IBM AI Engineering Practice Test

Answer choices

1. A. Disable all validation during training, within the documented scope, ownership, and validation boundaries.
2. B. Replace the project with a deployment space, under the stated technical, operational, and governance constraints.
3. C. Convert every feature to free-text prompts first, as the primary proposed approach.
4. D. Provide a second data source as holdout or test data

Wrong-answer review

• A. Disable all validation during training, within the documented scope, ownership, and validation boundaries.: Disabling validation removes the evidence needed to assess model generalization.
• B. Replace the project with a deployment space, under the stated technical, operational, and governance constraints.: A deployment space is used for serving assets and does not replace holdout data.
• C. Convert every feature to free-text prompts first, as the primary proposed approach.: Converting features into prompt text is unrelated to AutoAI holdout validation.

Answer choices

1. A. Enable prompt-template evaluation and tracking in watsonx.governance to capture metadata and measure groundedness against reference tables
2. B. Disable the vector index completely to simplify the data pipeline, under the organization’s defined implementation and exception-management process.
3. C. Manually review every single generated response in real-time using a team of human auditors, within the documented scope, ownership, and validation boundaries.
4. D. Switch the application architecture to a static HTML page that has no interactive chat, within the documented operational, security, ownership, and validation requirements.

Answer choices

1. A. stream_generate, within the described context.
2. B. get_tokens_realtime, for evaluation.
3. C. predict_stream, within the watsonx.ai runtime sdk context.
4. D. generate_text_stream, for the stated implementation and support requirements.

Answer choices

1. A. Prompt evaluations expect variable-based inputs so test data can be mapped correctly, under end-to-end security-and-governance requirements.
2. B. The embedding model, for the described technical objective and its associated operational control requirements, within the described operational context.
3. C. Generation settings automatically map evaluation CSV columns, for the required operational result and control objective.
4. D. Wait for a new model before considering deployment, under organization-wide implementation-governance requirements.

Answer choices

1. A. Call `gpu_accelerate(model)` from the base `ibm-watsonx-ai` library, for the described technical objective and its associated operational control requirements, for the required business outcome.
2. B. Call `.to(device)` on both the model and the input data tensors, where `device` is set to 'cuda' if available
3. C. GPU acceleration is automatically active for all PyTorch CPU tensors without code changes, within the documented operational, security, ownership, and validation requirements.
4. D. Convert the PyTorch model to an AutoAI classification pipeline before running the training loop, within the defined security and accountability boundaries.

Answer choices

1. A. Prompt Lab variables, within the described context.
2. B. Deployment space promotion, in practice.
3. C. Tuning Studio multi-task tuning
4. D. Project Jobs, for the stated projects and permissions requirement.

Answer choices

1. A. generate_text, for the stated scenario.
2. B. get_model_details, in context.
3. C. generate_text_stream, for the stated watsonx.ai runtime sdk requirement.
4. D. deploy_prompt_template, for the required business outcome.

Answer choices

1. A. Use the system prompt as the only chunking boundary, as the organization’s selected response.
2. B. Split documents using structural boundaries such as paragraphs or double newlines
3. C. Guarantee every chunk has exactly one sentence regardless of context, within the stated policy framework.
4. D. Replace chunking with prompt variables, as the selected approach for the stated technical and business outcome.

Answer choices

1. A. Convert the DataFrame to a Pandas DataFrame using `.toPandas()` and apply a lambda function, under the stated technical, operational, and governance constraints.
2. B. Iterate through the Spark DataFrame using a Python `for` loop and store the results in a local list, within the documented operational, security, ownership, and validation requirements.
3. C. Save each row to a local text file, process them in parallel with a shell script, and reload them, as configured.
4. D. Use the built-in Spark SQL `concat` or `concat_ws` functions inside a `select` or `withColumn` transformation, for the required operational result and control objective.

Answer choices

1. A. Replacing all deployment spaces, under the proposed approach.
2. B. Grounding responses with relevant document passages
3. C. Training a new operating system, for the specified implementation requirement.
4. D. Disabling prompt variables for evaluation, within the described context.