happyz
/
llama.cpp
mirror of https://github.com/ggerganov/llama.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Merge 620b2c05d1 into d3416a4aa9

This commit is contained in:
Piotr Wilkin (ilintar) 2026-04-03 14:34:15 +02:00 committed by GitHub
parent d3416a4aa9 620b2c05d1
commit 3d0999bf70
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
7 changed files with 242 additions and 153 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

120
common/chat-auto-parser-generator.cpp
View File

@ -6,6 +6,7 @@
#include "json-schema-to-grammar.h"
#include "log.h"
#include "nlohmann/json.hpp"
#include "peg-parser.h"
#include <algorithm>
#include <stdexcept>
@ -317,6 +318,44 @@ common_peg_parser analyze_tools::build_tool_parser_json_native(parser_build_cont
p.end();
}
common_peg_parser analyze_tools::build_func_parser(common_chat_peg_builder & p, const std::string & name,
const common_peg_parser & call_id_section, bool have_call_id,
const common_peg_parser & args,
std::optional<common_peg_parser> atomic_peek) const {
auto open = p.tool_open(function.name_prefix + p.tool_name(p.literal(name)) + function.name_suffix);
bool matched_atomic = false;
common_peg_parser func_parser = p.eps();
if (!function.name_suffix.empty()) {
func_parser = open + call_id_section + p.space() + args;
matched_atomic = true;
} else if (have_call_id) {
func_parser = p.atomic(open + call_id_section) + p.space() + args;
matched_atomic = true;
} else if (atomic_peek.has_value()) {
func_parser = p.atomic(open + call_id_section + p.space() + *atomic_peek) + args;
matched_atomic = true;
} else {
func_parser = open + call_id_section + p.space() + args;
}
if (!function.close.empty()) {
func_parser = func_parser + p.space() + p.tool_close(p.literal(function.close));
} else if (!format.per_call_end.empty()) {
// When there's no func_close but there is a per_call_end marker, use peek() to ensure
// we only emit tool_close when we can actually see the closing marker. This prevents
// premature closing during partial parsing when we've seen e.g. "</" which could be
// either "</tool_call>" (end) or "<arg_key>" prefix that failed to match.
func_parser = func_parser + p.tool_close(p.peek(p.literal(format.per_call_end)));
} else {
func_parser = func_parser + p.tool_close(p.space()); // force this to process tool closing callbacks in mapper
}
if (!matched_atomic) {
func_parser = p.atomic(func_parser);
}
return func_parser;
}
common_peg_parser analyze_tools::build_tool_parser_tag_json(parser_build_context & ctx) const {
auto & p = ctx.p;
const auto & inputs = ctx.inputs;
@ -330,17 +369,27 @@ common_peg_parser analyze_tools::build_tool_parser_tag_json(parser_build_context
const auto & schema = func.contains("parameters") ? func.at("parameters") : json::object();
// Build call_id parser based on position (if supported)
bool have_call_id = false;
common_peg_parser call_id_section = p.eps();
if (call_id.pos == call_id_position::BETWEEN_FUNC_AND_ARGS && !call_id.prefix.empty() &&
!call_id.suffix.empty()) {
call_id_section = p.optional(call_id.prefix + p.tool_id(p.until(call_id.suffix))) + call_id.suffix;
(!call_id.suffix.empty() || !arguments.start.empty())) {
if (!call_id.suffix.empty()) {
call_id_section = p.optional(call_id.prefix + p.tool_id(p.until(call_id.suffix))) + call_id.suffix;
} else {
call_id_section = p.optional(call_id.prefix + p.tool_id(p.until(arguments.start)));
}
have_call_id = true;
}
auto args_parser = p.tool_args(p.schema(p.json(), "tool-" + name + "-schema", schema));
if (!arguments.start.empty()) {
args_parser = p.literal(arguments.start) + args_parser;
}
if (!arguments.end.empty()) {
args_parser = args_parser + p.literal(arguments.end);
}
auto func_parser = p.tool_open(function.name_prefix + p.tool_name(p.literal(name)) + function.name_suffix) +
