MLS-C01 Practice Test 30 — 15 Questions

Question 1

A company wants to use Amazon SageMaker to host a model that was trained using a custom algorithm. The model artifact is stored in Amazon S3. The company wants to ensure that the endpoint can automatically scale based on the number of incoming requests. Which configuration should the company use?

Choose an answer to begin — your selection is scored in the full session.

Accepted Answer

Create a SageMaker real-time endpoint and configure automatic scaling using a target tracking policy.. A SageMaker real-time endpoint with automatic scaling using a target tracking policy allows the endpoint to dynamically adjust the number of instances based on the incoming request load. This configuration is ideal for hosting a custom algorithm model artifact stored in S3, as it provides low-latency inference and can scale out or in based on a target metric like average CPU utilization or request count per instance.

Answer

Create a SageMaker multi-model endpoint with automatic scaling.

Answer

Use SageMaker Serverless Inference which scales automatically.

Answer

Use SageMaker Batch Transform with a scheduled job.

Question 2

Refer to the exhibit. A data scientist is training a PyTorch model on a SageMaker ml.p3.2xlarge instance (16 GB GPU memory). The training fails with the shown error. Which change should the scientist make to resolve the error?

Accepted Answer

Reduce the batch size in the training script.. The error is an out-of-memory (OOM) condition on the GPU. Reducing the batch size directly decreases the memory footprint per training step, allowing the model to fit within the 16 GB GPU memory of the ml.p3.2xlarge instance. This is the most immediate and effective fix for a GPU memory exhaustion error in PyTorch.

Answer

Increase the number of instances to 2.

Answer

Use SageMaker Managed Spot Training.

Answer

Increase the number of epochs.

Question 3

A data scientist is building a recommendation system using matrix factorization. The dataset has 1 million users and 100,000 items, with a sparse user-item interaction matrix. The scientist wants to minimize training time on Amazon SageMaker. Which algorithm would be most appropriate?

Accepted Answer

Factorization Machines. Factorization Machines (B) are specifically designed for sparse, high-dimensional datasets like the user-item interaction matrix in recommendation systems. They extend matrix factorization by modeling pairwise feature interactions, which is ideal for collaborative filtering tasks. On Amazon SageMaker, the built-in Factorization Machines algorithm is optimized for sparse data and can train efficiently on 1 million users and 100,000 items, minimizing training time compared to general-purpose algorithms.

Answer

Linear Learner

Answer

K-Means

Answer

XGBoost

Question 4

A machine learning engineer is deploying a PyTorch model on SageMaker for real-time inference. The model requires GPU for low latency. Which instance type and configuration should the engineer choose?

Accepted Answer

Deploy to an ml.p3.2xlarge instance with a SageMaker endpoint.. The ml.p3.2xlarge instance type includes a GPU (NVIDIA Tesla V100), which is necessary for low-latency GPU inference. Deploying with a SageMaker endpoint provides a real-time inference endpoint suitable for low-latency requirements. Batch transform is for offline, batch processing and does not provide real-time low latency.

Answer

Deploy to an ml.c5.4xlarge instance with SageMaker batch transform.

Answer

Deploy to an ml.m5.large instance with a SageMaker model endpoint.

Answer

Deploy to an ml.p3.2xlarge instance with SageMaker batch transform.

Question 5

A data scientist needs to select a model training infrastructure that supports distributed training across multiple GPUs and provides automatic model parallelism. Which TWO AWS services should the scientist consider?

Accepted Answer

Amazon EMR. Amazon EMR is correct because it supports distributed training across multiple GPUs using frameworks like TensorFlow, PyTorch, and Apache Spark, and it can automatically handle model parallelism through its integration with Horovod or custom distributed training scripts. Amazon SageMaker is correct because it provides built-in distributed training libraries (e.g., SageMaker Distributed Data Parallel and Model Parallel) that automatically partition model layers across multiple GPUs, enabling efficient training of large models.

Answer

AWS Glue

Answer

AWS Lambda

Answer

Amazon Redshift

Question 6

A research lab is training a large language model (LLM) on SageMaker using PyTorch. The model has 1 billion parameters and does not fit on a single GPU. They have access to a cluster of 16 p4d.24xlarge instances (each with 8 A100 GPUs). They need to train the model with minimal changes to the training script. Which SageMaker feature should they use?

