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

Merge branch 'master' into compilade/imatrix-batched-chunks

Browse source
This commit is contained in:
Francis Couture-Harpin 2025-02-09 12:06:24 -05:00
commit 1be357d990
149 changed files with 14407 additions and 5105 deletions
Download patch file Download diff file Expand all files Collapse all files

28
common/CMakeLists.txt
View file

@ -65,6 +65,7 @@ add_library(${TARGET} STATIC
console.h
json-schema-to-grammar.cpp
json.hpp
llguidance.cpp
log.cpp
log.h
minja.hpp
@ -91,6 +92,33 @@ if (LLAMA_CURL)
set(LLAMA_COMMON_EXTRA_LIBS ${LLAMA_COMMON_EXTRA_LIBS} ${CURL_LIBRARY})
endif ()
if (LLAMA_LLGUIDANCE)
include(ExternalProject)
set(LLGUIDANCE_SRC ${CMAKE_BINARY_DIR}/llguidance/source)
set(LLGUIDANCE_PATH ${LLGUIDANCE_SRC}/target/release)
ExternalProject_Add(llguidance_ext
GIT_REPOSITORY https://github.com/guidance-ai/llguidance
# v0.6.12:
GIT_TAG ced1c9023d47ec194fa977932d35ce65c2ebfc09
PREFIX ${CMAKE_BINARY_DIR}/llguidance
SOURCE_DIR ${LLGUIDANCE_SRC}
BUILD_IN_SOURCE TRUE
CONFIGURE_COMMAND ""
BUILD_COMMAND cargo build --release
INSTALL_COMMAND ""
BUILD_BYPRODUCTS ${LLGUIDANCE_PATH}/libllguidance.a ${LLGUIDANCE_PATH}/llguidance.h
UPDATE_COMMAND ""
)
target_compile_definitions(${TARGET} PUBLIC LLAMA_USE_LLGUIDANCE)
add_library(llguidance STATIC IMPORTED)
set_target_properties(llguidance PROPERTIES IMPORTED_LOCATION ${LLGUIDANCE_PATH}/libllguidance.a)
add_dependencies(llguidance llguidance_ext)
target_include_directories(${TARGET} PRIVATE ${LLGUIDANCE_PATH})
set(LLAMA_COMMON_EXTRA_LIBS ${LLAMA_COMMON_EXTRA_LIBS} llguidance)
endif ()
target_include_directories(${TARGET} PUBLIC .)
target_compile_features (${TARGET} PUBLIC cxx_std_17)
target_link_libraries (${TARGET} PRIVATE ${LLAMA_COMMON_EXTRA_LIBS} PUBLIC llama Threads::Threads)

63
common/arg.cpp
View file

@ -1465,15 +1465,28 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
{"--list-devices"},
"print list of available devices and exit",
[](common_params &) {
printf("Available devices:\n");
std::vector<ggml_backend_dev_t> rpc_devices;
std::vector<ggml_backend_dev_t> all_devices;
for (size_t i = 0; i < ggml_backend_dev_count(); ++i) {
auto * dev = ggml_backend_dev_get(i);
if (ggml_backend_dev_type(dev) == GGML_BACKEND_DEVICE_TYPE_GPU) {
size_t free, total;
ggml_backend_dev_memory(dev, &free, &total);
printf(" %s: %s (%zu MiB, %zu MiB free)\n", ggml_backend_dev_name(dev), ggml_backend_dev_description(dev), total / 1024 / 1024, free / 1024 / 1024);
ggml_backend_reg_t reg = ggml_backend_dev_backend_reg(dev);
if (ggml_backend_reg_name(reg) == std::string("RPC")) {
rpc_devices.push_back(dev);
} else {
all_devices.push_back(dev);
}
}
}
// insert RPC devices in front
all_devices.insert(all_devices.begin(), rpc_devices.begin(), rpc_devices.end());
printf("Available devices:\n");
for (size_t i = 0; i < all_devices.size(); ++i) {
auto * dev = all_devices[i];
size_t free, total;
ggml_backend_dev_memory(dev, &free, &total);
printf(" %s: %s (%zu MiB, %zu MiB free)\n", ggml_backend_dev_name(dev), ggml_backend_dev_description(dev), total / 1024 / 1024, free / 1024 / 1024);
}
exit(0);
}
));
@ -2311,5 +2324,47 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
}
).set_examples({LLAMA_EXAMPLE_TTS}));
add_opt(common_arg(
{"--embd-bge-small-en-default"},
string_format("use default bge-small-en-v1.5 model (note: can download weights from the internet)"),
[](common_params & params) {
params.hf_repo = "ggml-org/bge-small-en-v1.5-Q8_0-GGUF";
params.hf_file = "bge-small-en-v1.5-q8_0.gguf";
params.pooling_type = LLAMA_POOLING_TYPE_NONE;
params.embd_normalize = 2;
params.n_ctx = 512;
params.verbose_prompt = true;
params.embedding = true;
}
).set_examples({LLAMA_EXAMPLE_EMBEDDING, LLAMA_EXAMPLE_SERVER}));
add_opt(common_arg(
{"--embd-e5-small-en-default"},
string_format("use default e5-small-v2 model (note: can download weights from the internet)"),
[](common_params & params) {
params.hf_repo = "ggml-org/e5-small-v2-Q8_0-GGUF";
params.hf_file = "e5-small-v2-q8_0.gguf";
params.pooling_type = LLAMA_POOLING_TYPE_NONE;
params.embd_normalize = 2;
params.n_ctx = 512;
params.verbose_prompt = true;
params.embedding = true;
}
).set_examples({LLAMA_EXAMPLE_EMBEDDING, LLAMA_EXAMPLE_SERVER}));
add_opt(common_arg(
{"--embd-gte-small-default"},
string_format("use default gte-small model (note: can download weights from the internet)"),
[](common_params & params) {
params.hf_repo = "ggml-org/gte-small-Q8_0-GGUF";
params.hf_file = "gte-small-q8_0.gguf";
params.pooling_type = LLAMA_POOLING_TYPE_NONE;
params.embd_normalize = 2;
params.n_ctx = 512;
params.verbose_prompt = true;
params.embedding = true;
}
).set_examples({LLAMA_EXAMPLE_EMBEDDING, LLAMA_EXAMPLE_SERVER}));
return ctx_arg;
}

