Code

Orca sytorch#205

Code

Draft

trajore wants to merge 154 commits intompc-msri:mastermpc-msri/EzPC:masterfrom trajore:orca_sytorchtrajore/EzPC_1:orca_sytorchCopy head branch name to clipboard

+247,874-86

Commits

Commits on Oct 3, 2022

• init: torch like interface

  kanav99

  committed
  9ae2194

  Copy full SHA for 9ae2194

Commits on Oct 4, 2022

• feat: add forward pass for many layers

  kanav99

  committed
  2d826fe

  Copy full SHA for 2d826fe

• forward works, backward buggy

  kanav99

  committed
  8ee28db

  Copy full SHA for 8ee28db

• fixed matmul bug

  kanav99

  committed
  6b2b9ec

  Copy full SHA for 6b2b9ec

• fix backprop double free and start mnist training

  kanav99

  committed
  bcbd291

  Copy full SHA for bcbd291

Commits on Oct 5, 2022

• training works somehow

  kanav99

  committed
  2885cea

  Copy full SHA for 2885cea

• fix memory leak in tensor.h

  kanav99

  committed
  58eb082

  Copy full SHA for 58eb082

Commits on Oct 6, 2022

• fuck

  kanav99

  committed
  8ef6947

  Copy full SHA for 8ef6947

• no clue what's happening

  kanav99

  committed
  5803dc5

  Copy full SHA for 5803dc5

Commits on Oct 8, 2022

• save progress

  kanav99

  committed
  0e864e7

  Copy full SHA for 0e864e7

• fixed point trains now but initial accuracy is bad

  kanav99

  committed
  ee23a2b

  Copy full SHA for ee23a2b

• too much accuracy

  kanav99

  committed
  e4e262b

  Copy full SHA for e4e262b

Commits on Oct 10, 2022

• conv2d forward pass

  kanav99

  committed
  901e580

  Copy full SHA for 901e580

Commits on Oct 11, 2022

• partial conv backward implementation

  kanav99

  committed
  869e6c9

  Copy full SHA for 869e6c9

Commits on Oct 13, 2022

• buggy but mine

  kanav99

  committed
  adf2012

  Copy full SHA for adf2012

• lenet training works

  kanav99

  committed
  812c6c2

  Copy full SHA for 812c6c2

Commits on Oct 14, 2022

• fixed point lenet training works

  kanav99

  committed
  b654638

  Copy full SHA for b654638

• modularity but new sus behaviour discovered

  kanav99

  committed
  5952191

  Copy full SHA for 5952191

Commits on Oct 15, 2022

• linked minillama

  kanav99

  committed
  aecd1bf

  Copy full SHA for aecd1bf

• more cleanup

  kanav99

  committed
  6921865

  Copy full SHA for 6921865

• removed GroupElement, untested

  kanav99

  committed
  309e832

  Copy full SHA for 309e832

Commits on Oct 16, 2022

• some changes and keysize backend

  kanav99

  committed
  9d296c7

  Copy full SHA for 9d296c7

• add clear maxpool

  kanav99

  committed
  74e2381

  Copy full SHA for 74e2381

• forgot to add same code to cleartext.h

  kanav99

  committed
  ded3617

  Copy full SHA for ded3617

• relu maxpool optimization

  kanav99

  committed
  737a635

  Copy full SHA for 737a635

Commits on Oct 17, 2022

• llama integration

  kanav99

  committed
  ec03673

  Copy full SHA for ec03673

• key calculation

  kanav99

  committed
  c393b3c

  Copy full SHA for c393b3c

• added cifar10 dataset

  kanav99

  committed
  473b6a4

  Copy full SHA for 473b6a4

• some cleanup

  kanav99

  committed
  b331130

  Copy full SHA for b331130

Commits on Oct 18, 2022

• progress

  kanav99

  committed
  5719385

  Copy full SHA for 5719385

• progress bar and parallel loops

  kanav99

  committed
  049855c

  Copy full SHA for 049855c

• fix initialization but scale 16 still doesnt work in lenet

  kanav99

  committed
  57b3705

  Copy full SHA for 57b3705

• memory access fix in reshape

  kanav99

  committed
  c496ea5

  Copy full SHA for c496ea5

Commits on Oct 19, 2022

• why are there nans?

