Skip to content
This repository has been archived by the owner on Nov 17, 2023. It is now read-only.

[FFI] Tracking Numpy_extension operators packedfunc based API coverage. #20096

Open
19 of 21 tasks
barry-jin opened this issue Mar 25, 2021 · 0 comments
Open
19 of 21 tasks

[FFI] Tracking Numpy_extension operators packedfunc based API coverage. #20096

barry-jin opened this issue Mar 25, 2021 · 0 comments
Labels
Feature request

Comments

@barry-jin
Copy link
Contributor

barry-jin commented Mar 25, 2021 •
edited
Loading

Description

Adopting packedfunc on some frequently used numpy extension operators, like fully_connected, activation, to reduce invocation overhead. Most of the operators that do not require parameter inputs are covered by _np_imperative_invoke in #20006. The rests are in the following checklist.

checklist:

Implemented

#20087

  • fully_connected
  • activation
  • batch_norm
  • softmax(softmax, log_softmax, masked_softmax, masked_log_softmax)

#20101

  • pick
  • convolution
  • deconvolution

#20102

  • pooling
  • dropout
  • one_hot
  • rnn

#20006

  • batch_flatten

#20110

  • batch_dot
  • broadcast_like
  • arange_like

#20105

  • embedding
  • topk
  • layer_norm
  • leaky_relu

Not Implemented

  • roipooling
  • sequence_mask
Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Assignees
No one assigned
Labels
Feature request
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
1 participant
@barry-jin