call_id_section + p.tool_args(p.schema(p.json(), "tool-" + name + "-schema", schema));
if (!function.close.empty()) {
func_parser = func_parser + function.close;
}
auto atomic_peek = !arguments.start.empty() ? std::optional(p.peek(p.literal(arguments.start))) : std::nullopt;
auto func_parser = build_func_parser(p, name, call_id_section, have_call_id, args_parser, atomic_peek);
tool_choice |= p.rule("tool-" + name, func_parser);
});
@ -448,52 +497,31 @@ common_peg_parser analyze_tools::build_tool_parser_tag_tagged(parser_build_conte
args_seq = args_seq + p.repeat(p.space() + any_opt, 0, (int) optional_parsers.size());
}
if (!arguments.start.empty()) {
args_seq = p.literal(arguments.start) + args_seq;
}
if (!arguments.end.empty()) {
args_seq = args_seq + p.literal(arguments.end);
}
// Build call_id parser based on position (if supported)
common_peg_parser call_id_section = p.eps();
bool have_call_id = false;
if (call_id.pos == call_id_position::BETWEEN_FUNC_AND_ARGS && !call_id.prefix.empty() &&
!call_id.suffix.empty()) {
(!call_id.suffix.empty() || !arguments.start.empty())) {
have_call_id = true;
call_id_section = p.optional(call_id.prefix + p.tool_id(p.until(call_id.suffix)) + call_id.suffix);
}
bool matched_atomic = false;
common_peg_parser func_parser = p.eps();
if (!function.name_suffix.empty()) {
func_parser = p.tool_open(function.name_prefix + p.tool_name(p.literal(name)) + function.name_suffix) +
call_id_section + p.space() + args_seq;
matched_atomic = true;
} else if (have_call_id) {
func_parser = p.atomic(p.tool_open(function.name_prefix + p.tool_name(p.literal(name)) + function.name_suffix) +
call_id_section) + p.space() + args_seq;
matched_atomic = true;
} else if (!arguments.name_prefix.empty() && !required_parsers.empty()) {
// Only peek for an arg tag when there are required args that must follow.
// When all args are optional, the model may emit no arg tags at all (#20650).
func_parser = p.atomic(p.tool_open(function.name_prefix + p.tool_name(p.literal(name)) + function.name_suffix) +
call_id_section + p.space() + p.peek(p.literal(arguments.name_prefix))) + args_seq;
matched_atomic = true;
} else {
func_parser = p.tool_open(function.name_prefix + p.tool_name(p.literal(name)) + function.name_suffix) +
call_id_section + p.space() + args_seq;
}
if (!function.close.empty()) {
func_parser = func_parser + p.space() + p.tool_close(p.literal(function.close));
} else if (!format.per_call_end.empty()) {
// When there's no func_close but there is a per_call_end marker, use peek() to ensure
// we only emit tool_close when we can actually see the closing marker. This prevents
// premature closing during partial parsing when we've seen e.g. "</" which could be
// either "</tool_call>" (end) or "<arg_key>" prefix that failed to match.
func_parser = func_parser + p.tool_close(p.peek(p.literal(format.per_call_end)));
} else {
func_parser =
func_parser + p.tool_close(p.space()); // force this to process tool closing callbacks in mapper
}
if (!matched_atomic) {
func_parser = p.atomic(func_parser);
if (!call_id.suffix.empty()) {
call_id_section = p.optional(call_id.prefix + p.tool_id(p.until(call_id.suffix)) + call_id.suffix);
} else {
call_id_section = p.optional(call_id.prefix + p.tool_id(p.until(arguments.start)));
}
}
// Only peek for an arg tag when there are required args that must follow.
// When all args are optional, the model may emit no arg tags at all (#20650).
auto atomic_peek = (!arguments.name_prefix.empty() && !required_parsers.empty()) ?
std::optional(p.peek(p.literal(arguments.name_prefix))) : std::nullopt;
auto func_parser = build_func_parser(p, name, call_id_section, have_call_id, args_seq, atomic_peek);
tool_choice |= p.rule("tool-" + name, func_parser);
});