215
common/chat-template.hpp
View file

@ -33,6 +33,29 @@ struct chat_template_caps {
bool requires_typed_content = false;
};
struct chat_template_inputs {
nlohmann::ordered_json messages;
nlohmann::ordered_json tools;
bool add_generation_prompt = true;
nlohmann::ordered_json extra_context;
std::chrono::system_clock::time_point now = std::chrono::system_clock::now();
};
struct chat_template_options {
bool apply_polyfills = true;
bool use_bos_token = true;
bool use_eos_token = true;
bool define_strftime_now = true;
bool polyfill_tools = true;
bool polyfill_tool_call_examples = true;
bool polyfill_tool_calls = true;
bool polyfill_tool_responses = true;
bool polyfill_system_role = true;
bool polyfill_object_arguments = true;
bool polyfill_typed_content = true;
};
class chat_template {
private:
@ -41,6 +64,7 @@ class chat_template {
std::string bos_token_;
std::string eos_token_;
std::shared_ptr<minja::TemplateNode> template_root_;
std::string tool_call_example_;
std::string try_raw_render(
const nlohmann::ordered_json & messages,
@ -49,7 +73,18 @@ class chat_template {
const nlohmann::ordered_json & extra_context = nlohmann::ordered_json()) const
{
try {
auto prompt = apply(messages, tools, add_generation_prompt, extra_context, /* adjust_inputs= */ false);
chat_template_inputs inputs;
inputs.messages = messages;
inputs.tools = tools;
inputs.add_generation_prompt = add_generation_prompt;
inputs.extra_context = extra_context;
// Use fixed date for tests
inputs.now = std::chrono::system_clock::from_time_t(0);
chat_template_options opts;
opts.apply_polyfills = false;
auto prompt = apply(inputs, opts);
// fprintf(stderr, "try_raw_render: %s\n", prompt.c_str());
return prompt;
} catch (const std::exception & e) {
@ -176,6 +211,58 @@ class chat_template {
caps_.supports_tool_responses = contains(out, "Some response!");
caps_.supports_tool_call_id = contains(out, "call_911_");
}
try {
if (!caps_.supports_tools) {
const json user_msg {
{"role", "user"},
{"content", "Hey"},
};
const json args {
{"arg1", "some_value"},
};
const json tool_call_msg {
{"role", "assistant"},
{"content", nullptr},
{"tool_calls", json::array({
{
// TODO: detect if requires numerical id or fixed length == 6 like Nemo
{"id", "call_1___"},
{"type", "function"},
{"function", {
{"name", "tool_name"},
{"arguments", (caps_.requires_object_arguments ? args : json(minja::Value(args).dump(-1, /* to_json= */ true)))},
}},
},
})},
};
std::string prefix, full;
{
chat_template_inputs inputs;
inputs.messages = json::array({user_msg});
inputs.add_generation_prompt = true;
prefix = apply(inputs);
}
{
chat_template_inputs inputs;
inputs.messages = json::array({user_msg, tool_call_msg});
inputs.add_generation_prompt = false;
full = apply(inputs);
}
if (full.find(prefix) != 0) {
if (prefix.rfind(eos_token_) == prefix.size() - eos_token_.size()) {
prefix = prefix.substr(0, prefix.size() - eos_token_.size());
}
}
if (full.find(prefix) != 0) {
fprintf(stderr, "Failed to infer a tool call example (possible template bug)\n");
}
tool_call_example_ = full.substr(prefix.size());
}
} catch (const std::exception & e) {
fprintf(stderr, "Failed to generate tool call example: %s\n", e.what());
}
}
const std::string & source() const { return source_; }
@ -183,28 +270,72 @@ class chat_template {
const std::string & eos_token() const { return eos_token_; }
const chat_template_caps & original_caps() const { return caps_; }
// Deprecated, please use the form with chat_template_inputs and chat_template_options
std::string apply(
const nlohmann::ordered_json & messages,
const nlohmann::ordered_json & tools,
bool add_generation_prompt,
const nlohmann::ordered_json & extra_context = nlohmann::ordered_json(),
bool adjust_inputs = true) const
bool apply_polyfills = true)
{
fprintf(stderr, "[%s] Deprecated!\n", __func__);
chat_template_inputs inputs;
inputs.messages = messages;
inputs.tools = tools;
inputs.add_generation_prompt = add_generation_prompt;
inputs.extra_context = extra_context;
inputs.now = std::chrono::system_clock::now();
chat_template_options opts;
opts.apply_polyfills = apply_polyfills;
return apply(inputs, opts);
}
std::string apply(
const chat_template_inputs & inputs,
const chat_template_options & opts = chat_template_options()) const
{
json actual_messages;
auto needs_adjustments = adjust_inputs && (false
|| !caps_.supports_system_role
|| !caps_.supports_tools
|| !caps_.supports_tool_responses
|| !caps_.supports_tool_calls
|| caps_.requires_object_arguments
|| caps_.requires_typed_content
auto has_tools = inputs.tools.is_array() && !inputs.tools.empty();
auto has_tool_calls = false;
auto has_tool_responses = false;
auto has_string_content = false;
for (const auto & message : inputs.messages) {
if (message.contains("tool_calls") && !message["tool_calls"].is_null()) {
has_tool_calls = true;
}
if (message.contains("role") && message["role"] == "tool") {
has_tool_responses = true;
}
if (message.contains("content") && message["content"].is_string()) {
has_string_content = true;
}
}
auto polyfill_system_role = opts.polyfill_system_role && !caps_.supports_system_role;
auto polyfill_tools = opts.polyfill_tools && has_tools && !caps_.supports_tools;
auto polyfill_tool_call_example = polyfill_tools && opts.polyfill_tool_call_examples;
auto polyfill_tool_calls = opts.polyfill_tool_calls && has_tool_calls && !caps_.supports_tool_calls;
auto polyfill_tool_responses = opts.polyfill_tool_responses && has_tool_responses && !caps_.supports_tool_responses;
auto polyfill_object_arguments = opts.polyfill_object_arguments && has_tool_calls && caps_.requires_object_arguments;
auto polyfill_typed_content = opts.polyfill_typed_content && has_string_content && caps_.requires_typed_content;
auto needs_polyfills = opts.apply_polyfills && (false
|| polyfill_system_role
|| polyfill_tools
|| polyfill_tool_calls
|| polyfill_tool_responses
|| polyfill_object_arguments
|| polyfill_typed_content
);
if (needs_adjustments) {
if (needs_polyfills) {
actual_messages = json::array();
auto add_message = [&](const json & msg) {
if (caps_.requires_typed_content && msg.contains("content") && !msg.at("content").is_null() && msg.at("content").is_string()) {
if (polyfill_typed_content && msg.contains("content") && !msg.at("content").is_null() && msg.at("content").is_string()) {
actual_messages.push_back({
{"role", msg.at("role")},
{"content", {{
@ -227,9 +358,17 @@ class chat_template {
pending_system.clear();
}
};
auto needs_tools_in_system = !tools.is_null() && tools.size() > 0 && !caps_.supports_tools;
for (const auto & message_ : needs_tools_in_system ? add_system(messages, "Available tools: " + tools.dump(2)) : messages) {
json adjusted_messages;
if (polyfill_tools) {
adjusted_messages = add_system(inputs.messages,
"You can call any of the following tools to satisfy the user's requests: " + minja::Value(inputs.tools).dump(2, /* to_json= */ true) +
(!polyfill_tool_call_example || tool_call_example_.empty() ? "" : "\n\nExample tool call syntax:\n\n" + tool_call_example_));
} else {
adjusted_messages = inputs.messages;
}
for (const auto & message_ : adjusted_messages) {
auto message = message_;
if (!message.contains("role") || !message.contains("content")) {
throw std::runtime_error("message must have 'role' and 'content' fields: " + message.dump());
@ -237,7 +376,7 @@ class chat_template {
std::string role = message.at("role");
if (message.contains("tool_calls")) {
if (caps_.requires_object_arguments || !caps_.supports_tool_calls) {
if (polyfill_object_arguments || polyfill_tool_calls) {
for (auto & tool_call : message.at("tool_calls")) {
if (tool_call["type"] == "function") {
auto & function = tool_call.at("function");
@ -252,7 +391,7 @@ class chat_template {
}
}
}
if (!caps_.supports_tool_calls) {
if (polyfill_tool_calls) {
auto content = message.at("content");
auto tool_calls = json::array();
for (const auto & tool_call : message.at("tool_calls")) {
@ -279,14 +418,16 @@ class chat_template {
message.erase("tool_calls");
}
}
if (!caps_.supports_tool_responses && role == "tool") {
if (polyfill_tool_responses && role == "tool") {
message["role"] = "user";
auto obj = json {
{"tool_response", {
{"tool", message.at("name")},
{"content", message.at("content")},
}},
};
if (message.contains("name")) {
obj["tool_response"]["name"] = message.at("name");
}
if (message.contains("tool_call_id")) {
obj["tool_response"]["tool_call_id"] = message.at("tool_call_id");
}
@ -294,7 +435,7 @@ class chat_template {
message.erase("name");
}
if (!message["content"].is_null() && !caps_.supports_system_role) {
if (!message["content"].is_null() && polyfill_system_role) {
std::string content = message.at("content");
if (role == "system") {
if (!pending_system.empty()) pending_system += "\n";
@ -313,28 +454,36 @@ class chat_template {
}
add_message(message);
}
if (!caps_.supports_system_role) {
flush_sys();
}
flush_sys();
} else {
actual_messages = messages;
actual_messages = inputs.messages;
}
auto context = minja::Context::make(json({
{"messages", actual_messages},
{"add_generation_prompt", add_generation_prompt},
{"bos_token", bos_token_},
{"eos_token", eos_token_},
{"add_generation_prompt", inputs.add_generation_prompt},
}));
context->set("bos_token", opts.use_bos_token ? bos_token_ : "");
context->set("eos_token", opts.use_eos_token ? eos_token_ : "");
if (opts.define_strftime_now) {
auto now = inputs.now;
context->set("strftime_now", Value::callable([now](const std::shared_ptr<minja::Context> &, minja::ArgumentsValue & args) {
args.expectArgs("strftime_now", {1, 1}, {0, 0});
auto format = args.args[0].get<std::string>();
if (!tools.is_null()) {
auto tools_val = minja::Value(tools);
context->set("tools", tools_val);
auto time = std::chrono::system_clock::to_time_t(now);
auto local_time = *std::localtime(&time);
std::ostringstream ss;
ss << std::put_time(&local_time, format.c_str());
return ss.str();
}));
}
if (!extra_context.is_null()) {
for (auto & kv : extra_context.items()) {
minja::Value val(kv.value());
context->set(kv.key(), val);
if (!inputs.tools.is_null()) {
context->set("tools", minja::Value(inputs.tools));
}
if (!inputs.extra_context.is_null()) {
for (auto & kv : inputs.extra_context.items()) {
context->set(kv.key(), minja::Value(kv.value()));
}
}
@ -351,7 +500,7 @@ class chat_template {
std::string existing_system = messages_with_system.at(0).at("content");
messages_with_system[0] = json {
{"role", "system"},
{"content", existing_system + "\n" + system_prompt},
{"content", existing_system + "\n\n" + system_prompt},
};
} else {
messages_with_system.insert(messages_with_system.begin(), json {

