1f9b938 | Hugh Perkins | 17 May 2016, 13:33:35 UTC | Add some doc | 17 May 2016, 13:33:35 UTC |
7c6cdf0 | Soumith Chintala | 11 May 2016, 20:19:07 UTC | Merge pull request #269 from lukeyeager/rockspec-cxx-flags Add CMAKE_CXX_FLAGS to rockspec | 11 May 2016, 20:19:07 UTC |
37f9196 | Luke Yeager | 11 May 2016, 20:02:33 UTC | Add CMAKE_CXX_FLAGS to rockspec Helps with building on Ubuntu 16.04 | 11 May 2016, 20:10:35 UTC |
3495625 | Soumith Chintala | 01 May 2016, 18:48:42 UTC | Merge pull request #267 from apaszke/spatial_nll Add SpatialClassNLLCriterion (with pre-Kepler support) | 01 May 2016, 18:48:42 UTC |
ef166c8 | Adam Paszke | 14 April 2016, 13:50:36 UTC | Add SpatialClassNLLCriterion (with pre-Kepler support) | 01 May 2016, 18:37:05 UTC |
c9b0b3f | Soumith Chintala | 01 May 2016, 16:41:49 UTC | Merge pull request #266 from torch/revert-250-spatial_nll Revert "Add SpatialClassNLLCriterion" | 01 May 2016, 16:41:49 UTC |
c097e16 | Soumith Chintala | 01 May 2016, 16:41:28 UTC | Revert "Add SpatialClassNLLCriterion" | 01 May 2016, 16:41:28 UTC |
89cf5f1 | Soumith Chintala | 28 April 2016, 18:18:43 UTC | Merge pull request #261 from torch/dilated Adding SpatialDilatedConvolution | 28 April 2016, 18:18:43 UTC |
229a980 | Soumith Chintala | 28 April 2016, 03:46:40 UTC | Merge pull request #250 from apaszke/spatial_nll Add SpatialClassNLLCriterion | 28 April 2016, 03:46:40 UTC |
dca99d8 | soumith | 27 April 2016, 22:21:54 UTC | Adding SpatialDilatedConvolution | 27 April 2016, 22:21:54 UTC |
3d97620 | Soumith Chintala | 27 April 2016, 19:20:06 UTC | Merge pull request #260 from torch/multilabelmargin adding CUDA version of MultiLabelMarginCriterion + tests | 27 April 2016, 19:20:06 UTC |
2b9dff8 | soumith | 27 April 2016, 18:56:00 UTC | adding CUDA version of MultiLabelMarginCriterion + tests | 27 April 2016, 18:56:00 UTC |
9558c29 | Soumith Chintala | 27 April 2016, 18:54:20 UTC | Merge pull request #259 from torch/dptfixes2 fixes to DataParallelTable for no parameter networks | 27 April 2016, 18:54:20 UTC |
b0c106c | soumith | 27 April 2016, 18:52:02 UTC | fixes to DataParallelTable for no parameter networks | 27 April 2016, 18:52:02 UTC |
6474907 | Adam Paszke | 14 April 2016, 13:50:36 UTC | Add SpatialClassNLLCriterion | 26 April 2016, 19:04:05 UTC |
17038fd | Soumith Chintala | 25 April 2016, 21:53:10 UTC | Merge pull request #258 from apaszke/dpt_clearstate Fix clearState in DataParallelTable | 25 April 2016, 21:53:10 UTC |
8737702 | Adam Paszke | 25 April 2016, 21:51:39 UTC | Fix clearState in DataParallelTable Make it return the module, just like all other modules do. | 25 April 2016, 21:51:39 UTC |
24daf02 | Soumith Chintala | 25 April 2016, 16:33:26 UTC | Merge pull request #257 from ngimel/clearState fix clearState for DataParallelTable | 25 April 2016, 16:33:26 UTC |
d764ff2 | Natalia Gimelshein | 25 April 2016, 16:14:02 UTC | fix clearState for DataParallelTable | 25 April 2016, 16:14:02 UTC |
6c3c3f3 | Soumith Chintala | 20 April 2016, 20:02:11 UTC | fixing tests sizes in Reflection and Replication padding tests to be valid (#255) | 20 April 2016, 20:02:11 UTC |
ca1b6e8 | Adam Paszke | 19 April 2016, 21:45:35 UTC | Fix VolumetrixMaxPooling tests (#254) | 19 April 2016, 21:45:35 UTC |
f03b92f | Kaiyu Yang | 18 April 2016, 15:27:25 UTC | add noBias for nn.Linear and nn.SpatialConvolution (#252) add tests testing done add noBias for nn.Linear and nn.SpatialConvolution add check in THCUNN_assertSameGPU add noBias for nn.Linear and nn.SpatialConvolution minor edits add noBias for nn.Linear and nn.SpatialConvolution | 18 April 2016, 15:27:25 UTC |
