pytorch / pytorch

31 Pull requests merged by 18 people

• [v1.7][JIT] Improve class type annotation inference

  #46422 merged Oct 15, 2020

• [v1.7] Quick fix for view/inplace issue with DDP

  #46407 merged Oct 15, 2020

• [v1.7] Disable tcuda_fuser tests in Profiling Mode (#45638)

  #46419 merged Oct 15, 2020

• [v1.7] Pin XLA CI to use r1.7 release branch.

  #46425 merged Oct 15, 2020

• [v1.7] Stop running clang-tidy on torch/csrc/generic/*.cpp. (#46335)

  #46418 merged Oct 15, 2020

• Fix error message for scatter reduction

  #46408 merged Oct 15, 2020

• Workaround for bug in DistributedDataParallel

  #46385 merged Oct 15, 2020

• [v1.7 patch] Add warning on ProcessGroup and ProcessGroup::Work APIs

  #46366 merged Oct 15, 2020

• Remove Python version upper boundary check

  #46388 merged Oct 15, 2020

• Remove variable_excluded_from_dispatch() check for factory functions.

  #46303 merged Oct 15, 2020

• Update index.rst

  #46324 merged Oct 15, 2020

• Performance fix for torch.cat operator on ROCm (#46097)

  #46323 merged Oct 14, 2020

• [v1.7] Rocm skip test cases (#45782)

  #46333 merged Oct 14, 2020

• Complex Autograd Doc fix

  #46281 merged Oct 13, 2020

• [NNC] Fix two bugs in Cuda Half support

  #46214 merged Oct 13, 2020

• Cherrypick smooth l1 loss fixes

  #45759 merged Oct 13, 2020

• [v1.7] Update allowlist back compat date for min_values / max_values.

  #46262 merged Oct 13, 2020

• [dist_optim] serialize compilation when creating dist_optim (#45871)

  #46071 merged Oct 13, 2020

• qnnpack quantized activations: fix memory format issues

  #46217 merged Oct 13, 2020

• annotate torch.autograd.* modules

  #46206 merged Oct 12, 2020

• Improve error checking of Storage._writeFile.

  #46207 merged Oct 12, 2020

• [v1.7 cherry-pick] [fix] torch.kthvalue : handle non-contiguous CUDA tensor

  #46177 merged Oct 12, 2020

• [v1.7 patch] Prioritize raising error message about unused parameters when rebuild_buckets fails

  #46099 merged Oct 12, 2020

• [Release/1.7] [ONNX] Improve error handling for adaptive_pool

  #46100 merged Oct 12, 2020

• [v1.7 cherry-pick] [JIT] Dict Bug Fixes

  #46105 merged Oct 12, 2020

• [v1.7 cherry-pick] Revert "Add Rowwise Prune PyTorch op (#42708)"

  #46103 merged Oct 12, 2020

• [v1.7 patch] Disallow creation of ProcessGroupNCCL without GPUs. (#45…

  #46073 merged Oct 12, 2020

• [Release/1.7] aten::set_grad_enabled should not push as it does not return a value

  #46060 merged Oct 12, 2020

• [v1.7 patch] Remove object-based collective APIs from public docs

  #46109 merged Oct 10, 2020

• [v1.7 cherry-pick] Fix flakiness in caffe2/test:serialization - test_serialization_new_f…

  #46114 merged Oct 10, 2020

• Doc note update for complex autograd (CP-45270)

  #46072 merged Oct 9, 2020

127 Pull requests opened by 79 people

• [ONNX] Export and shape inference for prim uninitialized in If subblock

  #46094 opened Oct 9, 2020

• [DO NOT REVIEW] Test a custom class

  #46111 opened Oct 9, 2020

• [wip] fused fake_quant linear impl

  #46113 opened Oct 9, 2020

• [caffe2][not for land][testing] ErrorPlanWithCancellableStuckNet test with cancel count

  #46117 opened Oct 9, 2020

• Use pytorch multi tensor rmsprop optimizer for better performance

  #46118 opened Oct 9, 2020

• [caffe2] Use folly::F14FastMap for CaffeMap in fbcode

  #46123 opened Oct 9, 2020

• [JIT] Dont use specialized tensor type

  #46130 opened Oct 10, 2020

• Rowwise Prune op (Add the test to OSS run_test), Make the op private.

  #46131 opened Oct 10, 2020

• [TESTING]: add a new test job

  #46133 opened Oct 10, 2020

• [caffe2] Fix inplace ops in onnx::SsaRewrite

  #46134 opened Oct 10, 2020

• Added torch.linalg.tensorsolve

  #46142 opened Oct 10, 2020

• [WIP] [CUDA graphs] Option to maintain CUDA RNG states on the device

  #46148 opened Oct 10, 2020

• Fix Module backward hooks for all Tensor inputs/outputs

  #46163 opened Oct 11, 2020

• WIP - experiment with unique APIs

  #46167 opened Oct 11, 2020

• Suggest autograd.no_grad in nn.Module.{train,eval} (and vice versa)

  #46173 opened Oct 12, 2020

• Implement torch.igamma

  #46183 opened Oct 12, 2020

• [Testing] DataLoader new pushing index strategy

  #46189 opened Oct 12, 2020

• Update target determinator to point to release/1.7

  #46193 opened Oct 12, 2020

• [1.7] Remove torch.vmap (#45571)

  #46194 opened Oct 12, 2020

• Update target determinator to point to release/1.7

  #46195 opened Oct 12, 2020

• [1.7] Remove torch.vmap (#45571)

  #46196 opened Oct 12, 2020

• Update target determinator to point to release/1.7

  #46197 opened Oct 12, 2020

• [1.7] Remove torch.vmap (#45571)

  #46198 opened Oct 12, 2020

• Update target determinator to point to release/1.7

  #46200 opened Oct 12, 2020

• [1.7] Remove torch.vmap (#45571)

  #46201 opened Oct 12, 2020

• [vulkan] Remove vulkan v2 api ATen/native/vulkan/api/*.cpp from compi…

  #46203 opened Oct 12, 2020

• detect inplace modifications of views earlier (fix #21875)

  #46204 opened Oct 12, 2020

• Model parameters passed as argument in freezing

  #46211 opened Oct 12, 2020

• Added foreach_zero_ API

  #46215 opened Oct 12, 2020

• Push rocm to slow path

  #46216 opened Oct 12, 2020

• [StaticRuntime] Threading model

  #46219 opened Oct 12, 2020

• [WIP][JIT] Add CUDA stream support

  #46226 opened Oct 13, 2020

• Remove PyCFunction casts as much as possible.

  #46227 opened Oct 13, 2020

• [quant] Custom module fix

  #46228 opened Oct 13, 2020

• [quant] Statically quantized LSTM

  #46229 opened Oct 13, 2020

• [quant] Compute scale and zero point automatically in testing::_quantize

  #46232 opened Oct 13, 2020

• [WIP] BatchNorm channels last memory format optimization on CPU

  #46234 opened Oct 13, 2020

• [quant] Quantized flip dispatch

  #46235 opened Oct 13, 2020

• [pytorch] add script to run all codegen

  #46243 opened Oct 13, 2020

• [pytorch] rewrite of the python binding codegen with the v2 API

  #46244 opened Oct 13, 2020

• [WIP] Enable JIT C++ tests on ROCm

  #46247 opened Oct 13, 2020

• Avoid inlining kernel lambdas on mobile

  #46249 opened Oct 13, 2020

• Add max_src_column_width to autograd profiler

  #46257 opened Oct 13, 2020

• Create prototype for AST rewriter

  #46261 opened Oct 13, 2020

• [ONNX] Update peephole pass for prim::ListUnpack

  #46264 opened Oct 13, 2020

• Apply hip vs hipcc compilation flags correctly for building extensions

  #46273 opened Oct 13, 2020

• Providing more information while crashing process in async error handling

  #46274 opened Oct 13, 2020

• Update backward formulas (Re #44444)

  #46275 opened Oct 13, 2020

• Delete CUDAUnaryOps.cpp

  #46280 opened Oct 13, 2020

• Add CUDA 11.1 docker build

  #46283 opened Oct 13, 2020

• [WIP] make `torch.lu` differentiable.

  #46284 opened Oct 13, 2020

• Add utilities to support handling of nested python data structures

  #46287 opened Oct 13, 2020

• Implement `_broadcast_to_and_flatten(pytree, spec)`

  #46288 opened Oct 13, 2020

• Allow vmap to accept nested python data structures as inputs

  #46289 opened Oct 13, 2020

• Remove dilation restriction on cuDNN ConvTranspose2d

  #46290 opened Oct 13, 2020

• Fix inplace check logic to be triggered when written to Tensor does not require gradients

  #46296 opened Oct 13, 2020

• [quant][graphmode][fx] Support non_traceable_module/module_class

  #46298 opened Oct 14, 2020

• Avoid scatter for single-device case in DDP

  #46304 opened Oct 14, 2020

• [StaticRuntime] Add a 'speed of light' benchmark.

  #46308 opened Oct 14, 2020

• implement NumPy-like functionality column_stack, row_stack

  #46313 opened Oct 14, 2020

• [TESTING] Try to trigger clang-tidy error.

  #46329 opened Oct 14, 2020

• [pytorch] fix debug symbol flag for android clang

  #46331 opened Oct 14, 2020

• [RFC] Move fast pass functions into a centralized file.

  #46332 opened Oct 14, 2020

• [quant][refactor] Remove register api and rename get_*_mapping to get_default_*_mapping

  #46337 opened Oct 14, 2020

• [quant][graphmode][fx] Add additional_object_mapping argument to convert

  #46338 opened Oct 14, 2020

• Add a way to fetch the boxed kernel

  #46343 opened Oct 14, 2020

• [quant][graphmode][fx] Add additional_qat_module_mapping

  #46344 opened Oct 14, 2020

• [quant][graphmode][fx] Add support for additional_fuse_method_mapping

  #46345 opened Oct 14, 2020

• [quant][graphmode][fx] Add support for additional_{fusion/quant}_pattern

  #46346 opened Oct 14, 2020

• [ONNX] Update ONNX doc for indexing export

  #46349 opened Oct 14, 2020

• Remove catchAllKernel_.

  #46354 opened Oct 14, 2020

• [quant][eagermode] Add additional_fuser_method_mapping to config

  #46355 opened Oct 14, 2020

• Promote -Wcast-function-type to an error in builds.

  #46356 opened Oct 14, 2020

• Use unsafe_split/chunk to preserve same behavior as 1.6

  #46361 opened Oct 14, 2020

• Enable softmax and tiny norm FP16 tests on ROCm

  #46363 opened Oct 14, 2020

• [reland][quant][eagermode] Move custom_module registration to prepare/convert_custom_config_dict (#46293)

  #46364 opened Oct 14, 2020

• Add flatten loops transformation

  #46365 opened Oct 15, 2020

• [RPC] print exception message on workers that run python functions

  #46372 opened Oct 15, 2020

• [Draft]Try to enable Azure 2 GPUs test box working with circleci

  #46376 opened Oct 15, 2020

• [GPU] Introduce USE_PYTORCH_METAL

  #46383 opened Oct 15, 2020

• [Metal] Enable optimize_for_mobile on Linux

  #46384 opened Oct 15, 2020

• Distribute GPUs in round robin mode for distributed_test

  #46389 opened Oct 15, 2020

• add sparse_nn_partition

  #46390 opened Oct 15, 2020

• [WIP] Implement copysign

  #46396 opened Oct 15, 2020

• Fixed einsum compatibility issues

  #46398 opened Oct 15, 2020

• explicitly error out in comparison ops when the types don't match

  #46399 opened Oct 15, 2020

• [RFC] Add OperatorHandle overload to the RecordFunction::before() method

  #46401 opened Oct 15, 2020

• Ci all/taylorrobie/timer papercuts 1.7

  #46403 opened Oct 15, 2020

• Create prototype for AST rewriter

  #46410 opened Oct 15, 2020

• [c10d] add c10d_frontend to build, add getter/setter

  #46411 opened Oct 15, 2020

• [NNC] IRSimplifier rules for Compare, Mod, symbolic Div

  #46412 opened Oct 15, 2020

• use @mode/ndk_libcxx instead of mode/gnustl

  #46413 opened Oct 15, 2020

• Move irange to c10

  #46414 opened Oct 15, 2020

• Test pybind-2.6.0rc2

  #46415 opened Oct 15, 2020

• [dper3] fixing incorrect float field issue with exported learning rate op

  #46416 opened Oct 15, 2020

• Grammatically updated the tech docs

  #46417 opened Oct 15, 2020

• Make optional[T] sortable

  #46423 opened Oct 15, 2020

• caffe2/plan_executor: propagate exceptions from reporter substeps

  #46424 opened Oct 15, 2020

• [quant] Allow replacement of the custom modules in the "prepare" step

  #46426 opened Oct 15, 2020

• Add view test for tensor_split

  #46427 opened Oct 15, 2020

• enabling lcov for coverage tests

  #46429 opened Oct 15, 2020

• Remove CPUApplyUtils.h

  #46433 opened Oct 15, 2020

• dummy change

  #46435 opened Oct 15, 2020

• [WIP] Bump up windows cudnn version

  #46436 opened Oct 15, 2020

• Support hashing of various date types by implementing generic hashing for IValues

  #46441 opened Oct 16, 2020

• [ONNX] Add export of aten::is_floating point

  #46442 opened Oct 16, 2020

• Automated submodule update: FBGEMM

  #46443 opened Oct 16, 2020

• [DI] Allow explicit executor for torchscript interpreter

  #46444 opened Oct 16, 2020

• Fix implicit cast in custom_function

  #46445 opened Oct 16, 2020

• [TensorExpr] Re-enable test for torch.cat, add a test for torch.cat being a single node in a fusion group.

