Coursework consists of four machine problems (MP0–MP3).
This page contains links to MP documents and their due dates.
MP documents will be linked here as they are released.
| Assignment | Description | Release | Due |
|---|---|---|---|
| Team-up | Form MP teams | Aug 25 | Sep 7 |
| MP0 | Prerequisites (PyTorch / CUDA) | Aug 25 | Sep 15 |
| MP1 | Pre-training | Sep 8 | Oct 6 |
| MP1 CP1 | LLM Architecture | Sep 8 | Sep 22 |
| MP1 CP2 | LLM Training Logic | Sep 22 | Sep 29 |
| MP1 CP3 | Pre-training & Evaluation | Sep 29 | Oct 6 |
| MP2 | Post-training | Oct 6 | Nov 3 |
| MP2 CP1 | Supervised Fine-Tuning (SFT) | Oct 6 | Oct 13 |
| MP2 CP2 | Reinforcement Learning with Verifiable Rewards (RLVR) | Oct 13 | Oct 20 |
| MP2 CP3 | Reinforcement Learning with Human Feedback (RLHF) | Oct 20 | Oct 27 |
| MP2 CP4 | Open-ended Extension | Oct 27 | Nov 3 |
| MP3 | Inference engine | Nov 3 | Dec 10 |
| MP3 CP1 | Inference engine basics | Nov 3 | Nov 10 |
| MP3 CP2 | PagedAttention, chunked prefill, prefix caching | Nov 10 | Dec 1 |
| MP3 CP3 | Inference parallelisms, speculative decoding, FlashAttention, CUDA graph | Nov 19 | Dec 10 |
MPs typically release Tuesday after class; they are due Monday AoE before class.