File size: 2,600 Bytes
0ba9d09
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
# Replacing the current Hugging Face repository

We rebuilt `ChatterjeeLab/CIS6270` from the final lecture notes, and everything in this tree is
new except the four files listed under "keep" below.

## Keep from the current repository

| File | Why |
| --- | --- |
| `lecture_2/flow_unet_mnist.pt` | The published five-epoch checkpoint. `lecture_2/flow_matching_mnist.py` still defines `FlowUNet` with the same state-dict keys, so the file loads unchanged. Move it to `lecture_2/`. |
| `lecture_2/mnist_selected_examples.png` | The selected generations in the Lecture 2 guide. |
| `lecture_2/mnist_selected_trajectories.png` | The noise-to-image trajectories in the same guide. |
| `LICENSE` | Unchanged, MIT. |

We link the overview figure at the top of the README from the Hugging Face CDN and keep no copy in
the tree, so it survives the replacement on its own.

## Delete

Everything else. The old tree held 67 files per lecture folder of which most were generated:

- `run.py`, `run_all.py`, `lecture_core.py`, `common.py`, `diffusion.py`, `flows.py`, `examples/`.
  We replace the dispatcher with one file per framework, so you can read a method in one place.
- `verified_examples/` in lectures 4 to 7, about 300 JSON files, several over 200 KB. We now
  reproduce a seeded run by running the script, and each lecture's README lists the numbers a
  default run produces.
- `lecture_6/EQUATION_READINGS.md` (138 KB) and `lecture_7/lecture7_ot_sbm_notes.md` (148 KB),
  both of which you can read in the book.
- `MATHEMATICS.md`, `SOURCES.md`, `SLIDE_CODE_MAP.md`, `source_manifest.json`,
  `paper_manifest.json`, `lecture7_examples.ipynb`. The slide maps pointed at the decks as they
  were before the October rewrites.
- `lecture_4/data/dna_train.tsv` and `lecture_5/data/dna_train.tsv`, which were byte-identical
  copies of each other. `cis6270.data.synthetic_dna` generates the same data in one place, and
  `cis6270.data.write_dna_tsv` writes it out in the old format for anyone who wants the file.
- `lecture_3/esm2_example.csv`, `GUIDANCE_NOTES.md`, and the per-lecture `requirements.txt` files.

## What changed for a reader

- You now have a Lecture 1, covering the mathematical foundations chapter.
- `pip install -e .` puts `cis6270` on the path; otherwise we run from the repository root with
  `PYTHONPATH=.`.
- The shared flags are the same everywhere: `--steps`, `--batch-size`, `--lr`, `--width`,
  `--length`, `--samples`, `--sample-steps`, `--seed`, `--device`, `--out`, `--quiet`.
- `python -m pytest tests -q` runs 484 tests, which assert the numbers the notes work by hand.