  kanav99

  committed
  252fcb5

  Copy full SHA for 252fcb5

• fix nasty bias bug

  kanav99

  committed
  c28b480

  Copy full SHA for c28b480

Commits on Oct 20, 2022

• fixed point training is working

  kanav99

  committed
  a968137

  Copy full SHA for a968137

• read cifar10 in correct way

  kanav99

  committed
  2d515d8

  Copy full SHA for 2d515d8

Commits on Oct 27, 2022

• fairly stable code

  kanav99

  committed
  21a05ef

  Copy full SHA for 21a05ef

Commits on Oct 28, 2022

• secure fwd pass of p-vgg16 complete

  kanav99

  committed
  91c508c

  Copy full SHA for 91c508c

Commits on Oct 31, 2022

• RT backpass complete, maybe

  kanav99

  committed
  3cae24f

  Copy full SHA for 3cae24f

• FC and RT works, maybe

  kanav99

  committed
  6bda11c

  Copy full SHA for 6bda11c

Commits on Nov 1, 2022

• small example looks bitwise equivalent

  kanav99

  committed
  52fc6ca

  Copy full SHA for 52fc6ca

Commits on Nov 2, 2022

• piranha vgg bitwise equivalence in pt and ct

  kanav99

  committed
  4aa3580

  Copy full SHA for 4aa3580

• added 2 round relu api

  kanav99

  committed
  1285474

  Copy full SHA for 1285474

Commits on Nov 3, 2022

• wild west now

  kanav99

  committed
  63687a5

  Copy full SHA for 63687a5

Commits on Nov 4, 2022

• llamaextended

  kanav99

  committed
  94aff00

  Copy full SHA for 94aff00

Commits on Nov 5, 2022

• stochastic relutruncate

  kanav99

  committed
  1def44d

  Copy full SHA for 1def44d

• add sci backend

  kanav99

  committed
  8ddfe77

  Copy full SHA for 8ddfe77

Commits on Nov 6, 2022

• double round maxpool

  kanav99

  committed
  d46e612

  Copy full SHA for d46e612

Commits on Nov 8, 2022

• maxpool backpass works yay

  kanav99

  committed
  365c0c5

  Copy full SHA for 365c0c5

Commits on Nov 9, 2022

• things look stable

  kanav99

  committed
  ffa4f23

  Copy full SHA for ffa4f23

• stochastic ars in llama

  kanav99

  committed
  1dbb1fa

  Copy full SHA for 1dbb1fa

• bring back vgg code

  kanav99

  committed
  667c9ee

  Copy full SHA for 667c9ee

Commits on Dec 5, 2022

• fixtofloat done untested

  kanav99

  committed
  99165be

  Copy full SHA for 99165be

• successful link with secfloat

  kanav99

  committed
  5f6f5c8

  Copy full SHA for 5f6f5c8

Commits on Dec 6, 2022

• hard memory bug..

  kanav99

  committed
  7ddcbec

  Copy full SHA for 7ddcbec

Commits on Dec 8, 2022

• forgot to close file....