7
common/chat-auto-parser.h
View File

@ -356,6 +356,13 @@ struct analyze_tools : analyze_base {
common_peg_parser build_tool_parser_json_native(parser_build_context & ctx) const;
common_peg_parser build_tool_parser_tag_json(parser_build_context & ctx) const;
common_peg_parser build_tool_parser_tag_tagged(parser_build_context & ctx) const;
// Shared helper: builds func_parser from open+call_id+args, handling atomic wrapping and close.
// atomic_peek: if present, used as the peek expression in the third atomicity branch.
common_peg_parser build_func_parser(common_chat_peg_builder & p, const std::string & name,
const common_peg_parser & call_id_section, bool have_call_id,
const common_peg_parser & args,
std::optional<common_peg_parser> atomic_peek) const;
common_peg_parser build_tool_parser_tag_gemma4_dict(parser_build_context & ctx) const;
};

48
common/chat-diff-analyzer.cpp
View File

@ -25,6 +25,9 @@ static const std::string ARG_SECOND = "BB_ARG_SND_BB";
static const std::string USER_MSG = "U_USER_MSG Hello END_U";
static const std::string ASSISTANT_MSG = "A_ASST_MSG I can help END_A";
static const std::string THINKING_CONTENT = "REASON_PART I am thinking END_R";
static const std::string CALL_ID_001 = "call00001";
static const std::string CALL_ID_002 = "call00002";
static const std::string CALL_ID_999 = "call99999";
static std::vector<std::function<void(const common_chat_template & tmpl, autoparser &)>> workarounds(
{ // Old reasoning Qwen templates - they don't really display reasoning content, but we still want to
@ -131,6 +134,7 @@ static std::vector<std::function<void(const common_chat_template & tmpl, autopar
analysis.tools.function.name_prefix = "<｜tool▁sep｜>";
analysis.tools.format.per_call_end = "<｜tool▁call▁end｜>";
analysis.tools.function.close = "```";
LOG_DBG(ANSI_ORANGE "[Patch: DeepSeek-R1-Distill-Qwen]\n" ANSI_RESET);
}
}
});
@ -158,7 +162,7 @@ static json user_msg = json{
{ "content", USER_MSG }
};
static json build_tool_call(const std::string & name, const json & args, const std::string & id = "call00001") {
static json build_tool_call(const std::string & name, const json & args, const std::string & id = CALL_ID_001) {
return json{
{ "id", id },
{ "type", "function" },
@ -166,17 +170,17 @@ static json build_tool_call(const std::string & name, const json & args, const s
};
}
static json first_tool_call_zero_args = build_tool_call(FUN_FIRST, json::object(), "call00001");
static json first_tool_call_one_arg = build_tool_call(FUN_FIRST, {{ ARG_FIRST, "XXXX" }}, "call00001");
static json first_tool_call_one_arg_other_val = build_tool_call(FUN_FIRST, {{ ARG_FIRST, "YYYY" }}, "call00001");
static json first_tool_call_other_arg = build_tool_call(FUN_FIRST, {{ ARG_SECOND, "YYYY" }}, "call00001");
static json first_tool_call_zero_args = build_tool_call(FUN_FIRST, json::object(), CALL_ID_001);
static json first_tool_call_one_arg = build_tool_call(FUN_FIRST, {{ ARG_FIRST, "XXXX" }}, CALL_ID_001);
static json first_tool_call_one_arg_other_val = build_tool_call(FUN_FIRST, {{ ARG_FIRST, "YYYY" }}, CALL_ID_001);
static json first_tool_call_other_arg = build_tool_call(FUN_FIRST, {{ ARG_SECOND, "YYYY" }}, CALL_ID_001);
static json first_tool_call =
build_tool_call(FUN_FIRST, json{{ ARG_FIRST, "XXXX" }, { ARG_SECOND, "YYYY" }}, "call00001");
build_tool_call(FUN_FIRST, json{{ ARG_FIRST, "XXXX" }, { ARG_SECOND, "YYYY" }}, CALL_ID_001);
static json second_tool_call =
build_tool_call(FUN_SECOND, json{ { ARG_FIRST, "XXXX" }, { ARG_SECOND, "YYYY" }}, "call00002");
build_tool_call(FUN_SECOND, json{ { ARG_FIRST, "XXXX" }, { ARG_SECOND, "YYYY" }}, CALL_ID_002);
static json first_tool_call_alt_id =
