| 1ae6aa0 | Soumith Chintala | 07 September 2017, 06:15:28 UTC | fix static linkage and make THD statically linked | 07 September 2017, 06:41:45 UTC |
| 8980c40 | Alykhan Tejani | 24 July 2017, 18:58:43 UTC | Add numerically stable logsigmoid | 27 August 2017, 02:19:48 UTC |
| 283539e | Lu Fang | 30 July 2017, 18:22:02 UTC | Adding implicit padding for 3d average pooling | 26 August 2017, 18:44:59 UTC |
| f486923 | Zhou Mo | 30 July 2017, 12:38:00 UTC | Fix typos. | 25 August 2017, 18:25:58 UTC |
| a6522cf | Christian Sarofeen | 19 July 2017, 21:10:29 UTC | Updates for CUDA 9 | 25 August 2017, 11:27:25 UTC |
| 90afcbf | Nicholas Léonard | 03 August 2017, 20:46:50 UTC | Merge pull request #480 from nicholas-leonard/BN-batchsize1 BN supports batchsize=1 | 03 August 2017, 20:46:50 UTC |
| ab0ee1f | Nicholas Leonard | 03 August 2017, 19:22:55 UTC | BN supports batchsize=1 | 03 August 2017, 19:22:55 UTC |
| e9ef2d5 | Soumith Chintala | 03 August 2017, 02:03:45 UTC | remove limitations on output_padding in Conv* routines | 03 August 2017, 02:46:24 UTC |
| bbebfdc | Soumith Chintala | 28 July 2017, 03:17:44 UTC | add 2d and 3d dilated full Convolution | 03 August 2017, 02:44:59 UTC |
| 8d9e956 | Soumith Chintala | 25 July 2017, 21:01:59 UTC | Merge pull request #477 from wickedfoo/feature_lp_pooling GPU implementation of L_p feature pooling | 25 July 2017, 21:01:59 UTC |
| b336dc9 | Soumith Chintala | 24 July 2017, 15:22:18 UTC | Merge pull request #479 from mikepound/upsampling Added cunn tests for UpSampling module. | 24 July 2017, 15:22:18 UTC |
| 528a871 | Michael Pound | 24 July 2017, 13:07:08 UTC | Added cunn tests for UpSampling module. | 24 July 2017, 13:07:08 UTC |
| e9d54e1 | Soumith Chintala | 20 July 2017, 15:01:52 UTC | Merge pull request #478 from singam-sanjay/correct_README Update README | 20 July 2017, 15:01:52 UTC |
| 203a547 | Singapuram Sanjay Srivallabh | 20 July 2017, 09:56:15 UTC | Update README Clarify control-flow oddities in control-flow terminology. | 20 July 2017, 09:56:15 UTC |
| a4381e1 | Soumith Chintala | 14 July 2017, 20:27:20 UTC | Static linking against libstdc++ in Binary Build mode | 19 July 2017, 15:23:21 UTC |
| 27d79db | Natalia Gimelshein | 13 July 2017, 22:12:29 UTC | add launch_bounds to greedy kernels | 13 July 2017, 23:23:29 UTC |
| 2c84b98 | Jeff Johnson | 26 June 2017, 16:26:26 UTC | LP pooling kernels | 13 July 2017, 02:31:06 UTC |
| c3c0d9b | Soumith Chintala | 03 July 2017, 04:36:49 UTC | Merge pull request #476 from lospooky/SpatialDepthWiseConvolution-segfault Fix segfault in SpatialDepthWiseConvolution w/o bias | 03 July 2017, 04:36:49 UTC |
| 28e9b2b | Simone Cirillo | 23 June 2017, 09:09:28 UTC | Fix segfault in SpatialDepthWiseConvolution w/o bias | 23 June 2017, 09:14:00 UTC |
| fd47807 | cph | 21 June 2017, 10:34:26 UTC | add asserts to BCECriterion | 22 June 2017, 17:20:25 UTC |
| 9cffa0e | Soumith Chintala | 12 June 2017, 19:42:37 UTC | nn.EmbeddingBag to compute a bag of word embeddings (Embedding + Sum/Mean) | 15 June 2017, 16:32:28 UTC |
| 42c92bf | Sam Gross | 06 June 2017, 21:15:31 UTC | Added GLU (gated linear unit) From https://arxiv.org/abs/1612.08083 | 14 June 2017, 00:48:03 UTC |
| 3d484ec | Luca Antiga | 29 May 2017, 17:02:05 UTC | Add 3D upsampling (nearest and trilinear) with tests | 07 June 2017, 15:24:41 UTC |
| a9c4d64 | Soumith Chintala | 07 June 2017, 13:21:20 UTC | fix GRUFused signature | 07 June 2017, 13:21:20 UTC |
