pytorch / pytorch

1 Release published by 1 person

• v1.8.2 LTS 1.8.2, Wrap cub in its own namespace

  published Aug 17, 2021

5 Pull requests merged by 3 people

• [ONNX] Update instance_norm implementation and support training

  #60538 merged Aug 20, 2021

• [ONNX] Add log10 symbolic

  #63418 merged Aug 19, 2021

• [LTC] Fix some CI issues

  #63439 merged Aug 19, 2021

• [LTC] Add _log_softmax_backward_data TorchScript lowering

  #63306 merged Aug 18, 2021

• [LTC] Merge master into lazy_tensor_staging

  #63371 merged Aug 17, 2021

175 Pull requests opened by 95 people

• Add ONNX export support for argsort Fixes #33412

  #63283 opened Aug 14, 2021

• Change internal rep of methods from vector to map for faster lookup

  #63284 opened Aug 14, 2021

• Array API: Add torch.linalg.cross

  #63285 opened Aug 14, 2021

• Added updates to the tech docs to improve clarity

  #63286 opened Aug 15, 2021

• Test clang build on Windows

  #63288 opened Aug 15, 2021

• Enable Intel® VTune™ Profiler's Instrumentation and Tracing Technology APIs (ITT) to PyTorch

  #63289 opened Aug 15, 2021

• Improve `distributed.get_rank()` API docstring

  #63296 opened Aug 15, 2021

• Ported std/var to ReductionOpInfo

  #63300 opened Aug 16, 2021

• Add low level torch.profiler.profiler base class

  #63302 opened Aug 16, 2021

• add BFloat16 operators on CPU: cummax, cummin

  #63307 opened Aug 16, 2021

• Tensor docs

  #63308 opened Aug 16, 2021

• Port `gather` to structured kernel

  #63312 opened Aug 16, 2021

• Rename DataPipe to Op-er

  #63325 opened Aug 16, 2021

• Support detect out arg in mobile

  #63329 opened Aug 16, 2021

• Move variabletype functions around

  #63330 opened Aug 16, 2021

• Updated documentation for all reduction ops

  #63334 opened Aug 16, 2021

• "(object)" is extraneous in Python 3

  #63336 opened Aug 16, 2021

• Expand `ArrayRef` with `MutableArrayRef`

  #63338 opened Aug 16, 2021

• Enable test_api IMethodTest in OSS

  #63345 opened Aug 16, 2021

• Generating aliases

  #63349 opened Aug 16, 2021

• DRAFT: shareded_tensor.ones

  #63358 opened Aug 16, 2021

• [ignore] benchmarking only

  #63362 opened Aug 16, 2021

• plumb c10::nullopt through python binding layer when dtype isn't specified

  #63364 opened Aug 16, 2021

• Add support for slice, selec twith int, index_select

  #63365 opened Aug 16, 2021

• [Pytorch lite predictor] Use KinetoEdgeCPUProfiler for operator profiling.

