v0.9.16

NNlib v0.9.16

Diff since v0.9.15

Merged pull requests:

• Enzyme: broaden tests (#583) (@wsmoses)
• Type stabilize logsoftmax (#584) (@wsmoses)

Closed issues:

• Adding EnzymeRules (#503)

v0.9.15

NNlib v0.9.15

Diff since v0.9.14

Merged pull requests:

• Documenter.jl update to 1.4.0 (#578) (@dom-linkevicius)
• Update the activation inplace for cudnn (#579) (@avik-pal)
• Enzyme: add derivatives for ∇conv_filter and ∇conv_data (#582) (@wsmoses)

Closed issues:

• Missing EnzymeRule for ∇conv_data! (#565)
• conv_bias_act! allocates for activation in CUDNN (#575)
• Updating Documenter.jl to a more recent version? (#577)

v0.9.14

NNlib v0.9.14

Diff since v0.9.13

Merged pull requests:

• Add export for imrotate gradient (#574) (@roflmaostc)
• Fix batchnorm in testmode without track stats (#576) (@paulnovo)

Closed issues:

• Differentiable Image Rotatation with CUDA (#552)

v0.9.13

NNlib v0.9.13

Diff since v0.9.12

Merged pull requests:

• Replace unsafe_convert with pointer (#564) (@mcabbott)
• Fix typo in dot product attention error message (#567) (@johannes-fischer)
• Make softmax Enzyme-friendly (#569) (@mcabbott)
• Bump EnzymeCore compat (#571) (@wsmoses)

Closed issues:

• batched_mul broken on Julia nightly (#563)

v0.9.12

NNlib v0.9.12

Diff since v0.9.11

Merged pull requests:

• CUDA: Fix compilation failure. (#566) (@maleadt)

v0.9.11

NNlib v0.9.11

Diff since v0.9.10

Merged pull requests:

• Add image rotation (#553) (@roflmaostc)
• Add rrule for oftf (#559) (@ToucheSir)
• Use julia-actions/cache (#560) (@ToucheSir)

Closed issues:

• Enzyme when computing gradients for conv from NNlib fails compiling (#557)
• NNlibCUDACUDNNExt precompilation doesn't get finish (#558)

v0.9.10

NNlib v0.9.10

Diff since v0.9.9

Merged pull requests:

• CompatHelper: bump compat for Adapt to 4, (keep existing compat) (#554) (@github-actions[bot])
• Fixes #555 (#556) (@re1san)

Closed issues:

• Broken rrule for pad_constant() (#555)

v0.9.9

NNlib v0.9.9

Diff since v0.9.8

Merged pull requests:

• Bump AMDGPU compat to 0.8 (#551) (@pxl-th)

v0.9.8

NNlib v0.9.8

Diff since v0.9.7

Merged pull requests:

• Don't run CPU tests on AMDGPU CI machines (#542) (@pxl-th)
• Allow non-Int64 indices in scatter (#543) (@pxl-th)
• Update AMDGPU compat to 0.7 (#545) (@pxl-th)
• make gather/scatter work with views (#546) (@CarloLucibello)
• CompatHelper: add new compat entry for Statistics at version 1, (keep existing compat) (#549) (@github-actions[bot])
• Set compatibilities for standard packages (#550) (@bicycle1885)

v0.9.7

NNlib v0.9.7

Diff since v0.9.6

Merged pull requests:

• Add EnzymeRule for conv!/gather!/scatter!/dropout!/pool! (#536) (@wsmoses)
• Fix grad conv im2col (#539) (@wsmoses)
• unbreak gpu conv test (#541) (@CarloLucibello)

Closed issues:

• Failed to precompile NNlib (#532)
• grad_conv! computes wrong answer (#538)