tensorgami 0.1.0__tar.gz

tensorgami-0.1.0/.gitignore

@@ -0,0 +1,20 @@
+# Python
+__pycache__/
+*.py[cod]
+*.egg-info/
+build/
+dist/
+
+# Environments
+.venv/
+
+# Tooling caches
+.pytest_cache/
+.ruff_cache/
+.coverage
+htmlcov/
+
+# OS / editors
+.DS_Store
+.vscode/
+.idea/

tensorgami-0.1.0/CHANGELOG.md

@@ -0,0 +1,54 @@
+# Changelog
+
+All notable changes to Tensorgami. The format follows
+[Keep a Changelog](https://keepachangelog.com/en/1.1.0/); versions follow
+[Semantic Versioning](https://semver.org/) (0.x: the API may still change between minor
+versions).
+
+## [Unreleased]
+
+## [0.1.0] — 2026-10-13
+
+The first release: a PyTorch model to a figure in one call.
+
+### Drawing
+
+- `tg.draw(model, example_input, …)` traces a model once (no gradients, eval mode; weights,
+  training flags and the random state are left as they were) and returns a `tg.Figure`: shown
+  inline in Jupyter, saved on request as PNG, SVG, PDF, JPG or TIFF (`fig.save`), drawn into
+  a panel of your own matplotlib figure (`fig.draw_on(ax)`).
+- Levels of detail: `level=0` (system) … `3` (every operation), by number or name.
+- Views: `block` (boxes and wires), `unrolled` (recurrence as a time × layer grid),
+  `volume` (CNN layers as tensor volumes), `cell` (an LSTM / GRU cell as its gates), and
+  `auto` (unrolled for traced loops over time, block otherwise).
+- Encoder–decoders with nested skips (U-Net, pix2pix) fold into a U; `u_levels` limits how
+  many levels are drawn.
+- Figures wider than `wrap_width` (1100 px by default) fold into rows that read like lines of
+  text, frames included: a module too wide for the page wraps inside its frame. A residual
+  block is never split.
+- Several models in one figure (`tg.draw({"generator": G, "discriminator": D}, …,
+  links=[…])`) and annotations: `tg.Loss` (with training arrows), `tg.Input`, `tg.Note`.
+- Models on PyTorch's `meta` device (`model.to("meta")`) are traced without allocating
+  memory for weights or activations (#2, @HarrisonKramer).
+- `overrides` set the label, category or leaf-ness of a module; clear errors before the
+  model runs (misspelt options, unknown styles, levels, views, references).
+
+### Styles
+
+- `default`, `origami` (upright cards left → right) and `origami-extended` (flat cards, top →
+  bottom); a style is one TOML file, and your own can be passed by path.
+
+### Examples
+
+- Notebooks, already run: AlexNet, ResNet-18/34/50, pix2pix, and a tour of one model per
+  family; `examples/alexnet.py` and `examples/resnet.py` define the models as torchvision
+  does, without needing it.
+
+### Under the hood
+
+- Importer (PyTorch tracing), framework-neutral IR, transforms, layered layout with block
+  layouts (unrolled grid, U-shape, LSTM cell), SVG and matplotlib renderers.
+- An acceptance gallery (`docs/gallery/`) of every model family in every view and style,
+  checked against committed SVGs and layout invariants by the test suite.
+- Development on Windows: `uv sync` (CPU PyTorch) and the tests work there too (#1,
+  @HarrisonKramer).

tensorgami-0.1.0/CITATION.cff

@@ -0,0 +1,16 @@
+cff-version: 1.2.0
+message: "If you use Tensorgami for figures in your work, please cite it as below."
+title: "Tensorgami: draw the architecture and information flow of PyTorch models"
+type: software
+authors:
+  - family-names: "Utina"
+    given-names: "Andrei"
+version: "0.1.0"
+date-released: "2026-10-13"
+license: MIT
+repository-code: "https://github.com/AndreiUtina/TensorGami"
+keywords:
+  - pytorch
+  - neural networks
+  - architecture diagrams
+  - visualization

tensorgami-0.1.0/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Andrei Utina
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

