llama-cpp-sys-2 0.1.150

#include "wrapper_common.h"

#include <cstdlib>
#include <cstring>
#include <exception>
#include <string>
#include <stdint.h>

#include "llama.cpp/common/common.h"
#include "llama.cpp/common/fit.h"
#include "llama.cpp/common/json-schema-to-grammar.h"
#include "llama.cpp/include/llama.h"
#include "wrapper_utils.h"

#include <nlohmann/json.hpp>

extern "C" llama_rs_status llama_rs_json_schema_to_grammar(
    const char * schema_json,
    bool force_gbnf,
    char ** out_grammar) {
    if (!schema_json || !out_grammar) {
        return LLAMA_RS_STATUS_INVALID_ARGUMENT;
    }

    *out_grammar = nullptr;
    try {
        const auto schema = nlohmann::ordered_json::parse(schema_json);
        const auto grammar = json_schema_to_grammar(schema, force_gbnf);
        *out_grammar = llama_rs_dup_string(grammar);
        return *out_grammar ? LLAMA_RS_STATUS_OK : LLAMA_RS_STATUS_ALLOCATION_FAILED;
    } catch (const std::exception &) {
        return LLAMA_RS_STATUS_EXCEPTION;
    }
}

extern "C" void llama_rs_string_free(char * ptr) {
    if (ptr) {
        std::free(ptr);
    }
}

extern "C" struct llama_sampler * llama_rs_sampler_init_grammar(
    const struct llama_vocab * vocab,
    const char * grammar_str,
    const char * grammar_root) {
    try {
        return llama_sampler_init_grammar(vocab, grammar_str, grammar_root);
    } catch (...) {
        return nullptr;
    }
}

extern "C" struct llama_sampler * llama_rs_sampler_init_grammar_lazy(
    const struct llama_vocab * vocab,
    const char * grammar_str,
    const char * grammar_root,
    const char ** trigger_words,
    size_t num_trigger_words,
    const llama_token * trigger_tokens,
    size_t num_trigger_tokens) {
    try {
        std::vector<std::string> trigger_patterns;
        trigger_patterns.reserve(num_trigger_words);
        for (size_t i = 0; i < num_trigger_words; ++i) {
            const char * word = trigger_words ? trigger_words[i] : nullptr;
            if (word && word[0] != '\0') {
                trigger_patterns.push_back(regex_escape(word));
            }
        }
        std::vector<const char *> trigger_patterns_c;
        trigger_patterns_c.reserve(trigger_patterns.size());
        for (const auto & pattern : trigger_patterns) {
            trigger_patterns_c.push_back(pattern.c_str());
        }
        return llama_sampler_init_grammar_lazy_patterns(
            vocab,
            grammar_str,
            grammar_root,
            trigger_patterns_c.data(),
            trigger_patterns_c.size(),
            trigger_tokens,
            num_trigger_tokens);
    } catch (...) {
        return nullptr;
    }
}

extern "C" struct llama_sampler * llama_rs_sampler_init_grammar_lazy_patterns(
    const struct llama_vocab * vocab,
    const char * grammar_str,
    const char * grammar_root,
    const char ** trigger_patterns,
    size_t num_trigger_patterns,
    const llama_token * trigger_tokens,
    size_t num_trigger_tokens) {
    try {
        return llama_sampler_init_grammar_lazy_patterns(
            vocab,
            grammar_str,
            grammar_root,
            trigger_patterns,
            num_trigger_patterns,
            trigger_tokens,
            num_trigger_tokens);
    } catch (...) {
        return nullptr;
    }
}

extern "C" llama_rs_status llama_rs_sampler_accept(struct llama_sampler * sampler, llama_token token) {
    if (!sampler) {
        return LLAMA_RS_STATUS_INVALID_ARGUMENT;
    }
    try {
        llama_sampler_accept(sampler, token);
        return LLAMA_RS_STATUS_OK;
    } catch (const std::exception &) {
        return LLAMA_RS_STATUS_EXCEPTION;
    } catch (...) {
        return LLAMA_RS_STATUS_EXCEPTION;
    }
}

// Thin pass-through to llama.cpp's common_fit_params (a C++ symbol in libcommon).
// Returns common_params_fit_status as an int: 0 = success, 1 = failure, 2 = error.
extern "C" int llama_rs_fit_params(
    const char * path_model,
    struct llama_model_params * mparams,
    struct llama_context_params * cparams,
    float * tensor_split,
    struct llama_model_tensor_buft_override * tensor_buft_overrides,
    size_t * margins,
    uint32_t n_ctx_min,
    enum ggml_log_level log_level) {
    return static_cast<int>(common_fit_params(
        path_model,
        mparams,
        cparams,
        tensor_split,
        tensor_buft_overrides,
        margins,
        n_ctx_min,
        log_level));
}

extern "C" void llama_rs_memory_breakdown_print(const struct llama_context * ctx) {
    common_memory_breakdown_print(ctx);
}