4e56a81 | Kaiyu Yang | 16 April 2016, 17:33:48 UTC | add backpropagation for bachnormalization in evaluation mode (#251) add tests for batchnormalization in evaluation mode fix bugs BatchNormalization: compute save_mean and save_std in evaluattion mode | 16 April 2016, 17:33:48 UTC |
3d75c1e | soumith | 13 April 2016, 15:41:41 UTC | distkldiv test fix | 13 April 2016, 15:41:41 UTC |
dd13411 | Thomas Köppe | 12 April 2016, 18:01:23 UTC | Remove use of broken THInf macro, which caused UB * [VolmetricMaxPooling] Remove use of broken THInf macro, which caused UB * [SpatialFractionalMaxPooling] Remove use of broken THInf macro, which caused UB | 12 April 2016, 18:01:23 UTC |
1b0c643 | Xianjie Chen | 11 April 2016, 20:40:44 UTC | [LookupTable] Add Max-norm constraints to LookupTable (#240) | 11 April 2016, 20:40:44 UTC |
2fee130 | Soumith Chintala | 08 April 2016, 15:13:10 UTC | Merge pull request #244 from hughperkins/cunn-test-alias Add alias cunn.test(), following eg format of cutorch.test() | 08 April 2016, 15:13:10 UTC |
d9e4072 | Soumith Chintala | 08 April 2016, 14:47:10 UTC | Merge pull request #243 from hughperkins/remove-testsuite-warning remove warnings about testsuite | 08 April 2016, 14:47:10 UTC |
0687c41 | Hugh Perkins | 08 April 2016, 13:37:02 UTC | Add alias cunn.test(), following eg format of cutorch.test() | 08 April 2016, 13:37:02 UTC |
d168951 | Hugh Perkins | 08 April 2016, 13:28:14 UTC | remove warnings about testsuite | 08 April 2016, 13:28:14 UTC |
c9f24c7 | Soumith Chintala | 22 March 2016, 13:45:00 UTC | Merge pull request #239 from caldweln/cuda make margin parameterizable | 22 March 2016, 13:45:00 UTC |
cb3122d | nick | 22 March 2016, 09:08:14 UTC | make margin parameterizable | 22 March 2016, 09:08:14 UTC |
64959b2 | Soumith Chintala | 18 March 2016, 18:33:51 UTC | Merge pull request #223 from ebetica/sparse_linear Adding SparseLinear with CUDA | 18 March 2016, 18:33:51 UTC |
cc683ed | Zeming Lin | 29 February 2016, 23:39:43 UTC | Adding SparseLinear for CUDA | 18 March 2016, 18:14:10 UTC |
667f47b | Soumith Chintala | 14 March 2016, 17:16:10 UTC | Merge pull request #238 from anibali/inplace_elu In-place ELU | 14 March 2016, 17:16:10 UTC |
908e558 | Aiden Nibali | 10 March 2016, 00:22:21 UTC | In-place ELU | 10 March 2016, 01:01:29 UTC |
3827fcd | Soumith Chintala | 09 March 2016, 12:46:04 UTC | Merge pull request #237 from apaszke/err_msg_fix THNN -> THCUNN in assertSameGPU | 09 March 2016, 12:46:04 UTC |
6ef8d65 | Adam Paszke | 09 March 2016, 09:15:17 UTC | THNN -> THCUNN in assertSameGPU | 09 March 2016, 09:15:17 UTC |
8a5347d | Soumith Chintala | 08 March 2016, 21:57:02 UTC | Merge pull request #235 from apaszke/err_msg Add an error message for checkGPU assertion | 08 March 2016, 21:57:02 UTC |
ee63c24 | Adam Paszke | 08 March 2016, 21:01:24 UTC | Add THCUNN_h.lua to .gitignore | 08 March 2016, 21:01:24 UTC |
5b3dc54 | Adam Paszke | 08 March 2016, 21:00:59 UTC | Add an error message for checkGPU | 08 March 2016, 21:00:59 UTC |
c841015 | Soumith Chintala | 08 March 2016, 18:19:43 UTC | Merge pull request #234 from fmassa/THCUNN_header Reuse headers from THCUNN.h in THCUNN.lua | 08 March 2016, 18:19:43 UTC |
5b02b61 | fsuzanomassa | 08 March 2016, 18:00:09 UTC | Reuse headers from THCUNN.h in THCUNN.lua | 08 March 2016, 18:00:09 UTC |
