https://github.com/Microsoft/CNTK

sort by:
Revision Author Date Message Commit Date
5f69174 fixed a memory bug for smb cpu summation 03 July 2017, 08:08:10 UTC
b422f6a use cpu smb sum. (need test) 01 July 2017, 17:36:57 UTC
b164c7d close some test function and clear up code 29 June 2017, 15:37:56 UTC
afe5133 add model save and load 27 June 2017, 15:36:01 UTC
91895f6 Calculate CRC code for inputData and gradients 27 June 2017, 13:02:00 UTC
287a68b use reduce sum for smb, fix bug 26 June 2017, 06:25:38 UTC
1c0cefb change smb add algorithm. 26 June 2017, 03:14:32 UTC
143cbe6 close warmup func 23 June 2017, 11:25:03 UTC
41c03a8 Revert "for 256 gpus warmup" This reverts commit 2e23d429f33ccb198afd31cd771b9f9ea88d9e42. 22 June 2017, 15:23:03 UTC
2e23d42 for 256 gpus warmup 22 June 2017, 15:12:38 UTC
a969507 fix a small lr bug for 16 real gpus 22 June 2017, 13:57:13 UTC
cd4b124 change 16 gpus display iter 22 June 2017, 09:38:09 UTC
2be9d9f warmup lr for 16gpus 22 June 2017, 09:36:25 UTC
d573785 1 22 June 2017, 09:26:41 UTC
a1bbd23 fix for 256 gpus 22 June 2017, 09:25:05 UTC
add4db3 print more lr 22 June 2017, 09:07:39 UTC
b8534e6 for 16 real gpus warmup 22 June 2017, 09:02:51 UTC
0a393a0 little print change 13 June 2017, 09:02:46 UTC
108c501 a sample version for ssgd warmup experiment this version code just could use for ImageNet-SSGD-warmup experiment 13 June 2017, 09:00:31 UTC
183ce3a delete tmp file. 07 June 2017, 09:13:45 UTC
f23e9bc Revert "Revert "use fixed convolution algorithm"" This reverts commit f68fcb856f50587e05993c669122491a1b1a1c6b. 07 June 2017, 09:12:44 UTC
f68fcb8 Revert "use fixed convolution algorithm" This reverts commit 1c9d444e6e8e4d9ec9b8891f0fe92bddc5bcecc2. 07 June 2017, 09:12:09 UTC
1c9d444 use fixed convolution algorithm this version use fixed convolution algorithm as follow: forward algorithm: CUDNN_CONVOLUTION_FWD_ALGO_IMPLICIT_GEMM backward data algorithm: CUDNN_CONVOLUTION_BWD_DATA_ALGO_1 backward filter algorithm: CUDNN_CONVOLUTION_BWD_FILTER_ALGO_1 07 June 2017, 09:12:00 UTC
4fa0825 close memory share 07 June 2017, 04:50:15 UTC
9d7481e fix broken link 03 April 2017, 15:56:05 UTC
67f7cbe updated readme 03 April 2017, 15:22:21 UTC
0c48a8c Make property at NDArrayView/Value: is_sparse, device, is_read_only; add dtype 03 April 2017, 05:42:46 UTC
fba356f CNTK v2 library: Several sparse data handling fixes 1) Handle conversion of a single scipy.csr_matrix to a Value object 2) Better error handling for conversion to csr matrices and fixed a perf issue in the conversion. 3) Fixed the AssignOneHot sparse matrix implementations 4) Add sparse/dense value type inference in the computation network and centralize matrix shape determination for sparse vs. dense matrices. 03 April 2017, 01:40:26 UTC
8148cab CNTK v2 library: Fix per_dim_mean_variance_normalize op. 02 April 2017, 18:34:12 UTC
bed0e33 CNTK v2 library: Fix conversion of sparse Value/NDArrayView objects to scipy csr format 02 April 2017, 18:34:12 UTC
e61af8c CNTK v2 library: Add str() for Axis 02 April 2017, 18:34:12 UTC
7c6304e Integrate wilrich/fix_default_options into master 02 April 2017, 08:05:58 UTC
5eca042 Integrate chenta/one_hot_bug_fix into master 01 April 2017, 00:48:43 UTC