  #46447 opened Oct 16, 2020

• [quant][graph] Move removeExtraWaitCalls from freezing code to quantization code

  #46448 opened Oct 16, 2020

• [nvFuser] Switching to `CudaFusionGuard` from `BailOut` for nvfuser - update 2

  #46452 opened Oct 16, 2020

• [TensorExpr] Correctly handle negative dimensions in aten::cat when lowering to tensor expressions.

  #46453 opened Oct 16, 2020

• Remove fbcode/caffe2/mode

  #46454 opened Oct 16, 2020

• [GPU] Add metal to code gen

  #46455 opened Oct 16, 2020

• [Metal] Add the Python binding for optimize_for_mobile

  #46456 opened Oct 16, 2020

• Replace list(map(...)) constructs by list comprehensions

  #46461 opened Oct 16, 2020

• Replace map(lambda constructs

  #46462 opened Oct 16, 2020

• Removed term blacklist in low_net_transform.cc, glow_net_transform.h, fakefp16_transform.cc

  #46463 opened Oct 16, 2020

• [WIP] fix issue by which pytorch_jni is not bundled in libtorch

  #46466 opened Oct 16, 2020

• Check if pin_memory() is overridden before pinning Tensors

  #46467 opened Oct 16, 2020

• [RFC] Decouple fast pass functions

  #46469 opened Oct 16, 2020

• Use libkineto in profiler

  #46470 opened Oct 16, 2020

• Adding profiling capability to c++ ddp collective functions

  #46471 opened Oct 16, 2020

• Fix fbgemm gcc10

  #46472 opened Oct 16, 2020

• Update backward formulas (Re #44444)

  #46473 opened Oct 16, 2020

• Ensure kernel launches are checked

  #46474 opened Oct 16, 2020

135 Issues closed by 43 people

• Call to `torch.cuda.memory_stats` before any allocating any tensors on GPU causes SEGFAULT

  #46437 closed Oct 16, 2020

• [JIT] TensorExpr causes segmentation fault

  #46440 closed Oct 16, 2020

• Enable torch.quantization.default_mappings typechecks during CI

  #42970 closed Oct 16, 2020

• Enable torch.quantization._numeric_suite typechecks during CI

  #42977 closed Oct 16, 2020

• PyTorch's div, which performs true division, cannot be exported to ONNX with consistent semantics

  #43787 closed Oct 16, 2020

• no pytorch_jni in java.library.path windows

  #44363 closed Oct 16, 2020

• sspaddmm yields inconsistent results

  #45113 closed Oct 15, 2020

• Compiling NCCL does not obey MAX_JOBS, makes machine unresponsive

  #20692 closed Oct 15, 2020

• [v1.7] Figure out what to do about https://github.com/pytorch/pytorch/pull/46285

  #46432 closed Oct 15, 2020

• [1.7

  #46431 closed Oct 15, 2020

• [jit] Problem saving nn.Module as a TorchScript module (DLRM model)

  #41961 closed Oct 15, 2020

• cdist allocates a huge amount of memory in the bachward pass (pytorch 1.2.0)

  #24345 closed Oct 15, 2020

• torch.pow doesn't check for tensors being on different devices

  #46037 closed Oct 15, 2020

• Migrate CPU_tensor_apply to TensorIterator in aten/src/ATen/native/Distributions.cpp:286

  #24496 closed Oct 15, 2020

• Migrate CPU_tensor_apply to TensorIterator in aten/src/ATen/native/Distributions.cpp:201

  #24492 closed Oct 15, 2020

• Migrate CPU_tensor_apply to TensorIterator in aten/src/ATen/native/Distributions.cpp:220

  #24493 closed Oct 15, 2020

• Migrate CPU_tensor_apply to TensorIterator in aten/src/ATen/native/Distributions.cpp:235

  #24494 closed Oct 15, 2020

• Migrate CPU_tensor_apply to TensorIterator in aten/src/ATen/native/Distributions.cpp:266

  #24495 closed Oct 15, 2020

• Migrate CPU_tensor_apply to TensorIterator in aten/src/ATen/native/Distributions.cpp:176

  #24490 closed Oct 15, 2020

• Migrate CPU_tensor_apply to TensorIterator in aten/src/ATen/native/Distributions.cpp:189

  #24491 closed Oct 15, 2020

• Migrate CPU_tensor_apply to TensorIterator in aten/src/ATen/native/Distributions.cpp:142

  #24488 closed Oct 15, 2020

• Migrate CPU_tensor_apply to TensorIterator in aten/src/ATen/native/Distributions.cpp:151

  #24489 closed Oct 15, 2020

• Migrate CPU_tensor_apply to TensorIterator in aten/src/ATen/native/Loss.cpp:80

  #24487 closed Oct 15, 2020

• torch.pinverse for complex tensors and general complex tensor support

  #45385 closed Oct 15, 2020

• Couldn't build multi scaled kernel nested model

  #46393 closed Oct 15, 2020

• A few question about sharing weights of convolution kernels and your dilation convlution

  #46387 closed Oct 15, 2020

• *NNPACK are required to build Java bindings

  #43909 closed Oct 15, 2020

• Freeze module pass fails on torchvision RCNN models

  #45902 closed Oct 15, 2020

• [RFC] Improve the CI docker image builder workflow

  #37855 closed Oct 15, 2020

• Java bindings for Windows

  #32516 closed Oct 15, 2020

• Wrong signature for torch.max_pool(1d, 2d, 3d) in get_testing_overrides()

  #45494 closed Oct 15, 2020

• Could not load library cudnn_ops_infer64_8.dll. Error code 126

  #46312 closed Oct 15, 2020

• Simple model cannot converge with fp16 precision using torch.cuda.amp

  #46238 closed Oct 15, 2020

• libtorch_cpu.a is huge (1 GB) with build_android.sh

  #46246 closed Oct 15, 2020

• torch.jit.script segfault

  #45860 closed Oct 15, 2020

• Bazel build has warnings

  #43212 closed Oct 15, 2020

• bucketize fails with inputs requiring gradient

  #45119 closed Oct 15, 2020

• COMPILE_TIME_MAX_DEVICE_TYPES is not changed as new DeviceType::Metal added.

  #46310 closed Oct 15, 2020

• DistributedSampler internal asserts if len(dataset) * 2 < number of GPUs

  #45324 closed Oct 15, 2020

• v1.7.0-rc1 4 data loader test failures: unexpected keyword argument 'persistent_workers'

  #46297 closed Oct 14, 2020

• DistributedDataParallel hangs when using NCCL backend

  #46259 closed Oct 14, 2020

• OneCycleLR does not follow the algorithm described by the original paper

  #40362 closed Oct 14, 2020

• [JIT] get_jit_def breaks with continued string literals

  #44842 closed Oct 14, 2020

• Producing Imagenet_classes.txt labels

  #46160 closed Oct 14, 2020

• Make ROCm convolution work with empty batches

  #26669 closed Oct 14, 2020

• CUDA irfft may be doing unnecessary cloning of input

  #38413 closed Oct 14, 2020

• Enable torch.quantization.observer typechecks during CI

  #42972 closed Oct 14, 2020

• Enable torch.quantization.quantize_jit typechecks during CI

  #42975 closed Oct 14, 2020

• Enable torch.quantization.fake_quantize typechecks during CI

  #42974 closed Oct 14, 2020

• Does the model support semi precision reasoning?

  #46316 closed Oct 14, 2020

• test_cpp_extensions_aot_* fail on Power

  #41261 closed Oct 14, 2020

• Illegal instruction

  #46154 closed Oct 14, 2020

• Clang-tidy failed to run due to Python 3.9.0 of setup-python action

  #46314 closed Oct 14, 2020

• help please. segment fault when destruction after dlopen torch lib

  #46171 closed Oct 14, 2020

• torch-1.6.0 contains AVX instruction in the default codepath

  #43300 closed Oct 14, 2020

• Custom learning rate scheduler LambdaLR only supports epoch as argument rather than step

  #45895 closed Oct 14, 2020

• calculate partial derivatives instead of total derivative like tf.gradients with parameter stop_gradients

  #46209 closed Oct 14, 2020

• NCCL connection refused when doing broadcast in several process groups simultaneously

  #46087 closed Oct 14, 2020

• I'm compiling the pytorch 1.6.0 stable version on CUDA 11.1 and CUDNN 8.04 with 3090 and failed.

  #46299 closed Oct 14, 2020

• Work Logging causes Hang in NCCL Async Error Handling

  #46278 closed Oct 13, 2020

• torch.save w/ _use_new_zipfile_serialization=True corrupts state_dict

  #46020 closed Oct 13, 2020

• the different train speed between pip install torch==1.2 and build torch1.2 from source

  #46245 closed Oct 13, 2020

• DISABLED test_debug_info (__main__.DistAutogradTestWithSpawn)

  #35575 closed Oct 13, 2020

• Update RPC examples/docs/tutorials to use RRef helper

  #37459 closed Oct 13, 2020

• RPC ThreadPool threads could be eaten up and cause deadlock on too many nested RPC calls.

  #26362 closed Oct 13, 2020

• DistributedDataParallel cannot handle multi-device runs with non-evenly divisible batch sizes

  #46175 closed Oct 13, 2020

• TCPStoreTest.test_numkeys_delkeys takes 5+ min to finish

  #46064 closed Oct 13, 2020

• torch.autograd.profiler.load_nvprof() crash

  #45490 closed Oct 13, 2020

• Move the record_stream to Aten operation

  #36556 closed Oct 13, 2020

• using torch.utils.data.Dataset to make my dataset, find the index is out of the len defined in the __len__

  #45040 closed Oct 13, 2020

• tensorboard multiple add_graph()

  #46239 closed Oct 13, 2020

• Add `add_done_callback(fn)` to torch.futures.Future

  #45552 closed Oct 13, 2020

• Exception type changed when it's raised in backward operation

  #46088 closed Oct 13, 2020

• NCCL group broadcast fails

  #46181 closed Oct 13, 2020

• Embedding max_norm wrong results on cuda when repeating indices are present

  #44792 closed Oct 13, 2020

• DDP fp16_compress_hook communication hook increases peak memory

  #45968 closed Oct 12, 2020

• Extend quantized::clamp to support optional min/max

  #45928 closed Oct 12, 2020

• Distributed Autograd hang if enclosed RPC outlived the context.

  #38513 closed Oct 12, 2020

• weight_variable.get_device() returns -1

  #46152 closed Oct 12, 2020

• Fix windows test skipped in test/test_serialization.py

  #45917 closed Oct 12, 2020

• Could not find a version that satisfies the requirement torch=1.6.0

  #46205 closed Oct 12, 2020

• torch.jit.script: RuntimeError: default_program(61): error: no suitable conversion function from "__half" to "float" exists

  #45953 closed Oct 12, 2020

• Complex Number support for distributed

  #45760 closed Oct 12, 2020

• TensorPipe is shipped as sharedlibrary

  #41153 closed Oct 12, 2020

• Confusing error message for torch.LongTensor([1], device = 'cuda')

  #46085 closed Oct 12, 2020

• The reproducibility note in `scatter_add_` documentation is misleading

  #45663 closed Oct 12, 2020

• Wrong in nn.ParameterList and nn.DataParallel, pytorch1.6

  #42327 closed Oct 12, 2020

• Enable torch.distributions typechecks during CI

  #42979 closed Oct 12, 2020

• Network-size Dependant Illegal Instruction

  #45838 closed Oct 12, 2020

• test_nccl_errors_blocking_clean_exit is flaky

  #31924 closed Oct 12, 2020

• about aten/src/ATen/native/cudnn/Conv.cpp issue

  #45372 closed Oct 12, 2020

• kthvalue incorrect with strided GPU tensor

  #45721 closed Oct 12, 2020

• Feedback about PyTorch register_backward_hook

  #12331 closed Oct 11, 2020

• batched cholesky decomposition GPU vs. CPU speed

  #20700 closed Oct 10, 2020

• Conv3d fails with bigger batch size

  #22496 closed Oct 10, 2020

• Performance issues with initial call of MultivariateNormal.log_pdf()

  #24008 closed Oct 10, 2020

• TensorIterator traverse order and write locality

  #26812 closed Oct 10, 2020

• It seems the documentation of torch.norm is a little wrong.