tensorgami-0.1.0/PKG-INFO

@@ -0,0 +1,172 @@
+Metadata-Version: 2.5
+Name: tensorgami
+Version: 0.1.0
+Summary: Draw the architecture and information flow of PyTorch models: figures for papers, reports and slides
+Project-URL: Homepage, https://github.com/AndreiUtina/TensorGami
+Project-URL: Repository, https://github.com/AndreiUtina/TensorGami
+Project-URL: Issues, https://github.com/AndreiUtina/TensorGami/issues
+Project-URL: Changelog, https://github.com/AndreiUtina/TensorGami/blob/main/CHANGELOG.md
+Project-URL: Gallery, https://github.com/AndreiUtina/TensorGami/blob/main/docs/gallery/README.md
+Author: Andrei Utina
+License-Expression: MIT
+License-File: LICENSE
+Keywords: architecture,deep learning,diagram,figure,matplotlib,neural network,pytorch,visualization
+Classifier: Development Status :: 3 - Alpha
+Classifier: Framework :: Jupyter
+Classifier: Intended Audience :: Education
+Classifier: Intended Audience :: Science/Research
+Classifier: Operating System :: OS Independent
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3 :: Only
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
+Classifier: Topic :: Multimedia :: Graphics :: Presentation
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Classifier: Topic :: Scientific/Engineering :: Visualization
+Requires-Python: >=3.11
+Requires-Dist: matplotlib>=3.7
+Provides-Extra: torch
+Requires-Dist: torch>=2.2; extra == 'torch'
+Description-Content-Type: text/markdown
+
+![TensorGami](https://raw.githubusercontent.com/AndreiUtina/TensorGami/main/docs/brand/logo-banner.png)
+
+# Tensorgami
+
+[![CI](https://github.com/AndreiUtina/TensorGami/actions/workflows/ci.yml/badge.svg)](https://github.com/AndreiUtina/TensorGami/actions/workflows/ci.yml)
+[![PyPI](https://img.shields.io/pypi/v/tensorgami)](https://pypi.org/project/tensorgami/)
+[![Python](https://img.shields.io/pypi/pyversions/tensorgami)](https://pypi.org/project/tensorgami/)
+[![License: MIT](https://img.shields.io/badge/license-MIT-blue)](https://github.com/AndreiUtina/TensorGami/blob/main/LICENSE)
+
+Draw the architecture and information flow of PyTorch models, locally, in Python: figures
+for papers, reports and slides. Call `tg.draw` on a loaded model in a script or a notebook,
+see the figure inline in Jupyter, and save it as PNG, SVG or PDF.
+
+![AlexNet in the origami style](https://raw.githubusercontent.com/AndreiUtina/TensorGami/main/docs/gallery/origami/alexnet.svg)
+
+![pix2pix: U-Net generator and PatchGAN discriminator as tensor volumes](https://raw.githubusercontent.com/AndreiUtina/TensorGami/main/docs/gallery/pix2pix-system.svg)
+
+## Install
+
+```sh
+pip install tensorgami
+```
+
+Tensorgami needs Python ≥ 3.11 and matplotlib. To draw a model it runs the model once with
+PyTorch (≥ 2.2), which it does not install itself, so your CPU or CUDA build stays as it is.
+If PyTorch isn't installed yet, install it first
+([pytorch.org](https://pytorch.org/get-started/locally/)), or let pip pick the default build
+with `pip install "tensorgami[torch]"`.
+
+The latest development version: `pip install "git+https://github.com/AndreiUtina/TensorGami"`.
+
+## Quick start
+
+```python
+import torch
+from torch import nn
+import tensorgami as tg
+
+model = nn.Sequential(
+    nn.Conv2d(3, 16, 3, padding=1), nn.ReLU(), nn.MaxPool2d(2),
+    nn.Conv2d(16, 32, 3, padding=1), nn.ReLU(), nn.AdaptiveAvgPool2d(1),
+    nn.Flatten(), nn.Linear(32, 10),
+)
+
+fig = tg.draw(model, torch.randn(1, 3, 32, 32))  # runs the model once, leaves it untouched
+fig                                              # shows inline in Jupyter
+fig.save("model.png", dpi=300)                   # or .svg / .pdf; nothing is saved before
+```
+
+The example input is what the model is called with: a tensor, a tuple of positional
+arguments, or a dict of keyword arguments, including the batch dimension.
+
+## What you can draw
+
+```python
+tg.draw(model, x, level="stages")                  # less detail: 0 system … 3 operators
+tg.draw(model, x, style="origami")                 # or "origami-extended", or a path to a TOML
+tg.draw(model, x, direction="TB", legend=False)    # top → bottom, no legend
+tg.draw(model, x, wrap_width=1600)                 # fold into rows beyond 1600 px (None: never)
+tg.draw(model, x, view="unrolled")                 # recurrence as a time × layer grid
+tg.draw(model, x, view="volume")                   # CNN layers as tensor volumes
+tg.draw(model, x, view="cell")                     # an LSTM / GRU cell as its gates
