FluxML/Optimisers.jl

Issues

• do not accumulate updates in presence of shared gradient

  #193 opened 5 days ago by CarloLucibello
  0

• Consistency in the type behavior of restructure

  #95 opened 2 years ago by ChrisRackauckas
  7

• Adam optimizer can produce NaNs with Float16 due to small epsilon

  #167 opened 7 days ago by pevnak
  3

• mark as public any non-exported but documented interface

  #189 opened 7 days ago by CarloLucibello
  0

• `nothing` does not correspond to updating the state with a zero gradient.

  #140 opened 8 days ago by CarloLucibello
  1

• Utility for walking a tree (e.g. gradients) w.r.t. a model

  #143 opened 8 days ago by darsnack
  7

• Rename or outsource `iswriteable`

  #99 opened 8 days ago by ToucheSir
  9

• Use `OptChain` as an alias for `OptimiserChain`?

  #138 opened 8 days ago by CarloLucibello
  2

• Wrong model update for BatchNorm for some specific synthax

  #123 opened 8 days ago by jeremiedb
  5

• Frozen parameters

  #107 opened 2 years ago by mcabbott
  7

• Split out the `rules.jl` as a sub-package (or a separate package) ?

  #108 opened 8 days ago by chengchingwen
  10

• doc improvement: working with custom model types

  #84 opened 8 days ago by CarloLucibello
  2

• Allow keyword arguments for optimisers

  #74 opened 8 days ago by theabhirath
  2

• tag v1.0.0

  #183 opened 19 days ago by CarloLucibello
  0

• AdamW optimizer implemented incorrectly - weight decay does not incorporate learning rate

  #182 opened 20 days ago by BioTurboNick
  2

• Optimiser state a not moving to GPU

  #179 opened a month ago by vpuri3
  7

• GPU kernels for optimizers

  #178 opened 4 months ago by vpuri3
  2

• Restructure is not type stable but could be made stable?

  #177 opened 5 months ago by Red-Portal
  4

• Grokfast exponential moving average Optimizer

  #176 opened 5 months ago by vpuri3
  0

• Restructure makes a copy

  #146 opened 2 years ago by linusheck
  4

• Document `destructure` handling shared parameters differently to ComponentArrays.jl

  #161 opened a year ago by mcabbott
  0

• `destructure` doesn't work on Dictionaries

  #154 opened a year ago by mcabbott
  1

• Documenter CI is failing

  #169 opened 7 months ago by CarloLucibello
  0

• Type instability in `Flux.setup`

  #162 opened a year ago by Vilin97
  7

• Port over rule changes from Flux

  #38 opened 3 years ago by ToucheSir
  8

• `reset!(optimiser_state)`

  #163 opened a year ago by Vilin97
  2

• Error in `update!` for Metal arrays and Adam optimiser

  #150 opened a year ago by CarloLucibello
  4

• How to handle long compile times?

  #153 opened a year ago by DrChainsaw
  4

• Adam(0) fails

  #119 opened a year ago by cossio
  4

• Implement Lion, up to 5x faster than Adam, and more accurate

  #156 opened a year ago by PallHaraldsson
  7

• `OptimiserChain(..., ClipNorm)` fails on GPU

  #127 opened a year ago by mcabbott
  1

• Interface for gradient accumulation

  #130 opened 2 years ago by chengchingwen
  7

• Optimisers.update fails with gradient of type `CUDA.CUSPARSE.CuSparseMatrixCSC`

  #141 opened 2 years ago by hsseung
  5

• use structs instead of functions for walks in destructure

  #124 opened 2 years ago by CarloLucibello
  4

• Documentation error

  #131 opened 2 years ago by erlebach
  1

• Add ArrowTypes.jl dependency to serialize optimizers?

  #77 opened 2 years ago by ericphanson
  15

• Investigate using a different AD for tests

  #96 opened 2 years ago by ToucheSir
  5

• Support dictionary of parameters

  #114 opened 2 years ago by freddycct
  3

• "Optimisers.jl does not at present handle tied weights, sorry."

  #97 opened 2 years ago by gdalle
  7

• `destructure`'s gradient is confused by `trainable`

  #72 opened 3 years ago by mcabbott
  0

• Optimizing scalars

  #92 opened 2 years ago by samanklesaria
  1

• update! is ambiguous with ComponentArrays

  #91 opened 2 years ago by IlyaOrson
  2

• support a nice way of changing the learning rate

  #88 opened 2 years ago by SobhanMP
  2

• Applying `Grads` fails

  #76 opened 2 years ago by femtomc
  6

• Using destructure with functions or anything without (trainable) parameters

  #67 opened 3 years ago by lungd
  1

• Chaining gpu and cpu models

  #69 opened 3 years ago by lungd
  8

• `destructure` doesn't work correctly with certain functors

  #62 opened 3 years ago by rejuvyesh
  3

• Register 0.2

  #52 opened 3 years ago by mcabbott
  13

• Correct the optimiser's eltype?

  #55 opened 3 years ago by mcabbott
  2

• Optimise a subset of parameters

  #35 opened 3 years ago by mcabbott
  7