build_tool_call(FUN_FIRST, json{{ ARG_FIRST, "XXXX" }, { ARG_SECOND, "YYYY" }}, "call99999");
build_tool_call(FUN_FIRST, json{{ ARG_FIRST, "XXXX" }, { ARG_SECOND, "YYYY" }}, CALL_ID_999);
template <typename T>
static std::string mode_to_str(T mode) {
@ -215,6 +219,11 @@ void autoparser::analyze_template(const common_chat_template & tmpl) {
LOG_DBG("func_name_prefix: '%s'\n", tools.function.name_prefix.c_str());
LOG_DBG("func_name_suffix: '%s'\n", tools.function.name_suffix.c_str());
LOG_DBG("func_close: '%s'\n", tools.function.close.c_str());
LOG_DBG("call_id_prefix: '%s'\n", tools.call_id.prefix.c_str());
LOG_DBG("call_id_suffix: '%s'\n", tools.call_id.suffix.c_str());
LOG_DBG("call_id_pos: '%s'\n", mode_to_str(tools.call_id.pos).c_str());
LOG_DBG("args_start: '%s'\n", tools.arguments.start.c_str());
LOG_DBG("args_end: '%s'\n", tools.arguments.end.c_str());
LOG_DBG("arg_name_prefix: '%s'\n", tools.arguments.name_prefix.c_str());
LOG_DBG("arg_name_suffix: '%s'\n", tools.arguments.name_suffix.c_str());
LOG_DBG("arg_value_prefix: '%s'\n", tools.arguments.value_prefix.c_str());
@ -583,12 +592,15 @@ analyze_tools::analyze_tools(const common_chat_template & tmpl,
if (caps.supports_parallel_tool_calls) {
check_per_call_markers();
}
LOG_DBG(ANSI_ORANGE "Phase 3a: Function call analysis\n" ANSI_RESET);
extract_function_markers();
LOG_DBG(ANSI_ORANGE "Phase 3b: Argument analysis\n" ANSI_RESET);
if (format.mode == tool_format::TAG_WITH_TAGGED) {
analyze_arguments();
}
extract_argument_separator();
extract_args_markers();
LOG_DBG(ANSI_ORANGE "Phase 3c: Call id analysis\n" ANSI_RESET);
extract_call_id_markers();
}
}
@ -979,8 +991,6 @@ void analyze_tools::extract_function_markers() {
}
void analyze_tools::analyze_arguments() {
LOG_DBG(ANSI_ORANGE "Phase 4: Argument analysis\n" ANSI_RESET);
extract_argument_name_markers();
extract_argument_value_markers();
}
@ -1189,7 +1199,7 @@ void analyze_tools::extract_args_markers() {
const auto & diff = comparison->diff;
if (format.mode != tool_format::JSON_NATIVE) {
if (format.mode == tool_format::JSON_NATIVE) {
std::string prefix_marker = !format.section_start.empty() ? format.section_start : format.per_call_start;
std::string suffix_marker = !format.section_end.empty() ? format.section_end : format.per_call_end;
// these might happen earlier in the tools section as an example or somewhere else, so we need to find the closest ones
@ -1211,6 +1221,10 @@ void analyze_tools::extract_args_markers() {
if (find_fun != std::string::npos) {
args_start = args_start.substr(find_fun + FUN_FIRST.size(), args_start.size() - find_fun - FUN_FIRST.size());
}
size_t find_call_id = args_start.find(CALL_ID_001);
if (find_call_id != std::string::npos) {
args_start = args_start.substr(find_call_id + CALL_ID_001.size(), args_start.size() - find_call_id - CALL_ID_001.size());
}
arguments.start = args_start;
arguments.end = args_end;
}
@ -1250,8 +1264,8 @@ void analyze_tools::extract_call_id_markers() {
return;
}
std::string id_value_1 = "call00001";
std::string id_value_2 = "call99999";
std::string id_value_1 = CALL_ID_001;
std::string id_value_2 = CALL_ID_999;
size_t common_id_prefix_len = 0;
for (size_t i = 0; i < std::min(id_value_1.length(), id_value_2.length()); i++) {
@ -1350,6 +1364,14 @@ void analyze_tools::extract_call_id_markers() {
call_id.suffix = find_first_marker(before_func);
}
if (call_id.prefix == arguments.end) {
call_id.prefix = "";
}
if (call_id.suffix == arguments.start) {
call_id.suffix = "";
}
// When call_id is detected, per_call_end may have been incorrectly set to include
// the call_id_suffix and sample args. Clear it if it starts with call_id_suffix.
if (call_id.pos != call_id_position::NONE && !call_id.suffix.empty() &&

