Skip to content

atomicarchitects/tinye3nn

Repository files navigation

tinye3nn

Caution

API is experimental and likely to break

License: MIT

e3nn + tinygrad

Installation

> python -m pip install .

Amuze-Bouche

import tinye3nn

# Create Irreps as per e3nn convention (currently not supporting other input formats except strings)
x_irreps = tinye3nn.so3.Irreps("32x0e + 32x1o + 32x2e")
y_irreps = tinye3nn.so3.Irreps("0e + 1o + 2e")


# Initialize arrays and move them to the GPU
x = x_irreps.randn()
y = y_irreps.randn()

# Initialize TP + Linear module with the option to specify the device
mod = tinye3nn.nn.TensorProductLinear(x_irreps, y_irreps)

# Run the model :)
out = mod(x, y)

Acknowledgement

  • e3nn
  • e3nn-jax
  • e3x
  • tinygrad

Releases

No releases published

Packages

No packages published

Languages