tool-call: fix Qwen 2.5 Coder support, add micro benchmarks, support trigger patterns for lazy grammars (#12034)

* sampler: turn lazy grammar trigger words to regexes

* add scripts/tool_bench.sh & .py

* constrain llama json output regardless of function name if matches at beginning

* update relaxed newline space rule in grammar tests

* support add_generation_prompt query parameter (useful for /apply_template)

* Update src/llama-grammar.cpp

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>

---------

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
This commit is contained in:
Olivier Chafik
2025-03-05 13:05:13 +00:00
committed by GitHub
parent fa31c438e0
commit 669912d9a5
26 changed files with 1314 additions and 408 deletions

View File

@ -110,9 +110,21 @@ enum common_conversation_mode {
COMMON_CONVERSATION_MODE_AUTO = 2,
};
enum common_grammar_trigger_type {
COMMON_GRAMMAR_TRIGGER_TYPE_TOKEN,
COMMON_GRAMMAR_TRIGGER_TYPE_WORD,
COMMON_GRAMMAR_TRIGGER_TYPE_PATTERN,
COMMON_GRAMMAR_TRIGGER_TYPE_PATTERN_START,
};
struct common_grammar_trigger {
std::string word;
bool at_start;
common_grammar_trigger_type type;
std::string value;
llama_token token = LLAMA_TOKEN_NULL;
// T can only be nlohmann::ordered_json
template <class T> T to_json() const;
template <class T> static common_grammar_trigger from_json(const T & in);
};
// sampling parameters
@ -163,8 +175,7 @@ struct common_params_sampling {
std::string grammar; // optional BNF-like grammar to constrain 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::vector<common_grammar_trigger> grammar_triggers; // optional triggers (for lazy grammars)
std::set<llama_token> preserved_tokens;
std::vector<llama_logit_bias> logit_bias; // logit biases to apply
@ -458,6 +469,8 @@ std::string string_repeat(const std::string & str, size_t n);
void string_replace_all(std::string & s, const std::string & search, const std::string & replace);
std::string regex_escape(const std::string & s);
template<class T>
static std::vector<T> string_split(const std::string & str, char delim) {
static_assert(!std::is_same<T, std::string>::value, "Please use the specialized version for std::string");