  #41388 closed Oct 10, 2020

• [TensorExpr] Fuser incorrectly constructs fusion groups if they start from aten::cat

  #46145 closed Oct 10, 2020

• Error with DistributedDataParallel

  #46140 closed Oct 10, 2020

• missing LAPACK

  #32344 closed Oct 10, 2020

• Add an argument 'beta' in smooth L1 loss?

  #16357 closed Oct 10, 2020

• Change function that truncate to scalar_t on each atomicAdd

  #20590 closed Oct 10, 2020

• SyncBN doesn't support fp16 inputs with fp32 parameters

  #20634 closed Oct 10, 2020

• May be a BUG in 'Conv2d' with "Padding_mode='circular', Padding=1"

  #20981 closed Oct 10, 2020

• RuntimeError: s__th_ne is not implemented for type SparseCPUType

  #21840 closed Oct 10, 2020

• Why aren't torch.functional.sigmoid and torch.nn.functional.relu deprecated like torch.nn.functional.tanh?

  #22117 closed Oct 10, 2020

• CUDA optimization of THCTensorMathPointwise.cuh

  #22612 closed Oct 10, 2020

• torch.mean() calculations are not consistent across CPU/GPU

  #23392 closed Oct 10, 2020

• Add Type Promotion for Bool types

  #23758 closed Oct 10, 2020

• count_nonzero

  #23907 closed Oct 10, 2020

• Documentation for torch.argsort includes an `out` argument

  #24158 closed Oct 10, 2020

• Migrate `_addr` and `_addr_` from the TH to Aten (CUDA)

  #24519 closed Oct 10, 2020

• inconsistent behavior while multiplying int and bool values

  #24288 closed Oct 10, 2020

• torch.utils.dlpack.from_dlpack causes conversion of CuPy float64 tensor to float 16

  #24213 closed Oct 10, 2020

• torch.rfft segfaults with CPU, but works with GPU

  #25640 closed Oct 10, 2020

• Inconsistent torch.norm() behavior

  #29723 closed Oct 10, 2020

• CUDA error: device-side assert triggered for BERT Multi Class Classification

  #31247 closed Oct 10, 2020

• About torch.constant_pad_nd

  #31687 closed Oct 10, 2020

• where is the Source Code of torch.mode operator?

  #46081 closed Oct 10, 2020

• Linspace+logspace support all integral types on CUDA but not CPU

  #31991 closed Oct 10, 2020

• [feature request] torch.cat to apply to as_tensor to inputs

  #33042 closed Oct 10, 2020

• Division by zero crashes for fmod, remainder and addcdiv operators

  #32699 closed Oct 10, 2020

• RuntimeError: iter.device(arg).is_cuda() INTERNAL ASSERT FAILED at /pytorch/aten/src/ATen/native/cuda/Loops.cuh:197

  #32680 closed Oct 10, 2020

• Is there a way to parallelize independent sequential steps?

  #33937 closed Oct 10, 2020

• How to build on Arch Linux

  #46137 closed Oct 10, 2020

• [quant] Quantized AdaptivePool3d is much slower for ChannelsLast3d.

  #42779 closed Oct 9, 2020

• Inconsistent result from FBGEMM and QNNPACK quantization backends

  #44939 closed Oct 9, 2020

• eager mode quantization: document custom qconfigs

  #44432 closed Oct 9, 2020

• Xcode 12 Build Error: LibTorch/install/lib/libpytorch_qnnpack.a , building for iOS, but linking in object file built for macOS

  #44143 closed Oct 9, 2020

• Wrong Description about torch.logspace

  #45417 closed Oct 9, 2020

• torch.autograd.gradcheck support for Tensor-like types (__torch_function__).

  #42942 closed Oct 9, 2020

• DistributedDataSampler converts NamedTuple to regular tuple

  #38028 closed Oct 9, 2020

• [JIT] Change the way strides are printed in types

  #41772 closed Oct 9, 2020

• test flakiness onoperator_test/locally_connected_op_test.py::TestLocallyConnectedOp::test_lc_1d

  #45925 closed Oct 9, 2020

91 Issues opened by 71 people

• Advanced indexing: allow combining Boolean & integer index

  #46468 opened Oct 16, 2020

• Random errors：module 'torch.utils.data' has no attribute 'IterableDataset'

  #46465 opened Oct 16, 2020

• [META] Assert that expected libraries appear in libtorch

  #46464 opened Oct 16, 2020

• C++ API: Training and inference of torchscript modules on multiple GPU

  #46460 opened Oct 16, 2020

• INTERNAL ASSERT FAILED at "/pytorch/aten/src/ATen/core/type.cpp":1029

  #46459 opened Oct 16, 2020

• Tensor expressions type promotion for torch.sin is incorrect after #45733

  #46458 opened Oct 16, 2020

• How can I use parallel cudastreams in multi threads?

  #46451 opened Oct 16, 2020

• How to use GPU Tensor in diffrent GPUStreams with multi threads

  #46450 opened Oct 16, 2020

• On crashes in DDP/C10d, dump relevant state to logs to aid debugging.

  #46438 opened Oct 15, 2020

• [RFC] A general C10D_DEBUG_LEVEL flag for verbose DDP/torch.distributed logging

  #46434 opened Oct 15, 2020

• module type is not preserved when symbolic tracing torch.nn.quantized.FloatFunctional

  #46430 opened Oct 15, 2020

• send_object/recv_object APIs for c10d

  #46428 opened Oct 15, 2020

• Custom ops get stuck in multiprocess data loader under certain environments

  #46409 opened Oct 15, 2020

• Allow downstream users to suppress Save Optimizer warnings

  #46405 opened Oct 15, 2020

• Make Undefined reprensentable in DispatchKeySet.

  #46400 opened Oct 15, 2020

• Replace list(map(lambda constructs by list comprehension

  #46392 opened Oct 15, 2020

• Warnings during compiling: floating-point value does not fit in required integral type

  #46391 opened Oct 15, 2020

• GPU memory leak when registering a forward hook with "self" access

  #46386 opened Oct 15, 2020

• Code asserts when register new aten operations' implementation thru "c10::RegisterOperators::op" API.

  #46382 opened Oct 15, 2020

• [Feature] Imbalanced batch scattering in th.nn.DataParallel

  #46381 opened Oct 15, 2020

• JIT LibTorch: Profiling Mode Enabled causes memory leak

  #46380 opened Oct 15, 2020

• Reduce "reserved" memory by PyTorch.

  #46379 opened Oct 15, 2020

• rtx3090 with half precision shows lower speed than single precision

  #46377 opened Oct 15, 2020

• JIT: error in LSTM with `flatten_parameters`

  #46375 opened Oct 15, 2020

• Complex Number support for torch.nn modules

  #46374 opened Oct 15, 2020

• Add `inputs` argument to `autograd.backward()`

  #46373 opened Oct 15, 2020

• [JIT] Inconsistent behavior of torchScript parser in python and c++

  #46368 opened Oct 15, 2020

• [chore] Test pybind11 2.6.0 (RC 2?)

  #46367 opened Oct 15, 2020

• torch.distributions.half_normal.HalfNormal.cdf returns negative values

  #46350 opened Oct 14, 2020

• C++ DDP Communication Hook

  #46348 opened Oct 14, 2020

• Enable batched gradient computation for torch.cdist

  #46342 opened Oct 14, 2020

• [JIT] Better Error Messages for mistyped Class Members

  #46341 opened Oct 14, 2020

• Add batching rule for real and imag

  #46340 opened Oct 14, 2020

• Fix batching rule for transpose

  #46339 opened Oct 14, 2020

• A race problem of JIT cpp extensions in distributed setting

  #46334 opened Oct 14, 2020

• [JIT] Unannotated NamedTuple Type should use TensorType::getInferred()

  #46326 opened Oct 14, 2020

• How can I fix NAN loss (or very large MSE losses)?

  #46322 opened Oct 14, 2020

• Add distributed examples into PyTorch CI tests

  #46321 opened Oct 14, 2020

• Remove header declarations for CPUType/TypeDefault

  #46319 opened Oct 14, 2020

• Issues with building Libtorch with /MTd in Windows

  #46318 opened Oct 14, 2020

• Compile error when use c10::RegisterOperators::Options::kernel with the return type "std::tuple<Tensor&, Tensor&>"

  #46311 opened Oct 14, 2020

• support a kind of elementwise mul for sparse tensor

  #46307 opened Oct 14, 2020

• CUDA elementwise ops should use common_dtype instead of output dtype to do computation

  #46302 opened Oct 14, 2020

• ProcessGroup::Work API changes

  #46291 opened Oct 13, 2020

• `__torch_function__` does not get call for torch.Tensor's `__getitem__` syntax

  #46277 opened Oct 13, 2020

• test_distributed_* does not show error details from the subprocess

  #46276 opened Oct 13, 2020

• test_distributed_* does not work with run_test.py -i option

  #46272 opened Oct 13, 2020

• Add dtype arg support to torch.linalg.norm with ord='fro' and ord='nuc'

  #46255 opened Oct 13, 2020

• Enable named tensor inputs to `torch.linalg.norm`

  #46254 opened Oct 13, 2020

• Add short guide for updating `torch.norm` calls to `torch.linalg.norm`

  #46253 opened Oct 13, 2020

• "distributed" NCCL tests fail when having more than 3 GPUs

  #46248 opened Oct 13, 2020

• Assigning a parameter to an indexed tensor that was produced by DDP no longer works in torch nightly (1.7)

  #46242 opened Oct 13, 2020

• ubuntu16.04+vscode compile libtorch error

  #46241 opened Oct 13, 2020

• Support "symmetric" reflection padding

  #46240 opened Oct 13, 2020

• RuntimeError: Unsupported: ONNX export of index_put in opset 9. Please try opset version 11.

  #46237 opened Oct 13, 2020

• Remove catchAllKernel_

  #46231 opened Oct 13, 2020

• torch.mode when input has nans

  #46225 opened Oct 13, 2020

• .to_sparse() should produce a tensor with contiguous indices

  #46224 opened Oct 12, 2020

• Use better tempfile creation mechanism to avoid skip windows test

  #46222 opened Oct 12, 2020

• Integrating CuDNN API for LSTMs with projections

  #46213 opened Oct 12, 2020

• Torchscript: Error passing custom class to module in c++

  #46212 opened Oct 12, 2020

• [FX]Add support for len() on Proxy

  #46210 opened Oct 12, 2020

• subprocess.CalledProcessError: Command '['cmake', '--build', '.', '--target', 'install', '--config', 'Release', '--', '/p:CL_MPCount=8']' returned non-zero exit status 1

  #46208 opened Oct 12, 2020

• pytorch not uploading data to tensorboard page

  #46202 opened Oct 12, 2020

• Better support for operators that return (named) tuples of tensors

  #46187 opened Oct 12, 2020

• redesign of Dropout

  #46184 opened Oct 12, 2020

• Training a pruned model makes it 10 times slower at inference

  #46180 opened Oct 12, 2020

• torch.cuda.amp! when I use @autocast() on DCN(DeformConv), the error "RuntimeError:expect scalar type Float but Half "

  #46178 opened Oct 12, 2020

• BucketSampler for easy variable-length input batching

  #46176 opened Oct 12, 2020

• Question about CUDA usage in libtorch

  #46172 opened Oct 12, 2020

• Squared 2-norm pdist (as available in SciPy / Faiss)

  #46169 opened Oct 11, 2020

• [discussion] In-place gradient (grad_input) computation for better memory utilisation

  #46168 opened Oct 11, 2020

• Error with DistributedDataParallel with specific model

  #46166 opened Oct 11, 2020

• Deprecate the usage of `to_dense` and `to_sparse` methods.

  #46165 opened Oct 11, 2020

• Deprecate spmm and dsmm functions

  #46164 opened Oct 11, 2020

• C++ project on Windows, linker links against torch_cpu.dll instead torch_cuda.dll

  #46161 opened Oct 11, 2020

• Publish a mechanism for disabling subclass preservation of Tensor subclasses

  #46159 opened Oct 11, 2020

• No improvement gain between sm_86 (cuda 11.1) and sm_80 (cuda 11.0) on 3090 or 3080 GPUs.

