Fine-Tuning
Free IBM Certified watsonx Generative AI Engineer - Associate practice — 6 questions on Fine-Tuning, with explanations. No sign-up. Full 12-question mixed test →
A team wants to adapt a foundation model in watsonx.ai for both sentiment classification and product-description generation using the same underlying base model instance, while minimizing storage costs and deployment complexity across multiple task-specific adapters. Which approach BEST meets these requirements?
Prompt tuning trains small, task-specific soft-prompt vectors while the base model weights remain frozen and shared, giving low storage overhead and simple multi-task deployment.
During a prompt tuning run in watsonx.ai Tuning Studio, the training loss steadily decreases across epochs while the validation loss begins increasing after epoch 15. What does this pattern most likely indicate, and what is the BEST corrective action?
Decreasing training loss combined with rising validation loss is the classic signature of overfitting; the fix is to reduce training epochs and/or add more diverse training examples.
An engineer needs to tune a foundation model in watsonx.ai to output a fixed set of predefined labels (e.g., 'positive', 'negative', 'neutral') for incoming customer reviews with high consistency. Which Tuning Studio experiment configuration is MOST appropriate?
The Classification experiment type in Tuning Studio is designed for constrained, fixed-label outputs, producing more consistent and exact-label results than open-ended generation.
Which of the following BEST describes the experiment/task types natively supported by watsonx.ai Tuning Studio for prompt tuning foundation models?
watsonx.ai Tuning Studio's prompt tuning experiments are configured as either Classification or Generation task types.
A data scientist is preparing a training file for a watsonx.ai prompt tuning experiment. Which data structure is required for each record in the training/validation dataset?
watsonx.ai Tuning Studio requires each training/validation example as a JSON/JSONL record with 'input' (the prompt) and 'output' (the desired completion) fields.
In watsonx.ai prompt tuning, what happens to the pre-trained foundation model's original weights during the tuning process?
Prompt tuning is a parameter-efficient technique where the base model's weights stay frozen; only the added soft-prompt (virtual token) embeddings are trained.
Ready for the real thing?
The full course has two full-length practice tests, video lessons for every exam domain, hands-on labs and detailed answer explanations.
Start my full course on Udemy →