pytorch / pytorch

21 Pull requests merged by 8 people

• [Release/1.7.1] Embed `libiomp5.dylib` into wheel package

  #48337 merged Nov 21, 2020

• Compare Weights FX Implementation

  #48056 merged Nov 21, 2020

• [Release/1.7.1] Fix mul cuda for bool

  #48310 merged Nov 21, 2020

• [v1.7.1] third_party: Update pybind to point to fork

  #48312 merged Nov 20, 2020

• [Release/1.7.1] Fix torch.version.debug generation

  #48319 merged Nov 20, 2020

• [Release/1.7.1] Make setup.py Python 2 friendly

  #48317 merged Nov 20, 2020

• [Release/1.7.1] Fix `collect_env.py` when PyTorch is not installed

  #48311 merged Nov 20, 2020

• Fix output type of torch.max for Tensor subclasses.

  #47735 merged Nov 20, 2020

• Fix incorrect signatures in get_testing_overrides for 1.7 release

  #47736 merged Nov 20, 2020

• [Release/1.7.1] Add max supported SM for nvrtc-11.0

  #48309 merged Nov 20, 2020

• Fix documentation to point to torch.overrides instead of _overrides.

  #47843 merged Nov 20, 2020

• [PyTorch] Lazily construct guts of RecordFunction

  #47550 merged Nov 20, 2020

• Tensor-expression fuser bugfixes for 1.7.1

  #48137 merged Nov 19, 2020

• [v1.7.1] Fix max_pool1d on discontiguous tensor (#47065)

  #48219 merged Nov 19, 2020

• [v1.7.1] [complex] torch.sqrt: fix edge values (#47424)

  #48216 merged Nov 19, 2020

• [v1.7.1] Make sure valid ParameterList/Dict don't warn on creation (#47772)

  #48215 merged Nov 19, 2020

• [v1.7.1] Various setup.py fixes

  #48220 merged Nov 19, 2020

• [v1.7.1] Enable Python 3.9 for Windows builds

  #48218 merged Nov 19, 2020

• T78750158 Support varying size input in numeric suite at 10/30/2020, 3:55:01 PM

  #47391 merged Nov 19, 2020

• Properly add quotes where needed

  #46927 merged Nov 18, 2020

• [v1.7.1] Add Python 3.9 support (linux / macOS)

  #48133 merged Nov 18, 2020

145 Pull requests opened by 79 people

• [Not for land] RPC double init issue

  #48001 opened Nov 16, 2020

• [numpy] `torch.log1p` : promote integer inputs to float

  #48002 opened Nov 16, 2020

• [complex] torch.exp : match output against numpy

  #48014 opened Nov 16, 2020

• [ONNX] Enable export af aten::__derive_index

  #48019 opened Nov 16, 2020

• Adding JIT support for cuda streams and events

  #48020 opened Nov 16, 2020

• [ROCm] Enable skipped distributed global tests

  #48023 opened Nov 16, 2020

• DO NOT MERGE

  #48025 opened Nov 16, 2020

• Revert "Disable test_distributed_for for multigpu test env (#47703)"

  #48026 opened Nov 16, 2020

• [JIT] Make type singletons const references

  #48027 opened Nov 16, 2020

• Added support for complex input for torch.lu_solve #2

  #48028 opened Nov 16, 2020

• MVAPICH2 Support and Force CUDA-Aware MPI Option

  #48030 opened Nov 16, 2020

• centos now installs cmake from conda

  #48035 opened Nov 16, 2020

• Compare Weights FX Implementation

  #48039 opened Nov 16, 2020

• Compare Weights FX Implementation

  #48040 opened Nov 16, 2020

• Add support for CorrCholeskyTransform

  #48041 opened Nov 16, 2020

• Adding operator benchmarks for CPU Fusion

  #48044 opened Nov 16, 2020

• [quant] AT_DISPATCH_FLOATING_AND_QINT_TYPES

  #48050 opened Nov 16, 2020

• [quant][refactor] Reflection pad now uses AT_DISPATCH_FLOATING_AND_QINT_TYPES

  #48051 opened Nov 16, 2020

• [vulkan] conv2d weights prepack from old api

  #48052 opened Nov 16, 2020

• only enable cudnn persistent RNN when batchsize % 8 == 0

  #48070 opened Nov 17, 2020

• [ONNX] Redesign onnx pass to enable shape type dependent pattern conversion

  #48071 opened Nov 17, 2020

• Remove unused six code for Python 2/3 compatibility

  #48077 opened Nov 17, 2020

• Revert "migrate export_caffe2_op_to_c10.h macros to the new dispatcher registration API (#47321)"

  #48078 opened Nov 17, 2020

• Revert "rename macro. TORCH_LIBRARY_FRAGMENT_THIS_API_IS_FOR_PER_OP_REGISTRATION_ONLY to TORCH_LIBRARY_FRAGMENT (#47320)"

  #48079 opened Nov 17, 2020

• Revert "update legacy dispatcher registration API tests to avoid duplicate def() calls (#47319)"

  #48080 opened Nov 17, 2020

• Revert "remove ops in the __caffe2 namespace (#47318)"

  #48081 opened Nov 17, 2020

• Add newProcessGroup method to c10 new frontend.

  #48086 opened Nov 17, 2020

• fix tests - update bc time for __caffe2 ops

  #48092 opened Nov 17, 2020

• Revert "remove ops in the __caffe2 namespace"

  #48095 opened Nov 17, 2020

• Revert "Revert "remove ops in the __caffe2 namespace""

  #48096 opened Nov 17, 2020

• make duplicate def() calls an error in the dispatcher

  #48098 opened Nov 17, 2020

• Fix jit doc model loading example

  #48104 opened Nov 17, 2020

• [quant] Enabling the perchannel quant in the convTranspose

  #48105 opened Nov 17, 2020

• Pruning codeowners who don't actual do code review.

  #48109 opened Nov 17, 2020

• [IGNORE] Opt swap

  #48111 opened Nov 17, 2020

• clone pytorch.github.io with depth 1

  #48115 opened Nov 17, 2020

• Structured kernels generate Meta registrations

  #48116 opened Nov 17, 2020

• [ROCm] remove builds for versions less than 3.8

  #48118 opened Nov 17, 2020

• Leak PythonEngine singleton

  #48121 opened Nov 17, 2020

• [WIP] `torch.prod` backward for complex types.

  #48125 opened Nov 17, 2020

• DO NOT MERGE, testing sm_86 with CUDA 11.0

  #48128 opened Nov 17, 2020

• [JIT] Modify Type::isSubtypeOf[Ext] to take in a const Type&

  #48131 opened Nov 17, 2020

• .circleci: Add python 3.9 builds for windows

  #48138 opened Nov 17, 2020

• DO NOT MERGE, testing python 3.9 for windows

  #48139 opened Nov 17, 2020

• [Gloo] Compute Local Rank for Single-host Multi-GPU training (#269)

  #48140 opened Nov 18, 2020

• Add type annotations to nnapi

  #48142 opened Nov 18, 2020

• Add process group creation logic in c10d new frontend

  #48147 opened Nov 18, 2020

• Add TorchBind-based Python and TorchScript binding for ProcessGroup

  #48148 opened Nov 18, 2020

• [WIP][FX] Revamp node unique name logic + add additional uniquing in codegen

  #48157 opened Nov 18, 2020

• [caffe2] Register BlackBoxPredictor AllocationArenaPool as CPUCachingAllocator

  #48161 opened Nov 18, 2020

• [ONNX] Remove usage of isCompleteTensor() in symbolic functions

  #48162 opened Nov 18, 2020

• Fix broadcast_all crashing on Tensor-likes

  #48169 opened Nov 18, 2020

• Add documentation for torch.overrides submodule.

  #48170 opened Nov 18, 2020

• [Feature] Allow user to specify a fraction of the GPU memory.

  #48172 opened Nov 18, 2020

• Fix typo in docstrings and displayed strings.

  #48175 opened Nov 18, 2020

• [ONNX] Handle dynamic input axes for prim_ConstantChunk

  #48176 opened Nov 18, 2020

• Updating all call-sites of the legacy dispatcher registration API in fbcode to the new API.

  #48178 opened Nov 18, 2020

• [NNC] Complete SimpleIREvaluator support for bitwise ops (#48053)

  #48179 opened Nov 18, 2020

• Partial Balanced Batch Sampler

  #48180 opened Nov 18, 2020

• Refactor argument fields in FunctionSchema to Arguments

  #48182 opened Nov 18, 2020

• [CUDA] Optimize sumproduct_pair for certain inputs

  #48184 opened Nov 18, 2020

• Tuple Slicing with step size.

