Delete README.md

README.md

@@ -1,94 +0,0 @@
-# CIS 6270 Course Code
-
-![CIS6270_Unified_Generative_Dynamics (1)](https://cdn-uploads.huggingface.co/production/uploads/64cd5b3f0494187a9e8b7c69/myMO3iMdYkquG6D5mhU4S.png)
-
-Here is the code for CIS 6270, Discrete Generative Models, at the University of Pennsylvania. We
-wrote one file per framework so that you can read it beside the lecture notes, with the corruption,
-the loss and the sampler of each method all in that one file, in the notes' own symbols, and a
-comment on almost every line.
-
-You can run everything offline on a laptop CPU in seconds. For `lecture_2/flow_matching_mnist.py` we
-download MNIST, as the name says, and `--dataset toy` keeps even that one offline. In Lecture 3 we
-stay offline by default, and you can ask for MNIST or ESM-2 with `--dataset mnist` or
-`--dataset esm2`.
-
-To get started, we can clone the repository and build an environment.
-
-```bash
-git clone https://huggingface.co/ChatterjeeLab/CIS6270
-cd CIS6270
-python3 -m venv .venv && source .venv/bin/activate
-python -m pip install -r requirements.txt
-
-python lecture_4/mdlm.py          # train masked diffusion on DNA and generate, about 20 seconds
-python -m pytest tests -q         # the notes' worked numbers, checked against the code
-```
-
-On Windows PowerShell, we activate with `.venv\Scripts\Activate.ps1`.
-
-## The lectures
-
-| | Lecture | What the code builds | Guide |
-| --- | --- | --- | --- |
-| 1 | Mathematical Foundations | Every worked example of the chapter, printed beside the chapter's own number | [lecture_1](lecture_1/README.md) |
-| 2 | Flow Matching | Velocity fields and the flow map, the conditional path, the objective, MNIST | [lecture_2](lecture_2/README.md) |
-| 3 | Score Matching and Diffusion | The forward SDE, the three score objectives, DDPM, three samplers, guidance, latent diffusion on protein embeddings | [lecture_3](lecture_3/README.md) |
-| 4 | Discrete Diffusion | Jump processes, MDLM, uniform replacement, D3PM, SEDD, block diffusion, guidance, planners, search | [lecture_4](lecture_4/README.md) |
-| 5 | Discrete Flow Matching | The mixture path and its velocity, correctors, kinetic-optimal paths, Edit Flows, Dirichlet, Fisher and Gumbel geometries, guidance, rectification | [lecture_5](lecture_5/README.md) |
-| 6 | Flow Maps | The three axioms, the three residuals, MeanFlow, Shortcut, consistency, the discrete flow maps, stochastic, expanding and posterior maps | [lecture_6](lecture_6/README.md) |
-| 7 | Optimal Transport and Schrödinger Bridges | Kantorovich and duality, Sinkhorn, displacement interpolation, static and dynamic bridges, DSB, DSBM, SF2M, the discrete bridges, branched and entangled | [lecture_7](lecture_7/README.md) |
-
-## How a file is laid out
-
-Every script follows the shape of `lecture_2/flow_matching_mnist.py`, so once you can read one of
-them you can read them all. We open with a docstring naming the framework and the chapter sections
-it implements, then number the `# %%` sections in the order the lecture develops them, and finish
-with a `main()` that prints a report.
-
-```
-cis6270/           the scaffolding you read once: the data, the four small networks, the training
-                   loop and the sample metrics
-lecture_1/ ... 7/  one file per framework, standalone and runnable
-tests/             the chapters' worked numbers, asserted against the code
-```
-
-Each `lecture_N/` folder stands on its own, and the only import we take from inside the repository
-is `cis6270`, so to read a method you can open its file and stay there.
-
-## The data
-
-| Data | Used by | Download |
-| --- | --- | --- |
-| Synthetic DNA, four letters, a planted ACGT motif, a GC label | Lectures 4 to 6 | no |
-| Four 2-D Gaussians, two moons, a checkerboard | Lectures 2, 3, 6, 7 | no |
-| Four-token phrases such as `red circle moves left` | Lecture 6 | no |
-| MNIST | Lectures 2, 3, with `--dataset mnist` | yes |
-| ESM-2 residue embeddings | Lecture 3, with `--dataset esm2` | yes |
-
-We generate the DNA ourselves, and we keep the two guidance objectives simple enough to compute in a
-line, so a training run finishes while you watch it. As you can see from the samples a run prints,
-you can also check by eye whether the sampler recovered the motif we planted.
-
-## The shared flags
-
-`--steps`, `--batch-size`, `--lr`, `--width`, `--length`, `--samples`, `--sample-steps`, `--seed`,
-`--device`, `--out`, `--quiet`. You can learn them once and use them everywhere. Add `--out <dir>`
-and you get `checkpoint.pt`, `config.json`, `report.json`, `losses.json` and `samples.txt`.
-
-We set the defaults to a short budget, enough to watch a method work in well under a minute. When
-you want a method to look its best, try `--steps 1500 --width 128`.
-
-## Tests
-
-```bash
-python -m pytest tests -q
-```
-
-The suite checks the numbers the notes work out by hand, among them the posterior average of the
-reverse rates on one DNA base, the Sinkhorn iterates at `epsilon = 2`, the cocycle and Jacobian of
-the Gaussian flow, and the conditional velocity of the mixture path. Run it after any edit and you
-can see at once whether a number in the code still agrees with the number in the book.
-
-## License
-
-[MIT](LICENSE). ESM-2 weights come from their original repository under its own terms.