4
common/chat.cpp
View File

@ -1631,7 +1631,7 @@ static json common_chat_extra_context() {
return ctx;
}
static std::optional<common_chat_params> try_specialized_template(
std::optional<common_chat_params> common_chat_try_specialized_template(
const common_chat_template & tmpl,
const std::string & src,
const autoparser::generation_params & params) {
@ -1778,7 +1778,7 @@ static common_chat_params common_chat_templates_apply_jinja(const struct common_
return data;
}
if (auto result = try_specialized_template(tmpl, src, params)) {
if (auto result = common_chat_try_specialized_template(tmpl, src, params)) {
result->generation_prompt = params.generation_prompt;
return *result;
}

5
common/chat.h
View File

@ -270,3 +270,8 @@ std::map<std::string, bool> common_chat_templates_get_caps(const common_chat_tem
std::string common_chat_template_direct_apply(
const common_chat_template & tmpl,
const autoparser::generation_params & inputs);
std::optional<common_chat_params> common_chat_try_specialized_template(
const common_chat_template & tmpl,
const std::string & src,
const autoparser::generation_params & params);

101
tests/test-chat.cpp
View File

@ -2541,55 +2541,57 @@ static void test_template_output_peg_parsers(bool detailed_debug) {
// #20424 introduced effective_input = generation_prompt + input, but the throw
// uses input.substr(result.end) where result.end is in effective_input space.
{
auto tmpls = common_chat_templates_ptr(
common_chat_templates_init(nullptr, read_file("models/templates/GLM-4.7-Flash.jinja")));
if (!g_template_filter.empty() && std::string("models/templates/GLM-4.7-Flash.jinja").find(g_template_filter) != std::string::npos) {
auto tmpls = common_chat_templates_ptr(
common_chat_templates_init(nullptr, read_file("models/templates/GLM-4.7-Flash.jinja")));
static common_chat_tool weather_tool{
"get_weather", "Get weather",
R"({"type":"object","properties":{"city":{"type":"string"}},"required":["city"]})",
};
static common_chat_tool weather_tool{
"get_weather", "Get weather",
R"({"type":"object","properties":{"city":{"type":"string"}},"required":["city"]})",
};
common_chat_templates_inputs inputs;
inputs.tools = { weather_tool };
inputs.enable_thinking = true;
inputs.reasoning_format = COMMON_REASONING_FORMAT_AUTO;
inputs.add_generation_prompt = true;
inputs.use_jinja = true;
common_chat_msg msg;
msg.role = "user";
msg.content = "get_weather";
inputs.messages = { msg };
common_chat_templates_inputs inputs;
inputs.tools = { weather_tool };
inputs.enable_thinking = true;
inputs.reasoning_format = COMMON_REASONING_FORMAT_AUTO;
inputs.add_generation_prompt = true;
inputs.use_jinja = true;
common_chat_msg msg;
msg.role = "user";
msg.content = "get_weather";
inputs.messages = { msg };
auto params = common_chat_templates_apply(tmpls.get(), inputs);
common_peg_arena arena;
arena.load(params.parser);
common_chat_parser_params pp(params);
auto params = common_chat_templates_apply(tmpls.get(), inputs);
common_peg_arena arena;
arena.load(params.parser);
common_chat_parser_params pp(params);
// generation_prompt is non-empty for thinking models, so result.end
// will be offset by generation_prompt.size() into effective_input space.
assert(!pp.generation_prompt.empty());
// generation_prompt is non-empty for thinking models, so result.end
// will be offset by generation_prompt.size() into effective_input space.
assert(!pp.generation_prompt.empty());
std::string bad_input =
"Thinking.\n"
"</think>"
"<tool_call>get_weather"
"<arg_key>city</arg_key><arg_value>Tokyo</arg_value>"
"</tool_call>\n";
std::string bad_input =
"Thinking.\n"
"</think>"
"<tool_call>get_weather"
"<arg_key>city</arg_key><arg_value>Tokyo</arg_value>"
"</tool_call>\n";
bool got_runtime_error = false;
bool got_out_of_range = false;
std::string error_msg;
try {
common_chat_peg_parse(arena, bad_input, /*is_partial=*/false, pp);
} catch (const std::out_of_range & e) {
got_out_of_range = true;
error_msg = e.what();
} catch (const std::runtime_error & e) {
got_runtime_error = true;
error_msg = e.what();
bool got_runtime_error = false;
bool got_out_of_range = false;
std::string error_msg;
try {
common_chat_peg_parse(arena, bad_input, /*is_partial=*/false, pp);
} catch (const std::out_of_range & e) {
got_out_of_range = true;
error_msg = e.what();
} catch (const std::runtime_error & e) {
got_runtime_error = true;
error_msg = e.what();
}
GGML_ASSERT(!got_out_of_range && "throw path crashed with out_of_range (input.substr in effective_input space)");
GGML_ASSERT(got_runtime_error && "throw path should produce std::runtime_error with parse position");
}
GGML_ASSERT(!got_out_of_range && "throw path crashed with out_of_range (input.substr in effective_input space)");
GGML_ASSERT(got_runtime_error && "throw path should produce std::runtime_error with parse position");
}
// Kimi-K2-Thinking tests - custom parser
@ -3169,6 +3171,21 @@ static void test_template_output_peg_parsers(bool detailed_debug) {
.expect(message_assist_call_id)
.expect_reconstruction()
.run();
tst.test("[TOOL_CALLS]special_function[CALL_ID]000000001[ARGS]{\"arg1\": 1}"
"[TOOL_CALLS]special_function_with_opt[CALL_ID]000000002[ARGS]{\"arg1\": 1, \"arg2\": 2}")
.parallel_tool_calls(true)
.tools({
special_function_tool, special_function_tool_with_optional_param
})
.expect_tool_calls({
{ "special_function", R"({"arg1": 1})", "000000001" },
{ "special_function_with_opt", R"({"arg1": 1, "arg2": 2})", "000000002" },
})
.expect_reconstruction()
.run();
}
// Devstral
{

