Generative AI with LLMs Week 2 quiz

29 Jun 2023

Here are course notes I am taking from the DeepLearning.ai course on Coursera: Generative AI with Large Language Models.

Week 2 quiz

Question 1

Fill in the blanks: ____ involves using many prompt-completion examples as the labeled training dataset to continue training the model by updating its weights. This is different from _______ where you provide prompt-completion examples during inference.

• Prompt engineering, Pre-training
• Instruction fine-tuning, In-context learning
• Pre-training, Instruction fine-tuning
• In-context learning, Instruction fine-tuning

Question 2

Fine-tuning a model on a single task can improve model performance specifically on that task; however, it can also degrade the performance of other tasks as a side effect. This phenomenon is known as:

• Model toxicity
• Catastrophic loss
• Instruction bias
• Catastrophic forgetting

Question 3

Which evaluation metric below focuses on precision in matching generated output to the reference text and is used for text translation?

• HELM
• ROUGE-2
• ROUGE-1
• BLEU

Question 4

Which of the following statements about multi-task finetuning is correct? Select all that apply:

• Multi-task finetuning can help prevent catastrophic forgetting.
• FLAN-T5 was trained with multi-task finetuning.
• Multi-task finetuning requires separate models for each task being performed.
• Performing multi-task finetuning may lead to slower inference.

Question 5

“Smaller LLMs can struggle with one-shot and few-shot inference:”

• True
• False

Question 6

Which of the following are Parameter Efficient Fine-Tuning (PEFT) methods? Select all that apply.

• Additive
• Reparameterization
• Subtractive
• Selective

Question 7

Which of the following best describes how LoRA works?

• LoRA decomposes weights into two smaller rank matrices and trains those instead of the full model weights.
• LoRA trains a smaller, distilled version of the pre-trained LLM to reduce model size
• LoRA freezes all weights in the original model layers and introduces new components which are trained on new data.
• LoRA continues the original pre-training objective on new data to update the weights of the original model.

Question 8

What is a soft prompt in the context of LLMs (Large Language Models)?

• A set of trainable tokens that are added to a prompt and whose values are updated during additional training to improve performance on specific tasks.
• A strict and explicit input text that serves as a starting point for the model’s generation.
• A technique to limit the creativity of the model and enforce specific output patterns.
• A method to control the model’s behavior by adjusting the learning rate during training.

Question 9

“Prompt Tuning is a technique used to adjust all hyperparameters of a language model.”

• True
• False

Question 10

“PEFT methods can reduce the memory needed for fine-tuning dramatically, sometimes to just 12-20% of the memory needed for full fine-tuning.”

Is this true or false?

• True
• False