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.
- In-context learning, Instruction fine-tuning
- Prompt engineering, Pre-training
- Pre-training, Instruction fine-tuning
- Instruction fine-tuning, In-context learning
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:
- Catastrophic loss
- Catastrophic forgetting
- Instruction bias
- Model toxicity
3. Which evaluation metric below focuses on precision in matching generated output to the reference text and is used for text translation?
- HELM
- BLEU
- ROUGE-2
- ROUGE-1
4. Which of the following statements about multi-task finetuning is correct? Select all that apply:
- Performing multi-task finetuning may lead to slower inference.
- 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.
5. "Smaller LLMs can struggle with one-shot and few-shot inference:"
Is this true or false?
- True
- False
6. Which of the following are Parameter Efficient Fine-Tuning (PEFT) methods? Select all that apply.
- Reparametrization
- Subtractive
- Selective
- Additive
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 freezes all weights in the original model layers and introduces new components which are trained on new data.
- LoRA trains a smaller, distilled version of the pre-trained LLM to reduce model size
- LoRA continues the original pre-training objective on new data to update the weights of the original model.
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.
9. "Prompt Tuning is a technique used to adjust all hyperparameters of a language model."
Is this true or false?
- True
- False
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