110
tools/parser/debug-template-parser.cpp
View File

@ -5,15 +5,15 @@
#include "gguf.h"
#include "jinja/runtime.h"
#include "log.h"
#include "nlohmann/json.hpp"
#include "peg-parser.h"
#include <fstream>
#include <numeric>
#include <optional>
#include <sstream>
#include <string>
#include "nlohmann/json.hpp"
#include "peg-parser.h"
using json = nlohmann::ordered_json;
enum class output_mode {
@ -34,14 +34,14 @@ enum class input_message_type {
};
struct debug_options {
std::string template_path;
bool with_tools = true;
bool generation_prompt = true;
bool enable_reasoning = true;
bool debug_jinja = false;
bool force_tool_call = false;
output_mode mode = output_mode::BOTH;
input_message_type input_message = input_message_type::NONE;
std::string template_path;
bool with_tools = true;
bool generation_prompt = true;
bool enable_reasoning = true;
bool debug_jinja = false;
bool force_tool_call = false;
output_mode mode = output_mode::BOTH;
input_message_type input_message = input_message_type::NONE;
};
static std::string read_file(const std::string & path) {
@ -274,7 +274,7 @@ static void render_scenario(const common_chat_template & tmpl,
json final_messages = messages;
if (add_generation_prompt && !messages.empty() && messages.back().value("role", "") == "assistant") {
final_messages.push_back(json{
{ "role", "user" },
{ "role", "user" },
{ "content", "Now please continue with another response." }
});
}
@ -305,7 +305,7 @@ static void render_all_scenarios(const common_chat_template & tmpl,
const json & tools,
bool add_generation_prompt,
bool enable_thinking,
input_message_type message_type) {
input_message_type message_type) {
json user_msg = build_user_message();
auto render_if = [&](input_message_type type, const std::string & name, const json & assistant_msg) {
@ -335,6 +335,24 @@ static void render_all_scenarios(const common_chat_template & tmpl,
}
}
static autoparser::generation_params prepare_params(const debug_options & opts, const json & tools) {
autoparser::generation_params params;
params.messages = json::array({ build_user_message() });
params.reasoning_format = opts.enable_reasoning ? COMMON_REASONING_FORMAT_DEEPSEEK : COMMON_REASONING_FORMAT_NONE;
params.enable_thinking = opts.enable_reasoning;
params.add_generation_prompt = opts.generation_prompt;
if (opts.with_tools) {
params.tools = tools;
params.tool_choice = opts.force_tool_call ? COMMON_CHAT_TOOL_CHOICE_REQUIRED : COMMON_CHAT_TOOL_CHOICE_AUTO;
} else {
params.tools = json();
params.tool_choice = COMMON_CHAT_TOOL_CHOICE_NONE;
}
params.parallel_tool_calls = false;
return params;
}
int main(int argc, char ** argv) {
// Set log level to most verbose to capture all debug output
common_log_set_verbosity_thold(99);
@ -369,49 +387,41 @@ int main(int argc, char ** argv) {
try {
common_chat_template chat_template(template_source, "", "");
// Build tools definition
json tools = opts.with_tools ? build_tools_definition() : json();
// Render template scenarios if requested
if (opts.input_message != input_message_type::NONE &&
