changelogs.md


sovrasov/flops-counter.pytorch

Repository  -  API  -  Source

0.5.2

  • Fix handling of intermediate dimensions in the Linear layer hook.

0.5.0

  • Add per sequential number of parameters estimation.
  • Fix sample doesn't work without GPU.
  • Clarified output in sample.

0.4.0

  • Allocate temporal blobs on the same device as model's parameters are located.

0.3.0

  • Add 1d operators: batch norm, poolings, convolution.
  • Add ability to output extended report to any output stream.

0.2.0

  • Add new operations: Conv3d, BatchNorm3d, MaxPool3d, AvgPool3d, ConvTranspose2d.
  • Add some results on widespread models to the README.
  • Minor bugfixes.

0.1.0

  • Initial release with basic functionality