  #63367 opened Aug 16, 2021

• [iOS][GPU] Consolidate array and non-array kernel for hardswish

  #63369 opened Aug 16, 2021

• [EASY] Update the clang-tidy error message

  #63370 opened Aug 16, 2021

• [do not merge] Test clang-tidy

  #63381 opened Aug 17, 2021

• Small fixes to the Contributing.txt

  #63385 opened Aug 17, 2021

• fix batchnorm2d issue when input is non contiguous

  #63392 opened Aug 17, 2021

• Promote integral inputs to floating for `torch.logsumexp`

  #63393 opened Aug 17, 2021

• fix typo in torch/onnx/utils.py

  #63396 opened Aug 17, 2021

• add a mechanism to use our load_tests with imported test cases

  #63399 opened Aug 17, 2021

• extract TestAutogradComplex into its own test file

  #63400 opened Aug 17, 2021

• Add permute, arange

  #63407 opened Aug 17, 2021

• fix git diff issue

  #63408 opened Aug 17, 2021

• extract TestAutogradComplex into its own test file

  #63409 opened Aug 17, 2021

• make target det list export

  #63412 opened Aug 17, 2021

• [jit] Make operation call accept Stack& instead of Stack*

  #63414 opened Aug 17, 2021

• Remove overloads for torch::jit::Stack*

  #63415 opened Aug 17, 2021

• [NOT FOR SUBMIT] Threaded SIMD runtime for CUDA

  #63417 opened Aug 17, 2021

• layernorm

  #63419 opened Aug 17, 2021

• layernom inplace

  #63420 opened Aug 17, 2021

• [Pytorch Edge][tracing-based] reorganize model tracer dependency

  #63421 opened Aug 17, 2021

• [DO NOT LAND] test rig to trigger *_broadcast1 failures

  #63423 opened Aug 17, 2021

• [WIP] Prototype checkpoint_sequential with hooks

  #63425 opened Aug 17, 2021

• Kill THCUNN

  #63429 opened Aug 17, 2021

• Allow implementing either backward or vjp for Function

  #63434 opened Aug 17, 2021

• [JIT][DRAFT] profile guided memory planning

  #63438 opened Aug 17, 2021

• .github: Enable with-ssh for Windows

  #63440 opened Aug 17, 2021

• [fx2trt] Add a test for quantized resnet18

  #63446 opened Aug 18, 2021

• [package] selective intern

  #63449 opened Aug 18, 2021

• less strict size check in scatter

  #63452 opened Aug 18, 2021

• Add autograd not implemented boxed fallback

  #63458 opened Aug 18, 2021

• Update codegen to use boxed kernel

  #63459 opened Aug 18, 2021

• [docs] Add "See Also" blocks for `torch.t` and `torch.transpose`

  #63461 opened Aug 18, 2021

• [BE] Enable functional optim tests for windows

  #63462 opened Aug 18, 2021

• [BE] Enable PostLocalSGD tests on windows

  #63463 opened Aug 18, 2021

• [RFC] [PyTorch Edge] Buffet of changes to speed up model loading for on-device use

  #63465 opened Aug 18, 2021

• continue debug

  #63467 opened Aug 18, 2021

• Upgrade to ubuntu:trusty-20190515

  #63468 opened Aug 18, 2021

• Fix torch.istft length mismatch and window runtime error

  #63469 opened Aug 18, 2021

• Add const keyword in jit/python/init.cpp

  #63473 opened Aug 18, 2021

• Create linalg.tensordot

  #63478 opened Aug 18, 2021

• [ROCm] hipFree() may not not synchronize non-blocking streams

  #63484 opened Aug 18, 2021

• To add Combined Scheduler to the list of PyTorch schedulers.

  #63491 opened Aug 18, 2021

• Fix bug in `check_empty_containers`

  #63492 opened Aug 18, 2021

• Add python mode

  #63496 opened Aug 18, 2021

• [quant][graphmode][fx] Make maxpool and flatten produce the reference pattern

  #63501 opened Aug 18, 2021

• Migrate legacy lstsq from THC to ATen (CUDA)

  #63504 opened Aug 18, 2021

• Migrate THCTensor_copyIgnoringOverlaps to ATen

  #63505 opened Aug 18, 2021

• [docs] Replace "See Also" sentences with proper blocks

  #63506 opened Aug 18, 2021

• Updated LTS version to 1.8.3

  #63507 opened Aug 18, 2021

• Sparse CSR: Add torch.empty

  #63509 opened Aug 18, 2021

• Sparse CSR: Add tensor.resize_ and tensor.copy_

  #63510 opened Aug 18, 2021

• Sparse CSR CUDA: add `torch.addmm` with all inputs sparse

  #63511 opened Aug 18, 2021

• [Static Runtime] Mark pure ops with AliasAnalysisKind::PURE_FUNCTION

  #63512 opened Aug 18, 2021

• Add view with negative dim

  #63516 opened Aug 18, 2021

• OpInfo: nn.functional.conv2d

  #63517 opened Aug 18, 2021

• [pruner] add support for pruning BatchNorm2d

  #63519 opened Aug 18, 2021

• [pruner] add getter for pruned outputs in base pruner

  #63520 opened Aug 18, 2021

• [WIP] Replace checkpointing implementation

  #63524 opened Aug 18, 2021

• [caffe2][easy] Save heap allocation in ConcatOp

  #63529 opened Aug 18, 2021

• Add Maxpool to shape analysis / Opinfo

  #63530 opened Aug 18, 2021

• add operation list for AutocastCPU

  #63534 opened Aug 18, 2021

• [JIT] linear scan memory planning strategy

  #63537 opened Aug 19, 2021

• [JIT] greedy by size memory planning strategy

  #63538 opened Aug 19, 2021

• [JIT] greedy by breadth memory planning strategy

  #63539 opened Aug 19, 2021

• [PyTorch Edge] Support default args with out arg, flag off

  #63540 opened Aug 19, 2021

• [nvFuser] [WIP] Normalization support

  #63542 opened Aug 19, 2021

• [draft] Profiler as a service tls

  #63543 opened Aug 19, 2021

• OpInfo for `nn.functional.cross_entropy`

  #63547 opened Aug 19, 2021

• fix: typo fixing

  #63548 opened Aug 19, 2021

• [wip] Use python to run linters

  #63550 opened Aug 19, 2021

• [Testing] check builder change

  #63551 opened Aug 19, 2021

• Enable jit trace in autocast

  #63552 opened Aug 19, 2021

• add BFloat16 operators on CPU: nextafter, atan2

  #63553 opened Aug 19, 2021

• deprecate dtype getters from `torch.testing` namespace

  #63554 opened Aug 19, 2021

• Use trsm for triangular_solve in CPU

  #63567 opened Aug 19, 2021

• Add linalg.solve_triangular

  #63568 opened Aug 19, 2021

• Modify LU_backward and lu_solve_backward to use linalg_solve_triangular

  #63569 opened Aug 19, 2021

• Deprecate torch.triangular_solve

  #63570 opened Aug 19, 2021

• enable equal_nan for complex values in isclose

  #63571 opened Aug 19, 2021

• remove componentwise comparison of complex values in TestCase.assertEqual

  #63572 opened Aug 19, 2021

• add python API to print all operators that have kernels registered to a particular DispatchKey