150
common/chat.cpp
View file

@ -16,6 +16,7 @@ std::string common_chat_format_name(common_chat_format format) {
case COMMON_CHAT_FORMAT_FUNCTIONARY_V3_2: return "Functionary v3.2";
case COMMON_CHAT_FORMAT_FUNCTIONARY_V3_1_LLAMA_3_1: return "Functionary v3.1 Llama 3.1";
case COMMON_CHAT_FORMAT_HERMES_2_PRO: return "Hermes 2 Pro";
case COMMON_CHAT_FORMAT_COMMAND_R7B: return "Command R7B";
default:
throw std::runtime_error("Unknown chat format");
}
@ -162,6 +163,28 @@ static void foreach_function(const json & tools, const std::function<void(const
}
}
static std::string apply(
const common_chat_template & tmpl,
const nlohmann::ordered_json & messages,
const nlohmann::ordered_json & tools,
bool add_generation_prompt,
const nlohmann::ordered_json & extra_context = nlohmann::ordered_json())
{
minja::chat_template_inputs tmpl_inputs;
tmpl_inputs.messages = messages;
tmpl_inputs.tools = tools;
tmpl_inputs.add_generation_prompt = add_generation_prompt;
tmpl_inputs.extra_context = extra_context;
// TODO: add flag to control date/time, if only for testing purposes.
// tmpl_inputs.now = std::chrono::system_clock::now();
minja::chat_template_options tmpl_opts;
tmpl_opts.use_bos_token = false;
tmpl_opts.use_eos_token = false;
return tmpl.apply(tmpl_inputs, tmpl_opts);
}
static common_chat_params common_chat_params_init_generic(const common_chat_template & tmpl, const struct common_chat_inputs & inputs) {
common_chat_params data;
@ -243,7 +266,7 @@ static common_chat_params common_chat_params_init_generic(const common_chat_temp
inputs.messages,
"Respond in JSON format, either with `tool_call` (a request to call tools) or with `response` reply to the user's request");
data.prompt = tmpl.apply(tweaked_messages, inputs.tools.empty() ? json() : inputs.tools, inputs.add_generation_prompt);
data.prompt = apply(tmpl, tweaked_messages, inputs.tools.empty() ? json() : inputs.tools, inputs.add_generation_prompt);
data.format = COMMON_CHAT_FORMAT_GENERIC;
return data;
}
@ -309,7 +332,7 @@ static common_chat_params common_chat_params_init_mistral_nemo(const common_chat
builder.add_rule("root", "\"[TOOL_CALLS]\" " + builder.add_schema("tool_calls", schema));
}, grammar_options);
data.grammar_triggers.push_back({"[TOOL_CALLS]", /* .at_start = */ true});
data.prompt = tmpl.apply(inputs.messages, inputs.tools.empty() ? json() : inputs.tools, inputs.add_generation_prompt);
data.prompt = apply(tmpl, inputs.messages, inputs.tools.empty() ? json() : inputs.tools, inputs.add_generation_prompt);
data.format = COMMON_CHAT_FORMAT_MISTRAL_NEMO;
return data;
}
@ -317,6 +340,79 @@ static common_chat_msg common_chat_parse_mistral_nemo(const std::string & input)
return parse_prefixed_json_tool_call_array(input, "[TOOL_CALLS]");
}
static common_chat_params common_chat_params_init_command_r7b(const common_chat_template & tmpl, const struct common_chat_inputs & inputs) {
common_chat_params data;
data.grammar_lazy = inputs.tool_choice != "required";
data.grammar = build_grammar([&](const common_grammar_builder & builder) {
auto schemas = json::array();
foreach_function(inputs.tools, [&](const json & tool) {
const auto & function = tool["function"];
schemas.push_back({
{"type", "object"},
{"properties", {
{"tool_call_id", {
{"type", "string"},
// Command-R's template expects an integer string.
{"pattern", "^[0-9]{1,10}$"},
}},
{"tool_name", {
{"type", "string"},
{"const", function["name"]},
}},
{"parameters", function["parameters"]},
}},
{"required", json::array({"tool_call_id", "tool_name", "parameters"})},
});
});
auto schema = json {
{"type", "array"},
{"items", schemas.size() == 1 ? schemas[0] : json {{"anyOf", schemas}}},
{"minItems", 1},
};
if (!inputs.parallel_tool_calls) {
schema["maxItems"] = 1;
}
builder.add_rule("root", "\"<|START_ACTION|>\" " + builder.add_schema("tool_calls", schema) + " \"<|END_ACTION|>\"");
}, grammar_options);
data.grammar_triggers.push_back({"<|START_ACTION|>", /* .at_start = */ false});
data.preserved_tokens = {
"<|START_RESPONSE|>",
"<|END_RESPONSE|>",
"<|START_THINKING|>",
"<|END_THINKING|>",
"<|END_ACTION|>",
};
data.prompt = apply(tmpl, inputs.messages, inputs.tools.empty() ? json() : inputs.tools, inputs.add_generation_prompt);
data.format = COMMON_CHAT_FORMAT_COMMAND_R7B;
return data;
}
static common_chat_msg common_chat_parse_command_r7b(const std::string & input) {
static std::regex response_regex("<\\|START_RESPONSE\\|>([\\s\\S\\n\\r]*?)<\\|END_RESPONSE\\|>");
static std::regex thought_action_regex("<\\|START_THINKING\\|>([\\s\\S\\n\\r]*?)<\\|END_THINKING\\|><\\|START_ACTION\\|>([\\s\\S\\n\\r]*?)<\\|END_ACTION\\|>");
std::smatch match;
common_chat_msg result;
result.role = "assistant";
if (std::regex_match(input, match, response_regex)) {
result.content = match[1].str();
} else if (std::regex_match(input, match, thought_action_regex)) {
result.tool_plan = match[1].str();
auto actions_str = match[2].str();
auto actions = json::parse(actions_str);
for (const auto & action : actions) {
result.tool_calls.push_back({
/* .name = */ action["tool_name"],
/* .arguments = */ action["parameters"].dump(),
/* .id = */ action["tool_call_id"],
});
}
} else {
LOG_ERR("Failed to parse command_r output");
result.content = input;
}
return result;
}
static void expect_tool_parameters(const std::string & name, const json & parameters, const std::vector<std::string> & expected_properties) {
if (!parameters.is_object() || !parameters.contains("type") || parameters["type"] != "object" || !parameters.contains("properties") || !parameters.contains("required")) {
throw std::runtime_error("Parameters of tool " + name + " must be an object w/ required properties");
@ -384,21 +480,26 @@ static common_chat_params common_chat_params_init_llama_3_1_tool_calls(const com
tool_rules.push_back(
builder.add_rule(
name + "-call",
"\"{\" ( \"\\\"type\\\": \\\"function\\\", \" | space ) "
"\"{\" space "
"( \"\\\"type\\\":\" space \"\\\"function\\\",\" space )? "
"\"\\\"name\\\": \\\"" + name + "\\\", \\\"parameters\\\": \" " +
builder.add_schema(name + "-args", parameters) +
" \"}\""));
data.grammar_triggers.push_back({"{\"name\": \"" + name + "\"", /* .at_start = */ true});
});
data.grammar_triggers.push_back({"{\"name\":", /* .at_start = */ true});
data.grammar_triggers.push_back({"{\n \"name\":", /* .at_start = */ true});
data.grammar_triggers.push_back({"{\n \"name\":", /* .at_start = */ true});
data.grammar_triggers.push_back({"{\"type\": \"function\"", /* .at_start = */ true});
data.grammar_triggers.push_back({"{\n \"type\": \"function\"", /* .at_start = */ true});
data.grammar_triggers.push_back({"{\n \"type\": \"function\"", /* .at_start = */ true});
