chat_template(model, /* name */ "tool_use");
if (str) {
template_tool_use_src = str;
has_explicit_template = true;
}
}
if (default_template_src.empty() || default_template_src == "chatml") {
if (!template_tool_use_src.empty()) {
default_template_src = template_tool_use_src;
} else {
default_template_src = R"(
{%- for message in messages -%}
{{- "<|im_start|>" + message.role + "\n" + message.content + "<|im_end|>\n" -}}
{%- endfor -%}
{%- if add_generation_prompt -%}
{{- "<|im_start|>assistant\n" -}}
{%- endif -%}
)";
}
}
const auto get_token = [&](llama_token token, const char * name, const char * jinja_variable_name) {
if (token == LLAMA_TOKEN_NULL) {
if (default_template_src.find(jinja_variable_name) != std::string::npos
|| template_tool_use_src.find(jinja_variable_name) != std::string::npos) {
LOG_WRN("%s: warning: vocab does not have a %s token, jinja template won't work as intended.\n", __func__, name);
}
return std::string();
} else {
return common_token_to_piece(vocab, token, true);
}
};
auto token_bos = get_token(llama_vocab_bos(vocab), "BOS", "bos_token");
auto token_eos = get_token(llama_vocab_eos(vocab), "EOS", "eos_token");
return {
has_explicit_template,
std::make_unique<minja::chat_template>(default_template_src, token_bos, token_eos),
template_tool_use_src.empty()
? nullptr
: std::make_unique<minja::chat_template>(template_tool_use_src, token_bos, token_eos)
};
}
//