mirror of
https://github.com/ikawrakow/ik_llama.cpp.git
synced 2026-01-26 17:20:01 +00:00
* Implement function calling / tools for ik_llama.cpp for Kimi K2
* Implement basic tool choice
* Backport llama.cpp tool calls support
* Enhance function calls with improved chat parser and string utilities
- Add new chat.h/chat.cpp and chat-parser.h/chat-parser.cpp for better chat handling
- Improve function calls parsing with fallback to llama.cpp builder pattern
- Add string utility functions (starts_with, ends_with, find_partial_stop)
- Update README with function calls testing instructions
- Enhance Kimi K2 parser and function calls documentation
- Add comprehensive test suite for function calls
- Update CMakeLists.txt and Makefile for new components
* Enhance function calling with unified streaming and parser improvements
- Fix streaming content cleanup to prevent function syntax in output
- Unify content extraction patterns with llama.cpp approach
- Improve Kimi K2 parser robustness and partial content handling
- Add comprehensive test coverage for function call scenarios
- Optimize chat message parsing and diff computation
* Replace hardcoded values in kimi_k2_parser.hpp with named constants
- Add compile-time constants for all token format markers
- Add compile-time constants for XML format markers
- Add compile-time constants for simple format patterns
- Replace all hardcoded string literals with named constants
- Use compile-time length calculation to avoid manual counting
- Improve maintainability and reduce magic numbers throughout parser
* Fix duplicate common_chat_parse definition
- Remove duplicate implementation from chat-parser.cpp
- Keep single implementation in chat.cpp following llama.cpp patterns
- Resolves linker error: multiple definition of common_chat_parse
* Fix JSON assertion failure in function call parsing
- Add proper validation that 'function' field is an object before accessing nested keys
- Handle missing 'arguments' field gracefully with default "{}"
- Prevents crash when parsing malformed tool call JSON structures
* Add comprehensive Qwen3 XML tool calling support with unit tests
- Implement Qwen3 XML parser with <tool_call>{"name": "func", "arguments": {...}}</tool_call> format
- Add model detection and routing for Qwen3 vs Kimi-K2 formats
- Create 8 comprehensive unit tests covering parsing, streaming, error handling
- Fix token format cleaning bug in kimi_k2_parser.hpp processing order
- Remove progressive parsing code and related utilities
- Add tool injection support for Qwen3 format in server utils
* Add DeepSeek R1 function calling support with comprehensive unit tests
- Implement complete DeepSeek R1 tool call parsing in common_chat_parser.cpp
- Add DeepSeek R1 model detection and tool injection in deepseek_r1_tools.hpp
- Update function_calls.hpp with DeepSeek R1 integration and content extraction
- Update documentation to reflect support for Kimi-K2, Qwen3, and DeepSeek R1 models
- Add comprehensive unit tests for DeepSeek R1 reasoning, tool calls, and integration
- Port exact implementation patterns from original llama.cpp for compatibility
Key features:
- Native DeepSeek R1 format: <|tool▁calls▁begin|>function<|tool▁sep|>name```json{}```<|tool▁call▁end|><|tool▁calls▁end|>
- Reasoning content extraction from <think>...</think> tags
- Multiple tool calls support with separate call blocks
- Model detection for deepseek-r1, deepseek_r1 naming patterns
- Integration with incremental parsing and streaming support
* Add partial parsing support for JSON and regex
- json-partial.h/cpp: JSON partial parsing functionality
- regex-partial.h/cpp: Regex partial parsing functionality
* Add format_chat integration tests for Qwen3 tool injection
- Add test_qwen3_format_chat_integration() to validate tool injection pipeline
- Test tool injection conditions and system message enhancement
- Verify JSON formatting and anti-preamble instructions
- Add comprehensive test documentation
Tests confirm tool injection works correctly - conversational preamble
issue is not in ik_llama.cpp but likely in UI configuration.
* Fix Qwen3 tool call parsing - pass model name to parser
Server was not passing model name to parse_chat_message_incremental(),
causing Qwen3 to fall back to Kimi-K2 parser and return tool calls
as content instead of proper tool_calls array.
* Fix non-streaming path to use model-specific parsing
Non-streaming responses were hardcoded to use Kimi-K2 format,
causing Qwen3 XML tool calls to be returned as content instead
of proper tool_calls array. Now uses same model detection as
streaming path for consistency.