if (!builtin_tools.empty()) {
data.grammar_triggers.push_back({"<|python_tag|>", /* .at_start = */ false});
}
builder.add_rule("root", string_join(tool_rules, " | "));
}, grammar_options);
data.additional_stops.push_back("<|eom_id|>");
data.prompt = tmpl.apply(inputs.messages, inputs.tools.empty() ? json() : inputs.tools, inputs.add_generation_prompt, {
data.prompt = apply(tmpl, inputs.messages, inputs.tools.empty() ? json() : inputs.tools, inputs.add_generation_prompt, {
{"tools_in_user_message", false},
{"builtin_tools", builtin_tools.empty() ? json() : builtin_tools},
});
@ -457,9 +558,14 @@ static common_chat_params common_chat_params_init_deepseek_r1(const common_chat_
"\"<｜tool▁call▁begin｜>function<｜tool▁sep｜>" + name + "\\n```json\\n\" " + args_rule + " \"```<｜tool▁call▁end｜>\""));
});
data.grammar_triggers.push_back({"<｜tool▁calls▁begin｜>", /* .at_start = */ false});
data.preserved_tokens = {
"<｜tool▁sep｜>",
"<｜tool▁call▁end｜>",
};
builder.add_rule("root", "\"<｜tool▁calls▁begin｜>\" (" + string_join(tool_rules, " | ") + ")" + (inputs.parallel_tool_calls ? "*" : "") + " space");
}, grammar_options);
data.prompt = tmpl.apply(inputs.messages, inputs.tools.empty() ? json() : inputs.tools, inputs.add_generation_prompt);
auto prompt = apply(tmpl, inputs.messages, inputs.tools.empty() ? json() : inputs.tools, inputs.add_generation_prompt);
data.prompt = prompt;
data.format = COMMON_CHAT_FORMAT_DEEPSEEK_R1;
return data;
}
@ -473,10 +579,10 @@ static common_chat_msg common_chat_parse_deepseek_r1(const std::string & input)
static common_chat_params common_chat_params_init_firefunction_v2(const common_chat_template & tmpl, const struct common_chat_inputs & inputs) {
fprintf(stderr, "%s\n", __func__);
common_chat_params data;
data.prompt = tmpl.apply(inputs.messages, /* tools= */ nullptr, inputs.add_generation_prompt, {
data.prompt = apply(tmpl, inputs.messages, /* tools= */ nullptr, inputs.add_generation_prompt, {
{"datetime", "Jan 29 2025 13:00:00 GMT"},
{"functions", json(inputs.tools.empty() ? "" : inputs.tools.dump(2))},
}, /* adjust_inputs= */ false);
});
if (!inputs.tools.is_null() && !inputs.tools.empty()) {
data.grammar_lazy = inputs.tool_choice != "required";
data.grammar = build_grammar([&](const common_grammar_builder & builder) {
@ -520,7 +626,7 @@ static common_chat_params common_chat_params_init_functionary_v3_2(const common_
// >>>all\nlet's call functions>>>fn1\n{"arg1": 1...}\n>>>fn2\n{"arg1": 1...}...
// Using ">>>f1\n", ">>>f2\n"... as trigger words for the grammar
common_chat_params data;
data.prompt = tmpl.apply(inputs.messages, inputs.tools.empty() ? json() : inputs.tools, inputs.add_generation_prompt);
data.prompt = apply(tmpl, inputs.messages, inputs.tools.empty() ? json() : inputs.tools, inputs.add_generation_prompt);
data.format = COMMON_CHAT_FORMAT_FUNCTIONARY_V3_2;
if (!inputs.tools.is_null() && !inputs.tools.empty()) {
data.grammar_lazy = inputs.tool_choice != "required";
@ -586,9 +692,17 @@ static common_chat_msg common_chat_parse_functionary_v3_2(const std::string & in
}
}
// TODO: tighten & simplify.
auto res = parse_json_tool_calls(std::string(it, end), std::nullopt, function_regex, close_regex);
res.content = content;
return res;
try {
auto res = parse_json_tool_calls(std::string(it, end), std::nullopt, function_regex, close_regex);
res.content = content + res.content;
return res;
} catch (const std::exception & e) {
LOG_ERR("Failed to parse functionary v3.2 input: %s\n", e.what());
common_chat_msg res;
res.role = "assistant";
res.content = input;
return res;
}
}
static common_chat_params common_chat_params_init_functionary_v3_1_llama_3_1(const common_chat_template & tmpl, const struct common_chat_inputs & inputs) {
@ -639,7 +753,7 @@ static common_chat_params common_chat_params_init_functionary_v3_1_llama_3_1(con
data.grammar_triggers.push_back({"<function=", /* .at_start = */ false});
}, grammar_options);
data.prompt = tmpl.apply(inputs.messages, inputs.tools.empty() ? json() : inputs.tools, inputs.add_generation_prompt);
data.prompt = apply(tmpl, inputs.messages, inputs.tools.empty() ? json() : inputs.tools, inputs.add_generation_prompt);
// TODO: if (has_raw_python)
data.format = COMMON_CHAT_FORMAT_FUNCTIONARY_V3_1_LLAMA_3_1;
return data;
@ -691,11 +805,10 @@ static common_chat_params common_chat_params_init_hermes_2_pro(const common_chat
auto tool_call = "\"<tool_call>\" space " + builder.add_rule("tool_call", string_join(tool_rules, " | ")) + " \"</tool_call>\" space";
builder.add_rule("root", inputs.parallel_tool_calls ? "(" + tool_call + ")+" : tool_call);
data.grammar_triggers.push_back({"<tool_call>", /* .at_start = */ false});
// Not really a trigger but need to print this special token to get a successful parse.
data.grammar_triggers.push_back({"</tool_call>", /* .at_start = */ false});
data.preserved_tokens = { "</tool_call>" };
}, grammar_options);
data.prompt = tmpl.apply(inputs.messages, inputs.tools.empty() ? json() : inputs.tools, inputs.add_generation_prompt);
data.prompt = apply(tmpl, inputs.messages, inputs.tools.empty() ? json() : inputs.tools, inputs.add_generation_prompt);
data.format = COMMON_CHAT_FORMAT_HERMES_2_PRO;
return data;
}
@ -756,7 +869,7 @@ static common_chat_msg common_chat_parse_hermes_2_pro(const std::string & input)
static common_chat_params common_chat_params_init_without_tools(const common_chat_template & tmpl, const struct common_chat_inputs & inputs) {
common_chat_params data;
data.prompt = tmpl.apply(inputs.messages, inputs.tools.empty() ? json() : inputs.tools, inputs.add_generation_prompt);
data.prompt = apply(tmpl, inputs.messages, inputs.tools.empty() ? json() : inputs.tools, inputs.add_generation_prompt);
data.format = COMMON_CHAT_FORMAT_CONTENT_ONLY;
data.grammar_lazy = false;
if (!inputs.json_schema.is_null()) {
@ -809,6 +922,9 @@ common_chat_params common_chat_params_init(const common_chat_template & tmpl, co
if (src.find("[TOOL_CALLS]") != std::string::npos) {
return common_chat_params_init_mistral_nemo(tmpl, inputs);
}
if (src.find("<|END_THINKING|><|START_ACTION|>") != std::string::npos) {
return common_chat_params_init_command_r7b(tmpl, inputs);
}
return common_chat_params_init_generic(tmpl, inputs);
}
@ -842,6 +958,8 @@ common_chat_msg common_chat_parse(const std::string & input, common_chat_format
return common_chat_parse_hermes_2_pro(input);
case COMMON_CHAT_FORMAT_FIREFUNCTION_V2:
return common_chat_parse_firefunction_v2(input);
case COMMON_CHAT_FORMAT_COMMAND_R7B:
return common_chat_parse_command_r7b(input);
default:
throw std::runtime_error("Unsupported format: " + common_chat_format_name(format));
}