+tg.draw(unet, x, u_shape=False)                    # keep a U-Net on one line (default: a U)
+tg.draw(model, x, overrides={"classifier.6": {"label": "logits"}})       # rename a box
+tg.draw(vae, x, annotations=[tg.Loss("KL divergence", "mu", "logvar")])  # a loss term
+tg.draw({"generator": G, "discriminator": D}, {"generator": (z, y), "discriminator": (x, y)},
+        links=[("generator", "discriminator.input 0")],     # several models, wired together
+        annotations=[tg.Loss("adversarial", "discriminator", trains=["generator", "discriminator"])])
+```
+
+- **Levels of detail**: the whole system, stages, layers or single operators. Repeated blocks
+  fold into one frame marked ×N, activations fold into the layer before them, and shapes
+  appear on the wires where they change.
+- **Views**: block diagrams (`view="block"`), recurrent networks unrolled over time
+  (`"unrolled"`), CNNs as 3-D tensor volumes (`"volume"`) and recurrent cells as their gates
+  (`"cell"`). Encoder–decoders with nested skips (U-Net, pix2pix) fold into a U.
+- **Styles**: [`default`](https://github.com/AndreiUtina/TensorGami/blob/main/docs/style/style-baseline.md),
+  and [`origami` and `origami-extended`](https://github.com/AndreiUtina/TensorGami/blob/main/docs/style/origami.md)
+  (narrow boxes in the style of the survey figures). A style is a TOML file; your own can be
+  passed by path.
+- **Several models in one figure**: GANs, VAEs and other training set-ups, with links between
+  the models, extra inputs, loss terms with dashed training arrows, and notes.
+
+Model families covered by the tests: CNN (AlexNet, ResNet), LSTM, ConvLSTM, cGAN, VAE,
+Transformer and pix2pix. The [acceptance gallery](https://github.com/AndreiUtina/TensorGami/blob/main/docs/gallery/README.md)
+shows each of them in every view and style. Every figure there is drawn automatically, and
+the tests check it.
+
+## Examples
+
+Notebooks with every option, already run, are in
+[`examples/`](https://github.com/AndreiUtina/TensorGami/blob/main/examples):
+
+- [`alexnet.ipynb`](https://github.com/AndreiUtina/TensorGami/blob/main/examples/alexnet.ipynb):
+  levels, views, styles, layout options, saving, panels;
+- [`resnet.ipynb`](https://github.com/AndreiUtina/TensorGami/blob/main/examples/resnet.ipynb):
+  ResNet-18/34/50, how to read the figure, wide figures folded into rows;
+- [`pix2pix.ipynb`](https://github.com/AndreiUtina/TensorGami/blob/main/examples/pix2pix.ipynb):
+  a U-Net generator, a PatchGAN discriminator and both together;
+- [`tour.ipynb`](https://github.com/AndreiUtina/TensorGami/blob/main/examples/tour.ipynb):
+  one model per family.
+
+The pipeline's stages can also be used one by one, e.g. to inspect what was captured:
+
+```python
+from tensorgami.importers.torch import trace
+from tensorgami.transforms import simplify
+
+graph = trace(model, torch.randn(1, 3, 32, 32))
+print(graph.summary())                     # every operation, its module and shape
+print(simplify(graph, level=2).summary())  # layers: activations folded, repeats ×N
+```
+
+Background: the
+[literature survey](https://github.com/AndreiUtina/TensorGami/blob/main/docs/survey/tensorgami-literature-survey.pdf)
+on how neural-network architectures are drawn, and the
+[architecture](https://github.com/AndreiUtina/TensorGami/blob/main/docs/architecture.md)
+of Tensorgami itself.
+
+## Development
+
+```sh
+git clone https://github.com/AndreiUtina/TensorGami && cd TensorGami
+uv sync                                        # the environment, with CPU PyTorch
+uv run pytest                                  # the tests, including the gallery
+uv run ruff check . && uv run ruff format .    # lint / format
+uv run python docs/gallery/build_gallery.py    # accept changed drawings: rebuild the gallery
+cd examples && uv run --with nbclient --with nbformat --with ipykernel jupyter execute --inplace *.ipynb
+```
+
+Releases are described in [docs/RELEASING.md](https://github.com/AndreiUtina/TensorGami/blob/main/docs/RELEASING.md),
+changes in the [changelog](https://github.com/AndreiUtina/TensorGami/blob/main/CHANGELOG.md).
+
+## License
+
+MIT, see [LICENSE](https://github.com/AndreiUtina/TensorGami/blob/main/LICENSE). The example
+and test models adapted from torchvision and pytorch-CycleGAN-and-pix2pix keep their BSD
+licenses, see [THIRD_PARTY_NOTICES.md](https://github.com/AndreiUtina/TensorGami/blob/main/THIRD_PARTY_NOTICES.md).
+To cite Tensorgami, use [CITATION.cff](https://github.com/AndreiUtina/TensorGami/blob/main/CITATION.cff)
+(GitHub's "Cite this repository").