Accepted Answer

SageMaker's model parallelism with automatic partitioning. SageMaker's model parallelism library automatically partitions a large model across multiple GPUs and instances, and it integrates with PyTorch via a simple wrapper (smdistributed.modelparallel.torch) that requires minimal changes to the training script. With 1B parameters and 128 A100 GPUs available, model parallelism is the correct choice because the model does not fit on a single GPU.

Answer

SageMaker's distributed data parallelism with Horovod

Answer

Use SageMaker's built-in BlazingText algorithm

Answer

SageMaker's managed spot training with checkpointing

Question 7

A company uses SageMaker to host a real-time inference endpoint. The endpoint is receiving a large number of requests, but the latency is higher than expected. The data scientist observes that the CPU utilization is low but memory utilization is high. Which action should be taken to reduce latency?

Accepted Answer

Switch to an instance type with more memory or optimize the model to reduce memory footprint.. High memory utilization with low CPU utilization indicates the instance is memory-bound, not CPU-bound. Adding more memory or reducing the model's memory footprint (e.g., quantization, pruning) will alleviate the bottleneck and reduce latency. Scaling out instances or adding vCPUs won't help because the constraint is memory per instance.

Answer

Enable VPC traffic mirroring to diagnose network issues.

Answer

Use an instance type with more vCPUs.

Answer

Increase the number of instances in the endpoint.

Question 8

Which THREE measures can help reduce inference latency for a deep learning model deployed on SageMaker real-time endpoints? (Select THREE.)

Accepted Answer

Enable SageMaker Neo to compile the model.. Option A is correct because SageMaker Neo compiles the trained model into an optimized executable for the target hardware, applying operator fusion, quantization, and framework-specific optimizations that reduce per-inference compute and thus lower latency. Option C is correct because GPU instances provide massive parallel throughput for the matrix/tensor operations in deep learning, so inference completes faster than on comparable CPU instances for most models. Option D is correct because reducing input data size (e.g., lower-resolution images) shrinks the tensors that must be transferred and processed through every layer, directly cutting the compute and memory-bandwidth cost of each forward pass. Option B is not correct because increasing batch size improves throughput (requests per second) but typically increases the latency of an individual inference, since the endpoint must wait to assemble and process a larger batch. Option E is not correct because multi-model endpoints share one instance across multiple models to save hosting cost; they do not inherently reduce the latency of a single model's inference and can even add loading overhead.

Answer

Increase the batch size for inference.

Answer

Use a multi-model endpoint to share the instance.

Question 9

A company uses Amazon SageMaker to host a model for real-time inference. The model is a large ensemble that takes 2 seconds to load into memory. To reduce cold start latency, the data scientist uses SageMaker's managed warm pools. However, they notice that during a sudden traffic spike, new instances still experience high latency. What is the BEST way to ensure consistently low latency for all requests?

Accepted Answer

Configure auto scaling based on the number of active invocations to maintain a buffer of warmed instances.. SageMaker managed warm pools keep instances initialized with the model loaded, but they only help when instances are already in the pool. During a traffic spike, auto scaling must proactively maintain a buffer of warmed instances based on invocation metrics so new capacity is ready before requests arrive. Configuring target tracking on InvocationsPerInstance with a sufficient warm pool size ensures pre-warmed instances absorb the spike without cold starts.

Answer

Use a larger instance type to reduce model loading time.

Answer

Reduce the number of instances to minimize cold start frequency.

Answer

Switch to SageMaker Serverless Inference.

Question 10

Which TWO techniques are used for feature scaling? (Choose 2.)

Accepted Answer

Standardization (Z-score normalization). Standardization (Z-score normalization) (B) is a feature-scaling technique that transforms each feature by subtracting its mean and dividing by its standard deviation, producing values with mean 0 and standard deviation 1, which is essential for algorithms sensitive to feature magnitude such as SVM and k-means. Min-Max scaling (C) is also a feature-scaling technique that rescales each feature to a fixed range, typically [0, 1], using the formula (x - min) / (max - min), preserving the shape of the original distribution. One-hot encoding (A) is a categorical encoding method that creates binary columns, not a scaling technique, so it does not belong. Principal Component Analysis (D) is a dimensionality-reduction technique that projects data onto principal components, not a scaling method. Label encoding (E) assigns integer values to categorical labels and is an encoding technique, not feature scaling.