| 3087b87 | Soumith Chintala | 07 June 2017, 03:31:54 UTC | Merge pull request #470 from qqning/master Fix the mix-up of height and width on depth-wise convolution | 07 June 2017, 03:31:54 UTC |
| 8e3364c | Christian Sarofeen | 07 June 2017, 03:20:14 UTC | Remove clone in fused rnn | 07 June 2017, 03:20:14 UTC |
| 8c6df2a | Soumith Chintala | 06 June 2017, 19:06:41 UTC | Merge pull request #472 from twitter-forks/indexlinear-fix Fixing incorrect normalized values in IndexLinear during training | 06 June 2017, 19:06:41 UTC |
| 36d1b76 | Pavan Yalamanchili | 06 June 2017, 18:44:11 UTC | Fixing the issue with incorrect normalized values in IndexLinear | 06 June 2017, 18:44:11 UTC |
| a55fb73 | ningqingqun | 24 May 2017, 13:05:08 UTC | Fix the mix-up of height and width on depth-wise convolution | 24 May 2017, 13:05:08 UTC |
| f2b2286 | Soumith Chintala | 21 May 2017, 17:49:04 UTC | Merge pull request #468 from nicholas-leonard/ClassNLLCriterion ClassNLLCriterion ignoreIndex | 21 May 2017, 17:49:04 UTC |
| 53f7b25 | Nicholas Leonard | 16 May 2017, 02:27:00 UTC | ClassNLLCriterion ignoreIndex | 16 May 2017, 02:27:00 UTC |
| 501b31c | Soumith Chintala | 15 May 2017, 01:15:42 UTC | Merge pull request #467 from torch/revert-458-master Revert "Update to ignore zero targets" | 15 May 2017, 01:15:42 UTC |
| a5ae723 | Soumith Chintala | 15 May 2017, 01:15:30 UTC | Revert "Update to ignore zero targets" | 15 May 2017, 01:15:30 UTC |
| e97095d | stooloveu | 05 January 2017, 03:57:30 UTC | SpatialDepthWiseConvolution.cu added | 12 May 2017, 20:02:14 UTC |
| 3845f85 | Nicholas Léonard | 12 May 2017, 14:38:18 UTC | Merge pull request #458 from jnhwkim/master Update to ignore zero targets | 12 May 2017, 14:38:18 UTC |
| 8252eb7 | Gregory Chanan | 03 May 2017, 18:03:58 UTC | Add a keepdim parameter for reduction functions over a single dimension. By default, this parameter is False -- a backwards incompatible change, but one that follows numpy semantics, e.g. numpy.sum (numpy names the parameter "keepdims" since you can pass multiple dims to reduction functions). The old behavior seems desired for normalization type operations where the tensor will immediately be expanded out again, e.g.: probs.sum(1).expand_as(probs) which no longer works because the dimension to expand is missing. This can be fixed by simply passing True as "keepdim" argument to the reduction operation, e.g: probs.sum(1, keepdim=True).expand_as(probs) | 09 May 2017, 18:55:42 UTC |
| b9ab0f7 | Soumith Chintala | 22 April 2017, 08:14:09 UTC | fix typo | 22 April 2017, 08:14:09 UTC |
| 93d3167 | Christian Sarofeen | 21 April 2017, 20:53:50 UTC | Indexing fix for fused GRU/LSTM kernels when all tensors are not contiguous. | 22 April 2017, 08:09:46 UTC |
| 455e488 | Soumith Chintala | 22 April 2017, 08:05:43 UTC | Merge pull request #465 from torch/cunnchecks add contiguous checks | 22 April 2017, 08:05:43 UTC |
| 0112e3f | Soumith Chintala | 22 April 2017, 07:57:36 UTC | add contiguous checks | 22 April 2017, 07:57:36 UTC |
| ea15d0d | Christian Sarofeen | 18 April 2017, 17:37:25 UTC | Fused RNN kernel remove explicit instantiation, isn't needed. | 18 April 2017, 18:07:58 UTC |
| 9560085 | Soumith Chintala | 18 April 2017, 17:31:08 UTC | Merge pull request #463 from apaszke/sig_tanh Remove double precision math from LogSigmoid too | 18 April 2017, 17:31:08 UTC |
| 61dcc2d | Adam Paszke | 18 April 2017, 17:28:13 UTC | Remove double precision math from LogSigmoid too | 18 April 2017, 17:28:13 UTC |
