Neural programs
Typed Formula programs, named resource graphs and configured host integration, with explicit parameter and state ownership.
Program guideThe project behind the guides
Release and support.
The public package, implementation revision, component lifecycle and qualified workflow boundaries.
01 / PUBLIC PACKAGE
ARTI 3.3.0
The guides use a pinned package snapshot. Individual neural components retain their own contract versions and lifecycle status.
python -m pip install "arti-fit==3.3.0"- Distribution
- arti-fit
- Python import
- arti
- Requirements
- Python ≥ 3.10 · PyTorch ≥ 2.2
- Source revision
- dcb321d6e108d0167c5fbca0a77501f34b154af2
02 / INCLUDED SURFACE
From numerical programs to native tasks.
These are documented entrances in the pinned public package. Select the scope your application uses and verify its own model and data.
Native task lifecycle
Custom scalar objectives, immutable assets, supported checkpoint recovery, committed prediction, independent evaluation and selected result export.
Task walkthroughNative JAX / Flax
Optional native execution of the shared bounded/static IR, with array PyTrees and separate init/apply state contracts.
Backend scope03 / SOURCE AND SCOPE