  #48186 opened Nov 18, 2020

• add type annotations to common_nn.py

  #48190 opened Nov 18, 2020

• [DO NOT MERGE] Warn about assertEqual(actual, expected)

  #48193 opened Nov 18, 2020

• [WIP] Fix and re-enable distributed profiling tests

  #48194 opened Nov 18, 2020

• Move argument grouping into FunctionSchema

  #48195 opened Nov 18, 2020

• [WIP][CI-all] Fix profiling tests

  #48196 opened Nov 18, 2020

• Added linalg.matrix_rank

  #48206 opened Nov 18, 2020

• [PyTorch] avoid unnecessary call to empty_tensor_restride in empty()

  #48211 opened Nov 18, 2020

• [WIP] Enabled Scalar lists

  #48222 opened Nov 19, 2020

• [WIP] Update foreach APIs to use scalar lists

  #48223 opened Nov 19, 2020

• [WIP] Pushed unsupported scenarios to slow path

  #48224 opened Nov 19, 2020

• Testing, do not merge

  #48229 opened Nov 19, 2020

• add compiler code

  #48233 opened Nov 19, 2020

• Automated submodule update: FBGEMM

  #48238 opened Nov 19, 2020

• Add warning message to cuDNN convolution for specific memory formats and GPU arch

  #48239 opened Nov 19, 2020

• Add SparseLengthsSum4BitRowwiseSparse in c2_pt_converter

  #48240 opened Nov 19, 2020

• enable cuda11.1 and cudnn 8.0.5 in CI

  #48242 opened Nov 19, 2020

• Add new backend type for Intel heterogeneous computation platform.

  #48247 opened Nov 19, 2020

• Run RPC tests with TCP init in addition to file

  #48248 opened Nov 19, 2020

• Fix bad error message when int overflow

  #48250 opened Nov 19, 2020

• disable kthvalue overlap

  #48254 opened Nov 19, 2020

• fix move default not compile correctly on cuda92

  #48257 opened Nov 19, 2020

• Added linalg.inv

  #48261 opened Nov 19, 2020

• [POC] Class-based scaffolding for structured functions

  #48262 opened Nov 19, 2020

• Remove inferred from tensor type ctors

  #48263 opened Nov 19, 2020

• [NNC] Preserve strided output

  #48264 opened Nov 19, 2020

• [vulkan] Distribute weight prepacking along y dimension for conv2d

  #48266 opened Nov 19, 2020

• Add more list peephole idioms

  #48268 opened Nov 19, 2020

• Add Removing Dict Mutation support to Remove Mutation

  #48269 opened Nov 19, 2020

• caffe2/core.Net: is_external_input rebuild lookup tables when necessary

  #48271 opened Nov 19, 2020

• Suppress unsigned warning

  #48272 opened Nov 19, 2020

• [PyTorch] Make RecordFunctionCallback::should_run_ a function pointer

  #48274 opened Nov 19, 2020

• Fix PyTorch compilation on Apple M1

  #48275 opened Nov 19, 2020

• [FX] Pull value names from Python interpreter state

  #48276 opened Nov 19, 2020

• [WIP] Fix `fmod` type promotion

  #48278 opened Nov 19, 2020

• New profiler API

  #48280 opened Nov 20, 2020

• Replace kernel resource strings with real .cu source files

  #48283 opened Nov 20, 2020

• WIP: Add missing complex support for torch.norm and torch.linalg.norm

  #48284 opened Nov 20, 2020

• Implement Kumaraswamy Distribution

  #48285 opened Nov 20, 2020

• [PyTorch] Inline RecordFunctionCallback::shouldRun

  #48286 opened Nov 20, 2020

• [WIP] Fix cholesky TF32 tests [v2]

  #48288 opened Nov 20, 2020

• Extra sampling of record function events

  #48289 opened Nov 20, 2020

• [WIP] add RoutedDecoderIterDataset prototype

  #48295 opened Nov 20, 2020

• Fix cmake warning

  #48296 opened Nov 20, 2020

• Alias _size_N_t to BroadcastingListN[int]

  #48297 opened Nov 20, 2020

• [WIP] Min max 0 dim

  #48300 opened Nov 20, 2020

• Describe columns of profiler table

  #48301 opened Nov 20, 2020

• [numpy] `torch.digamma` : promote integer inputs to float

  #48302 opened Nov 20, 2020

• fix: migrate export_caffe2_op_to_c10.h macros to the new dispatcher registration API"

  #48308 opened Nov 20, 2020

• [vulkan] Fix Addmm prepacking to persist after GPU flush

  #48313 opened Nov 20, 2020

• Vulkan conv2d old cpu shaders

  #48314 opened Nov 20, 2020

• Fix incorrect warnings in ParameterList/Dict

  #48315 opened Nov 20, 2020

• Fix typos in comments

  #48316 opened Nov 20, 2020

• Revert "Revert D25056091: migrate export_caffe2_op_to_c10.h macros to…

  #48323 opened Nov 20, 2020

• use sysconfig.BASE_PREFIX to search conda installed libs

  #48327 opened Nov 20, 2020

• Grammatically update the readme docs

  #48328 opened Nov 20, 2020

• [vulkan] convolution old prepacking via cpu-shader

  #48330 opened Nov 20, 2020

• Implement JIT serialization of ProcessGroup

  #48333 opened Nov 20, 2020

• add aot_based_partition

  #48336 opened Nov 20, 2020

• RRef proxy support for ScriptModule methods

  #48339 opened Nov 20, 2020

• caffe2: refactor context to allow being typed

  #48340 opened Nov 20, 2020

• [WIP] foldIfNode

  #48342 opened Nov 20, 2020

• [packaging] missing quotation in graphviz printout

  #48344 opened Nov 20, 2020

• enable extend_distributed alltoall torchscript support

  #48345 opened Nov 21, 2020

• [Onnxifi] Warmup cache of output shapes

  #48346 opened Nov 21, 2020

• Add CUDA kernel checks to fbcode/caffe2/caffe2/sgd

  #48347 opened Nov 21, 2020

• quant: add type annotations on quantization.fx.Quantizer matches

  #48350 opened Nov 21, 2020

• [TensorExpr] IREval: fix div for Half dtype.

  #48354 opened Nov 21, 2020

• [TensorExpr] Add a envvar to disable LLVM backend and use IR Eval instead.

  #48355 opened Nov 21, 2020

• quant fx: fix typo

  #48356 opened Nov 21, 2020

• quant: make each line of fx/quantize.py <=80 chars

  #48357 opened Nov 21, 2020

• Dev5 draft

  #48362 opened Nov 21, 2020

• matrix_exp + matrix_exp.backward complex support

  #48363 opened Nov 21, 2020

• document tf32 on the reproducibility page

  #48367 opened Nov 21, 2020

• Update sleef to 3.5.1 to remove unsupported use of __sizeless_struct

  #48368 opened Nov 22, 2020

• Implement JIT serialization of ProcessGroup

  #48369 opened Nov 22, 2020

• added docs to nn.rst

  #48374 opened Nov 22, 2020

• Dev5 metaclass

  #48375 opened Nov 22, 2020

• [docs] fix torch.swap{dim/axes} to showup in docs

  #48376 opened Nov 22, 2020

• [pytorch] prune based on custom importance scores

  #48378 opened Nov 22, 2020

• Implement JIT serialization of ProcessGroup

  #48379 opened Nov 22, 2020

• add new build settings to torch.__config__

  #48380 opened Nov 23, 2020

• Dev5 new

  #48383 opened Nov 23, 2020

• [te] Fix comparison ops on booleans

  #48384 opened Nov 23, 2020

74 Issues closed by 27 people

• assertEqual does not compare scalar numpy.ndarray with torch.Tensor correctly

  #47948 closed Nov 22, 2020

• [jit] metacompile if-statements with boolean literal or boolean constants

  #46703 closed Nov 20, 2020

• Wrong gradient for torch.norm(x, p=float('inf')) when input tensor has non-unique max values

  #41779 closed Nov 20, 2020

• torch_cuda unresolved symbols in Windows

  #47744 closed Nov 20, 2020

• addr bug for complex and inefficient implementation

  #47313 closed Nov 20, 2020

• [jit] torch.device type annotation not supported in methods

  #47405 closed Nov 20, 2020

• Core dump when checking that basic CNN works (Python 3.9)

  #47776 closed Nov 20, 2020

• Simplify RPC Message Request/Response type check

  #47145 closed Nov 20, 2020

• How to use torch.onnx.export with customed input datatype, like SparseTensor?

