Function main

external/ggml/examples/gpt-2/main-alloc.cpp:726–880

}

int main(int argc, char ** argv) {
    ggml_time_init();

    const int64_t t_main_start_us = ggml_time_us();

    gpt_params params;
    params.model = "models/gpt-2-117M/ggml-model.bin";

    if (gpt_params_parse(argc, argv, params) == false) {
        return 1;
    }

    if (params.seed < 0) {
        params.seed = time(NULL);
    }

    printf("%s: seed = %d\n", __func__, params.seed);

    std::mt19937 rng(params.seed);
    if (params.prompt.empty()) {
        params.prompt = gpt_random_prompt(rng);
    }

    int64_t t_load_us = 0;

    gpt_vocab vocab;
    gpt2_model model;

    // load the model
    {
        const int64_t t_start_us = ggml_time_us();

        if (!gpt2_model_load(params.model, model, vocab)) {
            fprintf(stderr, "%s: failed to load model from '%s'\n", __func__, params.model.c_str());
            return 1;
        }

        t_load_us = ggml_time_us() - t_start_us;

        test_gpt_tokenizer(vocab, params.token_test);
    }

    ggml_gallocr_t allocr = NULL;
    // allocate the compute buffer
    {
        allocr = ggml_gallocr_new(ggml_backend_cpu_buffer_type());

        // create the worst case graph for memory usage estimation
        int n_tokens = std::min(model.hparams.n_ctx, params.n_batch);
        int n_past = model.hparams.n_ctx - n_tokens;
        struct ggml_cgraph * gf = gpt2_graph(model, n_past, n_tokens);

        // pre-allocate the compute buffer for the worst case (optional)
        ggml_gallocr_reserve(allocr, gf);
        size_t mem_size =  ggml_gallocr_get_buffer_size(allocr, 0);
        fprintf(stderr, "%s: compute buffer size: %.2f MB\n", __func__, mem_size/1024.0/1024.0);
    }