Function clip_model_load

examples/llava/clip.cpp:454–681

read and create ggml_context containing the tensors and their data

// read and create ggml_context containing the tensors and their data
struct clip_ctx * clip_model_load(const char * fname, const int verbosity = 1) {

    struct ggml_context * meta = NULL;

    struct gguf_init_params params = {
        /*.no_alloc = */ true,
        /*.ctx      = */ &meta,
    };

    struct gguf_context * ctx = gguf_init_from_file(fname, params);
    if (!ctx) {
        throw std::runtime_error(format("%s: failed to load CLIP model from %s. Does this file exist?\n", __func__, fname));
    }

    if (verbosity >= 1) {
        const int n_tensors = gguf_get_n_tensors(ctx);
        const int n_kv = gguf_get_n_kv(ctx);
        const int ftype = get_u32(ctx, KEY_FTYPE);
        const std::string ftype_str = get_ftype(ftype);
        const int idx_desc = get_key_idx(ctx, KEY_DESCRIPTION);
        const std::string description = gguf_get_val_str(ctx, idx_desc);
        const int idx_name = gguf_find_key(ctx, KEY_NAME);
        if (idx_name != -1) { // make name optional temporarily as some of the uploaded models missing it due to a bug
            const std::string name = gguf_get_val_str(ctx, idx_name);
            printf("%s: model name:   %s\n", __func__, name.c_str());
        }
        printf("%s: description:  %s\n", __func__, description.c_str());
        printf("%s: GGUF version: %d\n", __func__, gguf_get_version(ctx));
        printf("%s: alignment:    %zu\n", __func__, gguf_get_alignment(ctx));
        printf("%s: n_tensors:    %d\n", __func__, n_tensors);
        printf("%s: n_kv:         %d\n", __func__, n_kv);
        printf("%s: ftype:        %s\n", __func__, ftype_str.c_str());
        printf("\n");
    }

    // kv
    if (verbosity >= 3) {
        const int n_kv = gguf_get_n_kv(ctx);

        for (int i = 0; i < n_kv; ++i) {
            const char * key = gguf_get_key(ctx, i);

            printf("%s: kv[%d]: key = %s\n", __func__, i, key);
        }
        printf("\n");
    }

    // data
    size_t ctx_size = 0;
    {
        const int n_tensors = gguf_get_n_tensors(ctx);

        for (int i = 0; i < n_tensors; ++i) {
            const char * name = gguf_get_tensor_name(ctx, i);
            const size_t offset = gguf_get_tensor_offset(ctx, i);

            struct ggml_tensor * cur = ggml_get_tensor(meta, name);
            ctx_size += sizeof(struct ggml_tensor) + GGML_OBJECT_SIZE;