EnzymeAD/Reactant.jl

Missing neural network operations

Open

#250 opened on Nov 9, 2024

View on GitHub
 (0 comments) (0 reactions) (0 assignees)Julia (64 forks)github user discovery
good first issue

Repository metrics

Stars
 (336 stars)
PR merge metrics
 (PR metrics pending)

Description

NNlib

  • activations
    • tanh_fast
    • sigmoid
    • sigmoid_fast
  • softmax!
  • logsoftmax!
  • conv / conv! (#258)
  • maxpool / maxpool! (#258)
  • meanpool / meanpool! (#258)
  • batched_mul / batched_mul! (#258)
  • padding
    • pad_constant
    • pad_circular
    • pad_repeat
    • pad_zeros
    • pad_reflect
    • pad_symmetric
  • ∇conv_data / ∇conv_data!
  • ∇conv_filter / ∇conv_filter!
  • gather / gather!
  • scatter / scatter! https://github.com/EnzymeAD/Reactant.jl/pull/1395
  • dot_product_attention
  • pixel_shuffle
  • batchnorm (has a special stablehlo impl) - this needs to wait for a corresponding adjoint on EnzymeJAX end
  • grid_sample / grid_sample!
  • upsampling
    • nearest
    • linear
    • bilinear
    • trilinear
  • fold / fold!
  • unfold / unfold!

Some of these might not even be needed, but we should test the generated IR. Strikethrough denotes we don't need specialized handling for these operations

Feel free to add missing operations to the list

Contributor guide

Research direction
Examine existing implementations of neural network operations like conv or batchnorm in the Reactant.jl codebase. Understand the pattern for registering a new operation and implementing its forward and backward passes. Then implement grid sample/grid sample! following that pattern.
Tech stack
None
Domain
backendmachine learning
Issue type
Feature
DifficultyEstimated implementation difficulty for a new contributor, from 1 for very small changes to 5 for expert-level work.
4
Estimated timeA rough time range for an experienced contributor to investigate, implement, test, and prepare a pull request.
Over 1 week
Activity statusHow available the issue appears right now: fresh, active, stale, blocked, or waiting on maintainer input.
Active
ClarityHow clearly the issue explains the expected change, acceptance criteria, and next step.
Clear
Prerequisites
Julia
Newbie friendlinessA 1-100 score estimating how approachable this issue is for first-time contributors.
30