Module M3B Quiz

Test your understanding of LoRA/QLoRA fine-tuning, when to fine-tune vs. other strategies, the Apple Silicon vs. NVIDIA toolchain split, and the container-as-experiment-record pattern.

Which situations are strong candidates for fine-tuning over RAG or prompting? (Select all that apply)

(select all that apply)

The lesson compares LoRA to sticky notes on a textbook. What does each part of the analogy map to?

Why can you not run a GPU-accelerated QLoRA fine-tune inside a Docker container on an Apple Silicon Mac?

After a successful Axolotl QLoRA run, what does the output directory contain, and how does it relate to the base model?

The lesson says the frozen container image is the reproducibility unit for a fine-tuning experiment. What does pinning an image tag guarantee that pinning a pip requirements.txt alone does not?