You are working with a large dataset in a GPU-accelerated environment, and one of the columns, revenue, contains numeric values representing the annual revenue for companies. The revenue values are in the billions of dollars.
Which of the following is the most memory-efficient data type for the revenue column in a cuDF DataFrame?

• A. df['revenue'] = df['revenue'].astype('int64')
• B. df['revenue'] = df['revenue'].astype('float64')
• C. df['revenue'] = df['revenue'].astype('uint32')
• D. df['revenue'] = df['revenue'].astype('float32')

Which of the following is the best approach for performing benchmarking and optimizing GPU- accelerated workflows for MLOps using Nvidia technologies?

• A. Rely exclusively on the nvidia-smi tool for monitoring GPU utilization and memory usage across multiple GPUs without making any other performance adjustments.
• B. Use Nvidia's nsight tools to benchmark only the model training phase and ignore the inference phase, as training is the primary bottleneck.
• C. Use TensorRT to optimize deep learning models by converting them into highly optimized inference engines, allowing faster execution with lower latency.
• D. Use Nvidia's nvprof tool to profile GPU resource usage and identify bottlenecks, then adjust the batch size to optimize throughput.

You have a pandas DataFrame with a column containing floating-point numbers, but it takes up too much memory. You want to convert it into a lower-precision type using CuDF or pandas while ensuring computational efficiency.
Which function would you use?

• A. df.to_float16()
• B. df['col'].apply(lambda x: np.float16(x))
• C. df.convert_dtypes()
• D. df.astype('float16')

You are training a convolutional neural network (CNN) model with a large dataset on a single GPU.
The model is consuming too much GPU memory, and training is slow.
Which of the following techniques would help you reduce GPU memory consumption while maintaining or improving the efficiency of training? (Select two)

• A. Use a smaller batch size to reduce the memory requirements per training iteration, allowing for faster training.
• B. Use batch normalization to reduce memory usage by skipping some of the computations during training.
• C. Decrease the size of the model by reducing the number of layers or the number of filters per layer.
• D. Implement mixed precision training to lower memory usage and speed up computation without losing accuracy.
• E. Use automatic differentiation for every operation to save memory during the backpropagation phase.

A data scientist is processing a dataset that is too large to fit into the memory of a single GPU. They decide to use Dask with cuDF to leverage multiple GPUs for accelerated computation.
Which of the following approaches ensures efficient parallelism when working with dask_cudf?

• A. Assign explicit chunks of data to each worker by manually dividing the dataset before passing it to Dask.
• B. Use Dask futures to asynchronously submit GPU tasks and collect results when completed.
• C. Convert the dataset into a Dask cuDF dataframe and use persist() to keep it in GPU memory.
• D. Use a single-threaded scheduler to avoid contention when running on multiple GPUs.