torch-named-linops 0.3.21

A flexible linear operator abstraction in pytorch.

torch-named-linops

A flexible linear operator abstraction implemented in PyTorch.

Heavily inspired by einops.

Unrelated to the (also good) torch_linops

Selected Feature List

• Dedicated abstraction for naming linear operator dimensions.
• A set of core linops, including:
  • Dense
  • Diagonal
  • FFT
  • ArrayToBlocks[^1] (similar to PyTorch's unfold but in 1D/2D/3D/arbitrary dimensions)
    • Useful for local patch extraction
  • Interpolate[^1] (similar to SigPy's interpolate/gridding)
    • Comes with kaiser_bessel and (1D) spline kernels.
• .H and .N properties for adjoint $A^H$ and normal $A^HA$ linop creation.
• Chain and Add for composing linops together.
• Batch and DistributedBatch wrappers for splitting linops temporally on a single GPU, or across multiple GPUs (via torch.multiprocessing).
• Full support for complex numbers. Adjoint takes the conjugate transpose.
• Full support for autograd-based automatic differentiation.

[^1]: Includes a functional interface and triton backend for 1D/2D/3D.

Installation

Via pip

$ pip install torch-named-linops

From source (recommended for developers)

1. Clone the repo with git clone
2. Run pip install -e . from the root directory.

• Or uv add path/to/cloned/repo

3. Pull upstream changes as required.

Via pip's git integration (deprecated)

Run the following, replacing <TAG> with the appropriate version (e.g. `0.3.7``)

• http version:

$ pip install git+https://github.com/nishi951/torch-named-linops.git@<TAG>

• ssh version:

$ pip install git+ssh://[e-mail]/nishi951/torch-named-linops.git@<TAG>