ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-04-26 01:19:20 +00:00
Code Issues Packages Projects Releases Wiki Activity

Tool calls support from mainline (#723)

Browse Source 
* Tool calls support from mainline

* update cmake

* revert api for /completions

* Fix broken thinking process for gpt-oss

* add missing args and fix webui bugs

* add missing args and fix webui bugs2

* Fix reasoning format error

* add usage

* change default post_sampling_probs to true

* add back generated_text

* Remove server endpoints tests

* add log

* Chat fixes

* Remove logs

* webui: revert extra handling of thinking process

---------

Co-authored-by: firecoperana <firecoperana>
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
This commit is contained in:
firecoperana
2025-09-01 00:38:49 -05:00
committed by GitHub
parent 8de297b795
commit d7882c3cf8
87 changed files with 13581 additions and 2224 deletions
Download Patch File Download Diff File Expand all files Collapse all files

32
common/sampling.h
View File

@@ -1,9 +1,8 @@
#pragma once
#include "llama.h"
#include "grammar-parser.h"
#include <set>
#include <random>
#include <string>
#include <unordered_map>
@@ -22,6 +21,23 @@ enum class llama_sampler_type : char {
TEMPERATURE = 't'
};
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_FULL,
};
struct common_grammar_trigger {
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
typedef struct llama_sampling_params {
int32_t n_prev = 64; // number of previous tokens to remember
@@ -67,8 +83,11 @@ typedef struct llama_sampling_params {
llama_sampler_type::TEMPERATURE
};
std::string grammar; // optional BNF-like grammar to constrain sampling
std::string grammar; // optional BNF-like grammar to constrain sampling
bool grammar_lazy = false;
std::vector<common_grammar_trigger> grammar_triggers; // optional triggers (for lazy grammars)
std::set<llama_token> preserved_tokens;
// Classifier-Free Guidance
// https://arxiv.org/abs/2306.17806
std::string cfg_negative_prompt; // string to help guidance
@@ -106,7 +125,7 @@ struct llama_sampling_context {
std::mt19937 rng;
};
#include "common.h"
// Create a new sampling context instance.
struct llama_sampling_context * llama_sampling_init(const struct llama_vocab* vocab, const struct llama_sampling_params & params);
@@ -116,7 +135,7 @@ void llama_sampling_free(struct llama_sampling_context * ctx);
// Reset the sampler context
// - clear prev tokens
// - reset grammar
void llama_sampling_reset(llama_sampling_context * ctx);
void llama_sampling_reset(const struct llama_vocab* vocab, llama_sampling_context * ctx);
// Set the sampler seed
void llama_sampling_set_rng_seed(struct llama_sampling_context * ctx, uint32_t seed);
@@ -186,3 +205,6 @@ llama_token_data_array * llama_sampling_get_candidates(struct llama_sampling_con
std::vector<llama_token> llama_sampling_sample_and_accept_n(struct llama_sampling_context * gsmpl, struct llama_context * ctx, const std::vector<llama_token> & draft);
std::vector<llama_token> llama_sampling_sample_and_accept_n(struct llama_sampling_context * gsmpl, struct llama_context * ctx, const std::vector<int> & idxs, const std::vector<llama_token> & draft);
llama_grammar* llama_sampler_init_llg(const llama_vocab* vocab,
const char* grammar_kind, const char* grammar_data);