  #63575 opened Aug 19, 2021

• [JIT] Move UseVariadicCat internals

  #63577 opened Aug 19, 2021

• [JIT] Add variadic stack op

  #63578 opened Aug 19, 2021

• [Static Runtime] Implement prim::VarStack out variant

  #63579 opened Aug 19, 2021

• fix mpi ssh runtime error

  #63580 opened Aug 19, 2021

• [ao][docs] Add description of qconfig and qengine to quantization page

  #63582 opened Aug 19, 2021

• [WIP][jit] shape propagation for prepack

  #63585 opened Aug 19, 2021

• [TensorExpr] Make 'Tensor' a value type.

  #63586 opened Aug 19, 2021

• [TensorExpr] Nuke KernelArena and KernelScope.

  #63587 opened Aug 19, 2021

• [ONNX] Fix gather squeeze axis in constant folding

  #63588 opened Aug 19, 2021

• [ONNX] Export nn.Module call as ONNX local function

  #63589 opened Aug 19, 2021

• [WIP] fix export

  #63590 opened Aug 19, 2021

• Spectral norm without nograd test

  #63593 opened Aug 19, 2021

• [PyTorch Edge] Add unittest with out arg

  #63594 opened Aug 19, 2021

• make sure weights and biases are always TensorTypes

  #63595 opened Aug 19, 2021

• [ONNX] Fix cuda test case

  #63597 opened Aug 19, 2021

• [quant] Making isPlaceholderObserver work post migration

  #63599 opened Aug 19, 2021

• Sort out hpc work returns

  #63600 opened Aug 19, 2021

• [Draft] Experiment with no-init

  #63601 opened Aug 19, 2021

• Handle shared memory cases in MathBithFallback

  #63602 opened Aug 19, 2021

• [bazel] GPU-support: add @local_config_cuda and @cuda

  #63604 opened Aug 19, 2021

• [pytorch] add per_sample_weights support for embedding_bag_4bit_rowwise_offsets

  #63605 opened Aug 19, 2021

• [ONNX] Fix creation of constants with empty list as shape

  #63607 opened Aug 19, 2021

• Minor OptionalTensorRef updates

  #63611 opened Aug 19, 2021

• Factor out TensorBase that doesn't depend on native operators

  #63612 opened Aug 19, 2021

• [JIT] Freeze unrolls constant loops

  #63614 opened Aug 19, 2021

• Generalize expand logic

  #63615 opened Aug 19, 2021

• Made FuncTorchBatched decompose CompositeImplicitAutograd

  #63616 opened Aug 19, 2021

• ROCm MIOpen NHWC support

  #63617 opened Aug 19, 2021

• Add a record scope around autograd::engine::evaluate_function

  #63619 opened Aug 19, 2021

• add BFloat16 operators on CPU: logaddexp, logaddexp2, remainder

  #63621 opened Aug 20, 2021

• [quant][graphmode][fx] Add reference quantized linear module

  #63627 opened Aug 20, 2021

• Shape Propagation Pass: Fix AdaptiveAveragePooling2d

  #63629 opened Aug 20, 2021

• [static runtime] port c2 argmin kernel

  #63632 opened Aug 20, 2021

• OpInfo `nn.functional.embedding`

  #63633 opened Aug 20, 2021

• Optimize PReLU forward and backward (fp32) in CPU path

  #63634 opened Aug 20, 2021

• [Static Runtime] Remove unused fusion patterns

  #63636 opened Aug 20, 2021

• replace `self.assertTrue(torch.allclose(..))` with `self.assertEqual(…)`

  #63637 opened Aug 20, 2021

• [skip ci] lint 2

  #63638 opened Aug 20, 2021

• [LTC] Add nll_loss_backward TorchScript lowering

  #63639 opened Aug 20, 2021

• [skip ci][do not merge] Edit config.yml to filter specific jobs

  #63640 opened Aug 20, 2021

• [wip] Use more shards

  #63641 opened Aug 20, 2021

• [WIP] Ops in Python

  #63643 opened Aug 20, 2021

• Convert generator in Sampler back to lazy construction

  #63646 opened Aug 20, 2021

• [DataPipe] implementing fork()

  #63649 opened Aug 20, 2021

• [DataPipe] implementing demux()

  #63650 opened Aug 20, 2021

• [PyTorch Edge] Backport function for defaults args with out args, flag on

  #63651 opened Aug 20, 2021

• [debugging only] Why is there an extra TensorImpl created in dispatcher

  #63652 opened Aug 20, 2021

• Microbenchmarking matrix mult (einsum, torch.mult, torch.mm)