af66b65 Update index.rst 31 March 2017, 22:51:54 UTC
0bd078f Update typing.py 31 March 2017, 22:51:01 UTC
cac9fb1 Integrate kedeng/fasterTimes2 into master 31 March 2017, 22:16:55 UTC
63ef0b3 add more unit tests 31 March 2017, 21:27:40 UTC
e2edc80 Import all from default_options in cntk namespace 31 March 2017, 21:07:14 UTC
72f7a41 remove useless code 31 March 2017, 18:47:20 UTC
59a2c26 Make sparse label CE backprop faster When minibatch size is big (like 10000), the diagonal matrix in times node for gradient could be big. This change implements a ColumnwiseScaleAndWeightedAdd to reduce the cost in that case. 31 March 2017, 18:07:11 UTC
5651d57 Integrate chazhang/conv_nv1 into master 31 March 2017, 17:20:01 UTC
c4b9dd5 Match Python script parameter with BrainScript. 31 March 2017, 14:45:12 UTC
0772f60 Autopep; shorten the device descrption 31 March 2017, 09:54:32 UTC
e1786b1 Typemap NDArrayView.device() 31 March 2017, 08:53:51 UTC
4ddb7f1 Bug fix and check back in NVidia's change for convolution workspace. 31 March 2017, 05:21:20 UTC
88ecaa4 Fix test caused by previous check-in. 31 March 2017, 05:17:50 UTC
20d473d Fix ConvNet CIFAR10 scripts for namespace change. 31 March 2017, 04:17:36 UTC
f8de646 Fixes of ConvNet examples for namespace change. 31 March 2017, 04:03:48 UTC
16e5acb Fixed Fast R-CNN after API changes 30 March 2017, 22:29:26 UTC
8a9020f CNTK v2 library: Migrate past_value and future_value to sequence 30 March 2017, 19:11:20 UTC
0d2879e CNTK v2 library: Remove remnants of the use of deprecated input_variable and placeholder_variable methods 30 March 2017, 19:11:20 UTC
f32fde5 CNTK v2 library: Add support for reporting distributed training synchronization updates in the ProgressWriter. 30 March 2017, 19:11:20 UTC
e5458e9 Integrate fseide/s2spy into master 30 March 2017, 17:21:47 UTC
08e1945 Integrate eldak/addVerbosityToTest into master 30 March 2017, 16:23:55 UTC
48a3f6b fixed a doc link error 30 March 2017, 16:18:05 UTC
ff3f92d Integrate eldak/philippsFix into master 30 March 2017, 16:16:05 UTC
2439ee9 Merge branch 'master' of https://github.com/Microsoft/cntk into fseide/s2spy 30 March 2017, 15:31:27 UTC
2d9f31d addressed CR feedback 30 March 2017, 15:31:06 UTC
219ff7f Integrate nikmil/upsampling_layer into master 30 March 2017, 14:32:16 UTC
729bd3e Adding more output in case of error 30 March 2017, 14:28:39 UTC
300d229 Pilipps fix 30 March 2017, 14:23:08 UTC
8f33920 Add bilinear upsampling layer Add bilinear upsampling layer implemented using transposed convolution with biliner initialization and learning rate multiplier set to zero. Conflicts: Source/CNTK/BrainScript/CNTKCoreLib/CNTK.core.bs 30 March 2017, 12:43:08 UTC
dc0a5e2 Removing deprecated stuff from training session 30 March 2017, 07:31:59 UTC
8af30f6 Integrate chazhang/convfix into master 30 March 2017, 07:00:38 UTC
e1695eb Bug fixes. 30 March 2017, 03:59:28 UTC
051a667 Some bug fixes. 30 March 2017, 00:38:54 UTC
e68b171 Integrate ebarsoum/cleanup into master 30 March 2017, 00:04:35 UTC
74c2a30 Revert Nvidia convolution code to assure convolution accuracy. 29 March 2017, 23:58:46 UTC
30941f2 Change poolPadMode to poolIncludePad. 29 March 2017, 23:56:46 UTC
