Method load_model

src/common/AutoModel/modeling_gemma4e.cpp:415–444

Gemma4e::Gemma4e(xrt::device* npu_device_inst) : AutoModel(npu_device_inst, "Gemma4e") {}

void Gemma4e::load_model(std::string model_path, json model_info, int default_context_length, bool enable_preemption) {
    
    this->_shared_load_model(model_path, model_info, default_context_length, enable_preemption);
    
    this->q4nx = std::make_unique<Q4NX>(this->model_path);
    this->lm_engine = std::make_unique<gemma4e_npu>(*this->lm_config, this->npu.get(), this->MAX_L);

    this->lm_engine->load_weights(*this->q4nx);
    //free the q4nx
    this->q4nx.reset();
    this->lm_engine->clear_context();
    this->setup_tokenizer(model_path);
    this->sampler.reset();

    this->enable_tool = (model_info["size"] > 800000000)? true : false;

    sampler_config config;
    config.top_k = 64;
    config.top_p = 0.95;
    config.min_p = 0.0;
    config.temperature = 1.0;
    config.rep_penalty = 1.0;
    config.freq_penalty = 1.0;
    config.pre_penalty = 1.0f;

    this->set_sampler(config);
    for (size_t i = 0; i < PROFILER_TYPE_NUM; i++) {
        this->profiler_list[i].reset();
    }
}

void Gemma4e::setup_tokenizer(std::string model_path) {
    auto tokenizer_config = this->_shared_setup_tokenizer(model_path);