2
common/chat.hpp
View file

@ -32,6 +32,7 @@ enum common_chat_format {
COMMON_CHAT_FORMAT_FUNCTIONARY_V3_2,
COMMON_CHAT_FORMAT_FUNCTIONARY_V3_1_LLAMA_3_1,
COMMON_CHAT_FORMAT_HERMES_2_PRO,
COMMON_CHAT_FORMAT_COMMAND_R7B,
COMMON_CHAT_FORMAT_COUNT, // Not a format, just the # formats
};
@ -42,6 +43,7 @@ struct common_chat_params {
std::string grammar;
bool grammar_lazy = false;
std::vector<common_grammar_trigger> grammar_triggers;
std::vector<std::string> preserved_tokens;
std::vector<std::string> additional_stops;
};

49
common/common.cpp
View file

@ -1869,11 +1869,19 @@ std::string common_chat_format_example(const common_chat_template & tmpl, bool u
return common_chat_apply_template(tmpl, msgs, true, use_jinja);
}
#define CHATML_TEMPLATE_SRC \
"{%- for message in messages -%}\n" \
" {{- '<|im_start|>' + message.role + '\n' + message.content + '<|im_end|>\n' -}}\n" \
"{%- endfor -%}\n" \
"{%- if add_generation_prompt -%}\n" \
" {{- '<|im_start|>assistant\n' -}}\n" \
"{%- endif -%}"
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;
std::string default_template_src;
std::string template_tool_use_src;
bool has_explicit_template = !chat_template_override.empty();
if (chat_template_override.empty()) {
auto str = llama_model_chat_template(model, /* name */ nullptr);
@ -1886,21 +1894,17 @@ common_chat_templates common_chat_templates_from_model(const struct llama_model
template_tool_use_src = str;
has_explicit_template = true;
}
} else {
default_template_src = chat_template_override;
}
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 -%}
)";
default_template_src = CHATML_TEMPLATE_SRC;
}
}
auto vocab = llama_model_get_vocab(model);
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
@ -1914,13 +1918,22 @@ common_chat_templates common_chat_templates_from_model(const struct llama_model
};
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)
};
try {
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),
};
} catch (const std::exception & e) {
LOG_ERR("%s: failed to parse chat template: %s\n", __func__, e.what());
return {
has_explicit_template,
std::make_unique<minja::chat_template>(CHATML_TEMPLATE_SRC, token_bos, token_eos),
nullptr,
};
}
}
//

3
common/common.h
View file

@ -4,6 +4,7 @@
#include "llama-cpp.h"
#include <set>
#include <string>
#include <vector>
#include <sstream>
@ -163,6 +164,7 @@ struct common_params_sampling {
bool grammar_lazy = false;
std::vector<common_grammar_trigger> grammar_trigger_words; // optional trigger words to trigger lazy grammar
std::vector<llama_token> grammar_trigger_tokens; // optional trigger tokens to trigger lazy grammar and print trigger special tokens.
std::set<llama_token> preserved_tokens;
std::vector<llama_logit_bias> logit_bias; // logit biases to apply
@ -621,6 +623,7 @@ struct common_chat_msg {
std::string role;
std::string content;
std::vector<common_tool_call> tool_calls;
std::string tool_plan = "";
};
// Check if the template supplied via "--chat-template" is supported or not. Returns true if it's valid

