Function gguf_write_out

subprojects/llama.cpp/ggml/src/gguf.cpp:1366–1402

template <typename writer_t>
static void gguf_write_out(const struct gguf_context * ctx, writer_t & gw, bool only_meta) {
    const int64_t n_kv      = gguf_get_n_kv(ctx);
    const int64_t n_tensors = gguf_get_n_tensors(ctx);

    // write header
    gw.write(GGUF_MAGIC[0]);
    gw.write(GGUF_MAGIC[1]);
    gw.write(GGUF_MAGIC[2]);
    gw.write(GGUF_MAGIC[3]);
    gw.write(ctx->version);
    gw.write(n_tensors);
    gw.write(n_kv);

    // write key-value pairs
    for (int64_t i = 0; i < n_kv; ++i) {
        gw.write(ctx->kv[i]);
    }

    // write tensor info
    for (int64_t i = 0; i < n_tensors; ++i) {
        gw.write_tensor_meta(ctx->info[i]);
    }

    // we require the data section to be aligned
    gw.pad(ctx->alignment);

    if (only_meta) {
        return;
    }

    const size_t offset_data = gw.written_bytes;

    // write tensor data
    for (int64_t i = 0; i < n_tensors; ++i) {
        gw.write_tensor_data(ctx->info[i], offset_data, ctx->alignment);
    }
}

void gguf_write_to_buf(const struct gguf_context * ctx, std::vector<int8_t> & buf, bool only_meta) {
    gguf_writer_buf gw(buf);