  #48241 closed Nov 19, 2020

• pytorch profiler to analyze memory consumption

  #48244 closed Nov 19, 2020

• torch.cumsum in-place

  #47193 closed Nov 19, 2020

• INTERNAL ASSERT FAILED for searchsorted with requires_grad inputs.

  #42328 closed Nov 19, 2020

• Wrong gradients when using DistributedDataParallel and autograd.grad.

  #47562 closed Nov 19, 2020

• nn.utils.spectral_norm() does not perform normalization of the weight matrix.

  #48181 closed Nov 19, 2020

• C++ ModuleDict

  #45896 closed Nov 19, 2020

• ONNX model export input/output shape error

  #47740 closed Nov 19, 2020

• [FX] Need docstring for Tracer.create_arg

  #45998 closed Nov 19, 2020

• [FX] Make Tensor lookup faster

  #47324 closed Nov 19, 2020

• Proper LSTM input

  #48155 closed Nov 19, 2020

• tf.nn.raw_rnn

  #48166 closed Nov 19, 2020

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

  #24779 closed Nov 18, 2020

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

  #24611 closed Nov 18, 2020

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

  #24552 closed Nov 18, 2020

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

  #24548 closed Nov 18, 2020

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

  #24547 closed Nov 18, 2020

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

  #24525 closed Nov 18, 2020

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

  #24524 closed Nov 18, 2020

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

  #24666 closed Nov 18, 2020

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

  #24522 closed Nov 18, 2020

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

  #24521 closed Nov 18, 2020

• DDP fails to sync params across processes on construction

  #48188 closed Nov 18, 2020

• [TensorExpr] Convert tensorexpr tests to gtests

  #45703 closed Nov 18, 2020

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

  #24553 closed Nov 18, 2020

• Performance improvements for `arange` CPU

  #48022 closed Nov 18, 2020

• multi_head_attention_forward bug using torch.equal

  #47979 closed Nov 18, 2020

• DISABLED test_reduce_sum_cuda_twice (__main__.TestDistBackendWithFork)

  #47674 closed Nov 18, 2020

• DISABLED test_reduce_multigpu (__main__.TestDistBackendWithFork)

  #47675 closed Nov 18, 2020

• DISABLED test_all_reduce_sum_cuda_async (__main__.TestDistBackendWithFork)

  #47676 closed Nov 18, 2020

• Enable torch.nn.modules.* typechecks during CI

  #46012 closed Nov 18, 2020

• inconsistent environment variable naming for setting NVTOOLEXT_HOME in TorchConfig.cmake

  #48032 closed Nov 18, 2020

• C10's make_optional is in the global namespace

  #47742 closed Nov 18, 2020

• The `normalize` function failed in the TorchScript interpreter with AMP enabled.

  #47966 closed Nov 18, 2020

• ImportError: numpy.core.multiarray failed to import

  #42441 closed Nov 18, 2020

• The return of torch.inverse contains nan sometime

  #47272 closed Nov 18, 2020

• C++ CUDA Project in Windows - Second call to DNN Module::forward() throws exception

  #47590 closed Nov 18, 2020

• Fail to find unused parameters when find_unused_parameters is True

  #41324 closed Nov 17, 2020

• How to convert syncbn to batchnormND?

  #48075 closed Nov 17, 2020

• Segmentation fault (core dumped) on torch.jit.optimize_function() with Amazon Elastic Inference and PyTorch

  #48015 closed Nov 17, 2020

• Pinned Memory, Non-blocking feature doesn't seem to work.

  #48065 closed Nov 17, 2020

• Modifying inplace views of leaf gives unclear error message

  #39150 closed Nov 17, 2020

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

  #40294 closed Nov 17, 2020

• jit/test_recursive_script.py is not Python 3.6+ compatible

  #47760 closed Nov 17, 2020

• How use libtorch(or other API) to implement "contiguous", "view", "permute", "transpose" in c++?

  #48074 closed Nov 17, 2020

• Move ninja build to pytorch-dockerfiles

  #7588 closed Nov 17, 2020

• [ppc64le pytorch] test_nn failure test_Softmax2d (test_nn.TestNN) ... pure virtual method called

  #7615 closed Nov 17, 2020

• Cannot INSTALL pytorch on ubuntu18.04(nvcc -ccbin clang-3.8)

  #7635 closed Nov 17, 2020

• Enable torch.distributed typechecks during CI

  #42967 closed Nov 17, 2020

• torch script error for fairseq language model

  #46517 closed Nov 17, 2020

• Constructing a ParameterDict raises a warning

  #46983 closed Nov 16, 2020

• Bitwise shorthands aren't supported in TorchScript while regular bitwise ops are.

  #42915 closed Nov 16, 2020

• Enable torch.multiprocessing typechecks during CI

  #47757 closed Nov 16, 2020

• INTERNAL ASSERT FAILED for `S.unique().shape`

  #47851 closed Nov 16, 2020

• Accessing constructors for function-specific outputs in torch.return_types

  #47962 closed Nov 16, 2020

• Improvement of autograd engine original task placement in worker threads

  #42890 closed Nov 16, 2020

• "No module named torch.version"

  #47968 closed Nov 16, 2020

• C10 error: from is out of bounds for float

  #48013 closed Nov 16, 2020

• Incorrect info about overriding torch tensors in version 1.7.0

  #47697 closed Nov 16, 2020

• Update how to build PyTorch with CUDA Windows instructions

  #47483 closed Nov 16, 2020

• Exporting the operator binary_cross_entropy_with_logits to ONNX opset version 12 is not supported

  #47998 closed Nov 16, 2020

• Replacement for THCudaBlas_SgemmBatched

  #47987 closed Nov 16, 2020

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

  #46242 closed Nov 16, 2020

• torch.arange numerics are different after 1.7 update on CPU

  #47043 closed Nov 16, 2020

• Linker error when building from source with CUDA 11.0

  #47837 closed Nov 16, 2020

• Make `torch.distributed.launch.py` search for a free port for TCPStore

  #47461 closed Nov 16, 2020

89 Issues opened by 68 people

• hangs indefinitely at os.waitpid()

  #48382 opened Nov 23, 2020

• libtorch 1.7.0 count_nonzero bug

  #48381 opened Nov 23, 2020

• Management of Gpu memory to avoid memory errors

  #48377 opened Nov 22, 2020

• swapdims & swapaxes documentation is not appearing properly

  #48372 opened Nov 22, 2020

• Unable to install PyTorch in Pycharm (macOS / python 3.9)

  #48371 opened Nov 22, 2020

• Error when dataloader has pinned memory and persistent workers on Windows

  #48370 opened Nov 22, 2020

• Add Go to high level interface

  #48366 opened Nov 21, 2020

• Custom exception for out of memory

  #48365 opened Nov 21, 2020

• Shuffle= True, the speed of loading dataset become slower,Why??

  #48364 opened Nov 21, 2020

• Cuda.is_available() return False on Docker image pytorch/pytorch:1.7.0-cuda11.0-cudnn8-devel

  #48361 opened Nov 21, 2020

• [FR] improve _generate_square_subsequent_mask function in nn.Transformer

  #48360 opened Nov 21, 2020

• C++ tensor cuda stream API docs have syntax issues

  #48358 opened Nov 21, 2020

• [FR] bool tensor should support basic arithmetics

  #48353 opened Nov 21, 2020

• [FR] .all and .any should have dim argument

  #48352 opened Nov 21, 2020

• Bool tensor should support product

  #48351 opened Nov 21, 2020

• Update default value of opset_version in torch.onnx.export

  #48341 opened Nov 20, 2020

• torch.float128 datatype

  #48338 opened Nov 20, 2020

• [JIT] Constant Propagation Shouldn't Run on Values that Escape the Graph

  #48325 opened Nov 20, 2020

• docker: adding environment variable PYTORCH_VERSION

  #48324 opened Nov 20, 2020

• race during build, missing generated file Declarations.yaml

  #48321 opened Nov 20, 2020

• Provide torchscript support for staticmethods for classes derived from Enum

  #48307 opened Nov 20, 2020

• [doc] example for pairwise distance matrix

  #48306 opened Nov 20, 2020

• [RFC] CUDA-aware future for distributed

  #48305 opened Nov 20, 2020

• Figure out a better way to find libuv on Windows

  #48304 opened Nov 20, 2020

• cosine_similarity not ONNX exportable

  #48303 opened Nov 20, 2020

• torch.matrix_exp gives incorrect results for complex tensors

  #48299 opened Nov 20, 2020

• Enable RRef proxy helpers to work with ScriptModule

  #48294 opened Nov 20, 2020

• pytorch is not linked with support for cuda devices

  #48291 opened Nov 20, 2020

• [doc] Error "You've reached a dead end" when opening torch.__config__ in the docs

  #48287 opened Nov 20, 2020

• Batched inplace mm changes stride when out size is correct

  #48281 opened Nov 20, 2020

• Pytorch streams API don't execute concurrently, However Same code in CUDA does.

