| # 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. |
|
|
