echodict/llama.cpp

1#include "arg.h"2#include "common.h"3#include "debug.h"4#include "log.h"5#include "llama.h"6#include "llama-cpp.h"7 8#include <clocale>9#include <string>10#include <vector>11 12static bool run(llama_context * ctx, const common_params & params) {13    const llama_model * model = llama_get_model(ctx);14    const llama_vocab * vocab = llama_model_get_vocab(model);15 16    const bool add_bos = llama_vocab_get_add_bos(vocab);17 18    std::vector<llama_token> tokens = common_tokenize(ctx, params.prompt, add_bos, true);19 20    if (tokens.empty()) {21        LOG_ERR("%s : there are not input tokens to process - (try to provide a prompt with '-p')\n", __func__);22        return false;23    }24 25    LOG_INF("number of input tokens = %zu\n", tokens.size());26    for (size_t i = 0; i < tokens.size(); ++i) {27        LOG_INF("  %d\n", tokens[i]);28    }29 30    if (llama_decode(ctx, llama_batch_get_one(tokens.data(), tokens.size()))) {31        LOG_ERR("%s : failed to eval\n", __func__);32        return false;33    }34 35    return true;36}37 38int main(int argc, char ** argv) {39    std::setlocale(LC_NUMERIC, "C");40 41    base_callback_data cb_data;42 43    common_params params;44 45    common_init();46 47    if (!common_params_parse(argc, argv, params, LLAMA_EXAMPLE_COMMON)) {48        return 1;49    }50 51    llama_backend_init();52    llama_numa_init(params.numa);53 54    // pass the callback to the backend scheduler55    // it will be executed for each node during the graph computation56    params.cb_eval = common_debug_cb_eval<false>;57    params.cb_eval_user_data = &cb_data;58    params.warmup = false;59 60    // init61    auto llama_init = common_init_from_params(params);62 63    auto * model = llama_init->model();64    auto * ctx   = llama_init->context();65 66    if (model == nullptr || ctx == nullptr) {67        LOG_ERR("%s : failed to init\n", __func__);68        return 1;69    }70 71    // print system information72    {73        LOG_INF("\n");74        LOG_INF("%s\n", common_params_get_system_info(params).c_str());75        LOG_INF("\n");76    }77 78    bool OK = run(ctx, params);79    if (!OK) {80        return 1;81    }82 83    LOG("\n");84    llama_perf_context_print(ctx);85 86    llama_backend_free();87 88    return 0;89}90