  #48279 opened Nov 19, 2020

• torch.multinomial example is incorrect

  #48273 opened Nov 19, 2020

• torch.eye(d) is slow and hogs cpu for d >= 182

  #48251 opened Nov 19, 2020

• [RFC] XPU device for PyTorch

  #48246 opened Nov 19, 2020

• RuntimeError: "threshold_cpu" not implemented for 'Half'

  #48245 opened Nov 19, 2020

• torch.save writes file even when pickle fails

  #48243 opened Nov 19, 2020

• Softplus differs depending on number of elements in tensor

  #48235 opened Nov 19, 2020

• [Bug] Modification of tests for one method causes failures for the others

  #48230 opened Nov 19, 2020

• [doc] Module.share_memory is not documented

  #48228 opened Nov 19, 2020

• Modifying values() tensor of COO tensor requiring grad throws an odd error message

  #48227 opened Nov 19, 2020

• Assigning to an index of a sparse tensor is a no-op

  #48225 opened Nov 19, 2020

• Incorrect answer when using scatter_add_ and broadcasting, Feature Request: scatter_add broadcasting

  #48214 opened Nov 18, 2020

• [FX] Write a GPU inference optimization pass

  #48209 opened Nov 18, 2020

• Document torch.distributed.destroy_process_group()

  #48203 opened Nov 18, 2020

• "Found no NVIDIA driver" produces too much log

  #48202 opened Nov 18, 2020

• @torch.jit.script causes compilation error on Ampere architecture (RTX 3090)

  #48200 opened Nov 18, 2020

• Doc for global module hooks is not added to the docs website

  #48198 opened Nov 18, 2020

• [JIT] Contained Module Attributes should be recursively compiled

  #48191 opened Nov 18, 2020

• Enable torch.testing._internal.common_nn.* typechecks during CI

  #48189 opened Nov 18, 2020

• Errors in Eigen library when building pytorch mobile for Android

  #48183 opened Nov 18, 2020

• Support for oneDNN / MKL-DNN on AArch64

  #48177 opened Nov 18, 2020

• [JIT] `aten::eq/ne` should support two bool inputs

  #48174 opened Nov 18, 2020

• ONNX model's input: []

  #48173 opened Nov 18, 2020

• [libtorch] Build shared library with libtorch and compile a static library

  #48167 opened Nov 18, 2020

• Libtorch 1.7,1.6, 1.5 did not work on visual studio 2015

  #48165 opened Nov 18, 2020

• nn.functional.interpolate backward in fp16 is extremely slow

  #48163 opened Nov 18, 2020

• More robust list comprehension

  #48153 opened Nov 18, 2020

• Allow torch.nn.Linear to have 0 in_features

  #48152 opened Nov 18, 2020

• Enable PyTorch compilation on Apple Silicon

  #48145 opened Nov 18, 2020

• Make orthonormal initialization the default

  #48144 opened Nov 18, 2020

• Enable torch.backends._nnapi typechecks during CI

  #48141 opened Nov 18, 2020

• Integral Tensor `fmod` to zero gives different result on AMD and TPU

  #48130 opened Nov 17, 2020

• Distributed collectives profiling tests are flaky

  #48127 opened Nov 17, 2020

• Bad error message when int overflows

  #48114 opened Nov 17, 2020

• [distributed_c10d] Provide option in barrier function to pass rank to GPU mapping

  #48110 opened Nov 17, 2020

• JIT should not force users to write ugly code

  #48108 opened Nov 17, 2020

• quantization - document get_default_qconfig

  #48106 opened Nov 17, 2020

• test_error_in_context (__main__.ProcessGroupDistAutogradTestWithSpawn) is flaky

  #48103 opened Nov 17, 2020

• [JIT][PE] Profiling Executor should re-profile `DifferentiableGraph`

  #48094 opened Nov 17, 2020

• ShuffleNet_v2: batch input issue after convert pytorch model to onnx model

  #48093 opened Nov 17, 2020

• [JIT] enable normalization in autodiff

  #48091 opened Nov 17, 2020

• [JIT] Constant folding on no-op `aten::mul`

  #48090 opened Nov 17, 2020

• [JIT] Optimization pass in profiling executor to fold away conditional branches

  #48089 opened Nov 17, 2020

• [dataloader] Worker threads to print the signal they received before they die

  #48088 opened Nov 17, 2020

• Add documentation for the torch.overrides submodule.

  #48087 opened Nov 17, 2020

• CPU Tensor with Python MP Freezing in Docker Container

  #48083 opened Nov 17, 2020

• Microsoft C++ exception: torch::jit::ErrorReport at memory location 0x0000001C0C3BBB40.

  #48082 opened Nov 17, 2020

• [JIT] Saving and Loading JIT Model Does Not Preserve __dir__

  #48066 opened Nov 17, 2020

• FAILED: test_api/CMakeFiles/test_api.dir/dataloader.cpp.o

  #48064 opened Nov 17, 2020

• In DistributedSampler docstring, should default value of `num_replicas` be `world_size` not `rank`

  #48055 opened Nov 16, 2020

• RuntimeError: NYI: Named tensors are not supported with the tracer

  #48054 opened Nov 16, 2020

• [NNC] Fix IR Evaluator Bugs

  #48053 opened Nov 16, 2020

• native amp consumes 10x gpu memory

  #48049 opened Nov 16, 2020

• [NNC] Improve "UNSUPPORTED DTYPE" error messages

  #48045 opened Nov 16, 2020

• [TensorExpr Fuser] Handle fusing a value which has uses that aren't profiled

  #48043 opened Nov 16, 2020

• Missing -fopenmp when used in torchvision

  #48033 opened Nov 16, 2020

• [complex] torch.{exp}: does not match numpy

  #48010 opened Nov 16, 2020

• Inconsistent complex results with NumPy when computing non-positive power of 0

  #48000 opened Nov 16, 2020

• Torch DDP seems not prefer IPv6 over IPv4?

  #47999 opened Nov 16, 2020

311 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.

• [WIP] Sparse GCS layout implementation and sparse matrix multiplication.

  #44190 commented on Nov 21, 2020 • 50 new comments

• [numpy] torch.{all/any} : output dtype is always bool

  #47878 commented on Nov 20, 2020 • 40 new comments

• An implementation of torch.tile as requested in pytorch/pytorch#38349

  #47974 commented on Nov 22, 2020 • 38 new comments

• Sparse-sparse matrix multiplication (CPU/CUDA)

  #39526 commented on Nov 21, 2020 • 37 new comments

• [Pytorch][Annotation] Use callstack to get module path

  #47416 commented on Nov 19, 2020 • 32 new comments

• Add tensor.view(dtype)

  #47951 commented on Nov 19, 2020 • 30 new comments

• Enable tcpstore win v1

  #47749 commented on Nov 20, 2020 • 29 new comments

• Use Unicode friendly API on Win32 in THAllocator

  #47905 commented on Nov 21, 2020 • 28 new comments

• Implement torch.linalg.qr

  #47764 commented on Nov 17, 2020 • 26 new comments

• Implement torch.linalg.svd

  #45562 commented on Nov 17, 2020 • 24 new comments

• Added computing matrix condition numbers (linalg.cond)

  #45832 commented on Nov 20, 2020 • 24 new comments

• ATen lu_unpack. Required for making `torch.lu_solve` differentiable.

  #46913 commented on Nov 19, 2020 • 22 new comments

• [ONNX] Track and list model params for scripting

  #47348 commented on Nov 21, 2020 • 20 new comments

• [fx] Add a pass to fetch attributes of nn.Module to fx.node

  #47935 commented on Nov 20, 2020 • 20 new comments

• [v1.7.1] Release Tracker

  #47622 commented on Nov 21, 2020 • 18 new comments

• Add LazyConvXd and LazyConvTransposeXd

  #47350 commented on Nov 22, 2020 • 17 new comments

• Add high level profiling trace for dataloading and optimizer

  #47655 commented on Nov 21, 2020 • 17 new comments

• [CUDA graphs] Make CUDAGeneratorImpl capturable

  #47989 commented on Nov 23, 2020 • 16 new comments

• [JIT] Allow module interface subtyping to skip argument name matching

  #47950 commented on Nov 20, 2020 • 14 new comments

• LibTorch cannot load PyTorch exported model

  #47917 commented on Nov 17, 2020 • 13 new comments

• Upgrade oneDNN (mkl-dnn) to v1.7

  #47853 commented on Nov 21, 2020 • 13 new comments

• Add support for NCCL alltoall

  #44374 commented on Nov 19, 2020 • 12 new comments

• build pytorch from source on ubuntu, building error from fbgemm::SparseAdaGradSignature

  #47993 commented on Nov 20, 2020 • 11 new comments

• Issues on MacOS 11.0

  #47978 commented on Nov 22, 2020 • 11 new comments

• GPU acceleration for Apple's M1 chip?