  kanav99

  committed
  7155ede

  Copy full SHA for 7155ede

Commits on Dec 13, 2022

• float2fix done, a bit bug

  kanav99

  committed
  5144e8a

  Copy full SHA for 5144e8a

Commits on Dec 16, 2022

• fixed float2fix

  kanav99

  committed
  f7ef5df

  Copy full SHA for f7ef5df

• timing float operations

  kanav99

  committed
  ff658fc

  Copy full SHA for ff658fc

Commits on Dec 17, 2022

• branch add

  kanav99

  committed
  01bfa22

  Copy full SHA for 01bfa22

Commits on Dec 18, 2022

• batchnorm.py and main.cpp outputs match

  kanav99

  committed
  320ae34

  Copy full SHA for 320ae34

• match again

  kanav99

  committed
  f67931f

  Copy full SHA for f67931f

Commits on Dec 19, 2022

• cifar10 float with batchnorm trains

  kanav99

  committed
  8e8c6d8

  Copy full SHA for 8e8c6d8

• ramdisk/mmap support

  kanav99

  committed
  0f767cd

  Copy full SHA for 0f767cd

Commits on Dec 20, 2022

• batchnorm cleartext progress save

  kanav99

  committed
  a726b1f

  Copy full SHA for a726b1f

• ramdisk fix

  kanav99

  committed
  757ded4

  Copy full SHA for 757ded4

Commits on Dec 22, 2022

• major refactor before GPU integration

  kanav99

  committed
  e3d7971

  Copy full SHA for e3d7971

Commits on Feb 15, 2023

• Refactor (mpc-msri#1)

  Show description for c68a656
  kanav99

  authored
  c68a656

  Copy full SHA for c68a656

Commits on Feb 17, 2023

• remove gpu code

  kanav99

  committed
  1078f4c

  Copy full SHA for 1078f4c

• sytorch refactor part 1

  kanav99

  committed
  62ee65c

  Copy full SHA for 62ee65c

• part 2

  kanav99

  committed
  ac4cb9b

  Copy full SHA for ac4cb9b

Commits on Feb 19, 2023

• works on macOS

  kanav99

  committed
  940a221

  Copy full SHA for 940a221

Commits on Feb 21, 2023

• cleartext resnet9 inference works

  kanav99

  committed
  de3f283

  Copy full SHA for de3f283

• llama ext works with resnet9

  kanav99

  committed
  c5e9033

  Copy full SHA for c5e9033

Commits on Feb 22, 2023

• graph generation from code

  kanav99

  committed
  c2a16a7

  Copy full SHA for c2a16a7

• llama ext truncation optimization

  kanav99

  committed
  38bc49c

  Copy full SHA for 38bc49c

• resnet 32,12 still incorrect in llamaext

  kanav99

  committed
  7f22698

  Copy full SHA for 7f22698

• when shall the sun rise? resnet 32,12 works

  kanav99

  committed
  7c7cf59

  Copy full SHA for 7c7cf59

Commits on Feb 23, 2023

• llama improved almost fully optimized, add is pita

  kanav99

  committed
  2025102

  Copy full SHA for 2025102

• minor refactor

  kanav99

  committed
  75ec790

  Copy full SHA for 75ec790

• minor refactor

  kanav99

  committed
  be77ed8

  Copy full SHA for be77ed8

• sytorch modules lets goooo

  kanav99

  committed
  1ae575a

  Copy full SHA for 1ae575a

• minor refactor

  kanav99

  committed
  9a93dea

  Copy full SHA for 9a93dea

Commits on Feb 24, 2023

• minor cosmetics

  kanav99

  committed
  fc04b85

  Copy full SHA for fc04b85

• basic garbage cleaning

  kanav99

  committed
  3990dec

  Copy full SHA for 3990dec

• sequential was fine but was nice to refactor

  kanav99

  committed
  c222b3b

  Copy full SHA for c222b3b

• variadic add

  kanav99

  committed
  f498b3b

  Copy full SHA for f498b3b

• concat works

  kanav99

  committed
  c1b07fa

  Copy full SHA for c1b07fa

Commits on Feb 25, 2023

• vgg16 and chexpert untested

  kanav99

  committed
  347aa66

  Copy full SHA for 347aa66

Commits on Feb 26, 2023

• small changes

  kanav99

  committed
  360e408

  Copy full SHA for 360e408

• div support

  kanav99

  committed
  c5509fc

  Copy full SHA for c5509fc

Commits on Feb 27, 2023

• fixed model loading for chexpert

  kanav99

  committed
  24b5a31

  Copy full SHA for 24b5a31

Commits on Mar 19, 2023

• Onnx sytorch (mpc-msri#4)

