Commits

Breadcrumbs

History for

onparallel

User selector

All users

All time

Commit History

Commits on Dec 24, 2014

• Distributed training

  

  cypof

  committed
  aa3b877

  Copy full SHA for aa3b877

Commits on Oct 17, 2014

• back-merge

  sergeyk

  committed
  1ed11d5

  Copy full SHA for 1ed11d5

Commits on Oct 16, 2014

• Fix init current_step

  sguada

  committed
  3d95ed3

  Copy full SHA for 3d95ed3

Commits on Oct 10, 2014

• Added Multistep, Poly and Sigmoid learning rate decay policies

  Show description for b025da7
  sguada

  committed
  b025da7

  Copy full SHA for b025da7

Commits on Oct 6, 2014

• correct naming in comment and message about average_loss

  longjon

  committed
  914da95

  Copy full SHA for 914da95

• [fix] check solver prototxt parsing

  longjon

  committed
  aaccb48

  Copy full SHA for aaccb48

Commits on Oct 1, 2014

• Back-merge branch 'master' into dev (for fix in PR #1203)

  jeffdonahue

  committed
  ceb3011

  Copy full SHA for ceb3011

• [fix] solver indexing of output blobs was incorrect for non-singleton

  Show description for ac8a03d
  jeffdonahue

  committed
  ac8a03d

  Copy full SHA for ac8a03d

Commits on Sep 18, 2014

• fix warning

  qipeng

  committed
  7a688c7

  Copy full SHA for 7a688c7

Commits on Sep 17, 2014

• Display averaged loss over the last several iterations

  qipeng

  committed
  03e0e01

  Copy full SHA for 03e0e01

Commits on Sep 16, 2014

• [fix] snapshot model weights as .caffemodel, solver state as .solverstate

  shelhamer

  committed
  4e6d977

  Copy full SHA for 4e6d977

Commits on Sep 15, 2014

• fix caffe train GPU initialization

  Show description for bbd166e
  longjon

  committed
  bbd166e

  Copy full SHA for bbd166e

Commits on Sep 5, 2014

• Merge pull request #1036 from longjon/test-initialization-param

  Show description for 50d9d0d
  shelhamer

  committed
  50d9d0d

  Copy full SHA for 50d9d0d

Commits on Sep 4, 2014

• add test_initialization option to allow skipping initial test

  longjon

  committed
  2bdf516

  Copy full SHA for 2bdf516

• snapshot model with caffemodel extension

  

  shelhamer

  authored and

  sergeyk

  committed
  bcc12ef

  Copy full SHA for bcc12ef

Commits on Sep 1, 2014

• added unit test for solvers and fixed solver bugs

  

  qipeng

  authored and

  jeffdonahue

  committed
  7f2e66e

  Copy full SHA for 7f2e66e

• fixes after rebase

  

  qipeng

  authored and

  jeffdonahue

  committed
  23d4430

  Copy full SHA for 23d4430

• Added L1 regularization support for the weights

  

  qipeng

  authored and

  jeffdonahue

  committed
  a683c40

  Copy full SHA for a683c40

• bugfixes for AdaGrad

  

  qipeng

  authored and

  jeffdonahue

  committed
  76ef2ca

  Copy full SHA for 76ef2ca

• improved numerical stability for AdaGrad

  

  qipeng

  authored and

  jeffdonahue

  committed
  0144de6

  Copy full SHA for 0144de6

• Solver switching support & implementation of Nesterov's accelerated gradient and AdaGrad

  

  qipeng

  authored and

  jeffdonahue

  committed
  9d10569

  Copy full SHA for 9d10569

Commits on Aug 13, 2014

• Store loss coefficients in layer; use for prettier training output.

  jeffdonahue

  committed
  415123b

  Copy full SHA for 415123b

• Also display outputs in the train net. (Otherwise, why have them?)

  jeffdonahue

  committed
  7d590a8

  Copy full SHA for 7d590a8

Commits on Aug 7, 2014

• LICENSE governs the whole project so strip file headers

  shelhamer

  committed
  b97b88f

  Copy full SHA for b97b88f

Commits on Aug 4, 2014

• Add 'snapshot_after_train' to SolverParameter to override the final

  Show description for 36fd64c
  jeffdonahue

  committed
  36fd64c

  Copy full SHA for 36fd64c

Commits on Jul 29, 2014

• Incorporate NetState{,Rule} into Solver/Net. Net::FilterNet

  Show description for cb4555c
  jeffdonahue

  committed
  cb4555c

  Copy full SHA for cb4555c

Commits on Jul 27, 2014

• Move loss display before lr display in ComputeUpdateValue.

  jeffdonahue

  committed
  91ed16b

  Copy full SHA for 91ed16b

• Make training iterations 0-indexed.

  jeffdonahue

  committed
  84c424b

  Copy full SHA for 84c424b

Commits on Jul 26, 2014

• Print blob L1 norms during forward/backward passes and updates if

  Show description for 149a176
  jeffdonahue

  committed
  149a176

  Copy full SHA for 149a176

Commits on Jul 18, 2014

• reapply namespace change

  

  Yangqing Jia

  committed
  7683986

  Copy full SHA for 7683986

Commits on Jul 17, 2014

• collect CUDA includes and calls, separate from CPU-only mode, leave out

  Show description for e52d91e
  shelhamer

  committed
  e52d91e

  Copy full SHA for e52d91e

Commits on Jul 4, 2014

• replace all memcpy by caffe_copy

  shelhamer

  committed
  1da30a5

  Copy full SHA for 1da30a5

Commits on Jun 19, 2014

• Set device_id at the begining of Solver.Init() to avoid using memory in the default GPU

  sguada

  committed
  3686f87

  Copy full SHA for 3686f87

Commits on May 10, 2014

• require either train_net or train_net_param to be specified

  jeffdonahue

  committed
  ba2875b

  Copy full SHA for ba2875b

• lint and two test_iters in lenet_consolidated_solver

  jeffdonahue

  committed
  0ec86f2

  Copy full SHA for 0ec86f2