  #46158 opened Oct 11, 2020

• Ability to disable cusolver

  #46157 opened Oct 11, 2020

• ONNX exporter related bug

  #46156 opened Oct 11, 2020

• Complex backward returns NaN values

  #46155 opened Oct 11, 2020

• Getting this error while installing torch and torchvision!

  #46153 opened Oct 11, 2020

• Windows 3090 Convolutional 2D network crashing

  #46150 opened Oct 11, 2020

• seeding does not work when I initialize a linear model even if I do not use it in the code

  #46149 opened Oct 10, 2020

• torch.angle() grads are wrong

  #46144 opened Oct 10, 2020

• Policy CMP0012 is not set, CMake, Building For C++ With PyTorch CUDA

  #46139 opened Oct 10, 2020

• Feature Request: Let "tensor like" objects substitute for tensors

  #46138 opened Oct 10, 2020

• bazel build failed

  #46136 opened Oct 10, 2020

• isinstance(value, Dict[str, torch.Tensor]) does not work

  #46122 opened Oct 9, 2020

• test_cat_cuda (__main__.TestTensorExprFuser) fails

  #46102 opened Oct 9, 2020

• UNET Quantization Issues (slower)

  #46096 opened Oct 9, 2020

517 Unresolved conversations

Sometimes conversations happen on old items that aren’t yet closed. Here is a list of all the Issues and Pull Requests with unresolved conversations.

• `torch.nn.modules.LazyModuleMixin` and `torch.nn.LazyLinear` (Shape Inference II)

  #44538 commented on Oct 16, 2020 • 91 new comments

• Added linalg.cholesky

  #46083 commented on Oct 16, 2020 • 71 new comments

• [JIT] adding torch.jit.isinstance support

  #46062 commented on Oct 16, 2020 • 57 new comments

• Revised sparse tensor documentation.

  #45400 commented on Oct 16, 2020 • 49 new comments

• [numpy] `torch.sin` : support and promote integer inputs to float

  #45733 commented on Oct 16, 2020 • 49 new comments

• Support %-based string formatting

  #45976 commented on Oct 16, 2020 • 39 new comments

• [pytorch][PR] Record FutureNCCL callback stream on CUDA caching allocator

  #45318 commented on Oct 16, 2020 • 35 new comments

• Implement all communication APIs in DistributedC10d new frontend

  #46053 commented on Oct 16, 2020 • 27 new comments

• [v1.7.0] Release Tracker

  #45592 commented on Oct 16, 2020 • 26 new comments

• Added linalg.tensorinv

  #45969 commented on Oct 15, 2020 • 26 new comments

• Make empty c10-full

  #46092 commented on Oct 15, 2020 • 24 new comments

• [RFC] Switch PyTorch Selective Build (Custom Build) to use the SelectiveBuilder abstraction

  #45722 commented on Oct 15, 2020 • 22 new comments

• Add basic ldexp operator for numpy compatibility

  #45370 commented on Oct 15, 2020 • 21 new comments

• GeForce RTX 3080 with CUDA capability sm_86 is not compatible with the current PyTorch installation.

  #45028 commented on Oct 15, 2020 • 20 new comments

• [nvFuser] Switching to `CudaFusionGuard` from `BailOut` for nvfuser

  #45706 commented on Oct 16, 2020 • 19 new comments

• Implement autograd functions for c10d communication operations

  #40762 commented on Oct 14, 2020 • 18 new comments

• Migrate `eig` from the TH to Aten (CUDA)

  #44105 commented on Oct 12, 2020 • 18 new comments

• [WIP][JIT] Enable ModuleDict non-literal indexing

  #45716 commented on Oct 16, 2020 • 18 new comments

• Avoid mat1 references in mm_mat1_backward

  #45777 commented on Oct 16, 2020 • 17 new comments

• torch.fft: Two dimensional FFT functions

  #45164 commented on Oct 16, 2020 • 16 new comments

• Allow converting parameters of nn.Module to complex dtypes

  #44788 commented on Oct 15, 2020 • 15 new comments

• Add `torch.swapdim`

  #46041 commented on Oct 12, 2020 • 15 new comments

• Make torch.lu_solve differentiable with respect to the LU-factorized matrix

  #22620 commented on Oct 13, 2020 • 13 new comments

• Implement hsplit, vsplit, dsplit

  #44868 commented on Oct 15, 2020 • 13 new comments

• quantize_tensor_per_channel ARM implementation

  #46018 commented on Oct 14, 2020 • 12 new comments

• [c10d] implement init_process_group in C++

  #45930 commented on Oct 16, 2020 • 11 new comments

• Accelerated 3D depthwise convolution implementation

  #40801 commented on Oct 15, 2020 • 10 new comments

• strange warning about mixed memory format

  #42300 commented on Oct 13, 2020 • 9 new comments

• Make `torch.det()` support complex input.

  #45980 commented on Oct 15, 2020 • 9 new comments

• Add type annotations for a few torch.nn.modules

  #46013 commented on Oct 16, 2020 • 9 new comments

• The support for 3080 or 3090

  #45021 commented on Oct 14, 2020 • 7 new comments

• Version 1.3 no longer supporting Tesla K40m?

  #30532 commented on Oct 13, 2020 • 6 new comments

• User-friendly handling of types and devices

  #32589 commented on Oct 13, 2020 • 6 new comments

• Build error due to unintended include path /usr/include

  #23482 commented on Oct 14, 2020 • 6 new comments

• Implementing numpy-like histogram

  #44485 commented on Oct 12, 2020 • 6 new comments

• [TorchScript] Support user defined classes as constants

  #45556 commented on Oct 16, 2020 • 6 new comments

• WIP: implement torch.linalg.svd

  #45562 commented on Oct 16, 2020 • 6 new comments

• Refactor scalar list APIs to use overloads

  #45673 commented on Oct 15, 2020 • 6 new comments

• [quant] Add check for asan and qnnpack in the conv tests

  #45913 commented on Oct 15, 2020 • 6 new comments

• with torch.cuda.amp.autocast() get out of memory error when using with torch.no_grad() during validation

  #45910 commented on Oct 12, 2020 • 5 new comments

• Singular value decomposition of complex valued matrices does not match mathematical definition

  #45821 commented on Oct 14, 2020 • 5 new comments

• [JIT] Add selective backend lowering API

  #43613 commented on Oct 16, 2020 • 5 new comments

• Choose test affinity based on current affinity

  #44369 commented on Oct 15, 2020 • 5 new comments

• [doc] add functional optimizer to doc

  #45513 commented on Oct 13, 2020 • 5 new comments

• [ONNX] Enable NoneType inputs to export API

  #45792 commented on Oct 14, 2020 • 5 new comments

• Move catchAll to Math

  #45939 commented on Oct 16, 2020 • 5 new comments

• Data loader struct pack issue(overflow)?

  #43467 commented on Oct 10, 2020 • 4 new comments

• [Feature Request] reduce CUDA runtime size by selectively compiling PyTorch GPU kernels

  #31881 commented on Oct 12, 2020 • 4 new comments

• Add support for colors (and maybe other attributes) to NVTX API

  #1750 commented on Oct 12, 2020 • 4 new comments

• Libtorch C++ multiple GPU performance slower than single GPU

  #40581 commented on Oct 14, 2020 • 4 new comments

• Why is RTX3080 slower than RTX2020-Ti?

  #45578 commented on Oct 15, 2020 • 4 new comments

• Linear Algebra tracking issue

  #42666 commented on Oct 16, 2020 • 4 new comments

• ONNX export : unrecognised node type

  #45816 commented on Oct 12, 2020 • 3 new comments

• CUDA problem with Linux Kernel 5.9

  #46034 commented on Oct 13, 2020 • 3 new comments

• nn.Conv2d allocate huge memory for some shapes and data type on V100

  #22741 commented on Oct 13, 2020 • 3 new comments

• Invalid cast to PyCFunction for function taking only 1 argument instead of 2

  #45419 commented on Oct 13, 2020 • 3 new comments

• Queue get function stucks when using multiprocessing

  #35660 commented on Oct 13, 2020 • 3 new comments

• Could not find a package configuration file provided by "Torch" with any of the following names:

  #12449 commented on Oct 14, 2020 • 3 new comments

• GroupNorm is slow stragely when calling backward() in pytorch 1.6.0

  #46086 commented on Oct 14, 2020 • 3 new comments

• MKLDNN Segmentation Fault on backward pass on CPU with Conv1D layer

  #45746 commented on Oct 14, 2020 • 3 new comments

• Comparison operators and logical operators would fail if the output tensor has a dtype different from both inputs

  #42660 commented on Oct 14, 2020 • 3 new comments

• High-level Event Tracing

  #45629 commented on Oct 15, 2020 • 3 new comments

• Mixed precision causes NaN loss

  #40497 commented on Oct 16, 2020 • 3 new comments

• Don't split oversize cached blocks

  #44742 commented on Oct 13, 2020 • 3 new comments

• CUDA BFloat16 gelu, hardswish, hardsigmoid

  #44997 commented on Oct 15, 2020 • 3 new comments

• [JIT] Add recursive scripting for class type module attributes

  #45234 commented on Oct 14, 2020 • 3 new comments

• Utility that loads a DP/DDP model state dict into a non-DDP model with the same architecture.

  #45643 commented on Oct 16, 2020 • 3 new comments

• Add Kineto submodule

  #45887 commented on Oct 16, 2020 • 3 new comments

• Make `torch.lu` support complex input on CUDA.

  #45898 commented on Oct 14, 2020 • 3 new comments

• Fix mypy type errors in module torch.distributed.* to fix #42967.

  #45987 commented on Oct 12, 2020 • 3 new comments

• [Flaky tests] Fix test_all_gather_timeout test

  #45989 commented on Oct 16, 2020 • 3 new comments

• Use CAFFE2_USE_MSVC_STATIC_RUNTIME to determine when to avoid waiting for global destructors on Windows

  #46014 commented on Oct 14, 2020 • 3 new comments

• [JIT] torchscript does not support torch.nonzero when as_tuple is True

  #38718 commented on Oct 9, 2020 • 2 new comments

• Multiple torch.load in one file

  #45136 commented on Oct 10, 2020 • 2 new comments

• Gamma distribution does not have a CDF implementation, which is very useful in Pyro

  #41637 commented on Oct 12, 2020 • 2 new comments

• TorchScript sets requires_grad to True under some circumstances

  #43193 commented on Oct 13, 2020 • 2 new comments

• TorchScript and freezing of module attributes broken, _C._jit_pass_lower_graph failing

  #43948 commented on Oct 13, 2020 • 2 new comments

• torchscript does not support unicode strings

  #42207 commented on Oct 13, 2020 • 2 new comments

• nn.Parameter{List,Dict} not copied to gpus in forward pass when nn.DataParallel is used

  #36035 commented on Oct 14, 2020 • 2 new comments

• In PyTorch Tutorials, RuntimeError: CUDA error: out of memory happen

  #46084 commented on Oct 14, 2020 • 2 new comments

• [RFC] Add Windows support to torch.distributed package

  #42095 commented on Oct 14, 2020 • 2 new comments

• Call to backward() obstructs script from terminating (maybe related to Windows 7)

  #29383 commented on Oct 14, 2020 • 2 new comments

• TestDataLoader.test_proper_exit takes 2.5min to finish

  #46065 commented on Oct 14, 2020 • 2 new comments

• [jit] Make optional[T] sortable

  #45985 commented on Oct 14, 2020 • 2 new comments

• Can't send/recv within rpc callback

  #45995 commented on Oct 14, 2020 • 2 new comments

• autograd functional APIs should enable_grad

  #44640 commented on Oct 15, 2020 • 2 new comments

• `copy.deepcopy` does not copy gradient buffers of `torch.autograd.Variable` instance

  #3307 commented on Oct 15, 2020 • 2 new comments

• test_autograd test fails with ROCm

  #44546 commented on Oct 15, 2020 • 2 new comments

• DataParallel doesn't properly handle kwargs

  #31460 commented on Oct 15, 2020 • 2 new comments

• CPU memory gradually leaks when num_workers > 0 in the DataLoader

  #13246 commented on Oct 15, 2020 • 2 new comments

• [Feature request] create sparse coo matrix w/o index check

  #14945 commented on Oct 15, 2020 • 2 new comments

• torch.sparse improvements - tracking issue

  #44634 commented on Oct 16, 2020 • 2 new comments

• Why doesn't amp improve training speed？

  #45420 commented on Oct 16, 2020 • 2 new comments

• ONNX and grid_sample layer

  #27212 commented on Oct 16, 2020 • 2 new comments

• test_nn failures on power

  #41469 commented on Oct 16, 2020 • 2 new comments

• Update the autograd notes in the doc

  #30253 commented on Oct 13, 2020 • 2 new comments

• Enabled gloo backend in test_distributed unit tests for ROCm

  #40395 commented on Oct 14, 2020 • 2 new comments

• Added context manager enabling all futures returned by rpc_async and custom build rpc functions to be automatically waited on