  #47702 commented on Nov 20, 2020 • 10 new comments

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

  #46158 commented on Nov 23, 2020 • 10 new comments

• Proposing some safety exceptions

  #46956 commented on Nov 19, 2020 • 10 new comments

• Replace kernel resource strings with real .cu source files

  #47770 commented on Nov 17, 2020 • 9 new comments

• Add support for timing C++ snippets.

  #47864 commented on Nov 20, 2020 • 9 new comments

• initial rpc benchmark code

  #46901 commented on Nov 19, 2020 • 8 new comments

• Add nondeterministic alert to index_copy, median CUDA and kthvalue CUDA

  #46942 commented on Nov 19, 2020 • 8 new comments

• Add memory overlap checks to codegen for out and inplace operations

  #47715 commented on Nov 19, 2020 • 8 new comments

• [WIP] Parametrization Functionality

  #33344 commented on Nov 19, 2020 • 7 new comments

• [ONNX] Update onnx submodule

  #47366 commented on Nov 20, 2020 • 7 new comments

• [JIT]Exporting a scripted LSTM failed with "Modules that are called during a trace must be registered as submodules of the thing being traced"

  #47887 commented on Nov 20, 2020 • 6 new comments

• Hipify revamp

  #45451 commented on Nov 20, 2020 • 6 new comments

• refactor profiling optional

  #47667 commented on Nov 20, 2020 • 6 new comments

• Skipping Device::validate() when not necessary

  #47894 commented on Nov 16, 2020 • 6 new comments

• [numpy] `torch.rsqrt` : promote integer inputs to float

  #47909 commented on Nov 17, 2020 • 6 new comments

• /usr/bin/ld: failed to convert GOTPCREL relocation; relink with --no-relax

  #39968 commented on Nov 16, 2020 • 5 new comments

• Wrong grad of torch.symeig for degenerate eigenvalues

  #47599 commented on Nov 18, 2020 • 5 new comments

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

  #46138 commented on Nov 20, 2020 • 5 new comments

• Implement NumPy-like function torch.float_power()

  #44937 commented on Nov 20, 2020 • 5 new comments

• Improve torch.fft n-dimensional transforms

  #46911 commented on Nov 21, 2020 • 5 new comments

• [fx]added prototype of to_folder

  #47544 commented on Nov 20, 2020 • 5 new comments

• Faithful out arguments

  #47712 commented on Nov 18, 2020 • 5 new comments

• Why is RTX3080 slower than RTX2020-Ti?

  #45578 commented on Nov 16, 2020 • 4 new comments

• Incorrect output loss value under specific CUDA version

  #47844 commented on Nov 17, 2020 • 4 new comments

• amp for custom op

  #47906 commented on Nov 18, 2020 • 4 new comments

• Create a mechanism to trigger tests that only warn once

  #47624 commented on Nov 20, 2020 • 4 new comments

• Reproducibility breaks down with weighted Cross Entropy loss

  #46024 commented on Nov 20, 2020 • 4 new comments

• Training slows down and memory usage increases when upgrading from PyTorch 1.6 to 1.7

  #47056 commented on Nov 21, 2020 • 4 new comments

• Fuzzing benchmark for FFT operators

  #47872 commented on Nov 17, 2020 • 4 new comments

• Retire usages of CUDA_tensor_apply helpers in ATen

  #45447 commented on Nov 16, 2020 • 3 new comments

• [quant] Add documentation for python functions

  #47652 commented on Nov 16, 2020 • 3 new comments

• [Feature] torch.bitwise_where

  #47954 commented on Nov 17, 2020 • 3 new comments

• Mixing Numpy's arrays and PyTorch tensors

  #46829 commented on Nov 17, 2020 • 3 new comments

• Support `torch.linalg.norm` for complex tensors on both CPU and CUDA

  #47833 commented on Nov 17, 2020 • 3 new comments

• Loaded TorchScript Model in C++, and inference taken longer time?

  #47737 commented on Nov 18, 2020 • 3 new comments

• [feature request] Set limit on GPU memory use

  #18626 commented on Nov 18, 2020 • 3 new comments

• torch.ceil not supported for sparse layout

  #46988 commented on Nov 19, 2020 • 3 new comments

• Pytorch 1.7/Cuda 11.1 binaries

  #47109 commented on Nov 19, 2020 • 3 new comments

• Support calls to super().function() by implementing (poor man's) inheritance

  #47988 commented on Nov 19, 2020 • 3 new comments

• torch.distributed and RPC cannot both be initialized with the same host:port pair.

  #41614 commented on Nov 20, 2020 • 3 new comments

• Vec256<int64_t> does not handle LONG_MAX on minimum

  #43254 commented on Nov 20, 2020 • 3 new comments

• nn.utils.rnn.pack_padded_sequence: RuntimeError: 'lengths' argument should be a 1D CPU int64 tensor, but got 1D cuda:0 Long tensor

  #43227 commented on Nov 20, 2020 • 3 new comments

• Incorrect Validation Accuracy Due to Distributed Sampler

  #25162 commented on Nov 20, 2020 • 3 new comments

• Issues with building Libtorch with /MTd in Windows

  #46318 commented on Nov 21, 2020 • 3 new comments

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

  #45823 commented on Nov 23, 2020 • 3 new comments

• Port `eig` CPU from TH to ATen

  #43215 commented on Nov 19, 2020 • 3 new comments

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

  #44105 commented on Nov 19, 2020 • 3 new comments

• Set caffe2::pthreadpool() size in ParallelOpenMP

  #45566 commented on Nov 17, 2020 • 3 new comments

• Use libkineto in profiler

  #46470 commented on Nov 22, 2020 • 3 new comments

• Server connects to its listen socket addr

  #46801 commented on Nov 17, 2020 • 3 new comments

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

  #47323 commented on Nov 20, 2020 • 3 new comments

• [WIP] web dataset prototype

  #47831 commented on Nov 20, 2020 • 3 new comments

• [JIT] Fix type annotations of pooling modules

  #47888 commented on Nov 16, 2020 • 3 new comments

• [ONNX] Update batch_norm symbolic to handle track_running_stats=False

  #47903 commented on Nov 19, 2020 • 3 new comments

• Cudnn header files should be copied into build package as well

  #47743 commented on Nov 16, 2020 • 2 new comments

• Code becomes more than x20 slower after upgrading torch version

  #47912 commented on Nov 16, 2020 • 2 new comments

• Tensor iteration changes behavior in 1.7 with no hint

  #47899 commented on Nov 16, 2020 • 2 new comments

• Throw error if `kthvalue` `out` argument overlaps with input

  #47934 commented on Nov 16, 2020 • 2 new comments

• Python autograd engine threads never terminate in Python 3.5-3.8

  #38231 commented on Nov 16, 2020 • 2 new comments

• RuntimeError: CUDA error: an illegal memory access was encountered

  #46974 commented on Nov 17, 2020 • 2 new comments

• torch.distributions.utils.broadcast_all does not support Tensor-like objects

  #37141 commented on Nov 17, 2020 • 2 new comments

• Is nn.Conv2d equivalent with Unfold + Matrix Multiplication + Fold ?

