Code

Redesign Mask#24

Code

Merged

chengchingwen merged 22 commits intomasterchengchingwen/NeuralAttentionlib.jl:masterfrom maskchengchingwen/NeuralAttentionlib.jl:maskCopy head branch name to clipboardJun 1, 2024

+1,265-617

Commits

Commits on Dec 31, 2023

• make indexer abstract array

  chengchingwen

  committed
  bfe23e1

  Copy full SHA for bfe23e1

• fix bi sequence length dest size

  chengchingwen

  committed
  e3a8b31

  Copy full SHA for e3a8b31

• refine indexer interface

  chengchingwen

  committed
  679cea7

  Copy full SHA for 679cea7

Commits on Jan 21, 2024

• redesign mask type

  chengchingwen

  committed
  3af9063

  Copy full SHA for 3af9063

Commits on Apr 20, 2024

• use pkg ext and drop lts julia (#26)

  Show description for 006a5be
  chengchingwen

  authored
  006a5be

  Copy full SHA for 006a5be

Commits on May 1, 2024

• AMDGPU support (#28)

  Show description for e7f47be
  

  chengchingwen

  and

  radudiaconu0

  authored
  e7f47be

  Copy full SHA for e7f47be

• make indexer abstract array

  chengchingwen

  committed
  e66d3ba

  Copy full SHA for e66d3ba

• fix bi sequence length dest size

  chengchingwen

  committed
  e258688

  Copy full SHA for e258688

• refine indexer interface

  chengchingwen

  committed
  6cf4e25

  Copy full SHA for 6cf4e25

• redesign mask type

  chengchingwen

  committed
  f49743b

  Copy full SHA for f49743b

Commits on May 9, 2024

• fix adapt generic sequence mask type stability

  chengchingwen

  committed
  170e4d5

  Copy full SHA for 170e4d5

• remove mask device adapt

  chengchingwen

  committed
  a170935

  Copy full SHA for a170935

Commits on May 28, 2024

• rework Indexer

  chengchingwen

  committed
  b4f03c5

  Copy full SHA for b4f03c5

• Merge branch 'mask' of github.com:chengchingwen/NeuralAttentionlib.jl into mask

  chengchingwen

  committed
  4312174

  Copy full SHA for 4312174

Commits on May 31, 2024

• fix adapt mask type stability

  chengchingwen

  committed
  8718e6c

  Copy full SHA for 8718e6c

• use custom rng for random mask

  chengchingwen

  committed
  c9e8109

  Copy full SHA for c9e8109

• make indexer an abstract array with a scale value and update broadcast with NNlib._fast_broadcast! for cpu performance

  chengchingwen

  committed
  b1e0676

  Copy full SHA for b1e0676

• update test

  chengchingwen

  committed
  678a1f5

  Copy full SHA for 678a1f5

Commits on Jun 1, 2024

• refine dropout to allow RNG and update test

  chengchingwen

  committed
  06617e0

  Copy full SHA for 06617e0

• improve random pattern

  chengchingwen

  committed
  5946149

  Copy full SHA for 5946149

• update docs

  chengchingwen

  committed
  571bdff

  Copy full SHA for 571bdff

• update env

  chengchingwen

  committed
  297cc23

  Copy full SHA for 297cc23