  #41807 commented on Oct 16, 2020 • 2 new comments

• caffe2: use at::mt19937 instead of std::mt19937 (10x speedup)

  #43987 commented on Oct 10, 2020 • 2 new comments

• [c10d] switch ProcessGroup::Work to be managed by intrusive_ptr

  #44046 commented on Oct 13, 2020 • 2 new comments

• Add support for NCCL alltoall

  #44374 commented on Oct 16, 2020 • 2 new comments

• Implement bicubic grid sampler

  #44780 commented on Oct 12, 2020 • 2 new comments

• CUDA BFloat16 kaiser_window

  #45155 commented on Oct 15, 2020 • 2 new comments

• Change largeCUDATensorTest to largeTensorTest+onlyCUDA; add a buffer to large cuda tensor test

  #45332 commented on Oct 14, 2020 • 2 new comments

• fix error message

  #45416 commented on Oct 13, 2020 • 2 new comments

• [ONNX] Add test for Batchnorm

  #45633 commented on Oct 16, 2020 • 2 new comments

• Run __setstate__ when cloning modules

  #45858 commented on Oct 14, 2020 • 2 new comments

• JavaScript (Web Assembly) target for trained models

  #25091 commented on Oct 9, 2020 • 1 new comment

• Audit quantization functions to ensure proper argument sizes

  #42336 commented on Oct 9, 2020 • 1 new comment

• AutoGrad overhead has increased significantly since 1.6

  #44484 commented on Oct 9, 2020 • 1 new comment

• investigate reported issues with quantized BatchNorm2d

  #43774 commented on Oct 9, 2020 • 1 new comment

• The `@` operation uses too much memory on GPU

  #28943 commented on Oct 9, 2020 • 1 new comment

• [jit] Shape hints for TorchScript

  #43501 commented on Oct 9, 2020 • 1 new comment

• FX should preserve type annotations and not break TorchScript-ability

  #45220 commented on Oct 10, 2020 • 1 new comment

• Tensor.sort()[:topk] is faster than Tensor.topk() on cuda

  #22812 commented on Oct 10, 2020 • 1 new comment

• Cannot update part of the parameters in DistributedDataParallel.

  #22049 commented on Oct 10, 2020 • 1 new comment

• Error installing on source: "src.cxx:1:10: fatal error: glog/stl_logging.h: No such file or directory"

  #46035 commented on Oct 10, 2020 • 1 new comment

• RuntimeError: Expected to have finished reduction in the prior iteration before starting a new one.

  #43259 commented on Oct 10, 2020 • 1 new comment

• Unexpected behaviour for affine_grid and grid_sample with 3D inputs

  #35775 commented on Oct 10, 2020 • 1 new comment

• _pad_circular incorrect padding when pads[-1], pads[-3], or pads[-5] = 0

  #37256 commented on Oct 10, 2020 • 1 new comment

• Behavior of view on out= functions

  #33351 commented on Oct 10, 2020 • 1 new comment

• torch.masked_select out argument can easily be misused, because output shape is dynamically computed

  #31895 commented on Oct 10, 2020 • 1 new comment

• Improve handling of precision in testing

  #31135 commented on Oct 10, 2020 • 1 new comment

• torch.unique is inconsistent with NumPy's unique

  #24168 commented on Oct 10, 2020 • 1 new comment

• torch.norm produces incorrect results

  #20551 commented on Oct 10, 2020 • 1 new comment

• LSTMCell consumes x1.5 more memory on CUDA on pytorch >=1.3 comparing to pytorch 1.2

  #37682 commented on Oct 10, 2020 • 1 new comment

• matrix multiply support for int type in cuda

  #20897 commented on Oct 10, 2020 • 1 new comment

• Traceback for "Warning: Mixed memory format inputs detected while calling the operator."

  #45823 commented on Oct 11, 2020 • 1 new comment

• JIT runtime error after upgrading to 1.6.0

  #42357 commented on Oct 11, 2020 • 1 new comment

• List[Any] support in TorchScript

  #41289 commented on Oct 11, 2020 • 1 new comment

• How to build libtorch static libraries on Windows?

  #42722 commented on Oct 11, 2020 • 1 new comment

• [pytorch][feature request] Cosine distance / simialrity between samples of own tensor or two tensors

  #11202 commented on Oct 12, 2020 • 1 new comment

• RuntimeError: each element in list of batch should be of equal size

  #42654 commented on Oct 12, 2020 • 1 new comment

• When running model forward with large batch size, it reports the error: THCudaTensor sizes too large for THCDeviceTensor conversion

  #24401 commented on Oct 12, 2020 • 1 new comment

• [feature request] Set limit on GPU memory use

  #18626 commented on Oct 12, 2020 • 1 new comment

• C++ ModuleDict

  #45896 commented on Oct 12, 2020 • 1 new comment

• [FR] torch context with default device & dtype

  #27878 commented on Oct 12, 2020 • 1 new comment

• [JIT] Trace the ResNet18 with MKLDNN layout

  #43039 commented on Oct 13, 2020 • 1 new comment

• Boolean indexing of an ndarray with a torch.tensor mask breaks for size=1

  #45125 commented on Oct 13, 2020 • 1 new comment

• Give clearer guidance about multithreading in PyTorch, and how to disable it

  #16899 commented on Oct 13, 2020 • 1 new comment

• Consolidate definition of operators/gradients where possible

  #22024 commented on Oct 13, 2020 • 1 new comment

• [JIT] Use of global value creates confusing error message

  #37104 commented on Oct 14, 2020 • 1 new comment

• torch.random.randperm stuck in multiprocess

  #42444 commented on Oct 14, 2020 • 1 new comment

• Pretrained model in libtorch gives different results each time

  #36461 commented on Oct 14, 2020 • 1 new comment

• [utils.bottleneck] List of improvements

  #6424 commented on Oct 14, 2020 • 1 new comment

• PackedSequence objects that are created within jit.script are not PackedSequence

  #43873 commented on Oct 14, 2020 • 1 new comment

• [RFC] DeepSpeed + PT Distributed Integration

  #42849 commented on Oct 14, 2020 • 1 new comment

• Bug in max_pool2d with ceil_mode under certain conditions

  #45357 commented on Oct 14, 2020 • 1 new comment

• Implementation of Group equivariant convolutions

  #21411 commented on Oct 15, 2020 • 1 new comment

• TripletMarginWithDistanceLoss example code doesn't work correctly

  #45210 commented on Oct 15, 2020 • 1 new comment

• backward(create_graph=True) should raise a warning for potential memory leak

  #4661 commented on Oct 15, 2020 • 1 new comment

• LR scheduler throws warning when using scaler.step instead of optimizer.step, and when saving optimizer state

  #44511 commented on Oct 15, 2020 • 1 new comment

• autodiff for user script functions aka torch.jit.script for autograd.Function

  #22329 commented on Oct 15, 2020 • 1 new comment

• fx graph mode quantization tutorials

  #45635 commented on Oct 15, 2020 • 1 new comment

• NCCL version upgrade for PyTorch

  #35363 commented on Oct 15, 2020 • 1 new comment

• [JIT] Doesn't support dict() / list()

  #36247 commented on Oct 15, 2020 • 1 new comment

• Inference performance regression caused by hacky_wrapper_for_legacy_signatures

  #43429 commented on Oct 16, 2020 • 1 new comment

• Pytorch 1.5.0 (installed from conda) errors with complaints about incompatibility between MKL and libgomp when using Pytorch's multiprocessing

  #37377 commented on Oct 16, 2020 • 1 new comment

• Unify matrix multiplications operations

  #45897 commented on Oct 16, 2020 • 1 new comment

• ICNR inizialization

  #5429 commented on Oct 13, 2020 • 1 new comment

• added beta argument to SmoothL1Loss and functional call

  #17447 commented on Oct 16, 2020 • 1 new comment

• bound nccl build to NCORES on the system

  #20693 commented on Oct 15, 2020 • 1 new comment

• Add option to log subprocess output to files in DDP launcher.

  #33193 commented on Oct 16, 2020 • 1 new comment

• Sparse-sparse matrix multiplication (CPU/CUDA)

  #39526 commented on Oct 15, 2020 • 1 new comment

• Fixes bug with ParameterList/Dict replication in DP (#36035)

  #40589 commented on Oct 12, 2020 • 1 new comment

• torch.nn.Unfold accepts 0-dim for batch size

  #40689 commented on Oct 16, 2020 • 1 new comment

• [WIP] Add color to NVTX markers

  #43720 commented on Oct 12, 2020 • 1 new comment

• DOC: Building libtorch using CMake

  #44196 commented on Oct 13, 2020 • 1 new comment

• Add HalfToFloat + FloatToHalf operators to PyTorch

  #45092 commented on Oct 15, 2020 • 1 new comment

• [fx] allow custom behavior for args, kwargs, and bool

  #45193 commented on Oct 14, 2020 • 1 new comment

• Add ReadAdapterInterface for Buffers

  #45393 commented on Oct 16, 2020 • 1 new comment

• [WIP] Detect whether to run ROCm tests rather than using environment variables

  #45570 commented on Oct 13, 2020 • 1 new comment

• [JIT] Add `__prepare_scriptable__` duck typing to allow replacing nn.modules with scriptable preparations

  #45645 commented on Oct 14, 2020 • 1 new comment

• [quant] slow path for the FBGEMM conv_transpose

  #45688 commented on Oct 15, 2020 • 1 new comment

• annotate a few torch.nn.modules.* modules

  #45772 commented on Oct 16, 2020 • 1 new comment

• [ONNX] Fix scripting rand/randn/where

  #45793 commented on Oct 16, 2020 • 1 new comment

• Fixed citation to ensure correct capitalization

  #45818 commented on Oct 10, 2020 • 1 new comment

• [WIP][ONNX] Change default of flag ONNX Shape inference in exporter to True

  #45861 commented on Oct 14, 2020 • 1 new comment

• Add USE_KINETO build option

  #45888 commented on Oct 16, 2020 • 1 new comment

• [1.7] DO NOT MERGE, Getting signal for release/1.7

  #45978 commented on Oct 16, 2020 • 1 new comment

• Add Vulkan job dispatch and flush.

  #46008 commented on Oct 16, 2020 • 1 new comment

• Add benchmark for per channel tensor quantization

  #46017 commented on Oct 14, 2020 • 1 new comment

• [ONNX] Add export of prim::dtype, prim::tolist

  #46019 commented on Oct 16, 2020 • 1 new comment

• Templateize functions in ATen/native/Math.h

  #46033 commented on Oct 9, 2020 • 1 new comment

• preserve non-dense or overlapping tensor's layout in *_like functions

  #46046 commented on Oct 14, 2020 • 1 new comment

• [NCCL] Modularize ncclCommWatchdog

  #46051 commented on Oct 15, 2020 • 1 new comment

• [PyTorch] [BUCK] Replace pt_deps.bzl with a YAML operator dependency file which is generated by the code analyser

  #46057 commented on Oct 15, 2020 • 1 new comment

• [pytorch][rampup] convenience function for loading a non-DP/DDP model from a DP/DDP state_dict

  #46063 commented on Oct 16, 2020 • 1 new comment

• [Proposal]Alias dispatch keys

  #44680 commented on Oct 9, 2020 • 0 new comments

• torch.Tensor.new_full ignores requires grad

  #36455 commented on Oct 9, 2020 • 0 new comments

• Memory Leak with Docker GPU

  #38910 commented on Oct 9, 2020 • 0 new comments

• ASAN build breaks when using third_party/protobuf

  #41918 commented on Oct 9, 2020 • 0 new comments

• ASAN build broken when using USE_ASAN=1

  #41916 commented on Oct 9, 2020 • 0 new comments

• Channels Last (NHWC) support plan.

