Issues

• Metal extension: switch to the public MPSGraphs API

  FluxML/NNlib.jl#801 · Filter by author maleadt opened on Oct 16, 2026

• revisit dispatches to cuDNN

  FluxML/NNlib.jl#796 · Filter by author CarloLucibello opened on Oct 14, 2026

• Parallelize all Ops?

  FluxML/NNlib.jl#790 · Filter by author CarloLucibello opened on Oct 13, 2026

• port Mooncake rules here

  FluxML/NNlib.jl#787 · Filter by author CarloLucibello opened on Oct 12, 2026

• Deprecate within_gradient(x) in favor of within_gradient()?

  FluxML/NNlib.jl#780 · Filter by author CarloLucibello opened on Oct 12, 2026

• Add MIOpen (AMDGPU) fast path for batchnorm

  FluxML/NNlib.jl#752 · Filter by author CarloLucibello opened on Aug 3, 2026

• Upstream candidates from LuxLib

  FluxML/NNlib.jl#712 · Filter by author CarloLucibello opened on Jun 10, 2026

• failing enzyme tests on amdgpu and julia v1.12

  FluxML/NNlib.jl#710 · Filter by author CarloLucibello opened on Jun 10, 2026

• Tracking: audit of broken/skipped/disabled tests in the test suite

  FluxML/NNlib.jl#702 · Filter by author CarloLucibello opened on Jun 9, 2026

• Batched dot(x, A, y)

  FluxML/NNlib.jl#641 · Filter by author 3f6a opened on Jul 12, 2025
  4

• gather on a sparse source should return a dense array

  FluxML/NNlib.jl#625 · Filter by author CarloLucibello opened on Jan 13, 2025

• batched_mul launches one kernel per matrix when eltypes differ rather than promoting

  FluxML/NNlib.jl#621 · Filter by author THargreaves opened on Jan 7, 2025
  2

• Exported symbol σ too generic

  breaking
  FluxML/NNlib.jl#604 · Filter by author nomadbl opened on Sep 15, 2024
  2

• batched_transpose with multiple batch dimensions

  FluxML/NNlib.jl#588 · Filter by author AntonOresten opened on May 27, 2024
  3

• [AMDGPU] Reconsider MIOpen softmax once it is fixed

  FluxML/NNlib.jl#586 · Filter by author pxl-th opened on May 19, 2024
  1

• NNlib.gather triggers InvalidIRError when doing Forward over Reverse for hessian computation on GPU

  CUDAhelp wanted
  FluxML/NNlib.jl#547 · Filter by author vpuri3 opened on Oct 26, 2023
  4

• Add warnings/errors on CPU kernels when called with GPU arrays

  good first issue
  FluxML/NNlib.jl#523 · Filter by author ToucheSir opened on Jul 26, 2023

• Inplace version of batched adjoint/transpose

  good first issue
  FluxML/NNlib.jl#502 · Filter by author chengchingwen opened on Jun 24, 2023
  7

• spconv ... sparse convolution

  enhancementhelp wanted
  FluxML/NNlib.jl#493 · Filter by author dorn-gerhard opened on Jun 6, 2023
  1

• type-piracy of materialize

  CUDAgood first issue
  FluxML/NNlib.jl#504 · Filter by author vtjnash opened on Apr 18, 2023
  1

• Migrate to KernelAbstractions.jl for handwritten kernels

  help wanted
  FluxML/NNlib.jl#479 · Filter by author pxl-th opened on Mar 1, 2023
  11

• warn on batched_mul_generic!?

  performance
  FluxML/NNlib.jl#478 · Filter by author chengchingwen opened on Feb 25, 2023
  2

• Reshape of a view breaks batched_mul on GPU (scalar indexing happens)

  CUDA
  FluxML/NNlib.jl#466 · Filter by author reachtarunhere opened on Jan 26, 2023
  3

• Maxpool misbehaving in some edge cases

  bugCUDA
  FluxML/NNlib.jl#508 · Filter by author LlewellynS96 opened on Jan 6, 2023
  5

• Replace the fallback mul! in batchedmul with Octavian.matmul!?

  performance
  FluxML/NNlib.jl#446 · Filter by author chengchingwen opened on Dec 23, 2022
  11