Skip to content

Extended module support and experimental receptive field computation
Compare
Choose a tag to compare
@frgfm frgfm released this 03 Aug 22:25
v0.1.1
900eb16
This commit was created on GitHub.com and signed with GitHub’s verified signature. The key has expired.
GPG key ID: 4AEE18F83AFDEB23
Expired
Verified
Learn about vigilant mode.

This release adds support of more modules by the crawler and enables receptive field computation for highway nets.

Note: torchscan 0.1.1 requires PyTorch 1.1 or newer.

Highlights

Modules

In-hook information extraction for supported torch.nn.Module
New

  • Add experimental support of receptive field estimation for the following torch.nn.Module: Identity, Linear, Identity, ReLU, ELU, LeakyReLU, ReLU6, Tanh, Sigmoid, _ConvTransposeNd, _ConvNd, _BatchNorm, _MaxPoolNd, _AvgPoolNd, _AdaptiveMaxPoolNd, _AdaptiveAvgPoolNd, Dropout (#21).

Fixes

  • Fixed transposed convolutions identification (#14)
  • Fixed flops, macs & dmas estimation for pooling operations (#19, #20)

## Crawler
Module hooking agent
New

  • Added an experimental feature receptive_field in the summary function (#21)

Test

Verifications of the package well-being before release
New

  • Updated test for torchscan.modules (#21)

Documentation

Online resources for potential users
Improvements

  • Add documentation website referencing (#13)
  • Updated documentation (#22)

Fixes

  • Fixed documentation deployment (#16)

Others

Other tools and implementations

  • Fixed the conda upload job (#11, #23)
Assets 5
Loading