fc2a0b3 | Soumith Chintala | 06 March 2016, 16:55:47 UTC | fix cuda arch for cuda 6.5 | 06 March 2016, 16:55:47 UTC |
514c0c0 | Soumith Chintala | 01 March 2016, 00:09:49 UTC | Merge pull request #230 from szagoruyko/patch-3 Remove CMakeLists CUDA 5 os x workaround | 01 March 2016, 00:09:49 UTC |
1e02bf9 | Sergey Zagoruyko | 29 February 2016, 23:46:37 UTC | remove CMakeLists CUDA 5 os x bugfix not relevant as we are on 6.5 at least | 29 February 2016, 23:46:37 UTC |
a974027 | Soumith Chintala | 29 February 2016, 23:36:43 UTC | Merge pull request #229 from torch/cleanup removing useless unit.cu, utils.h and utils.cu | 29 February 2016, 23:36:43 UTC |
a6aa130 | soumith | 29 February 2016, 23:36:20 UTC | removing useless unit.cu, utils.h and utils.cu | 29 February 2016, 23:36:20 UTC |
41c750f | Soumith Chintala | 29 February 2016, 21:19:08 UTC | Merge pull request #228 from torch/softmargincriterion SoftMarginCriterion | 29 February 2016, 21:19:08 UTC |
932bb48 | soumith | 29 February 2016, 21:17:44 UTC | SoftMarginCriterion | 29 February 2016, 21:17:44 UTC |
629f028 | Soumith Chintala | 29 February 2016, 19:02:56 UTC | Merge pull request #214 from Jerrynet/master Do flattenParameters() only once | 29 February 2016, 19:02:56 UTC |
fef7354 | Soumith Chintala | 27 February 2016, 16:22:56 UTC | Merge pull request #227 from fmassa/conv_assert Add extra checks for convolution parameters | 27 February 2016, 16:22:56 UTC |
e73c473 | Francisco Massa | 27 February 2016, 11:11:38 UTC | Add checks for convolution parameters | 27 February 2016, 11:11:38 UTC |
f3b07db | Soumith Chintala | 26 February 2016, 04:06:52 UTC | Merge pull request #224 from colesbury/bn Add per-activation BatchNormalization implementation. | 26 February 2016, 04:06:52 UTC |
8cd0609 | Sam Gross | 24 February 2016, 22:15:27 UTC | Add per-activation BatchNormalization implementation. The batch normalization code now works for 2D (per-activation), 4D (spatial) and 5D (volumetric) tensors. The per-activation case is not optimized. | 25 February 2016, 20:02:06 UTC |
9a297de | Soumith Chintala | 25 February 2016, 01:24:37 UTC | Merge pull request #225 from torch/reflect adding SpatialReflectionPadding and SpatialReplicationPadding | 25 February 2016, 01:24:37 UTC |
86ef4ae | soumith | 25 February 2016, 01:11:21 UTC | adding SpatialReflectionPadding and SpatialReplicationPadding | 25 February 2016, 01:23:10 UTC |
ef1f945 | Soumith Chintala | 23 February 2016, 19:05:57 UTC | Merge pull request #222 from fbesse/volumetric_full_fix Fixed VolumetricFullConvolution. | 23 February 2016, 19:05:57 UTC |
2b5dd14 | Soumith Chintala | 20 February 2016, 03:25:15 UTC | Merge pull request #219 from jhjin/master Remove redundant and unused headers | 20 February 2016, 03:25:15 UTC |
508eae3 | Jonghoon Jin | 20 February 2016, 02:33:27 UTC | Remove redundant and unused headers Clean-up after the THCUNN conversion | 20 February 2016, 02:34:52 UTC |
12ada98 | Soumith Chintala | 20 February 2016, 01:47:59 UTC | Merge pull request #218 from torch/multimargweight added weights to MultiMarginCriterion | 20 February 2016, 01:47:59 UTC |
56c3ed2 | soumith | 20 February 2016, 01:42:47 UTC | added weights to MultiMarginCriterion | 20 February 2016, 01:42:47 UTC |
de6acba | Soumith Chintala | 19 February 2016, 23:48:14 UTC | Merge pull request #217 from torch/padValue adding padValue to LookupTable | 19 February 2016, 23:48:14 UTC |
c00281f | soumith | 19 February 2016, 19:07:32 UTC | adding padValue to LookupTable | 19 February 2016, 23:36:40 UTC |