tensorgami-0.1.0/README.md

@@ -0,0 +1,140 @@
+![TensorGami](docs/brand/logo-banner.png)
+
+# Tensorgami
+
+[![CI](https://github.com/AndreiUtina/TensorGami/actions/workflows/ci.yml/badge.svg)](https://github.com/AndreiUtina/TensorGami/actions/workflows/ci.yml)
+[![PyPI](https://img.shields.io/pypi/v/tensorgami)](https://pypi.org/project/tensorgami/)
+[![Python](https://img.shields.io/pypi/pyversions/tensorgami)](https://pypi.org/project/tensorgami/)
+[![License: MIT](https://img.shields.io/badge/license-MIT-blue)](LICENSE)
+
+Draw the architecture and information flow of PyTorch models, locally, in Python: figures
+for papers, reports and slides. Call `tg.draw` on a loaded model in a script or a notebook,
+see the figure inline in Jupyter, and save it as PNG, SVG or PDF.
+
+![AlexNet in the origami style](docs/gallery/origami/alexnet.svg)
+
+![pix2pix: U-Net generator and PatchGAN discriminator as tensor volumes](docs/gallery/pix2pix-system.svg)
+
+## Install
+
+```sh
+pip install tensorgami
+```
+
+Tensorgami needs Python ≥ 3.11 and matplotlib. To draw a model it runs the model once with
+PyTorch (≥ 2.2), which it does not install itself, so your CPU or CUDA build stays as it is.
+If PyTorch isn't installed yet, install it first
+([pytorch.org](https://pytorch.org/get-started/locally/)), or let pip pick the default build
+with `pip install "tensorgami[torch]"`.
+
+The latest development version: `pip install "git+https://github.com/AndreiUtina/TensorGami"`.
+
+## Quick start
+
+```python
+import torch
+from torch import nn
+import tensorgami as tg
+
+model = nn.Sequential(
+    nn.Conv2d(3, 16, 3, padding=1), nn.ReLU(), nn.MaxPool2d(2),
+    nn.Conv2d(16, 32, 3, padding=1), nn.ReLU(), nn.AdaptiveAvgPool2d(1),
+    nn.Flatten(), nn.Linear(32, 10),
+)
+
+fig = tg.draw(model, torch.randn(1, 3, 32, 32))  # runs the model once, leaves it untouched
+fig                                              # shows inline in Jupyter
+fig.save("model.png", dpi=300)                   # or .svg / .pdf; nothing is saved before
+```
+
+The example input is what the model is called with: a tensor, a tuple of positional
+arguments, or a dict of keyword arguments, including the batch dimension.
+
+## What you can draw
+
+```python
+tg.draw(model, x, level="stages")                  # less detail: 0 system … 3 operators
+tg.draw(model, x, style="origami")                 # or "origami-extended", or a path to a TOML
+tg.draw(model, x, direction="TB", legend=False)    # top → bottom, no legend
+tg.draw(model, x, wrap_width=1600)                 # fold into rows beyond 1600 px (None: never)
+tg.draw(model, x, view="unrolled")                 # recurrence as a time × layer grid
+tg.draw(model, x, view="volume")                   # CNN layers as tensor volumes
+tg.draw(model, x, view="cell")                     # an LSTM / GRU cell as its gates
+tg.draw(unet, x, u_shape=False)                    # keep a U-Net on one line (default: a U)
+tg.draw(model, x, overrides={"classifier.6": {"label": "logits"}})       # rename a box
+tg.draw(vae, x, annotations=[tg.Loss("KL divergence", "mu", "logvar")])  # a loss term
+tg.draw({"generator": G, "discriminator": D}, {"generator": (z, y), "discriminator": (x, y)},
+        links=[("generator", "discriminator.input 0")],     # several models, wired together