  #63654 opened Aug 20, 2021

• [pytorch][quant][oss] Support 2-bit embedding_bag op "embedding_bag_2bit_rowwise_offsets"

  #63658 opened Aug 20, 2021

• Parallelize loop nest with multiple buffer writes

  #63659 opened Aug 20, 2021

• Update CMake minimum version to 3.10

  #63660 opened Aug 20, 2021

• NOT_FOR_LAND

  #63661 opened Aug 20, 2021

• [TESTING] [skip ci] ghexport smoketest

  #63664 opened Aug 20, 2021

• Revert "Revert D30417370: [nnc] Enable CPU fusion"

  #63665 opened Aug 20, 2021

• [Static Runtime] Add an out variant op for aten::abs

  #63675 opened Aug 20, 2021

• [Pytorch Edge] Quantized Ops Dtype Selective

  #63680 opened Aug 20, 2021

• [PyTorch] Outline destructor of CppFunction

  #63688 opened Aug 20, 2021

• Add equality constraints for some acc opeartions for symbolic inference

  #63689 opened Aug 20, 2021

• [PyTorch] Add Library::fallback overload for CppFunction

  #63697 opened Aug 20, 2021

• [Static Runtime] SR clones graph input

  #63704 opened Aug 20, 2021

• Remove `_fork_processes` from common_distributed.py

  #63711 opened Aug 20, 2021

• [PyTorch] Don't inline KernelFunction constructor

  #63712 opened Aug 20, 2021

• Fix pocketfft include path in mobile build

  #63714 opened Aug 20, 2021

• [Static Runtime] Add pass to split fast_gather

  #63715 opened Aug 20, 2021

• [ONNX] Update submodule to 1.10.1

  #63716 opened Aug 20, 2021

• [ONNX] add test images to repo

  #63717 opened Aug 20, 2021

• [DO NOT MERGE] Enable Metal in PyTorch MacOS nightly builds

  #63718 opened Aug 21, 2021

• To add RMSProp algorithm documentation

  #63721 opened Aug 21, 2021

• Enable CPU fuser but with no TLP

  #63722 opened Aug 21, 2021

• Adds return type annotation for fork_rng function

  #63724 opened Aug 21, 2021

78 Issues closed by 30 people

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

  #24644 closed Aug 21, 2021

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

  #24645 closed Aug 21, 2021

• nn.LayerNorm breaks gradient flow under specific situations

  #63541 closed Aug 21, 2021

• Deprecation notice at torch.autograd.grad - only_inputs argument

  #63544 closed Aug 20, 2021

• ../torch/csrc/distributed/c10d/ProcessGroupMPI.cpp:131:10: error: ‘cerr’ is not a member of ‘std’

  #63653 closed Aug 20, 2021

• ONNX export failed on ATen operator mv

  #27723 closed Aug 20, 2021

• option to save to disk when calling save_for_backward()

  #28997 closed Aug 20, 2021

• Random CI error 'CUDA xxx installed failed'

  #63139 closed Aug 20, 2021

• Simple model training failure: loss does not descent.

  #63626 closed Aug 20, 2021

• pytorch '1.8.0+cu111' x = torch.stack(x, dim=0)

  #54134 closed Aug 20, 2021

• PyTorch Bazel scripts are incompatible with bazel-4.x

  #62569 closed Aug 20, 2021

• bug in exported onnx model for nn.InstanceNorm2d with track_running_stats=True

  #53887 closed Aug 20, 2021

• DataLoader should accept Sampler[Any]

  #63483 closed Aug 19, 2021

• IndexError: Caught IndexError in DataLoader worker process 0. Original Traceback (most recent call last):

  #63566 closed Aug 19, 2021

• loss not descent, while tensorflow do! Do I code in wrong way? Help!

  #63546 closed Aug 19, 2021

• RuntimeError: Exporting the operator square to ONNX opset version 9 is not supported

  #61766 closed Aug 19, 2021

• `torch.mul` documentation has format issue and refers `value` as an argument

  #39329 closed Aug 19, 2021

• SavedVariable default hooks

  #58659 closed Aug 19, 2021

• [FX] `is None` condition check does not error out in symbolic tracing, and can produce silently incorrect results.

  #50178 closed Aug 19, 2021

• Poisson function failed on rate vector contain zeros.

  #53485 closed Aug 19, 2021

• tensordot behavior change?

