Skip to content

10. Interview Prep Guide

2-minute script

"I fine-tuned a DeepSeek-R1 distill model on Apple Silicon with MLX LoRA so its chain-of-thought becomes Grug-style — short, telegraphic, still correct. The interesting work is the data pipeline: generate raw traces, compress them with a style guide, validate with a Grug score, regularize SFT to stop prompt leakage, then measure GSM8K tokens, latency, and accuracy. We got large efficiency wins with an accuracy tax, documented across iterations and a small React visualizer."

5-minute script

  1. Motivation: reasoning tokens dominate latency/cost.
  2. Approach: style-transfer SFT, not RL.
  3. Pipeline stages and why each exists.
  4. Leakage failure → regularization fix.
  5. Metrics honesty: efficiency vs GSM8K drop.
  6. Tooling: MLX, HF artifacts, visualizer.
  7. Next: scale to 7B (config already moving).

Likely questions

Area Question Angle
ML Why LoRA not full FT? Memory on Mac; fast iteration
ML Why not DPO/RLHF? Scope; data pipeline first
Data How avoid contamination? Blocklist + fuzzy match
Systems Why MLX? Native Apple Silicon path
Product Who is the user? Researcher/reproducer, not end customers

STAR: prompt leakage

S: Fine-tuned model echoed the Grug system prompt.
T: Stop regurgitation without losing terseness.
A: Mixed negatives, dropout, dual negative prompting.
R: Leakage largely fixed in regularized iteration; documented in report.

Weak spots

  • Accuracy regression on GSM8K
  • No automated Python tests
  • 1.5B reports vs 7B config drift
  • Compression quality depends on external API model

Short answers

Hard? Getting compression to stay logical and stopping leakage.
Improve? Unit-test scorers; lock config to published model; add ARC eval; reduce accuracy tax.
Proud of? Treating data validation and regularization as first-class engineering, not afterthoughts.