  #28619 commented on Oct 9, 2020 • 0 new comments

• whitelist keyword to quantization.prepare is implemented incorrectly

  #41633 commented on Oct 9, 2020 • 0 new comments

• Observer in Quantization throws Warning

  #40278 commented on Oct 9, 2020 • 0 new comments

• Inconsistency loading quantized models when state_dict with Version==None

  #41684 commented on Oct 9, 2020 • 0 new comments

• Error out when parameters() is called on replicated models

  #38493 commented on Oct 9, 2020 • 0 new comments

• Type mismatch error with torch.nn.functional.grid_sample() under AMP

  #42218 commented on Oct 9, 2020 • 0 new comments

• torch.flatten returns a view, but isn't documented or tested for doing so

  #39474 commented on Oct 10, 2020 • 0 new comments

• 4D tensor support for torch.nn.functionnal.fold() (col2im)

  #38394 commented on Oct 10, 2020 • 0 new comments

• Deprecate type() and type_as() call

  #37863 commented on Oct 10, 2020 • 0 new comments

• qnnpack's quantized-add gives wrong result

  #37343 commented on Oct 10, 2020 • 0 new comments

• Dimension reducing variants of bitwise operations (bitwise_or, bitwise_and, bitwise_xor)

  #35641 commented on Oct 10, 2020 • 0 new comments

• caffe2 `DEPTHWISE3x3.Conv` test is broken

  #35553 commented on Oct 10, 2020 • 0 new comments

• max / min doesn't work on tensors with 0 elements

  #34907 commented on Oct 10, 2020 • 0 new comments

• Expanded tensors don't work as registered buffers

  #34880 commented on Oct 10, 2020 • 0 new comments

• Extend nn.functional.softmax for arbitrary dimensions

  #34874 commented on Oct 10, 2020 • 0 new comments

• keepdim for Tensor.select?

  #34788 commented on Oct 10, 2020 • 0 new comments

• [feature request]Support dilation parameter for unfold2d_* function (slow cpu maxpool2d #28733)

  #34675 commented on Oct 10, 2020 • 0 new comments

• Significant speed difference between P100 and V100

  #34576 commented on Oct 10, 2020 • 0 new comments

• Strange behaviour of F.interpolate with bicubic mode.

  #34544 commented on Oct 10, 2020 • 0 new comments

• RuntimeError with torch.unique: radix_sort: failed on 2nd step: invalid argument

  #34242 commented on Oct 10, 2020 • 0 new comments

• empty_sparse shouldn't be called with memory layout but is

  #33571 commented on Oct 10, 2020 • 0 new comments

• Inconsistency between torch.clamp() and numpy.clip() behavior for complex numbers

  #33568 commented on Oct 10, 2020 • 0 new comments

• Reverse Cumulative Sum

  #33520 commented on Oct 10, 2020 • 0 new comments

• Max-pooling uses implicit negative infinity padding, not zero-padding as indicated in documentation

  #33384 commented on Oct 10, 2020 • 0 new comments

• torch.clamp_ not inplace during backward

  #33373 commented on Oct 10, 2020 • 0 new comments

• F.max_pool1d and F.min_pool1d should support integer dtypes

  #33086 commented on Oct 10, 2020 • 0 new comments

• bytearray(tensor) behaves very differently from bytearray(tensor.numpy())

  #33041 commented on Oct 10, 2020 • 0 new comments

• Support batch linear transformation

  #33035 commented on Oct 10, 2020 • 0 new comments

• torch batchwise max with indices

  #32998 commented on Oct 10, 2020 • 0 new comments

• Optional seq_len argument to torch.nn.utils.rnn.pad_sequence

  #32912 commented on Oct 10, 2020 • 0 new comments

• Confusing error messages of tensor.scatter_ on both CPU and CUDA

  #32872 commented on Oct 10, 2020 • 0 new comments

• add_, addcmul_, addcdiv_ deprecated

  #32861 commented on Oct 10, 2020 • 0 new comments

• Logical AND and OR for Tensors in C++ API.

  #32614 commented on Oct 10, 2020 • 0 new comments

• `einsum` is ~20X slower than manually multiplying and summing

  #32591 commented on Oct 10, 2020 • 0 new comments

• [feature request] Out-variant and dtype argument for torch.argmax / torch.argmin / torch.argsort (and friends)

  #32570 commented on Oct 10, 2020 • 0 new comments

• "Tried to register multiple operators with the same name and the same overload name" error is confusing

  #32549 commented on Oct 10, 2020 • 0 new comments

• Batched torch.eig() and gradient of torch.eig() for real eigenvalues

  #32531 commented on Oct 10, 2020 • 0 new comments

• addition of attention based techniques to pytorch

  #32529 commented on Oct 10, 2020 • 0 new comments

• Gaps for making template-unboxing work for all operators

  #32366 commented on Oct 10, 2020 • 0 new comments

• [FR] bincount along arbitrary dimension

  #32306 commented on Oct 10, 2020 • 0 new comments

• [FR] histc should (optionally) return a long tensor

  #32305 commented on Oct 10, 2020 • 0 new comments

• Allow range in dim argument of reducing operations such as sum

  #32288 commented on Oct 10, 2020 • 0 new comments

• repeat_interleave Performance Issue

  #31980 commented on Oct 10, 2020 • 0 new comments

• Implement erfcx, logerfc, logerfcx special functions

  #31945 commented on Oct 10, 2020 • 0 new comments

• sparse torch.blkdiag method

  #31942 commented on Oct 10, 2020 • 0 new comments

• Improve gradient stability of logsumexp, softmax, log_softmax, logsigmoid at -inf (replace nan by zero)

  #31829 commented on Oct 10, 2020 • 0 new comments

• Documentation for `scatter` incorrectly states that index values must be unique

  #31776 commented on Oct 10, 2020 • 0 new comments

• broadcast randperm with dim specification

  #31679 commented on Oct 10, 2020 • 0 new comments

• pytorch out of memory when calculation squared difference of unfold

  #31655 commented on Oct 10, 2020 • 0 new comments

• More dynamic PyTorch APIs

  #31558 commented on Oct 10, 2020 • 0 new comments

• Slow clip_grad_norm_ because of .item() calls when run on device

  #31474 commented on Oct 10, 2020 • 0 new comments

• Integrate `torch.xxx` and `Tensor.xxx`

  #31458 commented on Oct 10, 2020 • 0 new comments

• torch.scatter_logsumexp

  #31394 commented on Oct 10, 2020 • 0 new comments

• nn.MultiHeadAttention with different similarity measures

  #31277 commented on Oct 10, 2020 • 0 new comments

• pinned memory requires DeviceGuard in multi-process envs

  #31261 commented on Oct 10, 2020 • 0 new comments

• Memory leak with LayerNorm and 2nd derivatives (gradient penalty WGAN)

  #30592 commented on Oct 10, 2020 • 0 new comments

• `F.interpolate` returns unexpected result when dealing with output size `1`

  #30565 commented on Oct 10, 2020 • 0 new comments

• pairwise_dist eps argument is confusing

  #30263 commented on Oct 10, 2020 • 0 new comments

• Support broadcasting in einsum

  #30194 commented on Oct 10, 2020 • 0 new comments

• torch.stack: bad shape error message

  #29510 commented on Oct 10, 2020 • 0 new comments

• torch.nn.functional.cosine_similarity(x, y) result greater than 1.0, and x != y.

  #29442 commented on Oct 10, 2020 • 0 new comments

• Specifying `pos_weight` in F.binary_cross_entropy_with_logits leads to RuntimeError: class size not match

  #27936 commented on Oct 10, 2020 • 0 new comments

• [Feature Request]: Batchwise torch.lstsq

  #27749 commented on Oct 10, 2020 • 0 new comments

• Issue with Exporting Interpolate to ONNX

  #27176 commented on Oct 10, 2020 • 0 new comments

• Stop binding in-place methods to `torch.*`

  #27147 commented on Oct 10, 2020 • 0 new comments

• Convert manually bound `cuda` `cpu` `byte` `float` operators to native_functions

  #28223 commented on Oct 10, 2020 • 0 new comments

• reflective padding for 5D tensor

  #28214 commented on Oct 10, 2020 • 0 new comments

• Provide a way to select SVD algorithm in PyTorch?

  #25978 commented on Oct 10, 2020 • 0 new comments

• [feature request] symmetric matrix square root

  #25481 commented on Oct 10, 2020 • 0 new comments

• euclidean distance between two tensors

  #25156 commented on Oct 10, 2020 • 0 new comments

• [feature req. for torch.inverse] optional return for indicating that whether input is not invertable or not.

  #25095 commented on Oct 10, 2020 • 0 new comments

• Add new interpolation modes to `grid_sample`

  #25039 commented on Oct 10, 2020 • 0 new comments

• [FYI] NestedTensor Project Progress

  #25032 commented on Oct 10, 2020 • 0 new comments

• torch.multinomial samples same elements multiple times when replacement=False

  #25030 commented on Oct 10, 2020 • 0 new comments

• Migrate `ormqr` from the TH to Aten (CPU)

  #24748 commented on Oct 10, 2020 • 0 new comments

• Migrate `orgqr` from the TH to Aten (CPU)

  #24747 commented on Oct 10, 2020 • 0 new comments

• Migrate `nonzero` from the TH to Aten (CPU)

  #24745 commented on Oct 10, 2020 • 0 new comments

• Migrate `mode` from the TH to Aten (CPU)

  #24731 commented on Oct 10, 2020 • 0 new comments

• Migrate `lstsq` from the TH to Aten (CPU)

  #24726 commented on Oct 10, 2020 • 0 new comments

• Migrate `index_fill_` from the TH to Aten (CPU)

  #24714 commented on Oct 10, 2020 • 0 new comments

• Migrate `histc` from the TH to Aten (CPU)

  #24712 commented on Oct 10, 2020 • 0 new comments

• Migrate `geqrf` from the TH to Aten (CPU)

  #24705 commented on Oct 10, 2020 • 0 new comments

• Migrate `equal` from the TH to Aten (CPU)

  #24697 commented on Oct 10, 2020 • 0 new comments

• Migrate `eig` from the TH to Aten (CPU)

  #24693 commented on Oct 10, 2020 • 0 new comments

• Migrate `cholesky_inverse` from the TH to Aten (CPU)

  #24685 commented on Oct 10, 2020 • 0 new comments

• Migrate `_multinomial_alias_draw` from the TH to Aten (CPU)

  #24674 commented on Oct 10, 2020 • 0 new comments

• Migrate `_mode` from the TH to Aten (CPU)

  #24673 commented on Oct 10, 2020 • 0 new comments

• Migrate `_index_copy_` from the TH to Aten (CPU)

  #24670 commented on Oct 10, 2020 • 0 new comments

• Migrate `thnn_conv_depthwise2d_forward` from the TH to Aten (CUDA)

  #24647 commented on Oct 10, 2020 • 0 new comments

• Migrate `thnn_conv_depthwise2d_backward` from the TH to Aten (CUDA)

  #24646 commented on Oct 10, 2020 • 0 new comments

• Migrate `thnn_conv2d_forward` from the TH to Aten (CUDA)

  #24645 commented on Oct 10, 2020 • 0 new comments

• Migrate `thnn_conv2d_backward` from the TH to Aten (CUDA)

  #24644 commented on Oct 10, 2020 • 0 new comments

• Migrate `tanh_backward` from the TH to Aten (CUDA)

  #24643 commented on Oct 10, 2020 • 0 new comments

• Migrate `take` from the TH to Aten (CUDA)

  #24640 commented on Oct 10, 2020 • 0 new comments

• Migrate `sort` from the TH to Aten (CUDA)

  #24637 commented on Oct 10, 2020 • 0 new comments

• Migrate `sigmoid_backward` from the TH to Aten (CUDA)

  #24625 commented on Oct 10, 2020 • 0 new comments

• Migrate `rrelu_with_noise` and `rrelu_with_noise_` from the TH to Aten (CUDA)

  #24618 commented on Oct 10, 2020 • 0 new comments

• Migrate `renorm` and `renorm_` from the TH to Aten (CUDA)

  #24616 commented on Oct 10, 2020 • 0 new comments

• Migrate `nll_loss_forward` from the TH to Aten (CUDA)

  #24610 commented on Oct 10, 2020 • 0 new comments

• Migrate `nll_loss_backward` from the TH to Aten (CUDA)

  #24609 commented on Oct 10, 2020 • 0 new comments

• Migrate `nll_loss2d_forward` from the TH to Aten (CUDA)

  #24608 commented on Oct 10, 2020 • 0 new comments

• Migrate `nll_loss2d_backward` from the TH to Aten (CUDA)

  #24607 commented on Oct 10, 2020 • 0 new comments

• Migrate `multilabel_margin_loss_forward` from the TH to Aten (CUDA)

  #24603 commented on Oct 10, 2020 • 0 new comments

• Migrate `multilabel_margin_loss_backward` from the TH to Aten (CUDA)

  #24602 commented on Oct 10, 2020 • 0 new comments

• Migrate `multi_margin_loss_backward` from the TH to Aten (CUDA)

  #24601 commented on Oct 10, 2020 • 0 new comments

• Migrate `multi_margin_loss` from the TH to Aten (CUDA)

  #24600 commented on Oct 10, 2020 • 0 new comments

• Migrate `geqrf` from the TH to Aten (CUDA)

  #24569 commented on Oct 10, 2020 • 0 new comments

• Migrate `glu` from the TH to Aten (CUDA)

  #24571 commented on Oct 10, 2020 • 0 new comments

• Migrate `glu_backward` from the TH to Aten (CUDA)