  #63573 closed Aug 19, 2021

• No-op full backward hooks modify the input

  #61446 closed Aug 19, 2021

• Add wheels for Python 3.9

  #45731 closed Aug 19, 2021

• support comparisons between types `c10::optional<T>` and `U` where `T` is comparable to `U`

  #62565 closed Aug 19, 2021

• tensor_split does not handle non-contiguous indices

  #63281 closed Aug 18, 2021

• Ambiguous description of torch.finfo.tiny

  #44010 closed Aug 18, 2021

• distributed/test_c10d_spawn segfaults after [Sharing CUDA tensors] note

  #51961 closed Aug 18, 2021

• QNNPACK convolution race condition

  #58055 closed Aug 18, 2021

• document why `torch.functional.meshgrid` exists

  #62844 closed Aug 18, 2021

• common_device_type.expectedFailure should report as expected failure

  #63363 closed Aug 18, 2021

• Prototype of conjugate flag

  #43563 closed Aug 18, 2021

• torch.utils.checkpoint should not only allow variables

  #63394 closed Aug 18, 2021

• find issues in Reproducibility doc

  #62531 closed Aug 18, 2021

• DataChunk should support `sort` method

  #63095 closed Aug 18, 2021

• tensorinv and other linear algebra operators should not do a universal catch all block

  #53739 closed Aug 18, 2021

• BFloat16 doesn't work with CUDA yet

  #23149 closed Aug 18, 2021

• [op backward] torch.linalg.pinv

  #63474 closed Aug 18, 2021

• Help

  #63472 closed Aug 18, 2021

• Specify device_ids in barrier() to force use of a particular device

  #62802 closed Aug 18, 2021

• dataloader multiprocessing start method might influence performance?

  #63311 closed Aug 18, 2021

• [complex] torch.polar accepts negative values for `abs` argument

  #52919 closed Aug 17, 2021

• torch.matmul doesn't work on certain tensors with zero-sized dimensions

  #63352 closed Aug 17, 2021

• [quant] Remove dtype arg requirement from `torch.empty_quantized`

  #63234 closed Aug 17, 2021

• Running PyTorch program on RISCV Architecture

  #63377 closed Aug 17, 2021

• assertion fails when exporting to onnx

  #16895 closed Aug 17, 2021

• RNN bidirectional parameter non-bool input behavior not coherent to v0.4

  #16933 closed Aug 17, 2021

• pytorch was blocked at optimizer.step

  #16934 closed Aug 17, 2021

• Using dataloader RecursionError: maximum recursion depth exceeded while calling a Python object

  #16992 closed Aug 17, 2021

• E: Could not get lock /var/lib/apt/lists/lock - open (11: Resource temporarily unavailable)

  #16981 closed Aug 17, 2021

• Good way to take inverse on large matrix

  #16963 closed Aug 17, 2021

• Unexpected set_num_threads behavior

  #16912 closed Aug 17, 2021

• Building from source fails on `Building CXX object c10/test/CMa...ptr_test.dir/util/intrusive_ptr_test.cpp.o`

  #17091 closed Aug 17, 2021

• Crash in cudnnConvolutionBackwardFilter "free(): invalid next size (normal)"

  #17060 closed Aug 17, 2021

• CI for CuDNN 7.5

  #17063 closed Aug 17, 2021

• [ONNX/Doc] Update the ONNX related docs

  #17098 closed Aug 17, 2021

• [bug report] [ONNX] `torch.Size` is not recognized as `Iterable` in onnx-nization

  #17153 closed Aug 17, 2021

• request to clarify `set_default_dtype` docs

  #60560 closed Aug 17, 2021

• observer._PartialWrapper extends the lifetime of Tensors, causing leaks.

  #63326 closed Aug 17, 2021

• __torch_function__ overload resolution between subclasses doesn't work like in the RFC

  #63397 closed Aug 17, 2021

• hipErrorNoBinaryForGpu: Coudn't find binary for current devices!

  #53738 closed Aug 17, 2021

• `torch.meshgrid` requires an `OpInfo` test

  #62719 closed Aug 17, 2021

• torch.fft.rfft bug

  #63152 closed Aug 17, 2021

• codecoverage reporting does not work in GHA

  #63366 closed Aug 17, 2021

• Nightly windows CUDA builds failing on CUDA install step

  #63344 closed Aug 17, 2021

• When compiled with clang ScatterGather kernel perf is significantly worse

  #63000 closed Aug 16, 2021

• `viable/strict` is more than a week old

  #63335 closed Aug 16, 2021

• How to build a release version libtorch1.8.1 on windows

  #63304 closed Aug 16, 2021

• bug in bi-lstm's input h0 and c0 when use bi-lstm train netword

  #63323 closed Aug 16, 2021

• `clang-tidy` job doesn't print errors

  #61563 closed Aug 16, 2021

• Windows Azure multi-gpu tests are failing in CircleCI

  #59724 closed Aug 16, 2021

• Linter annotation are no longer displayed for PRs

  #62975 closed Aug 16, 2021

• Does dataloader creates a copy of the dataset and consequently use up more memory?