205 lines
8.2 KiB
C++
205 lines
8.2 KiB
C++
#include "regex-partial.h"
|
|
#include "common.h"
|
|
#include <functional>
|
|
#include <optional>
|
|
|
|
common_regex::common_regex(const std::string & pattern) :
|
|
pattern(pattern),
|
|
rx(pattern),
|
|
rx_reversed_partial(regex_to_reversed_partial_regex(pattern)) {}
|
|
|
|
common_regex_match common_regex::search(const std::string & input, size_t pos, bool as_match) const {
|
|
std::smatch match;
|
|
if (pos > input.size()) {
|
|
throw std::runtime_error("Position out of bounds");
|
|
}
|
|
auto start = input.begin() + pos;
|
|
auto found = as_match
|
|
? std::regex_match(start, input.end(), match, rx)
|
|
: std::regex_search(start, input.end(), match, rx);
|
|
if (found) {
|
|
common_regex_match res;
|
|
res.type = COMMON_REGEX_MATCH_TYPE_FULL;
|
|
for (size_t i = 0; i < match.size(); ++i) {
|
|
auto begin = pos + match.position(i);
|
|
res.groups.emplace_back(begin, begin + match.length(i));
|
|
}
|
|
return res;
|
|
}
|
|
std::match_results<std::string::const_reverse_iterator> srmatch;
|
|
if (std::regex_match(input.rbegin(), input.rend() - pos, srmatch, rx_reversed_partial)) {
|
|
auto group = srmatch[1].str();
|
|
if (group.length() != 0) {
|
|
auto it = srmatch[1].second.base();
|
|
// auto position = static_cast<size_t>(std::distance(input.begin(), it));
|
|
if ((!as_match) || it == input.begin()) {
|
|
common_regex_match res;
|
|
res.type = COMMON_REGEX_MATCH_TYPE_PARTIAL;
|
|
const size_t begin = std::distance(input.begin(), it);
|
|
const size_t end = input.size();
|
|
if (begin == std::string::npos || end == std::string::npos || begin > end) {
|
|
throw std::runtime_error("Invalid range");
|
|
}
|
|
res.groups.push_back({begin, end});
|
|
return res;
|
|
}
|
|
}
|
|
}
|
|
return {};
|
|
}
|
|
|
|
/*
|
|
Transforms a regex pattern to a partial match pattern that operates on a reversed input string to find partial final matches of the original pattern.
|
|
|
|
Ideally we'd like to use boost::match_partial (https://beta.boost.org/doc/libs/1_59_0/libs/regex/doc/html/boost_regex/partial_matches.html)
|
|
to see if a string ends with a partial regex match, but but it's not in std::regex yet.
|
|
Instead, we'll the regex into a partial match regex operating as a full match on the reverse iterators of the input.
|
|
|
|
- /abcd/ -> (dcba|cba|ba|a).* -> ((?:(?:(?:(?:d)?c)?b)?a).*
|
|
- /a|b/ -> (a|b).*
|
|
- /a*?/ -> error, could match ""
|
|
- /a*b/ -> ((?:b)?a*+).* (final repetitions become eager)
|
|
- /.*?ab/ -> ((?:b)?a).* (merge .*)
|
|
- /a.*?b/ -> ((?:b)?.*?a).* (keep reluctant matches)
|
|
- /a(bc)d/ -> ((?:(?:d)?(?:(?:c)?b))?a).*
|
|
- /a(bc|de)/ -> ((?:(?:(?:e)?d)?|(?:(?:c)?b)?)?a).*
|
|
- /ab{2,4}c/ -> abbb?b?c -> ((?:(?:(?:(?:(?:c)?b)?b)?b?)?b?)?a).*
|
|
|
|
The regex will match a reversed string fully, and the end of the first (And only) capturing group will indicate the reversed start of the original partial pattern
|
|
(i.e. just where the final .* starts in the inverted pattern; all other groups are turned into non-capturing groups, and reluctant quantifiers are ignored)
|
|
*/
|
|
std::string regex_to_reversed_partial_regex(const std::string & pattern) {
|
|
auto it = pattern.begin();
|
|
const auto end = pattern.end();
|
|
|
|
std::function<std::string()> process = [&]() {
|
|
std::vector<std::vector<std::string>> alternatives(1);
|
|
std::vector<std::string> * sequence = &alternatives.back();
|
|
|
|
while (it != end) {
|
