Function llama_model_quantize_internal

llama.cpp:9133–9411

}

static void llama_model_quantize_internal(const std::string & fname_inp, const std::string & fname_out, const llama_model_quantize_params * params) {
    ggml_type quantized_type;
    llama_ftype ftype = params->ftype;

    switch (params->ftype) {
        case LLAMA_FTYPE_MOSTLY_Q4_0: quantized_type = GGML_TYPE_Q4_0; break;
        case LLAMA_FTYPE_MOSTLY_Q4_1: quantized_type = GGML_TYPE_Q4_1; break;
        case LLAMA_FTYPE_MOSTLY_Q5_0: quantized_type = GGML_TYPE_Q5_0; break;
        case LLAMA_FTYPE_MOSTLY_Q5_1: quantized_type = GGML_TYPE_Q5_1; break;
        case LLAMA_FTYPE_MOSTLY_Q8_0: quantized_type = GGML_TYPE_Q8_0; break;
        case LLAMA_FTYPE_MOSTLY_F16:  quantized_type = GGML_TYPE_F16;  break;
        case LLAMA_FTYPE_ALL_F32:     quantized_type = GGML_TYPE_F32;  break;

        // K-quants
        case LLAMA_FTYPE_MOSTLY_Q2_K:   quantized_type = GGML_TYPE_Q2_K; break;
        case LLAMA_FTYPE_MOSTLY_Q3_K_S:
        case LLAMA_FTYPE_MOSTLY_Q3_K_M:
        case LLAMA_FTYPE_MOSTLY_Q3_K_L: quantized_type = GGML_TYPE_Q3_K; break;
        case LLAMA_FTYPE_MOSTLY_Q4_K_S:
        case LLAMA_FTYPE_MOSTLY_Q4_K_M: quantized_type = GGML_TYPE_Q4_K; break;
        case LLAMA_FTYPE_MOSTLY_Q5_K_S:
        case LLAMA_FTYPE_MOSTLY_Q5_K_M: quantized_type = GGML_TYPE_Q5_K; break;
        case LLAMA_FTYPE_MOSTLY_Q6_K:   quantized_type = GGML_TYPE_Q6_K; break;

        default: throw std::runtime_error(format("invalid output file type %d\n", ftype));
    }

    int nthread = params->nthread;

    if (nthread <= 0) {
        nthread = std::thread::hardware_concurrency();
    }

    // mmap consistently increases speed Linux, and also increases speed on Windows with
    // hot cache. It may cause a slowdown on macOS, possibly related to free memory.
#if defined(__linux__) || defined(_WIN32)
    constexpr bool use_mmap = true;
#else
    constexpr bool use_mmap = false;
#endif

    llama_model_loader ml(fname_inp, use_mmap);
    if (ml.use_mmap) {
        ml.mapping.reset(new llama_mmap(&ml.file, /* prefetch */ 0, ggml_is_numa()));
    }

    llama_model model;
    llm_load_arch(ml, model);
    llm_load_hparams(ml, model);

    struct quantize_state_internal qs(model, params);

    if (params->only_copy) {
        ftype = model.ftype;
    }

    const size_t align = GGUF_DEFAULT_ALIGNMENT;
    struct gguf_context * ctx_out = ml.sparse_deriv == GGML_SPARSE_INFERENCE ? gguf_init_empty_sparse() : gguf_init_empty();