Function main

subprojects/llama.cpp/tools/tts/tts.cpp:538–1093

}

int main(int argc, char ** argv) {
    common_params params;

    params.out_file = "output.wav";
    params.prompt = "";

    params.n_predict = 4096;
    params.n_batch   = 8192;
    params.n_ctx     = 8192;

    params.sampling.top_k = 4;
    params.sampling.samplers = { COMMON_SAMPLER_TYPE_TOP_K, };

    if (!common_params_parse(argc, argv, params, LLAMA_EXAMPLE_TTS, print_usage)) {
        return 1;
    }

    const int n_parallel = params.n_parallel;
    const int n_predict  = params.n_predict;

    common_init();

    // init LLM

    llama_backend_init();
    llama_numa_init(params.numa);

    llama_model * model_ttc = NULL; // text-to-codes
    llama_model * model_cts = NULL; // codes-to-speech

    llama_context * ctx_ttc = NULL;
    llama_context * ctx_cts = NULL;

    auto llama_init_ttc = common_init_from_params(params);

    model_ttc = llama_init_ttc->model();
    ctx_ttc   = llama_init_ttc->context();

    if (model_ttc == nullptr || ctx_ttc == nullptr) {
        return ENOENT;
    }

    const llama_vocab * vocab = llama_model_get_vocab(model_ttc);

    params.model = params.vocoder.model;
    params.embedding = true;
    params.n_ubatch = params.n_batch;

    auto llama_init_cts = common_init_from_params(params);

    model_cts = llama_init_cts->model();
    ctx_cts   = llama_init_cts->context();

    if (model_cts == nullptr || ctx_cts == nullptr) {
        return ENOENT;
    }

    std::vector<common_sampler *> smpl(n_parallel);