Method llama_model_loader

llama.cpp:1908–2009

    struct ggml_context * ctx_meta = NULL;

    llama_model_loader(const std::string & fname, bool use_mmap) : file(fname.c_str(), "rb") {
        struct gguf_init_params params = {
            /*.no_alloc = */ true,
            /*.ctx      = */ &ctx_meta,
        };

        ctx_gguf = gguf_init_from_file(fname.c_str(), params);
        if (!ctx_gguf) {
            throw std::runtime_error(format("%s: failed to load model from %s\n", __func__, fname.c_str()));
        }

        n_kv      = gguf_get_n_kv(ctx_gguf);
        n_tensors = gguf_get_n_tensors(ctx_gguf);
        sparse_deriv = gguf_get_sparse_deriv(ctx_gguf);
        fver = (enum llama_fver ) gguf_get_version(ctx_gguf);

        for (int i = 0; i < n_tensors; i++) {
            const char * name = gguf_get_tensor_name(ctx_gguf, i);
            struct ggml_tensor * t = ggml_get_tensor(ctx_meta, name);
            n_elements += ggml_nelements(t);
            n_bytes    += ggml_nbytes(t);
        }

        LLAMA_LOG_INFO("%s: loaded meta data with %d key-value pairs and %d tensors from %s (version %s)\n",
                __func__, n_kv, n_tensors, fname.c_str(), llama_file_version_name(fver));

        // determine file type based on the number of tensors for each quantization and print meta data
        // TODO: make optional
        {
            std::map<enum ggml_type, uint32_t> n_type;

            uint32_t n_type_max = 0;
            enum ggml_type type_max = GGML_TYPE_F32;

            for (int i = 0; i < n_tensors; i++) {
                const char * name = gguf_get_tensor_name(ctx_gguf, i);
                struct ggml_tensor * meta = ggml_get_tensor(ctx_meta, name);

                n_type[meta->type]++;

                if (n_type_max < n_type[meta->type]) {
                    n_type_max = n_type[meta->type];
                    type_max   = meta->type;
                }

                LLAMA_LOG_INFO("%s: - tensor %4d: %32s %-8s [ %s ]\n", __func__, i, name, ggml_type_name(meta->type), llama_format_tensor_shape(meta).c_str());
            }

            switch (type_max) {
                case GGML_TYPE_F32:  ftype = LLAMA_FTYPE_ALL_F32;       break;
                case GGML_TYPE_F16:  ftype = LLAMA_FTYPE_MOSTLY_F16;    break;
                case GGML_TYPE_Q4_0: ftype = LLAMA_FTYPE_MOSTLY_Q4_0;   break;
                case GGML_TYPE_Q4_1: ftype = LLAMA_FTYPE_MOSTLY_Q4_1;   break;
                case GGML_TYPE_Q5_0: ftype = LLAMA_FTYPE_MOSTLY_Q5_0;   break;
                case GGML_TYPE_Q5_1: ftype = LLAMA_FTYPE_MOSTLY_Q5_1;   break;
                case GGML_TYPE_Q8_0: ftype = LLAMA_FTYPE_MOSTLY_Q8_0;   break;
                case GGML_TYPE_Q2_K: ftype = LLAMA_FTYPE_MOSTLY_Q2_K;   break;
                case GGML_TYPE_Q3_K: ftype = LLAMA_FTYPE_MOSTLY_Q3_K_M; break;