  #24572 commented on Oct 10, 2020 • 0 new comments

• Migrate `index_fill_` from the TH to Aten (CUDA)

  #24577 commented on Oct 10, 2020 • 0 new comments

• Migrate `put_` from the TH to Aten (CUDA)

  #24614 commented on Oct 10, 2020 • 0 new comments

• Migrate `nonzero` from the TH to Aten (CUDA)

  #24611 commented on Oct 10, 2020 • 0 new comments

• Migrate `mode` from the TH to Aten (CUDA)

  #24597 commented on Oct 10, 2020 • 0 new comments

• Migrate `lstsq` from the TH to Aten (CUDA)

  #24592 commented on Oct 10, 2020 • 0 new comments

• Migrate `log_sigmoid_forward` from the TH to Aten (CUDA)

  #24591 commented on Oct 10, 2020 • 0 new comments

• Migrate `log_sigmoid_backward` from the TH to Aten (CUDA)

  #24590 commented on Oct 10, 2020 • 0 new comments

• Migrate `fmod` and `fmod_` from the TH to Aten (CUDA)

  #24565 commented on Oct 10, 2020 • 0 new comments

• Migrate `eig` from the TH to Aten (CUDA)

  #24553 commented on Oct 10, 2020 • 0 new comments

• Migrate `dot` from the TH to Aten (CUDA)

  #24552 commented on Oct 10, 2020 • 0 new comments

• Migrate `_mode` from the TH to Aten (CUDA)

  #24526 commented on Oct 10, 2020 • 0 new comments

• Migrate `_min` from the TH to Aten (CUDA)

  #24525 commented on Oct 10, 2020 • 0 new comments

• Migrate `_multinomial_alias_draw` from the TH to Aten (CUDA)

  #24527 commented on Oct 10, 2020 • 0 new comments

• Migrate `_max` from the TH to Aten (CUDA)

  #24524 commented on Oct 10, 2020 • 0 new comments

• Migrate `_index_copy_` from the TH to Aten (CUDA)

  #24523 commented on Oct 10, 2020 • 0 new comments

• Migrate `cumsum` from the TH to Aten (CUDA)

  #24548 commented on Oct 10, 2020 • 0 new comments

• Migrate `cumprod` from the TH to Aten (CUDA)

  #24547 commented on Oct 10, 2020 • 0 new comments

• Migrate `cholesky_inverse` from the TH to Aten (CUDA)

  #24543 commented on Oct 10, 2020 • 0 new comments

• Migrate `baddbmm` and `baddbmm_` from the TH to Aten (CUDA)

  #24539 commented on Oct 10, 2020 • 0 new comments

• Migrate `_multinomial_alias_setup` from the TH to Aten (CUDA)

  #24528 commented on Oct 10, 2020 • 0 new comments

• Migrate `_cumsum` from the TH to Aten (CUDA)

  #24522 commented on Oct 10, 2020 • 0 new comments

• Migrate `_cumprod` from the TH to Aten (CUDA)

  #24521 commented on Oct 10, 2020 • 0 new comments

• Value_select to perform region-wise selection

  #24313 commented on Oct 10, 2020 • 0 new comments

• Port `masked_fill` operator from the TH code to Aten

  #24303 commented on Oct 10, 2020 • 0 new comments

• "PyTorch core" thread local flag

  #24261 commented on Oct 10, 2020 • 0 new comments

• Migrate `__ilshift__` from the TH to Aten (CUDA)

  #24510 commented on Oct 10, 2020 • 0 new comments

• Port TH operators to Aten (umbrella issue)

  #24507 commented on Oct 10, 2020 • 0 new comments

• Benchmark cudnn version of grid sampler

  #24470 commented on Oct 10, 2020 • 0 new comments

• Consider changing the behavior of Tensor.__contains__(Tensor) to make more sense

  #24338 commented on Oct 10, 2020 • 0 new comments

• "To compact weights again call flatten_parameters()" is printed every step for every GPU

  #24155 commented on Oct 10, 2020 • 0 new comments

• tensor.var_mean variant for existing torch.var_mean (and same for std_mean)

  #24041 commented on Oct 10, 2020 • 0 new comments

• fractional_max_pool2d_with_indices silently ignores output_ratio if output_size is provided

  #24031 commented on Oct 10, 2020 • 0 new comments

• Fan out calculation broken for group (depthwise) convolution

  #23854 commented on Oct 10, 2020 • 0 new comments

• Construction of MultivariateNormal much slower on GPU than CPU

  #23780 commented on Oct 10, 2020 • 0 new comments

• logdet & slogdet is not numerically stable occasionally

  #23473 commented on Oct 10, 2020 • 0 new comments

• einsum equation with conditional mask works in numpy but not in PyTorch

  #23434 commented on Oct 10, 2020 • 0 new comments

• make torch.utils._download_url_from_file public and add a docstring

  #23305 commented on Oct 10, 2020 • 0 new comments

• Can't `torch.sum(tensor, dim)` where `dim >= 64`

  #23159 commented on Oct 10, 2020 • 0 new comments

• Deterministic mode for scatter_add operation

  #23151 commented on Oct 10, 2020 • 0 new comments

• Unify tensor shape formatting in shape checks

  #22919 commented on Oct 10, 2020 • 0 new comments

• [feature request] Log-determinant for symmetric positive definite matrices

  #22848 commented on Oct 10, 2020 • 0 new comments

• Port `fmod` operator from the TH code to Aten

  #22803 commented on Oct 10, 2020 • 0 new comments

• No assertion when using scatter_ on a non-contiguous tensor

  #22513 commented on Oct 10, 2020 • 0 new comments

• "Floating point exception" after trying the method from the issue #22382

  #22487 commented on Oct 10, 2020 • 0 new comments

• torch.stack is slower than torch.cat().view()

  #22462 commented on Oct 10, 2020 • 0 new comments

• Training CNNs with deconvolution

  #22330 commented on Oct 10, 2020 • 0 new comments

• torch.fill_() exists and modifies the input tensor: Expected or bug?

  #22707 commented on Oct 10, 2020 • 0 new comments

• Implement torch.xlogy

  #22656 commented on Oct 10, 2020 • 0 new comments

• Batched symeig and qr are very slow on GPU

  #22573 commented on Oct 10, 2020 • 0 new comments

• Eigen Tensor library for convolutions on CPU

  #22550 commented on Oct 10, 2020 • 0 new comments

• Mysterious Tensor Indexing Problem

  #22013 commented on Oct 10, 2020 • 0 new comments

• PyTorch Equivalent of Numpy's nanmean (or add exclude_nans to mean/var/other reductions)

  #21987 commented on Oct 10, 2020 • 0 new comments

• nn.init.orthogonal_ doesn't work with multiprocessing

  #21956 commented on Oct 10, 2020 • 0 new comments

• Batch Normalization axis

  #21856 commented on Oct 10, 2020 • 0 new comments

• [FR] faster reduce sum on expanded/unfolded tensors

  #21760 commented on Oct 10, 2020 • 0 new comments

• Batched Conv2d for sequence data

  #21688 commented on Oct 10, 2020 • 0 new comments

• Slow convolution with large kernels, should be using FFT

  #21462 commented on Oct 10, 2020 • 0 new comments

• downsampling with grid_sample doesn't match interpolate

  #21457 commented on Oct 10, 2020 • 0 new comments

• Support sublist arguments for torch.einsum

  #21412 commented on Oct 10, 2020 • 0 new comments

• Overhead performance regression over time umbrella issue.

  #20367 commented on Oct 10, 2020 • 0 new comments

• Support size to `torch.normal`

  #20323 commented on Oct 10, 2020 • 0 new comments

• Advanced indexing with uint8 tensor versus int64 tensor is inconsistent

  #20149 commented on Oct 10, 2020 • 0 new comments

• Creation of too big multidimensional array returns empty tensor.

  #20056 commented on Oct 10, 2020 • 0 new comments

• Feature Request: Earth Mover's Distance Loss

  #16599 commented on Oct 10, 2020 • 0 new comments

• Flip is much slower than advanced indexing

  #16424 commented on Oct 10, 2020 • 0 new comments

• torch.full_like missing documentation for out input variable

  #20935 commented on Oct 10, 2020 • 0 new comments

• [Feature Request] Unstructured Interpolation for PyTorch Tensor

  #1552 commented on Oct 10, 2020 • 0 new comments

• Unhelpful CrossEntropyLoss dimension error message

  #1328 commented on Oct 10, 2020 • 0 new comments

• PixelShuffle for downscaling. (Feature proposal)

  #1684 commented on Oct 10, 2020 • 0 new comments

• In-place bernoulli_ has more functionality than torch.bernoulli with output parameter

  #1747 commented on Oct 10, 2020 • 0 new comments

• Discrepancy in BCEWithLogitsLoss and ClassNLLLoss

  #2180 commented on Oct 10, 2020 • 0 new comments

• PyTorch with numpy syntax?

  #2228 commented on Oct 10, 2020 • 0 new comments

• Broadcasting and Additional Dimensions for pairwise_distance

  #15918 commented on Oct 10, 2020 • 0 new comments

• [feature request] Different interpolation algos for 'grid_sample' function

  #5565 commented on Oct 10, 2020 • 0 new comments

• [Feature Request] Extract glimpses from a batch of images (as in tf.image.extract_glimpse)

  #4958 commented on Oct 10, 2020 • 0 new comments

• CUDA multinomial is limited to 2^24 categories

  #2576 commented on Oct 10, 2020 • 0 new comments

• Add a hash function for tensor data

  #2569 commented on Oct 10, 2020 • 0 new comments

• Logsigmoid as chaining operation

  #42597 commented on Oct 10, 2020 • 0 new comments

• [docs] F.softmax, F.softmin, F.log_softmax: _stacklevel in argspec, but not described in Parameters section

  #36524 commented on Oct 10, 2020 • 0 new comments

• F.one_hot to return BoolTensor/ByteTensor instead of LongTensor?

  #33046 commented on Oct 10, 2020 • 0 new comments

• [debatable] Better infer dtype in torch.as_tensor

  #33040 commented on Oct 10, 2020 • 0 new comments

• Remove TensorOptions logic from generated code

  #26428 commented on Oct 10, 2020 • 0 new comments

• Tensor from mmaped storage loads the entire file into memory

  #23382 commented on Oct 10, 2020 • 0 new comments

• Bad overload order for zeros_like

  #19685 commented on Oct 10, 2020 • 0 new comments

• Improve one_hot

  #15457 commented on Oct 10, 2020 • 0 new comments

• [Feature request] Batch eig/symeig functions (for small matrices, with CUDA)

  #10172 commented on Oct 10, 2020 • 0 new comments

• Feature request: SSIM/MS-SSIM

  #6934 commented on Oct 10, 2020 • 0 new comments

• Feature Request: Add Pixel Unshuffle

  #2456 commented on Oct 10, 2020 • 0 new comments

• BatchNorm should use Bessel's correction consistently

  #1410 commented on Oct 10, 2020 • 0 new comments

• Complex Numbers Support

  #33152 commented on Oct 10, 2020 • 0 new comments

• torch.nn.functional.normalize epsilon too small for half precision

  #32137 commented on Oct 11, 2020 • 0 new comments

• Custom attention

  #46031 commented on Oct 12, 2020 • 0 new comments

• Support undispatched ops in codegen

  #46061 commented on Oct 12, 2020 • 0 new comments

• TestXNNPACKConv1dTransformPass.test_conv1d_with_relu_fc takes 2+ min to finsh

  #46066 commented on Oct 12, 2020 • 0 new comments

• Spawn wrapper should catch BaseException, not Exception

  #46093 commented on Oct 12, 2020 • 0 new comments

• Missing dilation from several pooling modules (AvgPool)

  #14907 commented on Oct 12, 2020 • 0 new comments

• [jit] Hook support tracking

  #34329 commented on Oct 13, 2020 • 0 new comments

• [JIT] Docstrings for custom C++ classes

  #46047 commented on Oct 13, 2020 • 0 new comments

• Support for defining/saving custom operators with dynamic schema

  #42882 commented on Oct 13, 2020 • 0 new comments

• Make `torch.jit._script_if_tracing` public

  #45921 commented on Oct 13, 2020 • 0 new comments

• Uninitialised buffers don't show up in `state_dict()`

  #45834 commented on Oct 13, 2020 • 0 new comments

• Generalized Helper for Parsing Environment Variables from Process Group

  #44205 commented on Oct 13, 2020 • 0 new comments

• The index overflow when running function 'multinomial'

  #44866 commented on Oct 13, 2020 • 0 new comments

• Complex Linear does not work

  #43477 commented on Oct 13, 2020 • 0 new comments

• JIT does not support class instance attribute type annotation

  #35116 commented on Oct 13, 2020 • 0 new comments

• libtorch 1.5.0 libiomp5.dylib contains erroneous link to /DLC/torch/libiomp5.dylib instead of using @rpath

