Schedule

Some of the later dates are not accurate, and are copy-pasted from earlier semesters.

When What
Week 1  
Fri, Sep 5 Topics: Generation hyperparameters; decoding algorithms; prompting with base models. Notes Out: Homework 0: Programming Capabilities
Week 2  
Tue, Sep 9 Topics: Basics of Transformers and GPUs; tensor practice.
Fri, Sep 12 Topics: Chain of thought prompting; introduction to conversational models.
Week 3  
Tue, Sep 16 Class cancelled: Arjun at Lawrence Livermore Laboratory Due: Homework 0: Programming Capabilities. Out: Homework 1: Math Word Problems
Fri, Sep 19 Topics: Numerical gradient descent; Word embeddings and vector similarity.
Week 4  
Tue, Sep 23 Topics: Training basics.
Fri, Sep 26 Topics:
Week 5 TODO: this and below.
Tue, Sep 30 Topics: DeepSeek results
Fri, Oct 3 Topics: Introduction to agents In-Class Notebook No reading this week. Reading: SWE-Bench.
Week 6  
Tue, Oct 7 Topics: The Act pattern In-Class Notebook Due: Math Word Problems
Fri, Oct 10 Topics: Information retrieval. Due: Reading 2. In-class Notebook Reading: s1: simple test-time scaling
Week 7  
Tue, Oct 14 Topics: Neural retrieval. In-class Notebook. Recommended Reading: Speech and Language Processing Chapter 14. Out: Thomas the Travel Agent
Fri, Oct 17 Due: Reading 3. Out: Project Proposal
Week 8  
Tue, Oct 21 Topics: Supervised fine-tuning.
Fri, Oct 24 Topics: More on GRPO. Due: Project Proposal
Week 9  
Tue, Oct 28 Due: Thomas the Travel Agent
Fri, Oct 31 Topics: Speculative decoding and proxy tuning.
Week 10  
Tue, Nov 4  
Fri, Nov 7  
Week 11  
Tue, Nov 11 Veterans Day Out: Obscure Questions
Fri, Nov 14 Topics: High-performance inference. PagedAttention; LLM.int8. Short class: Khoury College PhD Open House
Week 12  
Tue, Nov 18 Topics: Non-Local Control Flow, Containers
Fri, Nov 21 In-class activity: Programming in a language that you do not know.
Week 13  
Tue, Nov 25 Topics: Synthetic data generation. Self-Instruct; Magicoder; MultiPL-T. Due: Obscure Questions
Fri, Nov 28 Thanksgiving Break
Week 14  
Tue, Dec 2 Topics: Humans solve LLM benchmarks: ARC-AGI 1 and Social IQa.
Fri, Dec 5 Last day of class (Fall 2025) Reading: Demystifying CLIP Data and What Matters When Building vision-language models? (Questions)