Getting started

ARTI is distributed on PyPI as arti-fit. The Python import remains arti.

Requirements

• Python 3.10, 3.11, or 3.12
• PyTorch 2.2 or newer
• A CPU or CUDA PyTorch build selected by your application

Install

pip install arti-fit

With uv:

uv add arti-fit

ARTI 1.7.0 is a Stable Candidate. The supported 1.x surface is frozen for compatibility verification, but this release does not yet carry an LTS commitment. UnFold is supported core; FusionPulse and inspectable Web runs remain Alpha.

Use ARTI as a layer

import arti
import torch

layer = arti.nn.Layer(dim=32)
x = torch.randn(4, 16, 32)
mask = torch.ones(4, 16, dtype=torch.bool)

out = layer(x, mask=mask)

assert out.y.shape == (4, 16, 32)
assert out.pooled.shape == (4, 32)
print(out.diagnostics.keys())

For [B, D] inputs, ARTI treats each row as a single token and restores the original rank on output.

Enable capabilities explicitly

recall_layer = arti.nn.Layer(dim=32, profile="recall")
multisource = arti.nn.Layer(dim=32, profile="multisource", coord_dim=4)

Start with the smallest profile that matches the structure your data actually carries.