Code

Add LayerNorm support for Vivado#1110

Code

Open

rianbrooksflynn wants to merge 58 commits intofastmachinelearning:mainfastmachinelearning/hls4ml:mainfrom rianbrooksflynn:layernormrianbrooksflynn/hls4ml:layernormCopy head branch name to clipboard

+1,261-849

Commits

Commits on Sep 13, 2024

• paser_mht

  

  Ethan0Jiang

  authored and

  JanFSchulte

  committed
  c4c818b

• change parser and modify keras_to_hls

  

  Ethan0Jiang

  authored and

  JanFSchulte

  committed
  3ee64d1

• IR_mutihead_attention

  

  Ethan0Jiang

  authored and

  JanFSchulte

  committed
  5626a1a

• IR done

  

  Ethan0Jiang

  authored and

  JanFSchulte

  committed
  d51f8a9

• create mha file in template

  

  Ethan0Jiang

  authored and

  JanFSchulte

  committed
  89025a2

• mha .h file dummy algo

  

  Ethan0Jiang

  authored and

  JanFSchulte

  committed
  d76cf60

• config of mha

  

  Ethan0Jiang

  authored and

  JanFSchulte

  committed
  56811de

• update mha config

  

  Ethan0Jiang

  authored and

  JanFSchulte

  committed
  45cd493

• dummy mha

  

  Ethan0Jiang

  authored and

  JanFSchulte

  committed
  1402f48

• add transpose into mha

  

  Ethan0Jiang

  authored and

  JanFSchulte

  committed
  430b9ea

• projection_of_qkv_in_mha

  

  Ethan0Jiang

  authored and

  JanFSchulte

  committed
  97f3e8d

• mha_first_draft

  

  Ethan0Jiang

  authored and

  JanFSchulte

  committed
  52cc7e8

• able to predict model correct

  

  Ethan0Jiang

  authored and

  JanFSchulte

  committed
  3961f97

• delete some unnassary comments

  

  Ethan0Jiang

  authored and

  JanFSchulte

  committed
  3533999

• delete comments

  

  Ethan0Jiang

  authored and

  JanFSchulte

  committed
  d2f0df6

• resource strategy of transformer

  

  Ethan0Jiang

  authored and

  JanFSchulte

  committed
  6aaa5ed

• change sm lagacy

  

  Ethan0Jiang

  authored and

  JanFSchulte

  committed
  3b7a288

• update MHA, optimized

  

  Ethan0Jiang

  authored and

  JanFSchulte

  committed
  130092d

• support resource

  

  Ethan0Jiang

  authored and

  JanFSchulte

  committed
  09b0ba0

• update

  

  Ethan0Jiang

  authored and

  JanFSchulte

  committed
  b49fffd

• dense_muti_dim_support

  

  Ethan0Jiang

  authored and

  JanFSchulte

  committed
  5324a11

• parallel execute dense

  

  Ethan0Jiang

  authored and

  JanFSchulte

  committed
  bf8c788

• updates

  

  Ethan0Jiang

  authored and

  JanFSchulte

  committed
  b6be2c4

• add_layerNorm_support

  

  Ethan0Jiang

  authored and

  JanFSchulte

  committed
  2472b7d

• MHA updated

  

  Ethan0Jiang

  authored and

  JanFSchulte

  committed
  97e71e9

• LayerNorm_bug_fix

  

  Ethan0Jiang

  authored and

  JanFSchulte

  committed
  5ed4a76

• update bit precision

  

  Ethan0Jiang

  authored and

  JanFSchulte

  committed
  5d28f58

• config update

  

  Ethan0Jiang

  authored and

  JanFSchulte

  committed
  2fc68d0

• add some comment

  

  Ethan0Jiang

  authored and

  JanFSchulte

  committed
  b5c95cf

• run pre-commit

  JanFSchulte

  committed
  3b8aa8d

• Added support on QMultiHeadAttention, QLayerNormalization, and quantized_softmax

  

  LostEcho365

  authored and

  JanFSchulte

  committed
  d28b24c

• updated on hls4ml transformer

  

  LostEcho365

  authored and

  JanFSchulte

  committed
  de79bb9

• trying to clean the diff

  JanFSchulte

  committed
  6c23326

• trying to clean the diff

  JanFSchulte

  committed
  20a0199

Commits on Sep 17, 2024

• trying to clean the diff

  

  Rian Flynn

  committed
  ddccde2

• trying to clean the diff

  

  Rian Flynn

  committed
  afbe00b

• trying to clean the diff

  

  Rian Flynn

  committed
  dedf96c

Commits on Sep 18, 2024

• undo vhdl -> verilog change

  

  Rian Flynn

  committed
  a9de9cb

• halfway working layernorm + test

  

  Rian Flynn

  committed
  49313d3

• layernorm is now pretty functional

  

  Rian Flynn

  committed
  1156ba5

Commits on Sep 19, 2024

• layernorm on pytorch also

  

  Rian Flynn

  committed
  17e0048

• minor cleanup

  

  Rian Flynn

  committed
  63891fd

• more cleanup, pre-commit

  

  Rian Flynn

  committed
  8dccac6

• test for mha which kinda works maybe if you squint

  

  Rian Flynn

  committed
  595cc71

Commits on Sep 20, 2024

• multihead attention working on keras and pytorch

  

  Rian Flynn

  committed
  5f3ec00

Commits on Sep 25, 2024

• fiddly precision / accuracy changes for layernorm

  

  Rian Flynn

  committed
  5697334

Commits on Oct 11, 2024

• Merge remote-tracking branch 'upstream/main' into transformer

  rianbrooksflynn

  committed
  d2e27b8

Commits on Oct 22, 2024

• fix lookup table and label loops

  rianbrooksflynn

  committed
  a149f2e

Commits on Oct 23, 2024

• remove dense_seq

  rianbrooksflynn

  committed
  552fa83

• Merge remote-tracking branch 'upstream/main' into transformer

  rianbrooksflynn

  committed
  69f26bc

• undo qkeras changes

  rianbrooksflynn

  committed
  be5f5a4

Commits on Oct 24, 2024

• fix merge conflict residue

  rianbrooksflynn

  committed
  adf7356

Commits on Nov 4, 2024

• Merge remote-tracking branch 'upstream/main' into transformer

  rianbrooksflynn

  committed
  8437581

• remove non-layernorm changes

  rianbrooksflynn

  committed
  39ab36c

Commits on Dec 9, 2024

• change to uniform LUT and fix precision

  rianbrooksflynn

  committed
  b5b82e2

• Merge remote-tracking branch 'upstream/main' into layernorm

  rianbrooksflynn

  committed
  f3ff077

• [pre-commit.ci] auto fixes from pre-commit hooks

  pre-commit-ci[bot]

  committed
  0f08e7a

Commits on Dec 17, 2024

• Merge remote-tracking branch 'upstream/main' into layernorm

  rianbrooksflynn

  committed
  21049e7