Registry / ai-ml / fft-conv-pytorch

fft-conv-pytorch

library1.2.0pypypi✓ verified 91d ago

Implementation of 1D, 2D, and 3D FFT convolutions in PyTorch. Current version 1.2.0, supports padding='same' and half-precision input. Release cadence is irregular; latest updates in 2023.

ai-ml

pip install fft-conv-pytorch

github ↗package ↗

INSTALL

IMPORT

SIG · FFT-CONV-PYTORCH

fft-conv-pytorch

ai-mlpythonv1.2.0

harness data pending

Install & Compatibility

Where this runs

No compatibility data collected yet for this library.

Code

Verified usage

Verified import paths — ran on the pinned version, not inferred.

FFTConv1d

✓ from fft_conv_pytorch import FFTConv1d

Correct import path.

FFTConv2d

✓ from fft_conv_pytorch import FFTConv2d

Correct import path.

FFTConv3d

✓ from fft_conv_pytorch import FFTConv3d

Correct import path.

Create a 2D FFT convolutional layer with 'same' padding and run a forward pass.

import torch from fft_conv_pytorch import FFTConv2d conv = FFTConv2d(in_channels=3, out_channels=16, kernel_size=3, padding='same') x = torch.randn(1, 3, 32, 32) y = conv(x) print(y.shape) # torch.Size([1, 16, 32, 32])

Debug

Known issues

breakingRequires PyTorch >= 1.8 for gradient support on dilated convolutions; earlier versions may break gradient computation.

fix

Upgrade PyTorch to >= 1.8.

affects: 1.1.2+

breakingIn version 1.0.1, einsum usage was removed for efficiency; custom gradients from previous versions may not work.

fix

Ensure your code does not rely on gradient computation via einsum; re-implement custom operations if needed.

affects: >=1.0.1

breakingPadding='same' and half-precision support added in 1.2.0; using these with older versions will raise an error.

fix

Upgrade to 1.2.0 or use padding='valid' and full-precision tensors.

affects: <1.2.0

Errors

Common errors & fixes

ImportError: cannot import name 'FFTConv2d' from 'fft_conv_pytorch'

Incorrect import path or library not installed.

fix

Ensure you have installed fft-conv-pytorch and use: from fft_conv_pytorch import FFTConv2d

RuntimeError: Expected all tensors to be on the same device, but found at least two devices

Input tensor and kernel not on the same device (CPU/GPU).

fix

Move both input and model to the same device, e.g., conv.to(device) and x = x.to(device).

ValueError: padding must be 'valid' or 'same'

Padding argument is not one of the allowed values.

fix

Set padding='valid' or padding='same' (supported from v1.2.0+).

Upgrade

Version history

1.2.0latest on PyPI · released Sep 28, 2023

Audit

Dependencies

torchrequiredCore dependency; the library uses PyTorch tensors and FFT operations.

Agent activity

11 hits · last 30 days

node

OpenAI (training)

Resources

githubhttps://github.com/fkodom/fft-conv-pytorch ↗

packagehttps://pypi.org/project/fft-conv-pytorch/ ↗