Method generate

src/common/AutoModel/modeling_qwen3.cpp:102–217

}

std::string Qwen3::generate(chat_meta_info_t& meta_info, int length_limit, std::ostream& os, std::function<bool()> is_cancelled) {
    std::vector<int> sampled_tokens;
    std::string result;
    if (length_limit > 0){
        sampled_tokens.reserve(length_limit);
    }
    else{
        sampled_tokens.reserve(4096);
    }
    assert(this->last_token != -1);
    stop_reason_t reason = EOT_DETECTED;

    this->profiler_list[DECODING_TIME].reset();
    this->profiler_list[TKOEN_DECODE_TIME].reset();
    std::string token_str;
    int sampled_token;
    int last_sampled_token;
    if(!enable_think) {
        this->token_history.push_back(think_start_id);
        this->lm_engine->forward(think_start_id);
        token_str = this->tokenizer->run_time_decoder(think_start_id);

        // \n\n
        this->token_history.push_back(271);
        this->profiler_list[DECODING_TIME].start();
        this->lm_engine->forward(271);
        this->profiler_list[DECODING_TIME].stop(1);
        token_str = this->tokenizer->run_time_decoder(271);
        
        this->token_history.push_back(think_end_id);
        this->profiler_list[DECODING_TIME].start();
        this->lm_engine->forward(think_end_id);
        this->profiler_list[DECODING_TIME].stop(1);
        token_str = this->tokenizer->run_time_decoder(think_end_id);

        this->token_history.push_back(271);
        this->profiler_list[DECODING_TIME].start();
        buffer<bf16> y = this->lm_engine->forward(271);
        this->profiler_list[DECODING_TIME].stop(1);
        token_str = this->tokenizer->run_time_decoder(271);
        sampled_token = this->sampler->sample(y);

        this->total_tokens++;
        meta_info.generated_tokens++;
        last_sampled_token = sampled_token;
        token_str = this->tokenizer->run_time_decoder(last_sampled_token);
        result += token_str;
        os << token_str << std::flush;
    }
    else {
        last_sampled_token = this->last_token;
        this->token_history.push_back(this->last_token);
        if (this->is_normal_token(last_sampled_token) && last_sampled_token != -1){
            std::string token_str = this->tokenizer->run_time_decoder(last_sampled_token);
            result += token_str;
            os << token_str << std::flush;

        }