9
common/json-schema-to-grammar.cpp
View file

@ -991,7 +991,14 @@ public:
}
};
std::string json_schema_to_grammar(const json & schema) {
std::string json_schema_to_grammar(const json & schema, bool force_gbnf) {
#ifdef LLAMA_USE_LLGUIDANCE
if (!force_gbnf) {
return "%llguidance {}\nstart: %json " + schema.dump();
}
#else
(void)force_gbnf;
#endif // LLAMA_USE_LLGUIDANCE
return build_grammar([&](const common_grammar_builder & callbacks) {
auto copy = schema;
callbacks.resolve_refs(copy);

3
common/json-schema-to-grammar.h
View file

@ -5,7 +5,8 @@
#define JSON_ASSERT GGML_ASSERT
#include "json.hpp"
std::string json_schema_to_grammar(const nlohmann::ordered_json & schema);
std::string json_schema_to_grammar(const nlohmann::ordered_json & schema,
bool force_gbnf = false);
struct common_grammar_builder {
std::function<std::string(const std::string &, const std::string &)> add_rule;

270
common/llguidance.cpp Normal file
View file

@ -0,0 +1,270 @@
#include "sampling.h"
#include "log.h"
#ifdef LLAMA_USE_LLGUIDANCE
# include "llguidance.h"
# include <cmath>
struct llama_sampler_llg {
const llama_vocab * vocab;
std::string grammar_kind;
std::string grammar_data;
LlgTokenizer * tokenizer;
LlgConstraint * grammar;
LlgMaskResult llg_res;
bool has_llg_res;
};
static LlgConstraint * llama_sampler_llg_new(LlgTokenizer * tokenizer, const char * grammar_kind,
const char * grammar_data) {
LlgConstraintInit cinit;
llg_constraint_init_set_defaults(&cinit, tokenizer);
const char * log_level = getenv("LLGUIDANCE_LOG_LEVEL");
if (log_level && *log_level) {
cinit.log_stderr_level = atoi(log_level);
}
auto c = llg_new_constraint_any(&cinit, grammar_kind, grammar_data);
if (llg_get_error(c)) {
LOG_ERR("llg error: %s\n", llg_get_error(c));
llg_free_constraint(c);
return nullptr;
}
return c;
}
static const char * llama_sampler_llg_name(const llama_sampler * /*smpl*/) {
return "llguidance";
}
static void llama_sampler_llg_accept_impl(llama_sampler * smpl, llama_token token) {
auto * ctx = (llama_sampler_llg *) smpl->ctx;
if (ctx->grammar) {
LlgCommitResult res;
llg_commit_token(ctx->grammar, token, &res);
ctx->has_llg_res = false;
}
}
static void llama_sampler_llg_apply(llama_sampler * smpl, llama_token_data_array * cur_p) {
auto * ctx = (llama_sampler_llg *) smpl->ctx;
if (ctx->grammar) {
if (!ctx->has_llg_res) {
if (llg_compute_mask(ctx->grammar, &ctx->llg_res) == 0) {
ctx->has_llg_res = true;
} else {
LOG_ERR("llg error: %s\n", llg_get_error(ctx->grammar));
llg_free_constraint(ctx->grammar);
ctx->grammar = nullptr;
}
}
if (ctx->has_llg_res) {
if (ctx->llg_res.is_stop) {
for (size_t i = 0; i < cur_p->size; ++i) {
if (!llama_vocab_is_eog(ctx->vocab, cur_p->data[i].id)) {
cur_p->data[i].logit = -INFINITY;
}
}
} else {
const uint32_t * mask = ctx->llg_res.sample_mask;
for (size_t i = 0; i < cur_p->size; ++i) {
auto token = cur_p->data[i].id;
if ((mask[token / 32] & (1 << (token % 32))) == 0) {
cur_p->data[i].logit = -INFINITY;
}
}
}
}
}
}
static void llama_sampler_llg_reset(llama_sampler * smpl) {
auto * ctx = (llama_sampler_llg *) smpl->ctx;
if (!ctx->grammar) {
return;
}
auto * grammar_new = llama_sampler_llg_new(ctx->tokenizer, ctx->grammar_kind.c_str(), ctx->grammar_data.c_str());
llg_free_constraint(ctx->grammar);
ctx->grammar = grammar_new;
ctx->has_llg_res = false;
}
static llama_sampler * llama_sampler_llg_clone(const llama_sampler * smpl) {
const auto * ctx = (const llama_sampler_llg *) smpl->ctx;
auto * result = llama_sampler_init_llg(ctx->vocab, nullptr, nullptr);
// copy the state
{
auto * result_ctx = (llama_sampler_llg *) result->ctx;
if (ctx->grammar) {
result_ctx->grammar_kind = ctx->grammar_kind;
result_ctx->grammar_data = ctx->grammar_data;
result_ctx->grammar = llg_clone_constraint(ctx->grammar);
result_ctx->tokenizer = llg_clone_tokenizer(ctx->tokenizer);
}
}
return result;
}
static void llama_sampler_llg_free(llama_sampler * smpl) {
const auto * ctx = (llama_sampler_llg *) smpl->ctx;
if (ctx->grammar) {
llg_free_constraint(ctx->grammar);
llg_free_tokenizer(ctx->tokenizer);
}
delete ctx;
}
static llama_sampler_i llama_sampler_llg_i = {
/* .name = */ llama_sampler_llg_name,
/* .accept = */ llama_sampler_llg_accept_impl,
/* .apply = */ llama_sampler_llg_apply,
/* .reset = */ llama_sampler_llg_reset,
/* .clone = */ llama_sampler_llg_clone,
/* .free = */ llama_sampler_llg_free,
};
static size_t llama_sampler_llg_tokenize_fn(const void * user_data, const uint8_t * bytes, size_t bytes_len,
uint32_t * output_tokens, size_t output_tokens_len) {
const llama_vocab * vocab = (const llama_vocab *) user_data;
int r = 0;
try {
r = llama_tokenize(vocab, (const char *) bytes, bytes_len, (int32_t *) output_tokens, output_tokens_len, false,
true);
} catch (const std::exception & e) {
GGML_ABORT("llama_tokenize failed: %s\n", e.what());
}
if (r < 0) {
return -r;
}
return r;
}
static LlgTokenizer * llama_sampler_llg_new_tokenizer(const llama_vocab * vocab) {
// TODO store the tokenizer in the vocab somehow
static const llama_vocab * vocab_cache;
static LlgTokenizer * tokenizer_cache;
if (vocab_cache == vocab) {
return llg_clone_tokenizer(tokenizer_cache);
}
auto tok_eos = llama_vocab_eot(vocab);
if (tok_eos == LLAMA_TOKEN_NULL) {
tok_eos = llama_vocab_eos(vocab);
}
size_t vocab_size = llama_vocab_n_tokens(vocab);
auto token_lens = new uint32_t[vocab_size];
// we typically have ~7 bytes per token; let's go on the safe side here
auto token_bytes_size = vocab_size * 16 + 1024 * 1024;
auto token_bytes = new uint8_t[token_bytes_size];
size_t offset = 0;
for (size_t i = 0; i < vocab_size; i++) {
size_t max_token = 1024;
if (token_bytes_size - offset < max_token) {
GGML_ABORT("token_bytes buffer too small\n");
}
llama_token token = i;
auto dp = (char *) token_bytes + offset;
auto size = llama_detokenize(vocab, &token, 1, dp, max_token, false, false);
if (size < 0) {
GGML_ABORT("llama_detokenize failed\n");
}
if (size == 0) {
size = llama_detokenize(vocab, &token, 1, dp + 1, max_token - 1, false, true);
if (size < 0) {
GGML_ABORT("llama_detokenize failed\n");
}
if (size != 0) {
*dp = '\xff'; // special token prefix marker
size += 1;
}
}
token_lens[i] = size;
offset += size;
}
LlgTokenizerInit tinit = {
/* .vocab_size = */ (uint32_t) vocab_size,
/* .tok_eos = */ (uint32_t) tok_eos,
/* .token_lens = */ token_lens,
/* .token_bytes = */ token_bytes,
/* .tokenizer_json = */ nullptr,
/* .tokenize_assumes_string = */ true,
/* .tokenize_fn = */ llama_sampler_llg_tokenize_fn,
/* .use_approximate_greedy_tokenize_fn = */ false,
/* .tokenize_user_data = */ vocab,
};
char error_buffer[1024];
LlgTokenizer * tokenizer = llg_new_tokenizer(&tinit, error_buffer, sizeof(error_buffer));
delete[] token_bytes;
delete[] token_lens;
if (tokenizer == nullptr) {
LOG_ERR("llg tokenizer error: %s\n", error_buffer);
return tokenizer;
}
if (tokenizer_cache) {
llg_free_tokenizer(tokenizer_cache);
}
vocab_cache = vocab;
tokenizer_cache = tokenizer;
return llg_clone_tokenizer(tokenizer_cache);
}
llama_sampler * llama_sampler_init_llg(const llama_vocab * vocab, const char * grammar_kind,
const char * grammar_data) {
auto * ctx = new llama_sampler_llg;
if (grammar_kind != nullptr && grammar_kind[0] != '\0') {
auto tokenizer = llama_sampler_llg_new_tokenizer(vocab);
*ctx = {
/* .vocab = */ vocab,
/* .grammar_kind = */ grammar_kind,
/* .grammar_data = */ grammar_data,
/* .tokenizer = */ tokenizer,
/* .grammar = */ llama_sampler_llg_new(tokenizer, grammar_kind, grammar_data),
/* .llg_res = */ {},
/* .has_llg_res = */ false,
};
} else {
*ctx = {
/* .vocab = */ vocab,
/* .grammar_kind = */ {},
/* .grammar_data = */ {},
/* .tokenizer = */ nullptr,
/* .grammar = */ nullptr,
/* .llg_res = */ {},
/* .has_llg_res = */ false,
};
}
return llama_sampler_init(
/* .iface = */ &llama_sampler_llg_i,
/* .ctx = */ ctx
);
}
#else
llama_sampler * llama_sampler_init_llg(const llama_vocab *, const char *, const char *) {
LOG_WRN("llguidance (cmake -DLLAMA_LLGUIDANCE=ON) is not enabled");
return nullptr;
}
#endif // LLAMA_USE_LLGUIDANCE