| 5b4a2d5 | Soumith Chintala | 18 April 2017, 17:19:52 UTC | Merge pull request #462 from apaszke/sig_tanh Update ops for Sigmoid and Tanh | 18 April 2017, 17:19:52 UTC |
| 9c6905e | Adam Paszke | 18 April 2017, 16:45:39 UTC | Update ops for Sigmoid and Tanh | 18 April 2017, 16:55:11 UTC |
| e246952 | soumith | 11 April 2017, 21:55:09 UTC | fix THNN headers | 11 April 2017, 21:57:11 UTC |
| ca8fbb7 | Christian Sarofeen | 23 March 2017, 23:14:09 UTC | Fused pointwise kernels for GRU/LSTM | 11 April 2017, 20:36:38 UTC |
| 536f41a | Soumith Chintala | 09 April 2017, 17:52:56 UTC | Merge pull request #455 from twitter-forks/indexlinear Adding Indexlinear | 09 April 2017, 17:52:56 UTC |
| 14b181b | Soumith Chintala | 07 April 2017, 16:29:55 UTC | Merge pull request #459 from SYSTRAN/feature/support_TORCH_NVCC_FLAGS Support TORCH_NVCC_FLAGS environment variable | 07 April 2017, 16:29:55 UTC |
| 29161b3 | Thomas Riccardi | 07 April 2017, 16:23:22 UTC | Support TORCH_NVCC_FLAGS environment variable This is already supported in cutorch since august 2016, and is used in pytorch integration (to reduce the binary size). | 07 April 2017, 16:23:22 UTC |
| 5ea1105 | Jin-Hwa Kim | 05 April 2017, 18:51:54 UTC | Update to ignore zero targets If the target is zero, loss and gradient of input are set to zero. It is useful for variable-length natural language generation models. | 05 April 2017, 18:51:54 UTC |
| f225bb4 | Soumith Chintala | 31 March 2017, 18:44:07 UTC | Merge pull request #456 from twitter-forks/addmm-fixes Using temporary variables when performing transpose + addmm | 31 March 2017, 18:44:07 UTC |
| 54fdaa8 | Pavan Yalamanchili | 29 March 2017, 23:56:39 UTC | Using temporary variables when performing transpose + addmm | 29 March 2017, 23:56:39 UTC |
| 97940f0 | Pavan Yalamanchili | 15 February 2017, 23:42:13 UTC | Improving the performance of IndexLinear:updateOutput - Removes separate kernel for updateOutputTrain | 24 March 2017, 23:34:31 UTC |
| 73beb06 | Soumith Chintala | 24 March 2017, 00:29:17 UTC | Merge pull request #454 from ngimel/patch-1 Fix inconsistent in-place and out-of-place for HardTanh | 24 March 2017, 00:29:17 UTC |
| 67e30ae | ngimel | 24 March 2017, 00:27:29 UTC | Fix inconsistent in-place and out-of-place for HardTanh in-place and out-of-place updateGradOutput results are different where input=min_val or input=max_val | 24 March 2017, 00:27:29 UTC |
| e7783b3 | Pavan Yalamanchili | 14 February 2017, 01:00:00 UTC | Adding support for flattened inputs for IndexLinear - Adding relevant tests | 23 March 2017, 21:18:41 UTC |
| 0816bc2 | Pavan Yalamanchili | 13 October 2016, 20:00:11 UTC | IndexLinear support for cunn | 23 March 2017, 21:18:01 UTC |
| 6b6a462 | Soumith Chintala | 22 March 2017, 15:53:41 UTC | Merge pull request #453 from apaszke/lookup_renorm Cast accumulator in LookupTable renorm to accreal | 22 March 2017, 15:53:41 UTC |
| ab320e2 | Hardik | 05 March 2017, 23:21:14 UTC | Added support for multidimensional tensors in PReLU; Channel number now in second dimension | 22 March 2017, 15:45:02 UTC |
| 724f18b | Adam Paszke | 22 March 2017, 15:29:39 UTC | Cast accumulator in LookupTable renorm to accreal | 22 March 2017, 15:29:39 UTC |
| b610a35 | Soumith Chintala | 15 March 2017, 18:36:10 UTC | Merge pull request #449 from gchanan/precision_testing Improve precision testing | 15 March 2017, 18:36:10 UTC |
