Course Schedule

Class meets Wednesdays, 12–3pm, in UTA 1.212. Written work is due at 11:59pm on the listed date; in-class assessments and presentations take place during class. Schedule is subject to change; any changes will be announced in class and posted to Canvas.

Wk	Date	Lecture	Lab / Workshop	Due
1	Aug 26	Intro to post-training [slide]	TACC account setup; base vs. instruction-tuned model comparison	HW0 out
2	Sep 2	LLM landscape. Prompting and in-context learning as baselines	Chat templates, in-context learning, and cross-model comparison; project team formation
3	Sep 9	Evaluation I: task metrics, held-out design, what a benchmark can and cannot tell you	Proposal workshop	HW0 due
4	Sep 16	Evaluation II: LLM-as-a-judge, rubric design, judge bias and agreement	Build and validate a judge pipeline
5	Sep 23	Supervised fine-tuning: objective, chat templates, hyperparameters	First SFT run on TACC	Proposal due; HW1 out
6	Sep 30	Parameter-efficient fine-tuning: LoRA, QLoRA	LoRA rank ablation and cost comparison
7	Oct 7	Data for post-training: curation, synthetic generation, distillation	Synthetic data generation and filtering	HW1 due
8	Oct 14	RLHF foundations: Bradley–Terry reward models, PPO	Train and probe a reward model
9	Oct 21	Direct preference optimization (DPO)	Preference labeling and guided DPO workflow	HW2 out
10	Oct 28	RLAIF and Constitutional AI. Scalable oversight and its limits	AI-feedback pipeline; project checkpoint
11	Nov 4	Verifiable rewards and GRPO. Reasoning-oriented post-training	Guided GRPO lab	HW2 due
12	Nov 11	Failure modes: reward hacking, sycophancy, catastrophic forgetting, safety regressions	Concept check (in class, handwritten, closed-device), then project work
13	Nov 18	Project presentations	Presentation feedback and discussion	Final presentation
	Nov 25	No class (Thanksgiving break)
14	Dec 2	Feedback-based project revision and refinement		Final report due Dec 7