+        annotations=[tg.Loss("adversarial", "discriminator", trains=["generator", "discriminator"])])
+```
+
+- **Levels of detail**: the whole system, stages, layers or single operators. Repeated blocks
+  fold into one frame marked ×N, activations fold into the layer before them, and shapes
+  appear on the wires where they change.
+- **Views**: block diagrams (`view="block"`), recurrent networks unrolled over time
+  (`"unrolled"`), CNNs as 3-D tensor volumes (`"volume"`) and recurrent cells as their gates
+  (`"cell"`). Encoder–decoders with nested skips (U-Net, pix2pix) fold into a U.
+- **Styles**: [`default`](docs/style/style-baseline.md),
+  and [`origami` and `origami-extended`](docs/style/origami.md)
+  (narrow boxes in the style of the survey figures). A style is a TOML file; your own can be
+  passed by path.
+- **Several models in one figure**: GANs, VAEs and other training set-ups, with links between
+  the models, extra inputs, loss terms with dashed training arrows, and notes.
+
+Model families covered by the tests: CNN (AlexNet, ResNet), LSTM, ConvLSTM, cGAN, VAE,
+Transformer and pix2pix. The [acceptance gallery](docs/gallery/README.md)
+shows each of them in every view and style. Every figure there is drawn automatically, and
+the tests check it.
+
+## Examples
+
+Notebooks with every option, already run, are in
+[`examples/`](examples):
+
+- [`alexnet.ipynb`](examples/alexnet.ipynb):
+  levels, views, styles, layout options, saving, panels;
+- [`resnet.ipynb`](examples/resnet.ipynb):
+  ResNet-18/34/50, how to read the figure, wide figures folded into rows;
+- [`pix2pix.ipynb`](examples/pix2pix.ipynb):
+  a U-Net generator, a PatchGAN discriminator and both together;
+- [`tour.ipynb`](examples/tour.ipynb):
+  one model per family.
+
+The pipeline's stages can also be used one by one, e.g. to inspect what was captured:
+
+```python
+from tensorgami.importers.torch import trace
+from tensorgami.transforms import simplify
+
+graph = trace(model, torch.randn(1, 3, 32, 32))
+print(graph.summary())                     # every operation, its module and shape
+print(simplify(graph, level=2).summary())  # layers: activations folded, repeats ×N
+```
+
+Background: the
+[literature survey](docs/survey/tensorgami-literature-survey.pdf)
+on how neural-network architectures are drawn, and the
+[architecture](docs/architecture.md)
+of Tensorgami itself.
+
+## Development
+
+```sh
+git clone https://github.com/AndreiUtina/TensorGami && cd TensorGami
+uv sync                                        # the environment, with CPU PyTorch
+uv run pytest                                  # the tests, including the gallery
+uv run ruff check . && uv run ruff format .    # lint / format
+uv run python docs/gallery/build_gallery.py    # accept changed drawings: rebuild the gallery
+cd examples && uv run --with nbclient --with nbformat --with ipykernel jupyter execute --inplace *.ipynb
+```
+
+Releases are described in [docs/RELEASING.md](docs/RELEASING.md),
+changes in the [changelog](CHANGELOG.md).
+
+## License
+
+MIT, see [LICENSE](LICENSE). The example
+and test models adapted from torchvision and pytorch-CycleGAN-and-pix2pix keep their BSD
+licenses, see [THIRD_PARTY_NOTICES.md](THIRD_PARTY_NOTICES.md).
+To cite Tensorgami, use [CITATION.cff](CITATION.cff)
+(GitHub's "Cite this repository").

