Code

meson: do not enable AVX in non-x64#2629

Code

Closed

myungjoo wants to merge 25 commits intonnstreamer:mainnnstreamer/nntrainer:mainfrom myungjoo:fix/meson/avxmyungjoo/nntrainer:fix/meson/avxCopy head branch name to clipboard

+2,418-430

Commits

Commits on Jun 10, 2024

• [ Weight ] Add Var32 Tensor in Weight.

  Show description for 1ecac99
  

  jijoongmoon

  authored and

  myungjoo

  committed
  1ecac99

  Copy full SHA for 1ecac99

• [ Mixed ] Create weight with var32 tensor

  Show description for a5491f1
  

  jijoongmoon

  authored and

  myungjoo

  committed
  a5491f1

  Copy full SHA for a5491f1

• [ Layers ] Update Layers to support FP16

  Show description for 8dab24c
  

  jijoongmoon

  authored and

  myungjoo

  committed
  8dab24c

  Copy full SHA for 8dab24c

• [ Test ] Mixed Precision Test Case

  Show description for 69bda1a
  

  jijoongmoon

  authored and

  myungjoo

  committed
  69bda1a

  Copy full SHA for 69bda1a

• [ Optimizer ] Update Optimizer / Adam to support Mixed training

  Show description for 82e74a9
  

  jijoongmoon

  authored and

  myungjoo

  committed
  82e74a9

  Copy full SHA for 82e74a9

• [ Tensor ] add is_NaN check in Tensor

  Show description for 727a810
  

  jijoongmoon

  authored and

  myungjoo

  committed
  727a810

  Copy full SHA for 727a810

• [ Context ] Add loss scale in Context & using mse loss

  Show description for e0c6d98
  

  jijoongmoon

  authored and

  myungjoo

  committed
  e0c6d98

  Copy full SHA for e0c6d98

• [ Mixed Precision ] Enable Mixed Precision

  Show description for 0792b84
  

  jijoongmoon

  authored and

  myungjoo

  committed
  0792b84

  Copy full SHA for 0792b84

• [ Tensor ] Add inifinity check in Tensor

  Show description for 88908bd
  

  jijoongmoon

  authored and

  myungjoo

  committed
  88908bd

  Copy full SHA for 88908bd

• [ MSE ] Fix for better MSE loss precision

  Show description for 9351134
  

  jijoongmoon

  authored and

  myungjoo

  committed
  9351134

  Copy full SHA for 9351134

• [ TEST ] Add Torch Mixed Precision Model Test

  Show description for 931879b
  

  jijoongmoon

  authored and

  myungjoo

  committed
  931879b

  Copy full SHA for 931879b

• [ TEST ] add torch input and output test data for mixed precision

  Show description for c333096
  

  jijoongmoon

  authored and

  myungjoo

  committed
  c333096

  Copy full SHA for c333096

• [ TEST ] Add more unittest and fixes for mixed precsion

  Show description for 9c51f23
  

  jijoongmoon

  authored and

  myungjoo

  committed
  9c51f23

  Copy full SHA for 9c51f23

• [ Layer ] Update Conv2D to support Mixed Precision

  Show description for 91850a4
  

  jijoongmoon

  authored and

  myungjoo

  committed
  91850a4

  Copy full SHA for 91850a4

• [ Layer ] enable Mixed Precision in LSTM Layer

  Show description for 4ac4588
  

  jijoongmoon

  authored and

  myungjoo

  committed
  4ac4588

  Copy full SHA for 4ac4588

• [ Model ] Add Execution Mode in Compile

  Show description for 4c376fd
  

  jijoongmoon

  authored and

  myungjoo

  committed
  4c376fd

  Copy full SHA for 4c376fd

• [ Layer ] Mixed Precision support for BN Layer

  Show description for 6088bd2
  

  jijoongmoon

  authored and

  myungjoo

  committed
  6088bd2

  Copy full SHA for 6088bd2

• [layer] enable mixed precision - reshape_layer

  Show description for d15aef8
  

  DonghakPark

  authored and

  myungjoo

  committed
  d15aef8

  Copy full SHA for d15aef8

• [Layer] Enable mixed precision - pooling2d_layer

  Show description for ab6124d
  

  DonghakPark

  authored and

  myungjoo

  committed
  ab6124d

  Copy full SHA for ab6124d

• [ Model ] Fix the gradient clipping for the FP16 or Low bit Gradient

  Show description for ce6650e
  

  jijoongmoon

  authored and

  myungjoo

  committed
  ce6650e

  Copy full SHA for ce6650e

• [ Layer ] Add mu and var backup up tensor.

  Show description for 15794ac
  

  jijoongmoon

  authored and

  myungjoo

  committed
  15794ac

  Copy full SHA for 15794ac

• [ Layer ] prevent randomize when it restore the data

  Show description for 9faaaad
  

  jijoongmoon

  authored and

  myungjoo

  committed
  9faaaad

  Copy full SHA for 9faaaad

• [ Context ] add check if it needs restore previous data

  Show description for 1f80b65
  

  jijoongmoon

  authored and

  myungjoo

  committed
  1f80b65

  Copy full SHA for 1f80b65

• [ Tensor ] remove sscal to set zero.

  Show description for 527becc
  

  jijoongmoon

  authored and

  myungjoo

  committed
  527becc

  Copy full SHA for 527becc

• meson: do not enable AVX in non-x64

  Show description for 13bce59
  myungjoo

  committed
  13bce59

  Copy full SHA for 13bce59