All pull requests

Pull requests

• Skip the RNG lookup in dropout(A, p) when p == 0

  #770·Parvm1102 opened Aug 30, 2026Contributor·

  ·

• Revamp CUDA extension based on new cuDNN wrappers

  #738·maleadt opened Jul 14, 2026Contributor·

  ·

• Add GPU multi-head attention benchmark (NNlib vs LuxLib vs NNkernels)

  #721·CarloLucibello opened Jun 11, 2026Member·

  ·

• Add scaled_dot_product_attention with explicit head axis

  #719·CarloLucibello opened Jun 10, 2026Member·

  ·

• Test upsample testsuite against multiple eltypes

  #716·CarloLucibello opened Jun 10, 2026Member·

  ·

• Remove cuDNN/MIOpen activation broadcast overloads (fixes #504, #509)

  #686·CarloLucibello opened Jun 7, 2026Member·

  ·
  3

  2

• Extending scatter! to work with CUDA sparse arrays (retry of #648)

  #672·alonsoC1s opened Jan 27, 2026Contributor·

  ·
  4

• Add batched_mul support for oneAPI

  #644·AntonOresten opened Aug 17, 2025Member·

  ·
  2

• Fixed the lpnormpool implementation

  #639·IBArbitrary opened Jun 6, 2025·

  ·
  4

• Add TeLU activation functions telu and telu_fast

  #622·zengmao opened Jan 7, 2025·

  ·
  11

• Add a switch to turn slow paths into errors

  #612·mcabbott opened Nov 8, 2024Member·

  ·
  1

  2

• Start replacing tests with DifferentiationInterface

  #570·gdalle opened Mar 26, 2024·

  ·
  3

• Possible way to implement a LoopVectorization extension for conv2d & meanpool2d & activations

  benchmark
  #540·jonas208 opened Sep 26, 2023·

  ·
  22

• Add EnzymeRules for batchnorm

  Enzyme
  #537·wsmoses opened Sep 25, 2023Contributor·

  ·
  4

• [Dummy] Slow down relu intentionally for testing benchmarking tools

  benchmark
  #527·skyleaworlder opened Aug 11, 2023Contributor·

  ·
  3

• Using AbstractFloat instead of Float64 in warning check for slow conv

  #475·gabrielpreviato opened Feb 18, 2023Contributor·

  ·
  5

• Add norm functions

  documentationenhancement
  #452·ToucheSir opened Jan 3, 2023Member·

  ·
  26

• add dims keyword to gather

  #448·CarloLucibello opened Dec 26, 2022Member·

  ·

• permit NNlibCUDA to use Float16

  enhancement
  #363·bjarthur opened Nov 16, 2021Contributor·

  ·
  5

• add topk

  enhancement
  #353·CarloLucibello opened Sep 30, 2021Member·

  ·
  1

  9

• add groupcount overload for DepthwiseConvDims

  #336·DhairyaLGandhi opened Jul 17, 2021Member·

  ·
  2

• Add benchmarks, part 1

  #304·dfdx opened Mar 22, 2021Contributor·

  ·
  10

• Doc corrections and unicode(theta)

  #184·AdarshKumar712 opened Mar 7, 2020Contributor·

  ·
  8

• Big interface overhaul

  #163·staticfloat opened Jan 31, 2020Contributor·

  ·
  1

• An attempt to combine dense, depthwise and groupwise conv through DenseConvDims

  #146·arhik opened Nov 30, 2019·

  ·
  5