AI Systems · advanced
LoRA & PEFT
Frozen base, low-rank adapters, rank/alpha.
Mental model
LoRA fine-tunes a big model cheaply by adding small trainable matrices alongside the frozen weights. You only update around 1% of the parameters yet get most of the quality of full fine-tuning, and you can swap adapters at inference time.
How to study LoRA & PEFT
Begin by restating the mental model in your own words, then connect it to a concrete system you have built or operated. Name the mechanism, the constraint it addresses, and the trade-off it introduces. Use Stanford CS336 — L15 Alignment: SFT & RLHF (Spring 2025), CS336 Assignment 5 — Alignment & reasoning RL, LoRA: Low-Rank Adaptation (Hu et al.) to check details, but close the source before writing your explanation. Retrieval is the learning step; rereading is only preparation.
Next, compare LoRA & PEFT with Training Data Engineering, RL Alignment (GRPO & Policy Gradient). Ask what changes in correctness, latency, resource use, operability, and failure recovery. Complete LoRA parameter count and preserve the command, input, output, and one failed attempt as evidence. Finish by explaining the idea without jargon to someone who has not studied the track.
Proof of understanding
- Explain the mechanism from first principles and identify the state it reads or changes.
- Give one situation where the concept is the right choice and one where it is not.
- Predict a realistic failure mode before running the drill, then compare the prediction with evidence.
- Connect the result to a roadmap or build artifact instead of treating the concept as isolated trivia.
Learn from primary sources
Practice and explain it back
LoRA parameter count
W is 4096×4096 (frozen). LoRA rank r=8. How many trainable params in A (4096×r) and B (r×4096)?
Expected evidence: 4096·8 + 8·4096 = 65,536 trainable vs 16M full matrix.
Open the interactive drill →Review prompts
- What do rank r and alpha each control in LoRA, and what does r actually limit?
Build evidence
Synthesize: AI Models & Training
Move from transformer foundations through pre-training, fine-tuning, post-training, compression, and evaluation. Produce one working system, benchmark, or evidence-backed design that integrates the path.
- Implements or precisely models the core mechanisms from all three milestones
- Includes at least one injected failure or adversarial case and demonstrates recovery
- Reports quality, latency, resource, reliability, or usability measurements relevant to the domain
- Ships a concise architecture note explaining decisions, trade-offs, and remaining risks
Synthesize: Inference & Serving
Build a production mental model for inference engines, memory, kernels, routing, hardware, and serving economics. Produce one working system, benchmark, or evidence-backed design that integrates the path.
- Implements or precisely models the core mechanisms from all three milestones
- Includes at least one injected failure or adversarial case and demonstrates recovery
- Reports quality, latency, resource, reliability, or usability measurements relevant to the domain
- Ships a concise architecture note explaining decisions, trade-offs, and remaining risks
Synthesize: Multimodal & Spatial Computing
Connect vision, audio, generation, on-device intelligence, robotics, spatial interfaces, and HCI. Produce one working system, benchmark, or evidence-backed design that integrates the path.
- Implements or precisely models the core mechanisms from all three milestones
- Includes at least one injected failure or adversarial case and demonstrates recovery
- Reports quality, latency, resource, reliability, or usability measurements relevant to the domain
- Ships a concise architecture note explaining decisions, trade-offs, and remaining risks