11
common/log.cpp
View file

@ -14,16 +14,6 @@ void common_log_set_verbosity_thold(int verbosity) {
common_log_verbosity_thold = verbosity;
}
#define LOG_COL_DEFAULT "\033[0m"
#define LOG_COL_BOLD "\033[1m"
#define LOG_COL_RED "\033[31m"
#define LOG_COL_GREEN "\033[32m"
#define LOG_COL_YELLOW "\033[33m"
#define LOG_COL_BLUE "\033[34m"
#define LOG_COL_MAGENTA "\033[35m"
#define LOG_COL_CYAN "\033[36m"
#define LOG_COL_WHITE "\033[37m"
static int64_t t_us() {
return std::chrono::duration_cast<std::chrono::microseconds>(std::chrono::system_clock::now().time_since_epoch()).count();
}
@ -206,6 +196,7 @@ public:
vsnprintf(entry.msg.data(), entry.msg.size(), ss.str().c_str(), args_copy);
}
#endif
va_end(args_copy);
}
entry.level = level;

11
common/log.h
View file

@ -2,6 +2,17 @@
#include "ggml.h" // for ggml_log_level
#define LOG_CLR_TO_EOL "\033[K\r"
#define LOG_COL_DEFAULT "\033[0m"
#define LOG_COL_BOLD "\033[1m"
#define LOG_COL_RED "\033[31m"
#define LOG_COL_GREEN "\033[32m"
#define LOG_COL_YELLOW "\033[33m"
#define LOG_COL_BLUE "\033[34m"
#define LOG_COL_MAGENTA "\033[35m"
#define LOG_COL_CYAN "\033[36m"
#define LOG_COL_WHITE "\033[37m"
#ifndef __GNUC__
# define LOG_ATTRIBUTE_FORMAT(...)
#elif defined(__MINGW32__)