tensorgami-0.1.0/THIRD_PARTY_NOTICES.md

@@ -0,0 +1,112 @@
+# Third-party notices
+
+Tensorgami itself is released under the [MIT License](LICENSE). The repository also contains
+model definitions adapted from other projects, used as examples and test models. They are not
+part of the installed package (`src/tensorgami`); their licenses are reproduced here, as those
+licenses require.
+
+## AlexNet and ResNet — torchvision
+
+`examples/alexnet.py` defines AlexNet exactly as `torchvision.models.alexnet` does, and
+`examples/resnet.py` ResNet-18/34/50 as `torchvision.models.resnet` does (without its options
+for grouped and dilated convolutions), from <https://github.com/pytorch/vision>, so that
+torchvision is not needed.
+
+```text
+BSD 3-Clause License
+
+Copyright (c) Soumith Chintala 2016,
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+* Redistributions of source code must retain the above copyright notice, this
+  list of conditions and the following disclaimer.
+
+* Redistributions in binary form must reproduce the above copyright notice,
+  this list of conditions and the following disclaimer in the documentation
+  and/or other materials provided with the distribution.
+
+* Neither the name of the copyright holder nor the names of its
+  contributors may be used to endorse or promote products derived from
+  this software without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
+FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+```
+
+## pix2pix — pytorch-CycleGAN-and-pix2pix
+
+`tests/models/pix2pix.py` and `examples/pix2pix.ipynb` contain the U-Net generator and the
+PatchGAN discriminator as defined in `models/networks.py` of the authors' implementation
+(<https://github.com/junyanz/pytorch-CycleGAN-and-pix2pix>).
+
+```text
+Copyright (c) 2017, Jun-Yan Zhu and Taesung Park
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+* Redistributions of source code must retain the above copyright notice, this
+  list of conditions and the following disclaimer.
+
+* Redistributions in binary form must reproduce the above copyright notice,
+  this list of conditions and the following disclaimer in the documentation
+  and/or other materials provided with the distribution.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
+FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+
+--------------------------- LICENSE FOR pix2pix --------------------------------
+BSD License
+
+For pix2pix software
+Copyright (c) 2016, Phillip Isola and Jun-Yan Zhu
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+* Redistributions of source code must retain the above copyright notice, this
+  list of conditions and the following disclaimer.
+
+* Redistributions in binary form must reproduce the above copyright notice,
+  this list of conditions and the following disclaimer in the documentation
+  and/or other materials provided with the distribution.
+
+----------------------------- LICENSE FOR DCGAN --------------------------------
+BSD License
+
+For dcgan.torch software
+
+Copyright (c) 2015, Facebook, Inc. All rights reserved.
+
+Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met:
+
+Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer.
+
+Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution.
+
+Neither the name Facebook nor the names of its contributors may be used to endorse or promote products derived from this software without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+```

