Metadata-Version: 2.4
Name: 3dsem
Version: 0.1.1
Summary: Install and run 3D semantic segmentation models on your own point clouds
License: MIT
License-File: LICENSE
Requires-Python: >=3.9
Description-Content-Type: text/markdown

# inference_tui

Standalone inference portal for trainer models. Users install a **model**
(weights + its architecture environment, one transaction) from the prefix.dev
channel and run inference locally — no repo checkout, no GUI.

Layout:

- `portal/` — the public, stdlib-only CLI + TUI (this subtree becomes the
  public repo root at publish time).
- `build/` — trainer-side tooling: the lock-driven recipe/metapackage
  generator (`generate.py`), the trainer-runtime packager
  (`build_runtime.py`), and the equivalence check. Reads
  `../trainer_gui/envs/pixi.lock`; never shipped to users.
- `recipes/` + `catalog.json` — generator output, regenerable, gitignored.
  Local channel output goes to `C:\rb\out` (short path — Windows MAX_PATH).
  At release, catalog.json versions are stamped into `portal/catalog.py`.

Pilot: utonia (torch 2.5.0+cu124), model `dales-utonia`.