  #63301 closed Aug 16, 2021

• Installation with pip/rocm fails

  #62538 closed Aug 16, 2021

• DataChunk needs to implement `__repr__` to visualize

  #63173 closed Aug 16, 2021

• `torch.lu` CUDA gives different results for batched and unbatched cases

  #56312 closed Aug 16, 2021

• [mypy] Incorrect imports in stub files

  #63279 closed Aug 16, 2021

• Function `torch.exp()` return float32 in case of amp float16 context

  #63292 closed Aug 15, 2021

• RuntimeError

  #63294 closed Aug 15, 2021

75 Issues opened by 66 people

• a problem happened in torch.randperm

  #63726 opened Aug 21, 2021

• [Feature Request] add graph to hparams for tensorboard

  #63725 opened Aug 21, 2021

• Missing type annotation for `fork_rng`

  #63723 opened Aug 21, 2021

• Torch.save with zip serialization causes memory bloat

  #63720 opened Aug 21, 2021

• FP16 with forward only and FP32 in backward

  #63719 opened Aug 21, 2021

• [Meta][WIP][Do-Not-Remove] HUD CI SEV Issues

  #63695 opened Aug 20, 2021

• SyncBatchNorm should behave like BatchNorm when running on single GPU.

  #63662 opened Aug 20, 2021

• DISABLED test_ind_worker_queue (__main__.TestIndividualWorkerQueue)

  #63657 opened Aug 20, 2021

• [JIT] Typing on math.ceil is inaccurate for Scalar input

  #63648 opened Aug 20, 2021

• Create a new OptionalArrayRef template class to replace Optional<ArrayRef>

  #63645 opened Aug 20, 2021

• LeakyReLU and Elu use more VRAM than needed

  #63644 opened Aug 20, 2021

• Raw saved tensors can survive the deletion of the underlying SavedVariable object

  #63624 opened Aug 20, 2021

• Expose a SavedTensorsHooks nn.Module for users to register saved tensors hooks

  #63623 opened Aug 20, 2021

• Error in git clone pytorch wiki

  #63622 opened Aug 20, 2021

• C++ torch::cuda::synchronize speeds up training

  #63618 opened Aug 19, 2021

• GHA CI failure intermittently with "local git repository checkout dirty"

  #63613 opened Aug 19, 2021

• Enable some amount of CI tests on lazy_tensor_staging branch

  #63610 opened Aug 19, 2021

• Sampler should be seeded lazily

  #63609 opened Aug 19, 2021

• torch.trapezoid broadcasting behavior is incorrect

  #63608 opened Aug 19, 2021

• with-ssh doesn't work when job is triggered by probot

  #63603 opened Aug 19, 2021

• Using FFT on Apple/M1

  #63592 opened Aug 19, 2021

• Split up test/test_nn.py

  #63591 opened Aug 19, 2021

• Cannot specify character encoding in cpp_extension.py for C++ compiler info

  #63584 opened Aug 19, 2021

• Add (at least a daily) DEBUG=1 tests on a common platform

  #63583 opened Aug 19, 2021

• Java NVIDIA GPU Inference Support

  #63581 opened Aug 19, 2021

• compile_commands.json doesn't contain nvcc invocations

  #63574 opened Aug 19, 2021

• Replace `self.assertTrue(torch.allclose ...` with `self.assertEqual ...` in the test suite

  #63565 opened Aug 19, 2021

• Why is the sequence dimension before batch size dimension for " nn.MultiheadAttention"

  #63560 opened Aug 19, 2021

• Functional grid_sample: example of padding_mode="reflection" description error

  #63559 opened Aug 19, 2021

• universal binaries for libtorch on mac (x86_64+arm)

  #63558 opened Aug 19, 2021

• [RFC] A Launch script with Best Recipe of Deep Learning on Intel Xeon CPU

  #63556 opened Aug 19, 2021

• pytorch 1.8.1+cu111 used much more CPU RAM than pytorch 1.8.1 after run `import torch`

  #63555 opened Aug 19, 2021

• sparse _values returns a new TensorImpl in some situations

  #63549 opened Aug 19, 2021

• [nn] conv2d : different dtype support when dilation argument is passed

  #63518 opened Aug 18, 2021

• Extend @ops decorator with kwarg to apply dynamic decorators

  #63494 opened Aug 18, 2021

• [FX] Data type of target should be validated on Node construction and lint

  #63490 opened Aug 18, 2021

• MultiStepLR with different gammas for each parameter group

  #63488 opened Aug 18, 2021

• Distributed / MPI / CUDA: Incorrect messages received from isend/irecv in PyTorch 1.9

  #63486 opened Aug 18, 2021

• Preserve tensor subclasses when unpacking a SavedTensor

  #63485 opened Aug 18, 2021

• Half division on low precision scalar doesn't work on CPU

  #63482 opened Aug 18, 2021

• (ROC) ImportError: libtinfo.so.5: cannot open shared object file: No such file or directory

  #63480 opened Aug 18, 2021

• Cannot include extension.h under Windows - linker error THPVariable_Wrap

  #63479 opened Aug 18, 2021

• Supporting Pytorch for Custom Compiler Backend

  #63477 opened Aug 18, 2021

• Error loading Pytorch lite model in Android

  #63476 opened Aug 18, 2021

• C++ Inference error causes Identity Module

  #63475 opened Aug 18, 2021

• Multi-gpu training got an unbanlanced memory and redundant processes!!!

