Age | Commit message (Collapse) | Author | |
---|---|---|---|
2017-09-07 | fix static linkage and make THD statically linkedHEADmaster | Soumith Chintala | |
2017-08-27 | Add numerically stable logsigmoid | Alykhan Tejani | |
2017-08-26 | Adding implicit padding for 3d average pooling | Lu Fang | |
2017-08-25 | Fix typos. | Zhou Mo | |
2017-08-25 | Updates for CUDA 9 | Christian Sarofeen | |
2017-08-03 | remove limitations on output_padding in Conv* routines | Soumith Chintala | |
2017-08-03 | add 2d and 3d dilated full Convolution | Soumith Chintala | |
2017-07-26 | Merge pull request #477 from wickedfoo/feature_lp_pooling | Soumith Chintala | |
GPU implementation of L_p feature pooling | |||
2017-07-19 | Static linking against libstdc++ in Binary Build mode | Soumith Chintala | |
2017-07-14 | add launch_bounds to greedy kernels | Natalia Gimelshein | |
2017-07-13 | LP pooling kernels | Jeff Johnson | |
2017-06-23 | Fix segfault in SpatialDepthWiseConvolution w/o bias | Simone Cirillo | |
2017-06-22 | add asserts to BCECriterion | cph | |
2017-06-15 | nn.EmbeddingBag to compute a bag of word embeddings (Embedding + Sum/Mean) | Soumith Chintala | |
2017-06-14 | Added GLU (gated linear unit) | Sam Gross | |
From https://arxiv.org/abs/1612.08083 | |||
2017-06-07 | Add 3D upsampling (nearest and trilinear) with tests | Luca Antiga | |
2017-06-07 | fix GRUFused signature | Soumith Chintala | |
2017-06-07 | Merge pull request #470 from qqning/master | Soumith Chintala | |
Fix the mix-up of height and width on depth-wise convolution | |||
2017-06-07 | Remove clone in fused rnn | Christian Sarofeen | |
2017-06-06 | Fixing the issue with incorrect normalized values in IndexLinear | Pavan Yalamanchili | |
2017-05-24 | Fix the mix-up of height and width on depth-wise convolution | ningqingqun | |
2017-05-16 | ClassNLLCriterion ignoreIndex | Nicholas Leonard | |
2017-05-15 | Revert "Update to ignore zero targets" | Soumith Chintala | |
2017-05-12 | SpatialDepthWiseConvolution.cu added | stooloveu | |
2017-05-12 | Merge pull request #458 from jnhwkim/master | Nicholas Léonard | |
Update to ignore zero targets | |||
2017-05-09 | Add a keepdim parameter for reduction functions over a single dimension. | Gregory Chanan | |
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) | |||
2017-04-22 | fix typo | Soumith Chintala | |
2017-04-22 | Indexing fix for fused GRU/LSTM kernels when all tensors are not contiguous. | Christian Sarofeen | |
2017-04-22 | add contiguous checks | Soumith Chintala | |
2017-04-18 | Fused RNN kernel remove explicit instantiation, isn't needed. | Christian Sarofeen | |
2017-04-18 | Remove double precision math from LogSigmoid too | Adam Paszke | |
2017-04-18 | Update ops for Sigmoid and Tanh | Adam Paszke | |
2017-04-12 | fix THNN headers | soumith | |
2017-04-11 | Fused pointwise kernels for GRU/LSTM | Christian Sarofeen | |
2017-04-09 | Merge pull request #455 from twitter-forks/indexlinear | Soumith Chintala | |
Adding Indexlinear | |||
2017-04-07 | Support TORCH_NVCC_FLAGS environment variable | Thomas Riccardi | |
This is already supported in cutorch since august 2016, and is used in pytorch integration (to reduce the binary size). | |||
2017-04-05 | Update to ignore zero targets | Jin-Hwa Kim | |
If the target is zero, loss and gradient of input are set to zero. It is useful for variable-length natural language generation models. | |||
2017-03-31 | Merge pull request #456 from twitter-forks/addmm-fixes | Soumith Chintala | |
Using temporary variables when performing transpose + addmm | |||
2017-03-30 | Using temporary variables when performing transpose + addmm | Pavan Yalamanchili | |
2017-03-25 | Improving the performance of IndexLinear:updateOutput | Pavan Yalamanchili | |
- Removes separate kernel for updateOutputTrain | |||
2017-03-24 | Fix inconsistent in-place and out-of-place for HardTanh | ngimel | |
in-place and out-of-place updateGradOutput results are different where input=min_val or input=max_val | |||
2017-03-24 | Adding support for flattened inputs for IndexLinear | Pavan Yalamanchili | |
- Adding relevant tests | |||
2017-03-24 | IndexLinear support for cunn | Pavan Yalamanchili | |
2017-03-22 | Merge pull request #453 from apaszke/lookup_renorm | Soumith Chintala | |
Cast accumulator in LookupTable renorm to accreal | |||
2017-03-22 | Added support for multidimensional tensors in PReLU; Channel number now in ↵ | Hardik | |
second dimension | |||
2017-03-22 | Cast accumulator in LookupTable renorm to accreal | Adam Paszke | |
2017-03-13 | change lookup table sort | Jeff Johnson | |
2017-02-21 | Merge pull request #418 from ruotianluo/adaptiveAverage | Soumith Chintala | |
Add SpatialAdaptiveAveragePooling. | |||
2017-02-21 | Merge pull request #434 from bottler/master | Soumith Chintala | |
VolumetricFractionalMaxPooling like spatial | |||
2017-02-21 | Merge pull request #442 from twitter-forks/half-fixes | Soumith Chintala | |
Convert real to accreal in libTHCUNN |