← Back to syllabus
Fine-Tuning & RLHF Intuition · Week 8 · Day 6/7
DAY 55 / 210

Intro to Parameter-Efficient LLM Fine-Tuning

This day launches the finetune phase by establishing why full fine-tuning is impractical for StartupTribunal workloads and how PEFT methods enable targeted adaptation without retraining entire models. It sets the foundation for all subsequent days that will actually modify model behavior inside the live Maku pipeline.

45 min target📝 3 quiz Qs

Resources

Deliverable

Journal entry containing a one-paragraph fine-tuning objective for StartupTribunal plus a minimal PEFT config sketch

Quiz · 3 questions

1. Why do most production teams prefer LoRA over full fine-tuning when adapting models for domain-specific chat?

2. Name one concrete risk of using full fine-tuning on the same dataset you already use for RAG retrieval.

3. Draft a single-sentence fine-tuning goal that would make the Maku brief generator more accurate for startup tribunal cases.

Journal