  Show description for 1c4ae86
  

  drunkenlegend

  and

  kanav99

  authored
  1c4ae86

  Copy full SHA for 1c4ae86

Commits on Mar 20, 2023

• avgpool changes

  kanav99

  committed
  ac568a2

  Copy full SHA for ac568a2

• fix llama improved

  kanav99

  committed
  221182b

  Copy full SHA for 221182b

• remove SCIConfig.cmake

  kanav99

  committed
  d8724d4

  Copy full SHA for d8724d4

• add SCIConfig.cmake to gitignore

  kanav99

  committed
  c2838cf

  Copy full SHA for c2838cf

Commits on Mar 29, 2023

• Orca dumping (mpc-msri#5)

  Show description for 7aeb352
  kanav99

  authored
  7aeb352

  Copy full SHA for 7aeb352

• add additional scripts

  kanav99

  committed
  fc67130

  Copy full SHA for fc67130

Commits on Apr 9, 2023

• conv/fc optimizations (mpc-msri#7)

  kanav99

  authored
  2a1e023

  Copy full SHA for 2a1e023

• sequential fixed

  kanav99

  committed
  6484937

  Copy full SHA for 6484937

Commits on Jul 31, 2023

• buggy dcf with et

  kanav99

  committed
  25377e9

  Copy full SHA for 25377e9

Commits on Aug 1, 2023

• fix one bug

  kanav99

  committed
  674fb10

  Copy full SHA for 674fb10

• fix bug in dcf et

  kanav99

  committed
  48a2bc1

  Copy full SHA for 48a2bc1

Commits on Aug 2, 2023

• more test for dcfet

  kanav99

  committed
  5aa9889

  Copy full SHA for 5aa9889

• security tight in dcf et

  kanav99

  committed
  4aff784

  Copy full SHA for 4aff784

• wip dcf et 2 bit

  kanav99

  committed
  7a2d9ee

  Copy full SHA for 7a2d9ee

• minor

  kanav99

  committed
  a73c40e

  Copy full SHA for a73c40e

• dcf 2et done, seems little slow

  kanav99

  committed
  1917e74

  Copy full SHA for 1917e74

• benchmark code in dcftest

  kanav99

  committed
  3b17210

  Copy full SHA for 3b17210

Commits on Aug 3, 2023

• rename

  kanav99

  committed
  3dbc9b7

  Copy full SHA for 3dbc9b7

• dcfet wrapper

  kanav99

  committed
  b7a74db

  Copy full SHA for b7a74db

• dcfet1 replaced in Relu2Round and MaxpoolDouble

  kanav99

  committed
  969fa9a

  Copy full SHA for 969fa9a

• dcfet2 in ReluExt

  kanav99

  committed
  996d55d

  Copy full SHA for 996d55d

• dcfet1 in SignExtend2

  kanav99

  committed
  d04ffba

  Copy full SHA for d04ffba

• add greater than support in dcfet1

  kanav99

  committed
  f46816b

  Copy full SHA for f46816b

Commits on Aug 6, 2023

• new orca str

  kanav99

  committed
  20bcb54

  Copy full SHA for 20bcb54

Commits on Aug 7, 2023

• use secure STR

  kanav99

  committed
  82fa741

  Copy full SHA for 82fa741

Commits on Aug 10, 2023

• secret sharing based truncation

  kanav99

  committed
  5b205f5

  Copy full SHA for 5b205f5

Commits on Aug 11, 2023

• some bugs fixed

  kanav99

  committed
  09e191e

  Copy full SHA for 09e191e

• minor

  kanav99

  committed
  e78f173

  Copy full SHA for e78f173

Commits on Aug 15, 2023

• microbenchmark for RT

  kanav99

  committed
  cbe8fe1

  Copy full SHA for cbe8fe1

• rtm microbenchmark script

  kanav99

  committed
  8725316

  Copy full SHA for 8725316

Commits on Sep 1, 2023

• Minor Changes Timing improvements with std::thread

  trajore

  committed
  5477ae1

  Copy full SHA for 5477ae1

• Update sytorchBackendRep.py

  trajore

  authored
  876fd0a

  Copy full SHA for 876fd0a

Commits on Sep 4, 2023

• module.h fix

  trajore