Answer

One-hot encoding

Answer

Principal Component Analysis (PCA)

Answer

Label encoding

Question 11

A data scientist is analyzing a dataset and notices that the distribution of a continuous feature is heavily right-skewed. Which transformation is most likely to make the distribution more symmetric?

Accepted Answer

Log transformation (natural log). A log transformation (natural log) compresses the upper tail of a right-skewed distribution, pulling in large values and making the distribution more symmetric. It is the standard remedy for positive right-skewed data because it reduces the influence of extreme high values. Min-Max scaling only rescales values to a fixed range without changing the shape of the distribution. One-hot encoding is for categorical variables, and squaring a right-skewed distribution would amplify the skew, not reduce it.

Answer

Min-Max scaling

Answer

One-hot encoding

Answer

Square transformation

Question 12

A company is building a recommendation system using Amazon SageMaker's Factorization Machines algorithm. The dataset includes user IDs, item IDs, and ratings. The data is sparse. Which data format should be used for training?

Accepted Answer

RecordIO-protobuf format with sparse feature vectors.. Factorization Machines (FM) in SageMaker are optimized for sparse, high-dimensional data. The RecordIO-protobuf format allows you to directly specify sparse feature vectors using integer keys and float values, which avoids the memory overhead of dense representations and enables efficient distributed training. This format is the recommended input for SageMaker's built-in FM algorithm.

Answer

CSV format with one row per rating.

Answer

JSON lines format with nested structures.

Answer

Parquet format with columns for each feature.

Question 13

A data scientist is using SageMaker Debugger to monitor a training job. The training loss is not decreasing as expected. Which Debugger feature can help identify the issue?

Accepted Answer

Built-in rules to detect training anomalies. SageMaker Debugger's built-in rules are designed to automatically monitor training jobs for common issues such as vanishing gradients, overfitting, and loss not decreasing. When the training loss plateaus or fails to decrease, a rule like 'LossNotDecreasing' can trigger a CloudWatch alarm or stop the training job, providing immediate insight into the problem without manual inspection of tensors.

Answer

Automatic hyperparameter tuning

Answer

Saving tensors every step

Answer

Deploying a model endpoint for real-time monitoring

Question 14

A company is using Amazon SageMaker to train and deploy a fraud detection model. The model is a gradient boosting machine (GBM) trained on a dataset with 10 million rows and 50 features. The training job runs on an ml.m5.2xlarge instance with 8 vCPUs and 32 GB memory. The training completes successfully, and the model is deployed to a real-time endpoint. After deployment, the inference latency is around 200 ms per request, which is acceptable. However, after a week, the company observes that latency increases to over 1 second during peak hours (12:00-13:00 UTC). CloudWatch metrics show CPU utilization on the endpoint instance reaches 95% during these peaks. The endpoint is configured with a single ml.m5.large instance. The company wants to maintain latency under 500 ms during peak hours without incurring unnecessary cost during off-peak hours. Which solution should the company implement?

Accepted Answer

Configure SageMaker endpoint auto scaling with a target CPU utilization of 70% and a minimum instance count of 1. Configuring auto scaling based on CPU utilization dynamically adds instances during peak hours (when CPU exceeds target) and removes them during off-peak, maintaining latency under 500 ms while minimizing cost. Option A is incorrect because reducing instances to zero and manually launching a new endpoint is not automated, causes downtime, and is impractical. Option B is incorrect because Batch Transform is designed for offline, batch inference, not for real-time requests. Option D is incorrect because replacing the instance with a larger, always-on type (ml.m5.4xlarge) would handle peak load but incur higher cost during off-peak hours, unlike auto scaling which scales in.

Answer

Reduce the number of instances to zero during off-peak hours and manually launch a new endpoint every day at 12:00

Answer

Switch to SageMaker Batch Transform and have the application send requests in batches

Answer

Replace the endpoint instance type with ml.m5.4xlarge to handle peak load

Question 15

A data scientist is analyzing a dataset with missing values. Which technique is most appropriate for imputing missing values in a numerical feature that follows a normal distribution?

Accepted Answer

Mean imputation. Mean imputation is suitable for normally distributed data as it preserves the mean. Median is robust to outliers, not normality. Mode is for categorical data. Standard deviation is not an imputation method. KNN imputation is non-parametric.

Answer

Standard deviation imputation

Answer

Mode imputation

Answer

Median imputation