Hardware and framework version tracking in MLOps - Practice Problems & Coding Challenges

Challenge - 5 Problems

🧠 Conceptual

intermediate

Why track hardware and framework versions in MLOps?

Which of the following is the main reason to track hardware and framework versions in an MLOps pipeline?

A To reduce the size of the trained model files

B To speed up the training by using older hardware versions

C To ensure model reproducibility and consistent performance across environments

D To avoid using any version control system

💡 Hint

Think about why knowing exact versions helps when running the same model again.

Tracking hardware and framework versions helps reproduce results exactly and avoid unexpected differences in model behavior.

💻 Command Output

intermediate

Output of hardware info command

What is the output of the following Linux command used to check GPU info?

nvidia-smi --query-gpu=name,memory.total --format=csv,noheader

A SyntaxError: invalid command

B Tesla V100, 16160 MiB

C No GPUs found

D Tesla V100 16160

💡 Hint

The command queries GPU name and total memory in CSV format without header.

The command outputs GPU name and memory separated by a comma, without header lines.

❓ Configuration

advanced

Correct YAML snippet for framework version tracking

Which YAML snippet correctly specifies TensorFlow version 2.12.0 and CUDA version 11.8 for an MLOps environment config?

A

framework:
  name: tensorflow
  version: 2.12.0
hardware:
  cuda_version: 11.8

B

framework:
  name: tensorflow
  version: 'tensorflow-2.12.0'
hardware:
  cuda: 11.8

C

framework:
  tensorflow_version: 2.12.0
hardware:
  cuda_version: 11.8

D

framework:
  name: tensorflow
  version: 2.12
hardware:
  cuda_version: 11.8

💡 Hint

Look for correct keys and exact version strings.

Option A uses clear keys and exact version strings matching common config standards.

❓ Troubleshoot

advanced

Why does model training fail after framework upgrade?

You upgraded TensorFlow from 2.11 to 2.12 on your training server. Suddenly, your model training script fails with an error about missing attributes. What is the most likely cause?

A The CUDA version was downgraded accidentally

B The GPU hardware is incompatible with TensorFlow 2.12

C Your Python version is too new for TensorFlow 2.12

D The new TensorFlow version changed or removed some APIs your code uses

💡 Hint

Think about what changes between framework versions can break code.

Framework upgrades can change APIs, causing code that depends on old APIs to fail.

🔀 Workflow

expert

Best workflow to track hardware and framework versions automatically

Which workflow best automates tracking hardware and framework versions during model training in an MLOps pipeline?

A Use a script that logs GPU info and framework version at training start, saving to experiment metadata

B Only track framework version in code comments

C Manually write hardware and framework versions in a text file before training

D Ignore hardware info and rely on cloud provider defaults

💡 Hint

Automation and metadata logging are key for reliable tracking.

Automated scripts that log hardware and framework info into experiment metadata ensure accurate and consistent tracking.