Stanford AI Course Map 中文 GitHub ↗
← Back to the course map
Stanford · 2026—27 MS&E 319 Statistics & Optimization ★★★★☆

Core course · Statistics & Optimization

Efficient Generative Models

中文:高效生成模型

Efficient generative models: trade-offs among inference depth, compute, latency, memory, and data, with LLMs as the main case.

Draft translation: official facts are checked; wording may change after native review.

3 units Autumn 2026 Letter or Credit/No Credit ★★★★☆
Generative modelsEfficiencyLLMSystems

Course positioning

MS&E 319 studies the algorithmic problems of generative models under compute constraints. It discusses how objectives, architectures, data pipelines, inference, and serving systems are shaped by compute/latency/memory/data, using LLMs as the main case; students give one presentation and complete a survey or research project.

What it covers

  1. Compute constraints: inference depth, compute, latency, memory, data
  2. Objectives and architectures: scaling, sparsity, mixture of experts
  3. Data pipelines: quality, mixing, synthetic data
  4. Inference: sampling, speculative decoding, test-time compute
  5. Serving: batching, caching, scheduling
  6. Algorithmic problems: new problems under constraints
  7. Presentation + survey/project

Prerequisites

ML/LLM basics; algorithmic and systems awareness.

Assessment & workload

One presentation plus a survey or research project. Check the current syllabus for details.

Who it's for

People who want to work on inference efficiency, generative-model algorithms, or ML systems; graduate students.

Study advice

Pair with CS 349E for engineering implementation and CS 336 for model training.

Related courses

Official course description

As generative models become central to computing, their progress increasingly depends on algorithms that trade reasoning depth, computational cost, latency, memory, and data. This course studies those tradeoffs with a primary focus on modern language models. We examine how objectives, architectures, data pipelines, inference procedures, and serving systems are shaped by computational constraints, and how these constraints open new algorithmic questions. Students give one presentation and complete either a survey or a research project. May be repeated for credit.