ServiceNow and StarCoder Quiz

Answers and explanations appear as you take the quiz.

1. Which two organizations co-led the BigCode project that produced StarCoder?
   • ServiceNow Research and Hugging Face
   • Google and Meta
   • OpenAI and Microsoft
   • Amazon and IBM
2. Which dataset was used for StarCoderBase pretraining?
   • The Pile
   • The Stack
   • Common Crawl
   • LAION-5B
3. What training technique lets StarCoder complete code using context on both sides of the cursor?
   • Reinforcement learning from human feedback
   • Masked language modeling
   • Fill-in-the-Middle
   • Chain-of-thought prompting
4. Roughly how large is the flagship StarCoderBase model?
   • 1.5 billion parameters
   • 70 billion parameters
   • 175 billion parameters
   • 15.5 billion parameters
5. Which attention optimization does StarCoder use to reduce memory and speed up inference?
   • Multi-query attention
   • Sliding-window attention
   • Sparse mixture of experts
   • Linear attention
6. Under what kind of license was StarCoder released to encourage responsible use?
   • GPL v3
   • OpenRAIL-M responsible AI license
   • A proprietary commercial license
   • MIT license
7. What context length did the original StarCoder paper report?
   • 800 tokens
   • 80,000 tokens
   • 8,000 tokens
   • One million tokens
8. How was StarCoder produced from StarCoderBase?
   • It was trained only on image captions
   • It replaced The Stack with a speech corpus
   • It was created by reducing the parameter count to 1.5B
   • It was fine-tuned on 35 billion Python tokens