Function prepare_host_graph_plan

src/framework/core/backend.cpp:399–407

    std::vector<ggml_fp16_t> fp16_values(count);
    ggml_fp32_to_fp16_row(values, fp16_values.data(), static_cast<int64_t>(count));
    ggml_backend_tensor_set(tensor.tensor, fp16_values.data(), 0, count * sizeof(ggml_fp16_t));
}

void write_tensor_f16(const TensorValue & tensor, const std::vector<float> & values) {
    write_tensor_f16(tensor, values.data(), values.size());
}

void write_tensor_bf16(const TensorValue & tensor, const float * values, size_t count) {
    if (tensor.type != GGML_TYPE_BF16) {
        throw std::runtime_error("write_tensor_bf16 requires GGML_TYPE_BF16 tensor");
    }
    if (tensor.shape.num_elements() != static_cast<int64_t>(count)) {