Quiz: Data Preparation
30 questions on dataset formats, quality, and synthetic data.
Question 1
What format is best for single-turn instructions?
A) ShareGPT
B) Alpaca
C) JSON Lines
D) CSV
Show Answer
Correct Answer: B
Alpaca format with instruction/input/output fields is designed for single-turn instruction following.
Question 2
What format supports multi-turn conversations?
A) Alpaca
B) ShareGPT
C) Plain text
D) Markdown
Show Answer
Correct Answer: B
ShareGPT format with conversations array supports multi-turn dialogues.
Question 3
What is more important: data quality or quantity?
A) Quantity
B) Quality
C) They're equally important
D) Neither matters
Show Answer
Correct Answer: B
Quality over quantity—500 high-quality examples often outperform 50,000 low-quality ones.
Question 4
What is tokenizer.apply_chat_template() used for?
A) Creating new tokens
B) Formatting messages in model's expected format
C) Removing special tokens
D) Training the tokenizer
Show Answer
Correct Answer: B
apply_chat_template formats messages according to the model's expected conversation format.
Question 5
What is Self-Instruct?
A) Model training itself
B) Using LLMs to generate instruction-response pairs
C) Self-attention mechanism
D) A type of optimizer
Show Answer
Correct Answer: B
Self-Instruct is a technique where LLMs generate new instruction-response training pairs.
(Questions 6-30 cover: deduplication importance, quality dimensions, recommended dataset sizes, Evol-Instruct, validation strategies, mixing real and synthetic data, format consistency, domain-specific generation, contamination prevention, and practical implementation)
Results
| Score | Level |
|---|---|
| 27-30 | Excellent |
| 21-26 | Good |
| 15-20 | Average |
| Below 15 | Review needed |