9ddb343 Fix convolution. Working but transfer learning accuracy low. 29 March 2017, 23:56:46 UTC
3da1e6b Merge branch 'master' of https://github.com/Microsoft/CNTK 29 March 2017, 23:51:50 UTC
e0f443b Merge branch 'master' of https://github.com/Microsoft/CNTK 29 March 2017, 23:19:00 UTC
559e92b (typos in docstrings) 29 March 2017, 22:43:07 UTC
ffa1b9d moved @Function docstring from _to_Function (incorrect place) to class constructor, and added doctest examples 29 March 2017, 22:38:28 UTC
51aa4fa added a module docstring to typing.py with doc tests; changed Tensor[] to save the single axis as a list to make it more uniform to inspect 29 March 2017, 22:25:31 UTC
13f2c0a CNTK v2 library: Removed a trainer test that expected an exception in a valid Function graph due to a bug that is now fixed. While Eldar (the test author) and I discuss how to replace this with a test that covers the intention of the test, removing this to commit the bug fixes. 29 March 2017, 22:24:05 UTC
d251977 Merge with master. 29 March 2017, 22:05:51 UTC
05147af More fixes and examples for Parameter and Constant. 29 March 2017, 21:52:37 UTC
c474026 CNTK v2 library: Improve error message for Constants will mismatching datatype that cannot be implicitly coerced to the required type 29 March 2017, 21:43:51 UTC
d302daf CNTK v2 library: Bug fixes 1) Fix Function::FindByName to properly traverse entire graph when traversing inside blocks 2) Fix handling of empty blocks that just pass through their inputs 29 March 2017, 21:43:51 UTC
02bf4de Fix graph_test. 29 March 2017, 21:39:11 UTC
5bbf422 Revert "fix workspace corner case and reduce memory footprint" This reverts commit 1acef6f9898fa3a7375a6e05ce7dbcae0b3a800c. 29 March 2017, 20:20:54 UTC
5269f46 Revert "More fix from Deyu" This reverts commit d53936c4efa45b044ba06337b3e00d652ec2d938. 29 March 2017, 20:20:36 UTC
d8a3d47 More Jenkins fixes. 29 March 2017, 20:08:02 UTC
cf569da Fix doc test errors and add them to build script. 29 March 2017, 18:33:35 UTC
efb20de Integrate wilrich/miscFixes into master 29 March 2017, 18:32:49 UTC
f339507 Fix VS autocorrect. 29 March 2017, 17:44:41 UTC
43cd200 Fix comments and remaining notebooks. 29 March 2017, 17:41:59 UTC
6b0d979 Fix jenkins reported issues. 29 March 2017, 16:50:55 UTC
9adade3 Adapting examples; set MinibatchData.data to be a property 29 March 2017, 15:09:13 UTC
cfc178c Integrate eldak/fixingBroadcastScope into master 29 March 2017, 14:56:54 UTC
1368b2d Make one_hot PY2.7 compliant 29 March 2017, 14:31:40 UTC
feab98f Rename to_seq() to as_sequences() 29 March 2017, 14:31:38 UTC
fb5091a Replace __array_interface__ by .asarray() member methods; repurpose cntk.asarray() 29 March 2017, 14:31:36 UTC
d2fddcc Supporting NumPy's int types in Value.one_hot 29 March 2017, 14:31:35 UTC
579fce1 Fixing initialization of acc eval 29 March 2017, 11:25:41 UTC
a8e961d Merge branch 'master' into ebarsoum/cleanup For Jenkins. 29 March 2017, 09:08:23 UTC
90287b6 Fix Notebook... 29 March 2017, 09:07:15 UTC
12d116c Fixing broadcast scope 29 March 2017, 08:57:58 UTC
1996f9e Integrate zhouwang/cseval-refactor-value into master 29 March 2017, 07:59:48 UTC
back to top