Function common_perf_print

common/sampling.cpp:475–518

}

void common_perf_print(const struct llama_context * ctx, const struct common_sampler * gsmpl) {
    // TODO: measure grammar performance

    const double t_sampling_ms = gsmpl ? 1e-3*gsmpl->t_total_us : 0;

    llama_perf_sampler_data data_smpl;
    llama_perf_context_data data_ctx;

    memset(&data_smpl, 0, sizeof(data_smpl));
    memset(&data_ctx,  0, sizeof(data_ctx));

    if (gsmpl) {
        auto & data = data_smpl;

        data = llama_perf_sampler(gsmpl->chain);

        // note: the sampling time includes the samplers time + extra time spent in common/sampling
        LOG_INF("%s:    sampling time = %10.2f ms\n", __func__, t_sampling_ms);
        LOG_INF("%s:    samplers time = %10.2f ms / %5d tokens\n", __func__, data.t_sample_ms, data.n_sample);
    }

    if (ctx) {
        auto & data = data_ctx;

        data = llama_perf_context(ctx);

        const double t_end_ms = 1e-3 * ggml_time_us();

        const double t_total_ms = t_end_ms - data.t_start_ms;
        const double t_unacc_ms = t_total_ms - (t_sampling_ms + data.t_p_eval_ms + data.t_eval_ms);
        const double t_unacc_pc = 100.0 * t_unacc_ms /  t_total_ms;

        LOG_INF("%s:        load time = %10.2f ms\n", __func__, data.t_load_ms);
        LOG_INF("%s: prompt eval time = %10.2f ms / %5d tokens (%8.2f ms per token, %8.2f tokens per second)\n",
                __func__, data.t_p_eval_ms, data.n_p_eval, data.t_p_eval_ms / data.n_p_eval, 1e3 / data.t_p_eval_ms * data.n_p_eval);
        LOG_INF("%s:        eval time = %10.2f ms / %5d runs   (%8.2f ms per token, %8.2f tokens per second)\n",
                __func__, data.t_eval_ms, data.n_eval, data.t_eval_ms / data.n_eval, 1e3 / data.t_eval_ms * data.n_eval);
        LOG_INF("%s:       total time = %10.2f ms / %5d tokens\n", __func__, (t_end_ms - data.t_start_ms), (data.n_p_eval + data.n_eval));
        LOG_INF("%s: unaccounted time = %10.2f ms / %5.1f %%      (total - sampling - prompt eval - eval) / (total)\n", __func__, t_unacc_ms, t_unacc_pc);
        LOG_INF("%s:    graphs reused = %10d\n", __func__, data.n_reused);

        common_memory_breakdown_print(ctx);
    }
}

struct llama_sampler * common_sampler_get(const struct common_sampler * gsmpl) {
    if (!gsmpl) {