  #47990 commented on Nov 17, 2020 • 2 new comments

• Fix the way imports are done to be more correct for static type checkers

  #47027 commented on Nov 17, 2020 • 2 new comments

• Inconsistent interactions of PyTorch tensors and NumPy ops

  #7610 commented on Nov 17, 2020 • 2 new comments

• [ONNX]RuntimeError: Unexpected node type: onnx::Cast

  #47182 commented on Nov 18, 2020 • 2 new comments

• nn.InstanceNorm behavior is not consistent with batch size for corner case of inputs with spatial dimension = 1

  #45687 commented on Nov 18, 2020 • 2 new comments

• RFC: torch.vmap

  #42368 commented on Nov 18, 2020 • 2 new comments

• Make torch.nn.Hardswish ONNX exportable

  #43665 commented on Nov 18, 2020 • 2 new comments

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

  #13246 commented on Nov 20, 2020 • 2 new comments

• Batched MAGMA calls illegally read CUDA memory

  #26996 commented on Nov 20, 2020 • 2 new comments

• Pytorch 1.6.0 linking error when linking `libtorch_cuda.dylib` on macOS 10.13.6 with CUDA support

  #42271 commented on Nov 20, 2020 • 2 new comments

• Backward for sparse tensor item select does not work

  #45459 commented on Nov 20, 2020 • 2 new comments

• Error compiling from source (OS X 10.12, CUDA 9.0, CUDNN 7.0.4)

  #7721 commented on Nov 20, 2020 • 2 new comments

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

  #5565 commented on Nov 21, 2020 • 2 new comments

• enhance _embedding_bag_backward_cpu path by adding sum fast path for DLRM

  #27804 commented on Nov 18, 2020 • 2 new comments

• add mkldnn relu backward and reshape backward

  #37199 commented on Nov 19, 2020 • 2 new comments

• fix error message

  #45416 commented on Nov 19, 2020 • 2 new comments

• BatchNorm channels last memory format optimization on CPU

  #46234 commented on Nov 19, 2020 • 2 new comments

• [ONNX] Handle sequence output shape and type inference

  #46542 commented on Nov 19, 2020 • 2 new comments

• Resolve bug with torch.multiprocessing.spawn when it hangs during huge exception stacktrace

  #46754 commented on Nov 20, 2020 • 2 new comments

• [ONNX] Add export for aten::__isnot_ and prim::unchecked_cast

  #47036 commented on Nov 18, 2020 • 2 new comments

• Support torch.distributed.irecv(src=None, ...)

  #47137 commented on Nov 21, 2020 • 2 new comments

• [numpy] Add unary-ufunc tests for `erf` variants

  #47155 commented on Nov 22, 2020 • 2 new comments

• [ONNX] Changes to export API to better handle named arguments

  #47367 commented on Nov 19, 2020 • 2 new comments

• Fix lite interpreter record function issue.

  #47457 commented on Nov 21, 2020 • 2 new comments

• add channels last support for AvgPool2d on CPU path

  #47490 commented on Nov 17, 2020 • 2 new comments

• [PyTorch] Change representation of SizesAndStrides

  #47508 commented on Nov 19, 2020 • 2 new comments

• Adding support for CuDNN-based LSTM with projections

  #47725 commented on Nov 18, 2020 • 2 new comments

• Modify Pipe to return an RRef.

  #47829 commented on Nov 19, 2020 • 2 new comments

• Fixed einsum compatibility/performance issues (#46398)

  #47860 commented on Nov 20, 2020 • 2 new comments

• Destroy autograd device threads at exit

  #47886 commented on Nov 17, 2020 • 2 new comments

• DO NOT MERGE YET Replacing nightly CUDA11.0 builds with 11.1

  #47938 commented on Nov 20, 2020 • 2 new comments

• [JIT] Fix clang-tidy warnings in jit/runtime

  #47992 commented on Nov 21, 2020 • 2 new comments

• Unable to install pytorch with cuda 10.0 using conda

  #28320 commented on Nov 16, 2020 • 1 new comment

• cannot call rpc.init_rpc twice within a single process

  #46491 commented on Nov 16, 2020 • 1 new comment

• build pytorch from source fauled: undefined reference to `cusparseGetErrorString(cusparseStatus_t)'

  #32083 commented on Nov 16, 2020 • 1 new comment

• sparse filter layers (more specifically convolutions)

  #47915 commented on Nov 16, 2020 • 1 new comment

• Get forward output, grad, hessian all at once

  #47901 commented on Nov 16, 2020 • 1 new comment

• Function request: Sparse matrix inverse

  #47918 commented on Nov 16, 2020 • 1 new comment

• QAT with DDP should have documentation

  #47857 commented on Nov 16, 2020 • 1 new comment

• FX quantization: we should preserve original model class name through the quantization passes

  #47363 commented on Nov 16, 2020 • 1 new comment

• How to avoid `torch.onnx.export` use INT64?

  #47980 commented on Nov 16, 2020 • 1 new comment

• sparse.mm(S, D) with autograd

  #2389 commented on Nov 16, 2020 • 1 new comment

• PowBackward always operates in Complex dtype

  #46936 commented on Nov 16, 2020 • 1 new comment

• Unknown DispatchKey when indexing tensor with named dimensions

  #47931 commented on Nov 16, 2020 • 1 new comment

• computing entropy of a tensor

  #15829 commented on Nov 16, 2020 • 1 new comment

• pack_padded_sequence requires lengths to be on CPU

  #47637 commented on Nov 16, 2020 • 1 new comment

• "sub_iter.strides(0)[0] == 0 INTERNAL ASSERT FAILED at Reduce.cuh" when doing .sum() on large cuda tensor

  #37583 commented on Nov 16, 2020 • 1 new comment

• Investigate Applying PowerSGD to Communication Hook for Gradient Compression

  #47202 commented on Nov 16, 2020 • 1 new comment

• support for multiple torch.cuda.max_memory_allocated() counters

  #16266 commented on Nov 17, 2020 • 1 new comment

• ONNX and grid_sample layer

  #27212 commented on Nov 17, 2020 • 1 new comment

• Factorial & Binomial Coefficient

  #47841 commented on Nov 17, 2020 • 1 new comment

• embeddings layer with IntTensor / cuda.IntTensor inputs

  #145 commented on Nov 17, 2020 • 1 new comment

• ARM Mac 16-core Neural Engine

  #47688 commented on Nov 17, 2020 • 1 new comment

• PyTorch to ONNX issue

  #36899 commented on Nov 17, 2020 • 1 new comment

• [FR] torch context with default device & dtype

  #27878 commented on Nov 17, 2020 • 1 new comment

• Error when loading jit traced FasterRCNN model in C++

  #35881 commented on Nov 17, 2020 • 1 new comment

• Linear algebra GPU backend tracking issue [magma/cusolver/cublas]

  #47953 commented on Nov 17, 2020 • 1 new comment

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

  #32591 commented on Nov 17, 2020 • 1 new comment

• TorchScript is unable to distinguish int and ScalarType

  #47964 commented on Nov 17, 2020 • 1 new comment

• keepdims doesn't work in JIT

  #47955 commented on Nov 17, 2020 • 1 new comment

• DataParallel support for scripted modules in C++

  #47650 commented on Nov 17, 2020 • 1 new comment

• [jit][script] torch.jit.script does not support nn.RNN

  #28418 commented on Nov 17, 2020 • 1 new comment

• [PyTorch][Feature Request] Label Smoothing for CrossEntropyLoss

  #7455 commented on Nov 17, 2020 • 1 new comment

• Enable PyTorch Bfloat16 for CPU and add MKL-DNN bfloat16 optimization for Cooper Lake

  #23509 commented on Nov 17, 2020 • 1 new comment

• [jit] support for torch.distributed primitives

  #41353 commented on Nov 17, 2020 • 1 new comment

• NCCL 2.7.8 errors on PyTorch distributed process group creation

  #47885 commented on Nov 17, 2020 • 1 new comment

• torch.autograd.backward() fails to sync with other stream

  #47028 commented on Nov 17, 2020 • 1 new comment

• test_cudart_register doesn't work on ROCm

  #45202 commented on Nov 17, 2020 • 1 new comment

• RuntimeError: Exporting the operator quantize_per_tensor to ONNX opset version 10 is not supported.

  #47217 commented on Nov 18, 2020 • 1 new comment

• Windows 10 pytorch installation with Python 3.8.6 not working

  #47444 commented on Nov 18, 2020 • 1 new comment

• Upsample with a trilinear interpolation works at least 10x slower using Mixed Precision than with FP32.