tensorgami-0.1.0/pyproject.toml

@@ -0,0 +1,117 @@
+[project]
+name = "tensorgami"
+dynamic = ["version", "readme"]  # version: src/tensorgami/__init__.py; readme: README.md, below
+description = "Draw the architecture and information flow of PyTorch models: figures for papers, reports and slides"
+license = "MIT"
+license-files = ["LICENSE"]
+authors = [{ name = "Andrei Utina" }]
+keywords = [
+    "pytorch", "neural network", "architecture", "diagram", "visualization", "deep learning",
+    "figure", "matplotlib",
+]
+classifiers = [
+    "Development Status :: 3 - Alpha",
+    "Framework :: Jupyter",
+    "Intended Audience :: Science/Research",
+    "Intended Audience :: Education",
+    "Operating System :: OS Independent",
+    "Programming Language :: Python :: 3",
+    "Programming Language :: Python :: 3 :: Only",
+    "Programming Language :: Python :: 3.11",
+    "Programming Language :: Python :: 3.12",
+    "Programming Language :: Python :: 3.13",
+    "Topic :: Scientific/Engineering :: Artificial Intelligence",
+    "Topic :: Scientific/Engineering :: Visualization",
+    "Topic :: Multimedia :: Graphics :: Presentation",
+]
+requires-python = ">=3.11"
+dependencies = [
+    "matplotlib>=3.7",  # figures, PNG, Jupyter display (ADR-0005); the only runtime dependency
+]
+
+[project.optional-dependencies]
+torch = ["torch>=2.2"]  # never a hard dependency: users keep their own CPU/CUDA build (ADR-0006)
+
+[project.urls]
+Homepage = "https://github.com/AndreiUtina/TensorGami"
+Repository = "https://github.com/AndreiUtina/TensorGami"
+Issues = "https://github.com/AndreiUtina/TensorGami/issues"
+Changelog = "https://github.com/AndreiUtina/TensorGami/blob/main/CHANGELOG.md"
+Gallery = "https://github.com/AndreiUtina/TensorGami/blob/main/docs/gallery/README.md"
+
+[dependency-groups]
+dev = [
+    "pytest>=8",
+    "ruff>=0.6",
+]
+docs = [
+    "typst>=0.12",
+]
+torch = [
+    # the newest torch with Intel-Mac wheels is 2.2.2, which needs numpy < 2 (ADR-0006)
+    "torch==2.2.2; sys_platform == 'darwin' and platform_machine == 'x86_64'",
+    "numpy<2; sys_platform == 'darwin' and platform_machine == 'x86_64'",
+    "torch>=2.2; sys_platform != 'darwin' or platform_machine != 'x86_64'",
+]
+
+[build-system]
+requires = ["hatchling", "hatch-fancy-pypi-readme>=24"]
+build-backend = "hatchling.build"
+
+[tool.hatch.version]
+path = "src/tensorgami/__init__.py"
+
+# README.md links to the repository's files by relative paths, which GitHub resolves in a
+# private repository and on every branch; on PyPI they become absolute (images: the raw files)
+[tool.hatch.metadata.hooks.fancy-pypi-readme]
+content-type = "text/markdown"
+
+[[tool.hatch.metadata.hooks.fancy-pypi-readme.fragments]]
+path = "README.md"
+
+[[tool.hatch.metadata.hooks.fancy-pypi-readme.substitutions]]
+pattern = '(!\[[^\]]*\]\()(?!https?://)([^)\s]+)\)'
+replacement = '\1https://raw.githubusercontent.com/AndreiUtina/TensorGami/main/\2)'
+
+[[tool.hatch.metadata.hooks.fancy-pypi-readme.substitutions]]
+pattern = '(\]\()(?!https?://|#)([^)\s]+)\)'
+replacement = '\1https://github.com/AndreiUtina/TensorGami/blob/main/\2)'
+
+[tool.hatch.build.targets.wheel]
+packages = ["src/tensorgami"]
+
+[tool.hatch.build.targets.sdist]  # the package and what describes it; docs and examples stay on GitHub
+include = [
+    "/src/tensorgami",
+    "/README.md",
+    "/LICENSE",
+    "/THIRD_PARTY_NOTICES.md",
+    "/CHANGELOG.md",
+    "/CITATION.cff",
+    "/pyproject.toml",
+]
+
+[tool.uv]
+default-groups = ["dev", "docs", "torch"]
+environments = ["sys_platform == 'darwin'", "sys_platform == 'linux'", "sys_platform == 'win32'"]
+
+[tool.uv.sources]
+torch = [{ index = "pytorch-cpu", marker = "sys_platform == 'linux' or sys_platform == 'win32'" }]  # no CUDA in dev/CI
+
+[[tool.uv.index]]
+name = "pytorch-cpu"
+url = "https://download.pytorch.org/whl/cpu"
+explicit = true
+
+[tool.pytest.ini_options]
+testpaths = ["tests"]
+
+[tool.ruff]
+line-length = 100
+target-version = "py311"
+
+[tool.ruff.lint]
+select = ["E", "F", "I", "B", "UP"]
+
+[tool.ruff.format]
+exclude = ["*.md", "docs/survey/*"]  # prose code blocks and the survey's build scripts stay as written