  #38607 commented on Oct 14, 2020 • 0 new comments

• torch.fft tracking issue

  #42175 commented on Oct 14, 2020 • 0 new comments

• Enable torch.quantization.stubs typechecks during CI

  #42973 commented on Oct 14, 2020 • 0 new comments

• CuDNN version not found

  #45448 commented on Oct 15, 2020 • 0 new comments

• [JIT] Implement a hash function for a tuple of ints

  #44038 commented on Oct 15, 2020 • 0 new comments

• [FX] Handling of asserts is unergonomic

  #45681 commented on Oct 15, 2020 • 0 new comments

• Make non-literal indexing for ModuleList / ModuleDict work with ModuleInterface type in TorchScript

  #43737 commented on Oct 15, 2020 • 0 new comments

• Code coverage tool not capturing ONNX tests

  #44120 commented on Oct 15, 2020 • 0 new comments

• backward of torch.repeat slower than for torch.repeat_interleave

  #43192 commented on Oct 15, 2020 • 0 new comments

• [jit] Support for union types

  #41412 commented on Oct 15, 2020 • 0 new comments

• torch.cdist returns inconsistent result

  #37734 commented on Oct 15, 2020 • 0 new comments

• torch.cdist() implementation without using contiguous() calls

  #37408 commented on Oct 15, 2020 • 0 new comments

• cdist docs imply only one batch dimension is possible, but actually arbitrarily many are allowed

  #37363 commented on Oct 15, 2020 • 0 new comments

• `cdist` backward for euclidian distance should be expressed through matrix multiplies

  #31599 commented on Oct 15, 2020 • 0 new comments

• Half precision cdist

  #29528 commented on Oct 15, 2020 • 0 new comments

• Provide function similar to cdist that returns dist_p^p

  #28119 commented on Oct 15, 2020 • 0 new comments

• [Feature Request] cdist: pairwise distances between two sets of tensors with batch mode

  #15253 commented on Oct 15, 2020 • 0 new comments

• [Distance functions] F.pdist backward CUDA invalid configuration

  #25045 commented on Oct 15, 2020 • 0 new comments

• Label tracking meta-issue (edit me to get automatically CC'ed on issues! cc bot)

  #24422 commented on Oct 15, 2020 • 0 new comments

• Please stop the codecov bot from posting comments

  #44187 commented on Oct 15, 2020 • 0 new comments

• RemoteModule enhancements

  #40550 commented on Oct 16, 2020 • 0 new comments

• changes for fmod, modf, remainder, acos

  #19404 commented on Oct 14, 2020 • 0 new comments

• [test] Introduce c10 dispatch path for ATen

  #20077 commented on Oct 10, 2020 • 0 new comments

• Libtorch android

  #20852 commented on Oct 10, 2020 • 0 new comments

• quantize_linear_per_channel using FBGEMM

  #21049 commented on Oct 10, 2020 • 0 new comments

• Moved THTensor_(cauchy) to ATen

  #21086 commented on Oct 10, 2020 • 0 new comments

• [cuda] faster kernelTransformReduceInnermostDimIndex using cub

  #21295 commented on Oct 10, 2020 • 0 new comments

• Make rshift implementation consistent between pytorch and numpy

  #21414 commented on Oct 10, 2020 • 0 new comments

• Moved THTensor_(exponential) to ATen

  #21504 commented on Oct 10, 2020 • 0 new comments

• Moved THTensor_(logNormal) to ATen

  #21505 commented on Oct 10, 2020 • 0 new comments

• Moved THTensor_(geometric) to ATen

  #21506 commented on Oct 10, 2020 • 0 new comments

• Deprecate unordered_map and vector in kernel API

  #21714 commented on Oct 10, 2020 • 0 new comments

• ptype propagation on torch functions

  #22235 commented on Oct 10, 2020 • 0 new comments

• ptype propagation on numpy functions

  #22247 commented on Oct 10, 2020 • 0 new comments

• Back out "[pytorch][PR] Optimization of the Embedding and Embedding-Bag CUDA Kernel"

  #22377 commented on Oct 10, 2020 • 0 new comments

• WIP: Debugging pytorch-complex1

  #22387 commented on Oct 10, 2020 • 0 new comments

• Implementation quantize operation for mkldnn tensor.

  #22390 commented on Oct 10, 2020 • 0 new comments

• WIP: Debugging pytorch-complex2

  #22396 commented on Oct 10, 2020 • 0 new comments

• Added "same" padding for conv*d

  #22484 commented on Oct 10, 2020 • 0 new comments

• Let add, sub, mul, and div on CPU support Half type.

  #22558 commented on Oct 10, 2020 • 0 new comments

• Disallow function variants of in-place operators other than existing ones

  #22720 commented on Oct 10, 2020 • 0 new comments

• Implement noise_shape keyword for Dropout layers

  #22735 commented on Oct 10, 2020 • 0 new comments

• Fixed all BINARY_COMP implicit signed/unsigned comparisons

  #22987 commented on Oct 10, 2020 • 0 new comments

• Fix some misc warnings

  #22988 commented on Oct 10, 2020 • 0 new comments

• Enhance add_out_dense_sparse_cpu for hybrid sparse tensor

  #23057 commented on Oct 10, 2020 • 0 new comments

• [pt1][quant] Unit Test for the Dynamic Quantized Linear operator

  #23139 commented on Oct 10, 2020 • 0 new comments

• Allow unicode (char32_t) einsum equation characters

  #23214 commented on Oct 10, 2020 • 0 new comments

• Fix adapative pooling kernel size bug on Power

  #23260 commented on Oct 10, 2020 • 0 new comments

• Allow uppercase letters in einsum equation string

  #23277 commented on Oct 10, 2020 • 0 new comments

• Correcting behaviour of torch.fmod on CPU and GPU

  #23478 commented on Oct 10, 2020 • 0 new comments

• [WIP] Support abs for half on CUDA and bool on both CPU and CUDA.

  #23579 commented on Oct 10, 2020 • 0 new comments

• WIP: Cpu strided complex support

  #25373 commented on Oct 10, 2020 • 0 new comments

• Name inference for tensor.permute

  #25713 commented on Oct 10, 2020 • 0 new comments

• [wip][lazy] A Lazy Tensor Implementation

  #25753 commented on Oct 10, 2020 • 0 new comments

• Cpu strided complex unaryop support

  #25964 commented on Oct 10, 2020 • 0 new comments

• Add a torch function similar to numpy.in1d

  #26144 commented on Oct 10, 2020 • 0 new comments

• Gh/nairbv/1/head

  #26187 commented on Oct 10, 2020 • 0 new comments

• WIP: Debugging a CI failure regarding changes in vml.h

  #26564 commented on Oct 10, 2020 • 0 new comments

• Cpu-strided-complex support for complex linear algebra ops

  #26635 commented on Oct 10, 2020 • 0 new comments

• Cumulative Maximum

  #26637 commented on Oct 10, 2020 • 0 new comments

• Test fbgemm D17514003

  #26672 commented on Oct 10, 2020 • 0 new comments

• Update Convolution.cpp (NNPACK batchsize condition remove)

  #26711 commented on Oct 10, 2020 • 0 new comments

• [WIP] Graceful cuda assert (no device-side assert triggered` error)

  #26776 commented on Oct 10, 2020 • 0 new comments

• Enable INT8 linear and conv op for mkldnn

  #26794 commented on Oct 10, 2020 • 0 new comments

• [DO NOT DELETE OR MERGE] Nightly release

  #26921 commented on Oct 16, 2020 • 0 new comments

• Add multinomial test with unnormalized probabilities

  #26962 commented on Oct 10, 2020 • 0 new comments

• Vectorize sinh and cosh on CPU.

  #27121 commented on Oct 10, 2020 • 0 new comments

• WIP: CPU-Strided-Complex Support for reduce ops

  #27127 commented on Oct 10, 2020 • 0 new comments

• [DO NOT DELETE OR MERGE] Postnightly release

  #27167 commented on Oct 16, 2020 • 0 new comments

• Back out "Revert D17599915: [pytorch][PR] Support 0-batch size for nn.Linear."

  #27650 commented on Oct 10, 2020 • 0 new comments

• enhance _embedding_bag_backward_cpu path by adding sum fast path for DLRM

  #27804 commented on Oct 10, 2020 • 0 new comments

• Remove specific processing of declarations of add, sub, mul, div, rsub

  #29238 commented on Oct 10, 2020 • 0 new comments

• Add functions add/sub/rsub/mul/div that accept a scalar as first arg

  #29602 commented on Oct 10, 2020 • 0 new comments

• Kill torch.range

  #33374 commented on Oct 10, 2020 • 0 new comments

• Batched torch.eig() and use of magma for CUDA torch.eig() operation

  #33423 commented on Oct 10, 2020 • 0 new comments

• fix min max zero element

  #35062 commented on Oct 10, 2020 • 0 new comments

• Profile IValues.

  #38836 commented on Oct 14, 2020 • 0 new comments

• Diff only fusible ops

  #41487 commented on Oct 14, 2020 • 0 new comments

• add process group UCX

  #42238 commented on Oct 15, 2020 • 0 new comments

• Port `eig` CPU from TH to ATen

  #43215 commented on Oct 9, 2020 • 0 new comments

• [quant] Factored out common code in batchnorm.py

  #43392 commented on Oct 15, 2020 • 0 new comments

• [quant] BatchNorm1d

  #43393 commented on Oct 15, 2020 • 0 new comments

• [quant][WIP] Fusion + tests + whole bunch of other things...

  #43398 commented on Oct 15, 2020 • 0 new comments

• Add Vulkan Tensor.

  #44015 commented on Oct 16, 2020 • 0 new comments

• Add Vulkan Tensor factory.

  #44016 commented on Oct 16, 2020 • 0 new comments

• Add Vulkan op Add.

  #44017 commented on Oct 16, 2020 • 0 new comments

• Fix a bug in test `torch/testing/_internal/common_utils.py`

  #44192 commented on Oct 15, 2020 • 0 new comments

• [JIT] Add reference semantics to TorchScript classes

  #44324 commented on Oct 16, 2020 • 0 new comments

• remove residual profiling nodes

  #44429 commented on Oct 14, 2020 • 0 new comments

• Enable python code coverage on windows

  #44548 commented on Oct 16, 2020 • 0 new comments

• [numpy] torch.{all, any} : Extend Dtype Support

  #44790 commented on Oct 14, 2020 • 0 new comments

• Implement NumPy-like function torch.float_power()

  #44937 commented on Oct 16, 2020 • 0 new comments

• CUDA BFloat16 batchnorm (non-cuDNN)

  #44994 commented on Oct 14, 2020 • 0 new comments

• CUDA BFloat16 Dropout

  #45005 commented on Oct 14, 2020 • 0 new comments

• pass TypeMeta by value

  #45026 commented on Oct 15, 2020 • 0 new comments

• track Half/ComplexHalf default dtype

  #45043 commented on Oct 15, 2020 • 0 new comments

• faster TensorOptions merging

  #45046 commented on Oct 15, 2020 • 0 new comments

• [ONNX] Support nd mask index in opset >= 11

  #45252 commented on Oct 13, 2020 • 0 new comments

• Add type annotations to torch.onnx.* modules

  #45258 commented on Oct 16, 2020 • 0 new comments

• use bitfield to shrink TensorImpl

  #45263 commented on Oct 15, 2020 • 0 new comments

• [PROTOTYPE] Structured kernel definitions

  #45277 commented on Oct 14, 2020 • 0 new comments

• [IGNORE] Improve optim._multi_tensor.sgd performance

  #45463 commented on Oct 12, 2020 • 0 new comments

• [quant][graphmode][fx][fix] Fix observer insert logic for ops that inherits quantization parameters for input

  #45473 commented on Oct 15, 2020 • 0 new comments

• [quant] Add FixedQParamsFakeQuantize module

  #45538 commented on Oct 15, 2020 • 0 new comments

• [quant][graphmode][fx][eagermode] Support sigmoid/hardsigmoid/tanh in eager and fx graph mode

  #45539 commented on Oct 15, 2020 • 0 new comments

• reland fast TypeMeta/ScalarType conversion

  #45544 commented on Oct 15, 2020 • 0 new comments

• [ONNX] Export var, var_mean and std_mean ops

  #45678 commented on Oct 15, 2020 • 0 new comments

• [ONNX] Add export of prim::data

  #45747 commented on Oct 13, 2020 • 0 new comments

• Fix freezing attribute not found

  #46054 commented on Oct 14, 2020 • 0 new comments

• [NCCL][debug only]

  #46069 commented on Oct 14, 2020 • 0 new comments