  #44206 commented on Nov 18, 2020 • 1 new comment

• LibTorch 1.5.0 not supporting GLIBC < 2.23

  #38765 commented on Nov 18, 2020 • 1 new comment

• torch.distributed.reduce hangs with nccl backend

  #47498 commented on Nov 18, 2020 • 1 new comment

• Flask report "RuntimeError: cuda runtime error (3) : initialization error "

  #36068 commented on Nov 18, 2020 • 1 new comment

• torch.as_tensor fails to create named tensors

  #28742 commented on Nov 18, 2020 • 1 new comment

• CI no longer runs ProcessGroup MPI Tests

  #47443 commented on Nov 18, 2020 • 1 new comment

• Cholesky: return info instead of RuntimeError

  #47608 commented on Nov 18, 2020 • 1 new comment

• LSTM is not properly exported to ONNX

  #16025 commented on Nov 19, 2020 • 1 new comment

• LinearOperator Abstraction / Structure-Exploiting LazyTensors for Linear Algebra

  #28341 commented on Nov 19, 2020 • 1 new comment

• [proposal] [discussion] An abstraction to facilitate building weight_norm/spectral_norm-like things aka Reparametrization

  #7313 commented on Nov 19, 2020 • 1 new comment

• nn.Orthogonal

  #42243 commented on Nov 19, 2020 • 1 new comment

• DataLoader converts cuda FloatTensor into cpu DoubleTensor when shape is (n,)

  #2539 commented on Nov 19, 2020 • 1 new comment

• Implement torch::cuda::synchronize()

  #47722 commented on Nov 19, 2020 • 1 new comment

• Force JIT to do type inference even when mypy annotated

  #39670 commented on Nov 19, 2020 • 1 new comment

• Deepcopy fails with nn.parallel.replicate

  #33785 commented on Nov 20, 2020 • 1 new comment

• 'tensorflow_core._api.v2.io.gfile' has no attribute 'get_filesystem'

  #30966 commented on Nov 20, 2020 • 1 new comment

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

  #46161 commented on Nov 20, 2020 • 1 new comment

• [Feature request] Stochastic Frank-Wolfe optimizer

  #47307 commented on Nov 20, 2020 • 1 new comment

• Matrix corresponding to convolution by a 2D kernel (convmtx2)

  #26781 commented on Nov 20, 2020 • 1 new comment

• nn.MultiheadAttention causes gradients to become NaN under some use cases

  #41508 commented on Nov 20, 2020 • 1 new comment

• Does this mean onnx do not support spatialBN operator?

  #10062 commented on Nov 21, 2020 • 1 new comment

• please add 'tensor.astype(dtype_string)' syntax for numpy interoperability

  #40471 commented on Nov 21, 2020 • 1 new comment

• Add TF32 Information to Reproducibility Page

  #47287 commented on Nov 22, 2020 • 1 new comment

• Cannot re-initialize CUDA in forked subprocess

  #40403 commented on Nov 22, 2020 • 1 new comment

• Inconsistent handling of torch.Size.__add__

  #40914 commented on Nov 22, 2020 • 1 new comment

• No padding_idx argument in the EmbeddingBag layer

  #3194 commented on Nov 22, 2020 • 1 new comment

• PyTorch 1.7.0 CUDA driver warning

  #47038 commented on Nov 22, 2020 • 1 new comment

• SVD very very slow and GELS gives nans, -inf

  #11174 commented on Nov 23, 2020 • 1 new comment

• Can't solve torch.lstsq() with specific values

  #44378 commented on Nov 23, 2020 • 1 new comment

• Enhance add_out_dense_sparse_cpu for hybrid sparse tensor

  #23057 commented on Nov 19, 2020 • 1 new comment

• Add quantized avgpool1d[WIP]

  #33312 commented on Nov 21, 2020 • 1 new comment

• BFloat16: enable prepacked weights's inference

  #37218 commented on Nov 17, 2020 • 1 new comment

• support non-contiguous tensor in bilinear

  #38409 commented on Nov 16, 2020 • 1 new comment

• Update gather documentation to allow index.shape[k] <= input.shape[k] rather than ==.

  #41887 commented on Nov 19, 2020 • 1 new comment

• Add a new op for converting the dense feature to sparse representation

  #42119 commented on Nov 21, 2020 • 1 new comment

• onnx export of per channel fake quantize functions

  #42835 commented on Nov 17, 2020 • 1 new comment

• Raise error if Tensor.numel() calculation overflows

  #43755 commented on Nov 16, 2020 • 1 new comment

• Add generator for dropout

  #44289 commented on Nov 22, 2020 • 1 new comment

• Extend pytorch multi tensor rmsprop optimizer to support lr_in_momentum and decoupled_decay.

  #46118 commented on Nov 21, 2020 • 1 new comment

• DOC: dynamically create a version pull-down using javascript

  #46652 commented on Nov 17, 2020 • 1 new comment

• Merge all vec256 tests into one framework

  #47294 commented on Nov 21, 2020 • 1 new comment

• Support varying size input in numeric suite at 10/30/2020, …

  #47392 commented on Nov 19, 2020 • 1 new comment

• error in all_reduce on unsupported ops for complex tensors

  #47522 commented on Nov 19, 2020 • 1 new comment

• Support the int32 index type

  #47536 commented on Nov 21, 2020 • 1 new comment

• Allow zero annealing epochs

  #47579 commented on Nov 18, 2020 • 1 new comment

• [v1.7.1] Getting signal for release branch

  #47623 commented on Nov 22, 2020 • 1 new comment

• adding profile_ivalue

  #47666 commented on Nov 20, 2020 • 1 new comment

• Add test for empty tensors for batch matmuls

  #47700 commented on Nov 17, 2020 • 1 new comment

• Trim long file paths in profiler output

  #47710 commented on Nov 20, 2020 • 1 new comment

• [NCCL] enable p2p tests (#47685)

  #47769 commented on Nov 19, 2020 • 1 new comment

• [WIP] tuple slice support

  #47771 commented on Nov 16, 2020 • 1 new comment

• enabling universal binary for c10 executables

  #47773 commented on Nov 17, 2020 • 1 new comment

• [quant][refactor] Moving the 'quantization/fx/utils.py' to 'quantization/utils.py'

  #47801 commented on Nov 17, 2020 • 1 new comment

• Rework compat bindings.

  #47863 commented on Nov 18, 2020 • 1 new comment

• Fix complex tensors and missing data in benchmark utility

  #47871 commented on Nov 17, 2020 • 1 new comment

• Be more careful with device ids

  #47920 commented on Nov 18, 2020 • 1 new comment

• [cmake] clean up blas discovery

  #47940 commented on Nov 20, 2020 • 1 new comment

• [PyTorch][JIT] Parameter passing & std::map API usage pass on ProfilingRecord::instrumentGraph

  #47960 commented on Nov 18, 2020 • 1 new comment

• [JIT] Fix clang-tidy warnings for jit/frontend

  #47982 commented on Nov 21, 2020 • 1 new comment

• [JIT] Fix clang-tidy warnings jit/ir

  #47983 commented on Nov 21, 2020 • 1 new comment

• [JIT] Fix clang-tidy warnings in jit/passes

  #47984 commented on Nov 21, 2020 • 1 new comment

• [JIT] Fix clang-tidy warnings in jit/testing

  #47986 commented on Nov 21, 2020 • 1 new comment

• [RFC]Enable TcpStore for DDP on Windows platform

  #47659 commented on Nov 16, 2020 • 0 new comments

• [complex] torch.sqrt: Incorrect value on MacOS

  #47500 commented on Nov 16, 2020 • 0 new comments

• Autograd view CreationMeta are not properly propagated when chaining views

  #36608 commented on Nov 16, 2020 • 0 new comments

• torch in cloud(randomly sche) may have server connect timeout problem

  #46800 commented on Nov 16, 2020 • 0 new comments

• Implement pep 503 Simple Repository API for deployment

  #25639 commented on Nov 16, 2020 • 0 new comments

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

  #23434 commented on Nov 16, 2020 • 0 new comments

• Non-deterministic results when indexing into/out of the same tensor

  #45964 commented on Nov 16, 2020 • 0 new comments

• Exporting the operator binary_cross_entropy_with_logits to ONNX opset version 12 is not supported

  #47997 commented on Nov 16, 2020 • 0 new comments

• Issues with new implementation of layernorm

  #47902 commented on Nov 16, 2020 • 0 new comments

• Follow the JAX or Tensorflow convention for meaning of grad() with complex inputs

  #41857 commented on Nov 17, 2020 • 0 new comments

• Can we apply Weight normalization hook to a method other than `forward`?