  #63470 opened Aug 18, 2021

• Add a GitHub actions workflow for Macos

  #63466 opened Aug 18, 2021

• Perf regression for 1D cuda sort

  #63456 opened Aug 18, 2021

• distributed_error

  #63455 opened Aug 18, 2021

• getNcclVersion() in NCCLUtils.cpp handles "2.10.3" incorrectly

  #63445 opened Aug 18, 2021

• test_variant_consistency intermittently fails on Windows

  #63444 opened Aug 17, 2021

• OpenCV causes backpropagation to get stuck

  #63441 opened Aug 17, 2021

• Norm functions don't call __torch_function__ for weights or bias

  #63435 opened Aug 17, 2021

• scatter's backward formula should call scatter instead of scatter_

  #63430 opened Aug 17, 2021

• Weird behavior of cuda::cub::sort_pairs

  #63427 opened Aug 17, 2021

• torch.arange has issue tracing with bool input

  #63405 opened Aug 17, 2021

• Consolidate distributed benchmark folders

  #63403 opened Aug 17, 2021

• How to efficiently (without looping) get data from tensor predicted by a torchscript in C++?

  #63395 opened Aug 17, 2021

• [feature request] Including NADAM optimizer in torch.optim

  #63391 opened Aug 17, 2021

• [python codegen] correctly plumb TensorOptions defaults through the python binding layer.

  #63356 opened Aug 16, 2021

• Nightly Docker image builds are failing with conda install issues

  #63342 opened Aug 16, 2021

• Extend OpInfo tests to automatically validate noncontiguous inputs

  #63341 opened Aug 16, 2021

• Property based testing like AFL

  #63337 opened Aug 16, 2021

• Using 'multiprocessing’ in PyTorch on Windows10 got error-``RuntimeError: Couldn’t open shared file mapping: <torch_13684_4004974554>, error code: <0>''

  #63331 opened Aug 16, 2021

• lstm's input h0 and c0 bug

  #63322 opened Aug 16, 2021

• Incorrect computation of gradients for transposed 2d convolution when stride=2 on CPU

  #63314 opened Aug 16, 2021

• Automatic mixed precision works worse for 3D neural networks.

  #63313 opened Aug 16, 2021

• Alternate Model Loading - Android

  #63305 opened Aug 16, 2021

• Channels last mode doesn't work with inputs with 0 batch dimension

  #63299 opened Aug 16, 2021

• Named tensor in tracer

  #63297 opened Aug 15, 2021

• pytorch 1.8.2 cuda test errors

  #63295 opened Aug 15, 2021

• Support `torch.linalg.outer`

  #63293 opened Aug 15, 2021

• [typing] new_ones has wrong signature

  #63291 opened Aug 15, 2021

• DistributedDataParallel documentation

  #63290 opened Aug 15, 2021

• Serialising `torch.bool` generates a warning about `np.bool` being deprecated.

  #63282 opened Aug 14, 2021

344 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.

• Add a Functionalization pass

  #63048 commented on Aug 20, 2021 • 72 new comments

• Add torch.pad function

  #61635 commented on Aug 20, 2021 • 49 new comments

• Implement Python Array API `asarray` function.

  #60627 commented on Aug 19, 2021 • 40 new comments

• [CUDA graphs] [WIP] Prototype API and documentation

  #63269 commented on Aug 21, 2021 • 38 new comments

• Added API tests to ReductionOpInfo and ported more tests

  #62899 commented on Aug 19, 2021 • 29 new comments

• add support for sending cpu sparse tensors over rpc

  #62794 commented on Aug 20, 2021 • 26 new comments

• Added reference tests to ReductionOpInfo and ported mean test

  #62900 commented on Aug 20, 2021 • 17 new comments

• Adds DLPack support

  #57110 commented on Aug 17, 2021 • 15 new comments

• Enhancement for smart serialization for out schemas

  #63096 commented on Aug 21, 2021 • 15 new comments

• [fix] fix scalar behavior for `torch.where`

  #62084 commented on Aug 20, 2021 • 14 new comments

• [JIT] memory planning base (no strategies)

  #62174 commented on Aug 21, 2021 • 14 new comments

• [doc][BE] prefix CI job with trunk-only

  #63187 commented on Aug 18, 2021 • 14 new comments

• OpInfo for `nn.functional.layer_norm`

  #63276 commented on Aug 20, 2021 • 14 new comments

• [quant][graphmode][fx] Add reference option support for binary ops

  #62698 commented on Aug 20, 2021 • 13 new comments

• [OpInfo] Added ReductionOpInfo subclass of OpInfo and ported sum test

  #62737 commented on Aug 19, 2021 • 13 new comments

• [qunat][graphmode][fx] Add a separate lower_to_native_backend function for relu

  #62861 commented on Aug 20, 2021 • 13 new comments

• [TensorExpr] Switch Exprs and Stmt from kernel-arena to shared_ptr.

