Each team should be with 2-3 students, one student should be designated as the leader. A student may work on 1-2 projects (at most one as leader).
10/8: 1 page pre-proposal due.
10/11: initial feedback & approval.
10/17: Pre-proposal presentation and team formation.
10/18: Leaders decide on team members.
10/20-24: Feedback sessions with instructor
10/19: Leaders/members finalize team (must discuss with instructor to change team after this date)
10/28: 3-page proposal due with detailed execution plan.
12/5: Project due
Code & final report due @ 11:59pm
Final's week
Presentation and demo
Presentation Schedule: (30 min presentation + 10 min Q&A)
1:05-1:45pm ADPMoE: Adaptive Parallelism for Mixture-of-Expert Models Inference
1:50-2:30pm Understanding the Performance Bottlenecks of Block-wise Diffusion Language Model Inference
2:35-3:15pm Toward Fair and Efficient Multi-Tenant LLM Serving: Non-Preemptive Prefill and Preemptive Decoding
3:15-3:55pm UniForward: Breaking Synchronization Barriers in Batched Speculative Decoding
3:55-4:20pm Work on reviews
Submit your reviews by Dec 18 11:59 pm. You may submit reviews even if it's not assigned to you.
Every student present their pre-proposal ideas.
Each student present 7 mins, with 3 mins Q&A. Total 10 mins allocated. Unplug laptop at 9 mins mark, so that the next presenter can prepare.
Presentation should include
Motivation (why this is an important problem?)
State-of-the-art (related work, baseline)
Key approach
Anticipated benefits
After about 2 hours, students express their interests on other people's projects via Google form (10 mins, during break).
The result of the Google form will be displayed on screen.
Students talks to the presenters to have further discussion, update the form as their interests change.
Pre-proposal document. 1 page US letter, 11 pt or larger. 1 inch margin for all sides.
Full-proposal document. 3 page US letter, 11 pt or larger. 1 inch margin for all sides.
Code: submit your implementation on Github classroom.
Report: 4-5 page report, describing motivation, design, implementation, and evaluation.