echodict/llama.cpp

1#include "arg.h"2#include "common.h"3#include "log.h"4#include "llama.h"5 6#include <clocale>7#include <cmath>8#include <cstdio>9#include <cstring>10#include <ctime>11#include <vector>12 13#if defined(_MSC_VER)14#pragma warning(disable: 4244 4267)  // possible loss of data15#endif16 17int main(int argc, char ** argv) {18    std::setlocale(LC_NUMERIC, "C");19 20    common_params params;21    params.escape = false;22 23    common_init();24 25    if (!common_params_parse(argc, argv, params, LLAMA_EXAMPLE_FINETUNE)) {26        return 1;27    }28 29    if (params.use_mmap) {30        LOG_INF("%s: force disabling memory mapping because it would result in-read-only pointers to the weights\n",31                __func__);32        params.use_mmap = false;33    }34    if (params.cache_type_k != GGML_TYPE_F32) {35        LOG_INF("%s: force changing k cache type to f32 due to a lack of f16 support for OUT_PROD\n", __func__);36        params.cache_type_k = GGML_TYPE_F32;37    }38    if (params.cache_type_v != GGML_TYPE_F32) {39        LOG_INF("%s: force changing v cache type to f32 due to a lack of f16 support for OUT_PROD\n", __func__);40        params.cache_type_v = GGML_TYPE_F32;41    }42 43    llama_backend_init();44    llama_numa_init(params.numa);45    // load the model and apply lora adapter, if any46    auto llama_init = common_init_from_params(params);47 48    auto * model = llama_init->model();49    auto * ctx   = llama_init->context();50 51    if (model == NULL) {52        LOG_ERR("%s: unable to load model\n", __func__);53        return 1;54    }55 56    // print system information57    {58        LOG_INF("\n");59        LOG_INF("%s\n", common_params_get_system_info(params).c_str());60    }61 62    std::vector<llama_token> tokens  = common_tokenize(ctx, params.prompt, true);63    ggml_opt_dataset_t       dataset = common_opt_dataset_init(ctx, tokens, llama_n_ctx(ctx) / 2);64 65    struct lr_opt & lr = params.lr;66    LOG_INF("-optimizer %s -lr0 %.2g -wd %.2g -lr-min %.2g -min-epochs %.2g -epochs %d -period %.2g -val %.2g\n",67            ggml_opt_optimizer_name(params.optimizer), (double) lr.lr0, (double) lr.wd, (double) lr.lr_min, (double) lr.decay_epochs,68            (unsigned) lr.epochs, (double) params.n_batch / params.n_ubatch, (double) params.val_split);69 70    struct llama_opt_params lopt_params{71        /*n_ctx_train     =*/0,72        /*param_filter    =*/llama_opt_param_filter_all,73        /*param_filter_ud =*/nullptr,74        /*get_opt_pars    =*/common_opt_lr_pars,75        /*get_opt_pars_ud =*/&params.lr,76        /*optimizer_type  =*/params.optimizer,77    };78    llama_opt_init(ctx, model, lopt_params);79 80    const int64_t idata_split = ggml_opt_dataset_ndata(dataset) * (1.0f - params.val_split);81 82    ggml_opt_result_t result_train = ggml_opt_result_init();83    ggml_opt_result_t result_eval  = ggml_opt_result_init();84 85    for (lr.epoch = 0; lr.epoch < lr.epochs; ++lr.epoch) {86        llama_opt_epoch(ctx, dataset, result_train, result_eval, idata_split,87                        ggml_opt_epoch_callback_progress_bar, ggml_opt_epoch_callback_progress_bar);88        fprintf(stderr, "\n");89 90        ggml_opt_result_reset(result_train);91        ggml_opt_result_reset(result_eval);92    }93    ggml_opt_result_free(result_train);94    ggml_opt_result_free(result_eval);95 96    llama_model_save_to_file(model, params.out_file.c_str());97 98    llama_backend_free();99 100    return 0;101}102