(opts.mode == output_mode::TEMPLATE || opts.mode == output_mode::BOTH)) {
autoparser::generation_params params = prepare_params(opts, tools);
common_chat_params parser_data;
if (std::optional<common_chat_params> spec_tmpl =
common_chat_try_specialized_template(chat_template, template_source, params)) {
LOG_ERR("\n");
LOG_ERR("================================================================================\n");
LOG_ERR(" TEMPLATE RENDERING OUTPUT\n");
LOG_ERR("================================================================================\n");
LOG_ERR("This template uses a specialized parser, analysis results will not be available.");
parser_data = *spec_tmpl;
} else {
// Render template scenarios if requested
if (opts.input_message != input_message_type::NONE &&
(opts.mode == output_mode::TEMPLATE || opts.mode == output_mode::BOTH)) {
LOG_ERR("\n");
LOG_ERR("================================================================================\n");
LOG_ERR(" TEMPLATE RENDERING OUTPUT\n");
LOG_ERR("================================================================================\n");
render_all_scenarios(chat_template, tools, opts.generation_prompt, opts.enable_reasoning,
opts.input_message);
}
// Output analysis if requested
if (opts.mode == output_mode::ANALYSIS || opts.mode == output_mode::BOTH) {
LOG_ERR("\n");
LOG_ERR("================================================================================\n");
LOG_ERR(" TEMPLATE ANALYSIS\n");
LOG_ERR("================================================================================\n");
autoparser::autoparser analysis;
analysis.analyze_template(chat_template);
// Generate Parser
autoparser::generation_params params;
params.messages = json::array({ build_user_message() });
params.reasoning_format =
opts.enable_reasoning ? COMMON_REASONING_FORMAT_DEEPSEEK : COMMON_REASONING_FORMAT_NONE;
params.enable_thinking = opts.enable_reasoning;
params.add_generation_prompt = opts.generation_prompt;
if (opts.with_tools) {
params.tools = tools;
params.tool_choice = opts.force_tool_call ? COMMON_CHAT_TOOL_CHOICE_REQUIRED : COMMON_CHAT_TOOL_CHOICE_AUTO;
} else {
params.tools = json();
params.tool_choice = COMMON_CHAT_TOOL_CHOICE_NONE;
render_all_scenarios(chat_template, tools, opts.generation_prompt, opts.enable_reasoning,
opts.input_message);
}
params.parallel_tool_calls = false;
auto parser_data = autoparser::peg_generator::generate_parser(chat_template, params, analysis);
// Output analysis if requested
if (opts.mode == output_mode::ANALYSIS || opts.mode == output_mode::BOTH) {
LOG_ERR("\n");
LOG_ERR("================================================================================\n");
LOG_ERR(" TEMPLATE ANALYSIS\n");
LOG_ERR("================================================================================\n");
autoparser::autoparser analysis;
analysis.analyze_template(chat_template);
// Generate Parser
parser_data = autoparser::peg_generator::generate_parser(chat_template, params, analysis);
}
LOG_ERR("\n=== Generated Parser ===\n");
common_peg_arena arena;