  #47588 commented on Nov 17, 2020 • 0 new comments

• [feature request] Update module parameters after __init__()

  #7620 commented on Nov 17, 2020 • 0 new comments

• ONNX - .view() function doesn't seem to export valid Reshape

  #7612 commented on Nov 17, 2020 • 0 new comments

• A problem about elu_backward

  #47671 commented on Nov 17, 2020 • 0 new comments

• [jit] add ignored attributed to scriptmodule

  #43021 commented on Nov 17, 2020 • 0 new comments

• [doc] jit.fork doc is badly formatted and missing example

  #47932 commented on Nov 17, 2020 • 0 new comments

• NumPy-like Functionality Request Rollup

  #38349 commented on Nov 17, 2020 • 0 new comments

• Pytorch 1.7.0 with cuda 11.1.1 and cudnn 8.0.5

  #47669 commented on Nov 18, 2020 • 0 new comments

• Yolov5 detect.py(ingerence) Error:Torch.nn.modules.module.ModuleAttributeError: 'Hardswish' object has no attribute 'inplace'

  #46971 commented on Nov 18, 2020 • 0 new comments

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

  #23482 commented on Nov 18, 2020 • 0 new comments

• Memory Leak with Docker GPU

  #38910 commented on Nov 18, 2020 • 0 new comments

• Port TH operators to Aten (umbrella issue)

  #24507 commented on Nov 18, 2020 • 0 new comments

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

  #45746 commented on Nov 18, 2020 • 0 new comments

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

  #42218 commented on Nov 18, 2020 • 0 new comments

• Remove .data subset1 for fixathon

  #33628 commented on Nov 19, 2020 • 0 new comments

• Model weights silently fail to load when model is on different gpu than when it was saved

  #25878 commented on Nov 19, 2020 • 0 new comments

• at::from_blob documentation is broken

  #47462 commented on Nov 19, 2020 • 0 new comments

• Unary Ufunc tracking issue

  #42515 commented on Nov 19, 2020 • 0 new comments

• GPU Vendor-Agnosticism via Vulkan

  #47996 commented on Nov 19, 2020 • 0 new comments

• Linear Algebra tracking issue

  #42666 commented on Nov 19, 2020 • 0 new comments

• Complex Tensor Support for DataParallel

  #47330 commented on Nov 20, 2020 • 0 new comments

• [FX] Make a subgraph rewriter API

  #46962 commented on Nov 20, 2020 • 0 new comments

• Conv3D can be optimized for cases when kernel is spatial (probably)

  #7740 commented on Nov 20, 2020 • 0 new comments

• Keyword-only function not allowed with TorchScript

  #47565 commented on Nov 20, 2020 • 0 new comments

• Record shaping assertions and use them for tracing / scripting optimization and codegen

  #46948 commented on Nov 20, 2020 • 0 new comments

• [DO NOT DELETE OR MERGE] Nightly release

  #26921 commented on Nov 22, 2020 • 0 new comments

• [DO NOT DELETE OR MERGE] Postnightly release

  #27167 commented on Nov 22, 2020 • 0 new comments

• Migrate nonzero from the TH to Aten(CPU)

  #27217 commented on Nov 16, 2020 • 0 new comments

• Clearly define accumulate type, data type and output type for shared reduce kernels

  #29280 commented on Nov 16, 2020 • 0 new comments

• [WIP] Implement `torch.version.cuda`

  #31563 commented on Nov 17, 2020 • 0 new comments

• [ONNX] Fix translation for -1 in Expand op

  #33614 commented on Nov 17, 2020 • 0 new comments

• Add nanprod to ReduceOps

  #38634 commented on Nov 22, 2020 • 0 new comments

• add channels last support for AdaptiveAvgPool2d on CPU path

  #42104 commented on Nov 17, 2020 • 0 new comments

• Implementing sinc

  #42285 commented on Nov 18, 2020 • 0 new comments

• add channels last support for MaxPool2d on CPU path

  #42719 commented on Nov 17, 2020 • 0 new comments

• torchhub for private repository

  #44174 commented on Nov 17, 2020 • 0 new comments

• [JIT] Add reference semantics to TorchScript classes

  #44324 commented on Nov 19, 2020 • 0 new comments

• Add Kineto submodule

  #45887 commented on Nov 20, 2020 • 0 new comments

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

  #45978 commented on Nov 22, 2020 • 0 new comments

• FLOPs Roofline Analysis Feature for PyTorch Profiler.

  #46506 commented on Nov 21, 2020 • 0 new comments

• Implemented torch.inner

  #46716 commented on Nov 21, 2020 • 0 new comments

• [WIP][JIT] Create a backward pass for Gelu Function in JIT

  #46785 commented on Nov 17, 2020 • 0 new comments

• Fix integer conversion warnings

  #46836 commented on Nov 18, 2020 • 0 new comments

• [ONNX][WIP] Enable opset 13 ORT tests

  #46903 commented on Nov 17, 2020 • 0 new comments

• Refactor mkl fft planning to not use Tensor objects

  #46910 commented on Nov 21, 2020 • 0 new comments

• Optimize torch.fft.fftn for real input with an r2c transform

  #46912 commented on Nov 21, 2020 • 0 new comments

• Accept input tensor with 0-dim batch size for MultiLabelMarginLoss

  #46975 commented on Nov 20, 2020 • 0 new comments

• support bool tensordot

  #47046 commented on Nov 18, 2020 • 0 new comments

• Added CUDA support for complex input for torch.cholesky_solve

  #47047 commented on Nov 16, 2020 • 0 new comments

• Support the `in` operator with str

  #47057 commented on Nov 17, 2020 • 0 new comments

• Preserve submodule with __set_state__ in freezing

  #47308 commented on Nov 20, 2020 • 0 new comments

• Added foreach_trunc, foreahc_reciprocal, foreach_sigmoid APIs

  #47385 commented on Nov 19, 2020 • 0 new comments

• [JIT] adding missing test cases for test_isinstance.py

  #47396 commented on Nov 16, 2020 • 0 new comments

• [PyTorch] Introduce packed SizesAndStrides abstraction

  #47507 commented on Nov 19, 2020 • 0 new comments

• Autograd functional API should enable_grad

  #47543 commented on Nov 18, 2020 • 0 new comments

• [numpy] `torch.sigmoid` : promote integer inputs to float

  #47551 commented on Nov 19, 2020 • 0 new comments

• Use new FFT operators in stft

  #47601 commented on Nov 21, 2020 • 0 new comments

• [Test-Only][Don't Review] Re-enable test_distributed_fork.py in multigpu test

  #47639 commented on Nov 16, 2020 • 0 new comments

• Clarify and test the behavior of logspace when dtype is integral

  #47647 commented on Nov 19, 2020 • 0 new comments

• [ONNX] Cast Gather index to Long if needed

  #47653 commented on Nov 19, 2020 • 0 new comments

• Remove output used only by sizes (#448)

  #47665 commented on Nov 20, 2020 • 0 new comments

• profile ivalue for nvfuser

  #47668 commented on Nov 20, 2020 • 0 new comments

• change block codegen to handle new inlining in NNC

  #47687 commented on Nov 17, 2020 • 0 new comments

• DO NOT MERGE, testing windows / macOS py3.9

  #47690 commented on Nov 17, 2020 • 0 new comments

• Fix generator exhaustion in SparseAdam

  #47724 commented on Nov 20, 2020 • 0 new comments

• Tweak Vulkan memory use.

  #47728 commented on Nov 21, 2020 • 0 new comments

• Updated derivative rules for complex svd and pinverse

  #47761 commented on Nov 17, 2020 • 0 new comments

• Add dict comprehension

  #47774 commented on Nov 17, 2020 • 0 new comments

• Dont use symbolic shapes check

  #47810 commented on Nov 20, 2020 • 0 new comments

• [TensorExpr] Cache use of fallback in kernel invocation

  #47812 commented on Nov 20, 2020 • 0 new comments

• [NNC] Compute Tensor Output Properties in ininitialization

  #47813 commented on Nov 20, 2020 • 0 new comments

• [TensorExpr Fuser] Add support for nodes which have tensor constant inputs

  #47814 commented on Nov 20, 2020 • 0 new comments

• [Test] make validate debug-only in Device ctr

  #47854 commented on Nov 20, 2020 • 0 new comments

• Expose CXX_FLAGS through __config__

  #47861 commented on Nov 18, 2020 • 0 new comments

• Add option for cpp_extensions to compile standalone executable

  #47862 commented on Nov 18, 2020 • 0 new comments

• Enable callgrind collection for C++ snippets

  #47865 commented on Nov 18, 2020 • 0 new comments

• [DO NOT MERGE] Will it work with patch pybind11?

  #47891 commented on Nov 16, 2020 • 0 new comments

• [WIP] Add test binary to compare model outputs

  #47933 commented on Nov 20, 2020 • 0 new comments

• Added Adabound.py

  #47957 commented on Nov 16, 2020 • 0 new comments

• Ignore MSVC's pdb file

  #47963 commented on Nov 16, 2020 • 0 new comments

• [JIT] Fix clang-tidy warnings for jit/api

  #47981 commented on Nov 17, 2020 • 0 new comments

• [JIT] Fix clang-tidy warnings in jit/python

  #47985 commented on Nov 21, 2020 • 0 new comments

• [JIT] Fix clang-tidy warnings in jit/serialization

  #47991 commented on Nov 21, 2020 • 0 new comments

• Compare Weights FX Implementation

  #47995 commented on Nov 16, 2020 • 0 new comments