57
common/minja.hpp
View file

@ -693,7 +693,7 @@ enum SpaceHandling { Keep, Strip, StripSpaces, StripNewline };
class TemplateToken {
public:
enum class Type { Text, Expression, If, Else, Elif, EndIf, For, EndFor, Generation, EndGeneration, Set, EndSet, Comment, Macro, EndMacro, Filter, EndFilter };
enum class Type { Text, Expression, If, Else, Elif, EndIf, For, EndFor, Generation, EndGeneration, Set, EndSet, Comment, Macro, EndMacro, Filter, EndFilter, Break, Continue };
static std::string typeToString(Type t) {
switch (t) {
@ -714,6 +714,8 @@ public:
case Type::EndFilter: return "endfilter";
case Type::Generation: return "generation";
case Type::EndGeneration: return "endgeneration";
case Type::Break: return "break";
case Type::Continue: return "continue";
}
return "Unknown";
}
@ -815,6 +817,22 @@ struct CommentTemplateToken : public TemplateToken {
CommentTemplateToken(const Location & location, SpaceHandling pre, SpaceHandling post, const std::string& t) : TemplateToken(Type::Comment, location, pre, post), text(t) {}
};
enum class LoopControlType { Break, Continue };
class LoopControlException : public std::runtime_error {
public:
LoopControlType control_type;
LoopControlException(const std::string & message, LoopControlType control_type) : std::runtime_error(message), control_type(control_type) {}
LoopControlException(LoopControlType control_type)
: std::runtime_error((control_type == LoopControlType::Continue ? "continue" : "break") + std::string(" outside of a loop")),
control_type(control_type) {}
};
struct LoopControlTemplateToken : public TemplateToken {
LoopControlType control_type;
LoopControlTemplateToken(const Location & location, SpaceHandling pre, SpaceHandling post, LoopControlType control_type) : TemplateToken(Type::Break, location, pre, post), control_type(control_type) {}
};
class TemplateNode {
Location location_;
protected:
@ -825,6 +843,12 @@ public:
void render(std::ostringstream & out, const std::shared_ptr<Context> & context) const {
try {
do_render(out, context);
} catch (const LoopControlException & e) {
// TODO: make stack creation lazy. Only needed if it was thrown outside of a loop.
std::ostringstream err;
err << e.what();
if (location_.source) err << error_location_suffix(*location_.source, location_.pos);
throw LoopControlException(err.str(), e.control_type);
} catch (const std::exception & e) {
std::ostringstream err;
err << e.what();
@ -897,6 +921,15 @@ public:
}
};
class LoopControlNode : public TemplateNode {
LoopControlType control_type_;
public:
LoopControlNode(const Location & location, LoopControlType control_type) : TemplateNode(location), control_type_(control_type) {}
void do_render(std::ostringstream &, const std::shared_ptr<Context> &) const override {
throw LoopControlException(control_type_);
}
};
class ForNode : public TemplateNode {
std::vector<std::string> var_names;
std::shared_ptr<Expression> iterable;
@ -961,7 +994,12 @@ public:
loop.set("last", i == (n - 1));
loop.set("previtem", i > 0 ? filtered_items.at(i - 1) : Value());
loop.set("nextitem", i < n - 1 ? filtered_items.at(i + 1) : Value());
body->render(out, loop_context);
try {
body->render(out, loop_context);
} catch (const LoopControlException & e) {
if (e.control_type == LoopControlType::Break) break;
if (e.control_type == LoopControlType::Continue) continue;
}
}
}
};
@ -2156,10 +2194,10 @@ private:
}
TemplateTokenVector tokenize() {
static std::regex comment_tok(R"(\{#([-~]?)(.*?)([-~]?)#\})");
static std::regex comment_tok(R"(\{#([-~]?)([\s\S\r\n]*?)([-~]?)#\})");
static std::regex expr_open_regex(R"(\{\{([-~])?)");
static std::regex block_open_regex(R"(^\{%([-~])?[\s\n\r]*)");
static std::regex block_keyword_tok(R"((if|else|elif|endif|for|endfor|generation|endgeneration|set|endset|block|endblock|macro|endmacro|filter|endfilter)\b)");
static std::regex block_keyword_tok(R"((if|else|elif|endif|for|endfor|generation|endgeneration|set|endset|block|endblock|macro|endmacro|filter|endfilter|break|continue)\b)");
static std::regex non_text_open_regex(R"(\{\{|\{%|\{#)");
static std::regex expr_close_regex(R"([\s\n\r]*([-~])?\}\})");
static std::regex block_close_regex(R"([\s\n\r]*([-~])?%\})");
@ -2291,6 +2329,9 @@ private:
} else if (keyword == "endfilter") {
auto post_space = parseBlockClose();
tokens.push_back(std::make_unique<EndFilterTemplateToken>(location, pre_space, post_space));
} else if (keyword == "break" || keyword == "continue") {
auto post_space = parseBlockClose();
tokens.push_back(std::make_unique<LoopControlTemplateToken>(location, pre_space, post_space, keyword == "break" ? LoopControlType::Break : LoopControlType::Continue));
} else {
throw std::runtime_error("Unexpected block: " + keyword);
}
@ -2414,6 +2455,8 @@ private:
children.emplace_back(std::make_shared<FilterNode>(token->location, std::move(filter_token->filter), std::move(body)));
} else if (dynamic_cast<CommentTemplateToken*>(token.get())) {
// Ignore comments
} else if (auto ctrl_token = dynamic_cast<LoopControlTemplateToken*>(token.get())) {
children.emplace_back(std::make_shared<LoopControlNode>(token->location, ctrl_token->control_type));
} else if (dynamic_cast<EndForTemplateToken*>(token.get())
|| dynamic_cast<EndSetTemplateToken*>(token.get())
|| dynamic_cast<EndMacroTemplateToken*>(token.get())
@ -2572,6 +2615,7 @@ inline std::shared_ptr<Context> Context::builtins() {
}));
globals.set("join", simple_function("join", { "items", "d" }, [](const std::shared_ptr<Context> &, Value & args) {
auto do_join = [](Value & items, const std::string & sep) {
if (!items.is_array()) throw std::runtime_error("object is not iterable: " + items.dump());
std::ostringstream oss;
auto first = true;
for (size_t i = 0, n = items.size(); i < n; ++i) {
@ -2652,6 +2696,10 @@ inline std::shared_ptr<Context> Context::builtins() {
return Value::callable([=](const std::shared_ptr<Context> & context, ArgumentsValue & args) {
args.expectArgs(is_select ? "select" : "reject", {2, (std::numeric_limits<size_t>::max)()}, {0, 0});
auto & items = args.args[0];
if (items.is_null())
return Value::array();
if (!items.is_array()) throw std::runtime_error("object is not iterable: " + items.dump());
auto filter_fn = context->get(args.args[1]);
if (filter_fn.is_null()) throw std::runtime_error("Undefined filter: " + args.args[1].dump());
@ -2729,6 +2777,7 @@ inline std::shared_ptr<Context> Context::builtins() {
auto & items = args.args[0];
if (items.is_null())
return Value::array();
if (!items.is_array()) throw std::runtime_error("object is not iterable: " + items.dump());
auto attr_name = args.args[1].get<std::string>();
bool has_test = false;

22
common/sampling.cpp
View file

@ -156,13 +156,25 @@ struct common_sampler * common_sampler_init(const struct llama_model * model, co
for (const auto & str : params.grammar_trigger_words) {
trigger_words.push_back(str.word.c_str());
}
struct llama_sampler * grmr;
if (params.grammar.compare(0, 11, "%llguidance") == 0) {
#ifdef LLAMA_USE_LLGUIDANCE
grmr = llama_sampler_init_llg(vocab, "lark", params.grammar.c_str());
#else
GGML_ABORT("llguidance (cmake -DLLAMA_LLGUIDANCE=ON) is not enabled");
#endif // LLAMA_USE_LLGUIDANCE
} else {
grmr = params.grammar_lazy
? llama_sampler_init_grammar_lazy(vocab, params.grammar.c_str(), "root",
trigger_words.data(), trigger_words.size(),
params.grammar_trigger_tokens.data(), params.grammar_trigger_tokens.size())
: llama_sampler_init_grammar(vocab, params.grammar.c_str(), "root");
}
auto * result = new common_sampler {
/* .params = */ params,
/* .grmr = */ params.grammar_lazy
? llama_sampler_init_grammar_lazy(vocab, params.grammar.c_str(), "root",
trigger_words.data(), trigger_words.size(),
params.grammar_trigger_tokens.data(), params.grammar_trigger_tokens.size())
: llama_sampler_init_grammar(vocab, params.grammar.c_str(), "root"),
/* .grmr = */ grmr,
/* .chain = */ llama_sampler_chain_init(lparams),
/* .prev = */ ring_buffer<llama_token>(std::max(32, params.n_prev)),
/* .cur = */ {},

3
common/sampling.h
View file

@ -102,3 +102,6 @@ std::string common_sampler_type_to_str(enum common_sampler_type cnstr);
std::vector<enum common_sampler_type> common_sampler_types_from_names(const std::vector<std::string> & names, bool allow_alt_names);
std::vector<enum common_sampler_type> common_sampler_types_from_chars(const std::string & chars);
llama_sampler * llama_sampler_init_llg(const llama_vocab * vocab,
const char * grammar_kind, const char * grammar_data);