Google Gemma Quiz

Answers and explanations appear as you take the quiz.

1. What does the original Gemma report say the model family is based on?
   • Research and technology used to create Gemini
   • An open-source copy of GPT-4
   • A speech-only model family
   • A model trained only on code
2. What sizes were in the first Gemma release in 2024?
   • 70B and 130B parameters
   • 2B and 7B parameters
   • 175B parameters only
   • 500M parameters only
3. Which Gemma variant is specialized for programming tasks?
   • PaliGemma
   • RecurrentGemma
   • CodeGemma
   • VisionGemma
4. What does 'open-weight' mean for a model like Gemma?
   • The full training dataset and source code are completely open
   • The model has no copyright or license
   • It can only be used for non-commercial research
   • The trained parameters are downloadable for fine-tuning and self-hosting, even if the training data is not fully released
5. Which technique helps the Gemma 2 generation achieve strong quality at small sizes?
   • Knowledge distillation from larger teacher models
   • Running only on supercomputers
   • Removing all attention layers
   • Using no training data at all
6. Which command does the official Ollama Gemma 2 library document for running the model?
   • gemma compile --serve
   • ollama run gemma2
   • python train.py --all
   • git checkout gemma2
7. Does “open-weight” by itself mean Gemma has no use restrictions?
   • Yes; downloaded weights have no terms
   • Yes; only the source code is restricted
   • No; users must follow the applicable Gemma Terms of Use
   • No; the weights cannot be downloaded
8. What should a developer check before selecting a Gemma model for local hardware?
   • Only the Gemma family name
   • Whether a different model ran on a phone
   • The number of downloads on a model hub
   • That version’s size, modality, and hardware requirements