  committed
  8af3668

  Copy full SHA for 8af3668

• fix module.h bug type casting

  trajore

  committed
  f4d9919

  Copy full SHA for f4d9919

• fixed bug in llama_base.h, input_prng.cpp, module.h, tensor.h

  trajore

  committed
  ef06ca2

  Copy full SHA for ef06ca2

• ramdisk_path option added to llama config file

  trajore

  committed
  4c08c32

  Copy full SHA for 4c08c32

• added the ramdrive scripts

  trajore

  committed
  7771631

  Copy full SHA for 7771631

• Update Toy example- single inference.md

  trajore

  authored
  fefdb40

  Copy full SHA for fefdb40

Commits on Sep 12, 2023

• commented out print statements in comms.cpp

  trajore

  committed
  847cb3f

  Copy full SHA for 847cb3f

• Merge branch 'master' of https://github.com/trajore/EzPC-1

  trajore

  committed
  3382240

  Copy full SHA for 3382240

• fix bug in llama/comms.cpp

  trajore

  committed
  790c151

  Copy full SHA for 790c151

• fix to mult_key

  trajore

  committed
  9fef89d

  Copy full SHA for 9fef89d

Commits on Sep 14, 2023

• Updated FixtoFloats. work in progress

  trajore

  committed
  1db9c13

  Copy full SHA for 1db9c13

Commits on Sep 15, 2023

• updated flaot to fix, a few more bugs to fix

  trajore

  committed
  4c9ec8a

  Copy full SHA for 4c9ec8a

Commits on Sep 25, 2023

• float to fix working

  trajore

  committed
  ff7a964

  Copy full SHA for ff7a964

Commits on Sep 26, 2023

• changed the random ge to be a random element again

  trajore

  committed
  135e67f

  Copy full SHA for 135e67f

Commits on Oct 10, 2023

• working fix (initialization)

  trajore

  committed
  5d9b7ba

  Copy full SHA for 5d9b7ba

• float to fix corrected

  trajore

  committed
  6642c7d

  Copy full SHA for 6642c7d

• removed debugging statements

  trajore

  committed
  2f5d985

  Copy full SHA for 2f5d985

Commits on Nov 7, 2023

• added the add function for mix bitwidth

  trajore

  committed
  93974f1

  Copy full SHA for 93974f1

Commits on Nov 10, 2023

• Update float.cpp

  trajore

  authored
  098395d

  Copy full SHA for 098395d

• Update api.cpp

  Show description for f9d6085
  trajore

  authored
  f9d6085

  Copy full SHA for f9d6085

• Update api.cpp

  Show description for 5b59e54
  trajore

  authored
  5b59e54

  Copy full SHA for 5b59e54

Commits on Nov 17, 2023

• Merge pull request mpc-msri#8 from trajore/early_term

  Show description for e31fde2
  trajore

  authored
  e31fde2

  Copy full SHA for e31fde2

Commits on Dec 11, 2023

• removed add mixbitwidth temporary changes

  trajore

  committed
  2965bf3

  Copy full SHA for 2965bf3

• Merge branch 'early_term' of https://github.com/trajore/sytorch into early_term

  trajore

  committed
  4efa831

  Copy full SHA for 4efa831

Commits on Dec 18, 2023

• Merge pull request mpc-msri#9 from trajore/early_term

  Show description for d03dd74
  trajore

  authored
  d03dd74

  Copy full SHA for d03dd74

Commits on Dec 27, 2023

• Merge branch 'mpc-msri:master' into master

  trajore

  authored
  37c6d71

  Copy full SHA for 37c6d71

• Merge remote-tracking branch 'kanav_sytorch/early_term' into orca_sytorch

  trajore

  committed
  7376d28

  Copy full SHA for 7376d28

Commits on Jan 2, 2024

• Update float.cpp

  Show description for a6b4cb3
  trajore

  authored
  a6b4cb3

  Copy full SHA for a6b4cb3