2456639 | soumith | 19 February 2016, 21:03:40 UTC | fix merge typo | 19 February 2016, 21:03:40 UTC |
7be711d | Soumith Chintala | 19 February 2016, 20:14:48 UTC | Merge pull request #215 from torch/multmargfix Fix sizeAverage bug in MultiMarginCriterion (divide by nbatch, not ndim) | 19 February 2016, 20:14:48 UTC |
6c19d69 | soumith | 19 February 2016, 20:11:55 UTC | multmargsizeavgfix | 19 February 2016, 20:11:55 UTC |
be370df | Jerry Lin | 19 February 2016, 13:11:55 UTC | Do flattenParameters() only once Prevent parameters to be flattened more than once | 19 February 2016, 13:11:55 UTC |
0e23843 | Frederic Besse | 18 February 2016, 15:33:00 UTC | Reimplemented VolumetricFullConvolution in the same fashion as SpatialFullConvolution. | 18 February 2016, 15:33:00 UTC |
c837a7f | Soumith Chintala | 17 February 2016, 23:52:59 UTC | Merge pull request #207 from fmassa/THCUNN_fmassa2 THCUNN Spatial* conversion | 17 February 2016, 23:52:59 UTC |
f25d5b6 | Soumith Chintala | 17 February 2016, 21:59:35 UTC | Merge pull request #212 from colesbury/master Zero gradients on secondary GPUs after accGradParameters | 17 February 2016, 21:59:35 UTC |
95f351f | Sam Gross | 17 February 2016, 21:28:11 UTC | Zero gradients on secondary GPUs after accGradParameters Before this change, calling forward & backward multiple times without calling zeroGradParameters would compute an incorrect gradient. Fixes #209 | 17 February 2016, 21:28:11 UTC |
c79b0c0 | Soumith Chintala | 15 February 2016, 21:59:15 UTC | Merge pull request #211 from apaszke/master Instantiate impl field in older models | 15 February 2016, 21:59:15 UTC |
656adc3 | Adam Paszke | 15 February 2016, 21:50:43 UTC | Instantiate impl field in older models | 15 February 2016, 21:56:41 UTC |
99c34f8 | Soumith Chintala | 13 February 2016, 00:32:41 UTC | Merge pull request #208 from torch/cunn include -> require | 13 February 2016, 00:32:41 UTC |
a6d720c | Soumith Chintala | 13 February 2016, 00:32:08 UTC | include -> require | 13 February 2016, 00:32:08 UTC |
f3c3420 | Francisco Massa | 11 February 2016, 10:08:17 UTC | THNN conversion of Spatial* modules | 11 February 2016, 14:33:42 UTC |
19c1333 | Francisco Massa | 11 February 2016, 07:51:53 UTC | Moving Spatial* to lib/THCUNN | 11 February 2016, 07:51:53 UTC |
42ee94a | Soumith Chintala | 10 February 2016, 16:34:19 UTC | Merge pull request #206 from torch/temporal_thcunn moving Temporal* to THCUNN | 10 February 2016, 16:34:19 UTC |
6e32345 | soumith | 10 February 2016, 16:30:28 UTC | moving Temporal* to THCUNN | 10 February 2016, 16:30:28 UTC |
ab2b009 | Soumith Chintala | 08 February 2016, 20:34:51 UTC | Merge pull request #205 from ivankreso/master Fix thread configuration in inference mode | 08 February 2016, 20:34:51 UTC |
b1dc789 | Ivan Kreso | 08 February 2016, 16:53:04 UTC | Fix thread configuration in inference mode | 08 February 2016, 16:53:04 UTC |
8fc9450 | Soumith Chintala | 03 February 2016, 02:40:22 UTC | Merge pull request #204 from andreaskoepf/THCUNN_koepf_5 TCUHNN functional module conversions batch 5 (Volumetric*) | 03 February 2016, 02:40:22 UTC |
a4bbf8d | Andreas Köpf | 03 February 2016, 00:57:40 UTC | Remove 'dimension aliases' from VolumetricConvolution | 03 February 2016, 00:57:40 UTC |
7473557 | Andreas Köpf | 02 February 2016, 00:43:32 UTC | Add THCUNN conversion of Volumetric* modules | 02 February 2016, 23:38:28 UTC |