| 7293be0 | Soumith Chintala | 15 March 2017, 18:35:59 UTC | Merge pull request #451 from wickedfoo/faster-lookup-table Improve cunn LookupTable performance for large batch sizes | 15 March 2017, 18:35:59 UTC |
| 8d74ea9 | Jeff Johnson | 13 March 2017, 20:55:16 UTC | change lookup table sort | 13 March 2017, 20:55:16 UTC |
| 008f4b9 | Gregory Chanan | 06 March 2017, 16:44:02 UTC | Improve precision testing 1) In cases where tests were failing with some regularity (SmoothL1, l1cost, mse, SoftShrink-backward) scale error bounds by (absolute) value being tested. 2) Fix some spacing issues in error messages. | 06 March 2017, 16:44:02 UTC |
| 12d9add | Soumith Chintala | 22 February 2017, 20:07:07 UTC | Merge pull request #447 from gchanan/accreal Remove accreal typedef because it is defined multiple times | 22 February 2017, 20:07:07 UTC |
| 172c82e | Gregory Chanan | 22 February 2017, 19:57:54 UTC | Remove accreal typedef because it is defined multiple times with nn. | 22 February 2017, 19:57:54 UTC |
| 491fccf | Soumith Chintala | 21 February 2017, 19:39:53 UTC | Merge pull request #446 from lvdmaaten/master [DPT] Add option for synchronous copy of model to GPUs | 21 February 2017, 19:39:53 UTC |
| 40cb39f | Laurens van der Maaten | 21 February 2017, 19:18:10 UTC | [DPT] Add option for synchronous copy of model to GPUs | 21 February 2017, 19:18:10 UTC |
| 0d01aaa | Soumith Chintala | 21 February 2017, 14:15:12 UTC | Merge pull request #418 from ruotianluo/adaptiveAverage Add SpatialAdaptiveAveragePooling. | 21 February 2017, 14:15:12 UTC |
| e2084c1 | Soumith Chintala | 21 February 2017, 14:13:59 UTC | Merge pull request #434 from bottler/master VolumetricFractionalMaxPooling like spatial | 21 February 2017, 14:13:59 UTC |
| 6d15802 | Soumith Chintala | 21 February 2017, 14:12:56 UTC | Merge pull request #442 from twitter-forks/half-fixes Convert real to accreal in libTHCUNN | 21 February 2017, 14:12:56 UTC |
| a0f3035 | Soumith Chintala | 20 February 2017, 15:18:56 UTC | Merge pull request #445 from apaszke/nll_criterion Do SpatialClassNLLCriterion sizeAverage in a separate kernel | 20 February 2017, 15:18:56 UTC |
| 56e66d2 | Adam Paszke | 20 February 2017, 14:38:48 UTC | Do SpatialClassNLLCriterion sizeAverage in a separate kernel | 20 February 2017, 14:54:23 UTC |
| 0665363 | Soumith Chintala | 18 February 2017, 03:10:32 UTC | Merge pull request #444 from apaszke/fixes Improvements for spatial functions | 18 February 2017, 03:10:32 UTC |
| 51cc7c5 | Adam Paszke | 17 February 2017, 21:02:25 UTC | Use a more stable formula for spatial LogSoftMax | 17 February 2017, 21:05:45 UTC |
| 6726d1a | Adam Paszke | 17 February 2017, 21:01:56 UTC | Add shape checks for SpatialClassNLLCriterion | 17 February 2017, 21:01:56 UTC |
| 3996dbb | Pavan Yalamanchili | 17 February 2017, 01:25:33 UTC | Convert real to accreal in libTHCUNN - This reverts commit 0d85922d116879448485ef88ae21e83a9255a0b0. - Includes fixes for TemporalRowConvolution | 17 February 2017, 01:33:03 UTC |
| 618f847 | soumith | 15 February 2017, 12:10:43 UTC | add local | 15 February 2017, 12:18:43 UTC |
| 251fcbf | Eli Stevens | 15 February 2017, 11:28:15 UTC | implementation of bias=False for VolConv.cu | 15 February 2017, 12:18:17 UTC |
| e56cea1 | Soumith Chintala | 08 February 2017, 16:23:16 UTC | Merge pull request #440 from gchanan/glugeneric Use THCTensor rather than THCudaTensor in GatedLinearUnit | 08 February 2017, 16:23:16 UTC |
