# When to fine-tune interview questions > 6 when to fine-tune interview questions for AI and ML engineer roles, each with a model answer. Canonical: https://llmdojo.dev/interview/topic/when-to-fine-tune Last updated: 2026-10-23 6 questions: 1 junior, 3 mid-level and 2 senior. Taught in [When to Fine-Tune, and a Production-Readiness Plan](https://llmdojo.dev/learn/production-ai-systems/when-to-fine-tune). ## [When should you fine-tune instead of prompting?](https://llmdojo.dev/interview/q/when-should-you-fine-tune-instead-of-prompting) Level: Junior Usually after systematic prompting, examples and retrieval. Fine-tuning needs far more data, compute and expertise. It's for **behaviour** failures prompting can't fix. Set up evaluation first so you know whether it helped. **Diagram: Adapting a model to a task, step by step** 1. AI Engineering: before any adaptation, define your evaluation criteria and pipeline. Every later step is measured with it. 2. Start with prompting alone, then add examples to the prompt: the book suggests anywhere from 1 to 50. 3. If the model fails for lack of information, connect it to data, starting with basic retrieval such as term-based search (Module 05). 4. Then follow the failures: information failures call for better retrieval; behaviour failures (irrelevant, malformatted or unsafe outputs) for fine-tuning. 5. The two can be combined. The book notes better retrieval adds inference complexity, while fine-tuning adds development complexity but leaves inference unchanged. ## [RAG or fine-tuning?](https://llmdojo.dev/interview/q/rag-or-fine-tuning) Level: Mid-level Classify the failures. **Information** failures (missing or outdated facts) call for RAG; **behaviour** failures (wrong format, style or relevance) for fine-tuning. The book: fine-tuning is for form, RAG is for facts. With both, start with RAG. ## [Give good reasons to fine-tune.](https://llmdojo.dev/interview/q/give-good-reasons-to-fine-tune) Level: Mid-level Output **structure** or a rare syntax, like an uncommon SQL dialect; task-specific behaviour; **bias** mitigation; and a **small, specialised model** that's cheaper to serve, possibly distilled. The book's example: Grammarly's fine-tuned Flan-T5 beat a GPT-3 text-editing variant at 60× smaller. ## [Why does full fine-tuning need so much memory, and what does LoRA change?](https://llmdojo.dev/interview/q/why-does-full-fine-tuning-need-so-much-memory-and-what-does) Level: Mid-level Besides the weights, you store gradients and optimizer states for every trained parameter: a 7B model in 16-bit with Adam is 14 GB plus 42 GB. **LoRA** trains small low-rank matrices instead, merged back without inference latency, so training memory shrinks. **Diagram: A LoRA update** 1. The pretrained weight matrix W stays frozen: no gradients, no optimizer state for it. 2. LoRA adds a thin side path. A squeezes the input down to r numbers and B expands it back; only A and B are trained. 3. After training, B·A can be added into W, so the fine-tuned model runs exactly as fast as the original. Hu et al. (2021), LoRA: Low-Rank Adaptation of Large Language Models. Counts are for one 4,096 × 4,096 matrix at rank 8. ## [Make the case against fine-tuning for a team that wants to.](https://llmdojo.dev/interview/q/make-the-case-against-fine-tuning-for-a-team-that-wants-to) Level: Senior It can degrade other tasks; it needs data, expertise, serving and **maintenance** as new base models arrive. Often prompt experiments weren't systematic. The book's cautionary tale: BloombergGPT, about 1.3M A100 hours, was outperformed on financial benchmarks by GPT-4-0314, released the same month. ## [How would you assess whether an AI feature is ready for production?](https://llmdojo.dev/interview/q/how-would-you-assess-whether-an-ai-feature-is-ready-for-production) Level: Senior Check each layer with evidence: evaluated and pinned models, routing and a gateway; independent fallbacks and degraded modes; enforced budgets; metrics, traces and an incident process; guardrails and flagged, canaried releases. Turn gaps into a list ranked by likelihood × harm against effort, each with an owner, a date and the evidence that will show it's done. ## Related - [All interview questions](https://llmdojo.dev/interview)