a3314eb | Andreas Köpf | 01 February 2016, 21:36:31 UTC | Move Volumetric*.cu -> lib/THCUNN | 02 February 2016, 23:38:28 UTC |
469710a | Soumith Chintala | 02 February 2016, 19:46:07 UTC | Merge pull request #189 from colesbury/master Add multi-threading support in DataParallelTable | 02 February 2016, 19:46:07 UTC |
97f66e1 | Sam Gross | 11 January 2016, 17:37:42 UTC | Add multi-threading support in DataParallelTable This lets us use the fastest CUDNN R4 kernels with DPT, which have much higher launch overhead. With single-threaded dispatch, the launch overhead from these kernels delays the second, third, and fourth GPUs so much that it actually increases the total time. I've also included some other optimizations: - gradInput can be set to nil when the DataParallelTable is the top-level module; this saves time and memory - input can optionally be on the CPU; this saves memory on the first GPU, and is faster when the input is in pinned memory since the copies to each GPU happen concurrently. This change is mostly backwards compatible, with the below caveats: - The preferred way to add modules is by a single call to add with the list of GPUs i.e. add(module, {1, 2, 3, 4}) - DataParallelTable:get() only works with the first module | 02 February 2016, 19:39:46 UTC |
a1cdf09 | Soumith Chintala | 02 February 2016, 14:16:03 UTC | Merge pull request #198 from andreaskoepf/THCUNN_koepf_4 THCUNN functional module conversions batch 4 | 02 February 2016, 14:16:03 UTC |
563b1e7 | soumith | 02 February 2016, 04:41:10 UTC | minor bufix in LeakyReLU | 02 February 2016, 04:41:10 UTC |
cd76372 | soumith | 02 February 2016, 04:18:29 UTC | adding 5.2 cuda arch to cmake | 02 February 2016, 04:18:29 UTC |
c7ee6c3 | Andreas Köpf | 28 January 2016, 22:45:07 UTC | Add THCUNN conversions of {RReLU, Sigmoid, SmoothL1Criterion, ...} Converted modules: RReLU Sigmoid SmoothL1Criterion SoftMax SoftPlus SoftShrink Sqrt Square Tanh Threshold | 01 February 2016, 21:22:28 UTC |
b382db0 | Andreas Köpf | 28 January 2016, 21:33:26 UTC | Move {RReLU.cu, Sigmoid.cu, SmoothL1Criterion.cu,..} to lib/THCUNN Files moved: RReLU.cu Sigmoid.cu SmoothL1Criterion.cu SoftMax.cu SoftPlus.cu SoftShrink.cu Sqrt.cu Square.cu Tanh.cu Threshold.cu | 01 February 2016, 20:55:50 UTC |
707e087 | Andreas Köpf | 24 January 2016, 23:56:43 UTC | Add THCUNN conversions of {MSE, Margin, MultiMargin}Criterion & PReLU | 01 February 2016, 20:55:50 UTC |
7a30e65 | Andreas Köpf | 23 January 2016, 23:47:28 UTC | Move {MarginCriterion, MSECriterion, MultiMarginCriterion, PReLU}.cu -> lib/THCUNN | 01 February 2016, 20:45:58 UTC |
01d6c74 | Soumith Chintala | 01 February 2016, 13:06:35 UTC | Merge pull request #203 from fmassa/THCUNN_fmassa_1 THCUNN functional conversion batch 3 | 01 February 2016, 13:06:35 UTC |
a02b65f | Francisco Massa | 26 January 2016, 22:17:39 UTC | Add THNN conversion of {Spatial(Adaptive,Average,Max)Pooling} and SpatialConvolutionMM | 27 January 2016, 22:02:53 UTC |
95724b4 | Francisco Massa | 17 January 2016, 12:04:14 UTC | Move {SpatialConvolutionMM, Spatial(Adaptive,Average,Max)Pooling} to lib/TCUNN | 27 January 2016, 21:59:39 UTC |
edd692a | Soumith Chintala | 27 January 2016, 20:18:31 UTC | Merge pull request #202 from kmul00/volmaxpool_pad Added padding option to Volumetric Max Pooling (CUDA) | 27 January 2016, 20:18:31 UTC |
eeff62e | kmul00 | 27 January 2016, 18:13:49 UTC | Added padding option to Volumetric Max Pooling (CUDA) modified: VolumetricMaxPooling.cu modified: test.lua | 27 January 2016, 18:13:49 UTC |