| 6a9f05d | Gregory Chanan | 08 February 2017, 15:54:10 UTC | Use THCTensor rather than THCudaTensor in THCUNN.h definition of GatedLinearUnit. | 08 February 2017, 15:54:10 UTC |
| e0f2ed6 | Soumith Chintala | 03 February 2017, 05:45:21 UTC | Merge pull request #438 from huihuifan/betterCudaGLU cuda implementation of Gated Linear Unit, fixed issues with genericiz… | 03 February 2017, 05:45:21 UTC |
| 32a0ada | Angela Fan | 27 January 2017, 06:56:33 UTC | cuda implementation of Gated Linear Unit, fixed issues with genericization | 03 February 2017, 05:38:25 UTC |
| b137006 | Soumith Chintala | 02 February 2017, 05:21:41 UTC | Merge pull request #437 from torch/revert-430-newCudaGLU Revert "cuda implementation of Gated Linear Unit, fixed issues with genericization" | 02 February 2017, 05:21:41 UTC |
| dc71d25 | Soumith Chintala | 02 February 2017, 05:20:47 UTC | Revert "cuda implementation of Gated Linear Unit, fixed issues with genericization" | 02 February 2017, 05:20:47 UTC |
| 2a8d396 | Soumith Chintala | 02 February 2017, 02:46:35 UTC | Merge pull request #430 from huihuifan/newCudaGLU cuda implementation of Gated Linear Unit, fixed issues with genericization | 02 February 2017, 02:46:35 UTC |
| 7571794 | jeremy | 01 February 2017, 11:52:49 UTC | VolumetricFractionalMaxPooling like spatial | 01 February 2017, 11:52:49 UTC |
| fc0dc59 | Soumith Chintala | 30 January 2017, 23:04:02 UTC | Merge pull request #433 from apaszke/unpooling_leak Fix memory leak in SpatialMaxUnpooling | 30 January 2017, 23:04:02 UTC |
| a3440e4 | Adam Paszke | 30 January 2017, 21:13:59 UTC | Fix memory leak in SpatialMaxUnpooling | 30 January 2017, 22:23:07 UTC |
| fbef6e7 | Soumith Chintala | 29 January 2017, 20:45:55 UTC | Merge pull request #431 from apaszke/asserts Compile with asserts by default | 29 January 2017, 20:45:55 UTC |
| 6568da8 | Adam Paszke | 29 January 2017, 20:21:59 UTC | Compile with asserts by default | 29 January 2017, 20:21:59 UTC |
| 37db7b8 | Will Frey | 27 January 2017, 18:30:25 UTC | Added cunn support for TemporalRowConvolutionMM (#415) * Added cunn TemporalRowConvolutionMM support | 27 January 2017, 18:30:25 UTC |
| d253af0 | Angela Fan | 27 January 2017, 06:56:33 UTC | cuda implementation of Gated Linear Unit, fixed issues with genericization | 27 January 2017, 06:56:33 UTC |
| 5fa193a | Soumith Chintala | 25 January 2017, 22:15:58 UTC | Merge pull request #428 from colesbury/CMake Install THCUNN.h and generic/THCUNN.h | 25 January 2017, 22:15:58 UTC |
| 22b109d | Sam Gross | 25 January 2017, 22:13:17 UTC | Install THCUNN.h and generic/THCUNN.h The THCApply.cuh is moved to the .cu files so that THCUNN.h can be compiled by a standard C compiler. | 25 January 2017, 22:13:17 UTC |
| df99d8c | Soumith Chintala | 25 January 2017, 21:13:40 UTC | Merge pull request #427 from torch/revert-416-half-fixes Revert "Convert real to accreal in libTHCUNN" | 25 January 2017, 21:13:40 UTC |
| 0d85922 | Soumith Chintala | 25 January 2017, 21:13:20 UTC | Revert "Convert real to accreal in libTHCUNN" | 25 January 2017, 21:13:20 UTC |
| 8722303 | Soumith Chintala | 25 January 2017, 18:16:44 UTC | Merge pull request #426 from torch/revert-421-cudaGLU Revert "cuda implementation of Gated Linear Unit" | 25 January 2017, 18:16:44 UTC |
| 1d010fc | Soumith Chintala | 25 January 2017, 18:15:22 UTC | Revert "cuda implementation of Gated Linear Unit" | 25 January 2017, 18:15:22 UTC |