|
if (*it == '[') {
|
|
auto start = it;
|
|
++it;
|
|
while (it != end) {
|
|
if ((*it == '\\') && (++it != end)) {
|
|
++it;
|
|
} else if ((it != end) && (*it == ']')) {
|
|
break;
|
|
} else {
|
|
++it;
|
|
}
|
|
}
|
|
if (it == end) {
|
|
throw std::runtime_error("Unmatched '[' in pattern");
|
|
}
|
|
++it;
|
|
sequence->push_back(std::string(start, it));
|
|
} else if (*it == '*' || *it == '?' || *it == '+') {
|
|
if (sequence->empty()) {
|
|
throw std::runtime_error("Quantifier without preceding element");
|
|
}
|
|
sequence->back() += *it;
|
|
auto is_star = *it == '*';
|
|
++it;
|
|
if (is_star) {
|
|
if (*it == '?') {
|
|
++it;
|
|
}
|
|
}
|
|
} else if (*it == '{') {
|
|
if (sequence->empty()) {
|
|
throw std::runtime_error("Repetition without preceding element");
|
|
}
|
|
++it;
|
|
auto start = it;
|
|
while (it != end && *it != '}') {
|
|
++it;
|
|
}
|
|
if (it == end) {
|
|
throw std::runtime_error("Unmatched '{' in pattern");
|
|
}
|
|
auto parts = string_split(std::string(start, it), ',');
|
|
++it;
|
|
if (parts.size() > 2) {
|
|
throw std::runtime_error("Invalid repetition range in pattern");
|
|
}
|
|
|
|
auto parseOptInt = [&](const std::string & s, const std::optional<int> & def = std::nullopt) -> std::optional<int> {
|
|
if (s.empty()) {
|
|
return def;
|
|
}
|
|
return std::stoi(s);
|
|
};
|
|
auto min = parseOptInt(parts[0], 0);
|
|
auto max = parts.size() == 1 ? min : parseOptInt(parts[1]);
|
|
if (min && max && *max < *min) {
|
|
throw std::runtime_error("Invalid repetition range in pattern");
|
|
}
|
|
// Brutal but... let's repeat at least min times, then ? for the delta between min & max (or * for unbounded)
|
|
auto part = sequence->back();
|
|
sequence->pop_back();
|
|
for (int i = 0; i < *min; i++) {
|
|
sequence->push_back(part);
|
|
}
|
|
if (max) {
|
|
for (int i = *min; i < *max; i++) {
|
|
sequence->push_back(part + "?");
|
|
}
|
|
} else {
|
|
sequence->push_back(part + "*");
|
|
}
|
|
} else if (*it == '(') {
|
|
++it;
|
|
if (it != end && *it == '?' && (it + 1 != end) && *(it + 1) == ':') {
|
|
it += 2;
|
|
}
|
|
auto sub = process();
|
|
if (*it != ')') {
|
|
throw std::runtime_error("Unmatched '(' in pattern");
|
|
}
|
|
++it;
|
|
auto & part = sequence->emplace_back("(?:");
|
|
part += sub;
|
|
part += ")";
|
|
} else if (*it == ')') {
|
|
break;
|
|
} else if (*it == '|') {
|
|
++it;
|
|
alternatives.emplace_back();
|
|
sequence = &alternatives.back();
|
|
} else if (*it == '\\' && (++it != end)) {
|
|
auto str = std::string("\\") + *it;
|
|
sequence->push_back(str);
|
|
++it;
|
|
} else if (it != end) {
|
|
sequence->push_back(std::string(1, *it));
|
|
++it;
|
|
}
|
|
}
|
|
|
|
// /abcd/ -> (dcba|cba|ba|a).* -> ((?:(?:(?:d)?c)?b)?a).*
|
|
// if n(=4) parts, opening n-1(=3) non-capturing groups after the 1 capturing group
|
|
// We'll do the outermost capturing group and final .* in the enclosing function.
|
|
std::vector<std::string> res_alts;
|
|
for (const auto & parts : alternatives) {
|
|
auto & res = res_alts.emplace_back();
|
|
for (size_t i = 0; i < parts.size() - 1; i++) {
|
|
res += "(?:";
|
|
}
|
|
for (auto it = parts.rbegin(); it != parts.rend(); ++it) {
|
|
res += *it;
|
|
if (it != parts.rend() - 1) {
|
|
res += ")?";
|
|
}
|
|
}
|
|
}
|
|
return string_join(res_alts, "|");
|
|
};
|
|
auto res = process();
|
|
if (it != end) {
|
|
throw std::runtime_error("Unmatched '(' in pattern");
|
|
}
|
|
|
|
return "(" + res + ")[\\s\\S]*";
|
|
}
|