Add static auto-layout graph visualization

[lgulich]

Summary

• Replace the old manual Matplotlib/NetworkX graph drawing path with a static layered auto-layout renderer.
• Emit both <graph_name>.svg and <graph_name>.png from leapp.compile_graph(visualize=True).
• Render graph inputs/outputs, node names, backend labels, port names, tensor shapes/dtypes, and semantic kind when available.
• Split the reusable graph model, layout, geometry, SVG, and PNG rendering code into a sibling leapp-visualization pip package.
• Keep visualization enabled by default: leapp now depends on leapp-visualization==0.5.2, resolved locally by uv from packages/leapp-visualization during development.
• Move visualization-only dependencies (fast-sugiyama, Pillow) into leapp-visualization.
• Update docs, examples, and tests for the new SVG+PNG visualization behavior.

Package layout

• packages/leapp-visualization/src/leapp_visualization/: generic visual graph dataclasses, layout, geometry, SVG renderer, PNG renderer, and render_graph(...).
• leapp/leapp_graph/visualization_adapter.py: LEAPP-specific adapter from LEAPP nodes/connections/tensor descriptors into leapp_visualization.VisualGraph.
• leapp/leapp_graph/graph_gui.py: compatibility wrapper used by compile_graph(visualize=True).

Example artifacts

Generated on this branch by leapp.compile_graph(visualize=True):

• SVG: graph_visualization_autolayout_example.svg
• PNG: graph_visualization_autolayout_example.png

GR00T Graph Example

Larger saved LEAPP export rendered with the same static auto-layout path:

• PNG: gr00t_unitree_g1_graph_visualization.png

Verification

• uv run pytest tests/unit_tests/test_graph_visualization_*.py tests/functional_tests/test_annotate.py::TestAnnotateTensor::test_annotate_traced_tensors_diamond_with_feedback tests/test_examples/test_getting_started.py::TestGettingStarted::test_getting_started_execution tests/test_examples/test_feedback_example.py::TestFeedbackExample::test_feedback_example_execution -q -> 38 passed, 4 warnings
• uv build packages/leapp-visualization && uv build -> built both distributions successfully

[frlai]

@lgulich instead of installing from pip, can we configure it to install from source for now? this library will grow with leapp for a bit and I think for now it's ok to essentially think of them as part of the same monorepo. This way we can pass the tests too.

[frlai]

I also just learned (from my agent) that the raster generation is a different process than the vector image generation. I think it makes sense to remove the svg in that case. this lessens the burden to maintain 2 rendering pipelines and make sure they are equivalent.

[lgulich]

Yeah that makes sense, we can remove it On 16 Jul 2026, at 01:42, Frank Lai NV ***@***.***> wrote: [https://avatars.githubusercontent.com/u/220254569?s=20&v=4]frlai left a comment (nvidia-isaac/leapp#8)<#8 (comment)> I also just learned (from my agent) that the raster generation is a different process than the vector image generation. I think it makes sense to remove the svg in that case. this lessens the burden to maintain 2 rendering pipelines and make sure they are equivalent. — Reply to this email directly, view it on GitHub<#8?email_source=notifications&email_token=AFLQOBBFI4NX6WCEAQ7I7KD5FAJFJA5CNFSNUABFM5UWIORPF5TWS5BNNB2WEL2JONZXKZKDN5WW2ZLOOQXTIOJYGYZTGNJXHE42M4TFMFZW63VHNVSW45DJN5XKKZLWMVXHJLDGN5XXIZLSL5RWY2LDNM#issuecomment-4986335799>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AFLQOBFUNM62AQBLPPYH6QT5FAJFJAVCNFSNUABGKJSXA33TNF2G64TZHMYTEMRQGI3TKOBRGY5US43TOVSTWNBXGUYDSMRVHAYTJILWAI>. You are receiving this because you were mentioned.Message ID: ***@***.***>