  #63216 commented on Aug 20, 2021 • 13 new comments

• torch.scatter_: support index.size(d) > src.size(d)

  #63265 commented on Aug 21, 2021 • 12 new comments

• [structured] Preserve computed elements from meta func to impl

  #61746 commented on Aug 20, 2021 • 12 new comments

• Support `torch.concat` alias, add `cat` OpInfo & remove OpInfo test_out skips {cat, stack, hstack, vtack, dstack}

  #62560 commented on Aug 20, 2021 • 12 new comments

• Adding BFP16 quantization/dequantization support to OSS

  #63059 commented on Aug 20, 2021 • 12 new comments

• [ONNX] Deprecated the example_outputs param from torch.onnx.export() function.

  #62815 commented on Aug 20, 2021 • 11 new comments

• delegate to the proper comparison operator

  #62889 commented on Aug 21, 2021 • 11 new comments

• int8 linear is slower than fp32 linear

  #62804 commented on Aug 19, 2021 • 10 new comments

• Add relative and absolute tolerances for matrix_rank, pinv

  #63102 commented on Aug 16, 2021 • 10 new comments

• Update optimize_for_mobile to preserve node's debug information

  #63106 commented on Aug 20, 2021 • 9 new comments

• Rewrite PyTorch CUDA backend in Triton

  #62661 commented on Aug 19, 2021 • 8 new comments

• [ONNX] Fix input sequence for pad op

  #60554 commented on Aug 20, 2021 • 8 new comments

• add support for indexing to meshgrid

  #62722 commented on Aug 21, 2021 • 8 new comments

• Fix jit can not recognize the imported function

  #63248 commented on Aug 19, 2021 • 8 new comments

• [feature request] Core API for invertible/inplace and flow-like ops

  #23756 commented on Aug 18, 2021 • 7 new comments

• Add support of variadic length/type argument in `result_type`

  #61168 commented on Aug 17, 2021 • 7 new comments

• Some way to specify expected failures for OpInfo-based tests

  #62354 commented on Aug 18, 2021 • 6 new comments

• ROCm build documentation lacks needed dependencies

  #63188 commented on Aug 18, 2021 • 6 new comments

• [bug] torch.distributed.elastic logging: Failed to print the first info statement

  #63223 commented on Aug 17, 2021 • 5 new comments

• Add Tensor.nbytes() / `Tensor.itemsize()`

  #12728 commented on Aug 18, 2021 • 5 new comments

• BatchNorm2d permutes data on 1.10 nightly

  #63224 commented on Aug 19, 2021 • 5 new comments

• LazyTensor cuda fallback

  #62616 commented on Aug 19, 2021 • 5 new comments

• Create linalg.matrix_exp

  #62715 commented on Aug 20, 2021 • 5 new comments

• OperatorEntry: Move assertSignatureIsCorrect out of line

  #63127 commented on Aug 19, 2021 • 5 new comments

• [ONNX] Add warning for inplace updates on tensor.shape in tracing mode

  #63170 commented on Aug 20, 2021 • 5 new comments

• Check clang++/g++ version when compiling CUDA extensions

  #63230 commented on Aug 20, 2021 • 5 new comments

• document that `torch.triangular_solve` has optional out= parameter

  #63253 commented on Aug 21, 2021 • 5 new comments

• ROCm build issues - 'use of undeclared identifier cudaStreamCaptureModeRelaxed'

  #63280 commented on Aug 16, 2021 • 4 new comments

• [local linter] mypy_wrapper doesn't filter out distributed file

  #57967 commented on Aug 17, 2021 • 4 new comments

• multi_head_attention_forward 3D attention mask incorrect behavior

  #63271 commented on Aug 20, 2021 • 4 new comments

• [NOT_FOR_COMMIT][android] Pytorch android build with USE_LAPACK and QML lib

  #35200 commented on Aug 18, 2021 • 4 new comments

• [ONNX] Remove use_external_data_format param from torch.onnx.export() function.

  #62257 commented on Aug 20, 2021 • 4 new comments

• Do not modify saved variables in-place for spectral norm during power iteration

  #62293 commented on Aug 19, 2021 • 4 new comments

• correlate forward and backward op

  #62553 commented on Aug 19, 2021 • 4 new comments

• Refactor RecordFunction to use CustomClass

  #62710 commented on Aug 20, 2021 • 4 new comments

• support using gradients named for outputs in derivatives

  #63040 commented on Aug 20, 2021 • 4 new comments

• [LTC] Fix linter issues for lazy_tensor_core subdirectory

  #63128 commented on Aug 20, 2021 • 4 new comments

• [pruner] amend base pruner API to match base sparsifier

  #63178 commented on Aug 21, 2021 • 4 new comments

• state_dict and load_state_dict methods for DataLoader and Sampler to continue training at specific epoch and batch

  #36650 commented on Aug 16, 2021 • 3 new comments

• PyTorch 1.9.0, test_optim fails on Nvidia A10.

  #63079 commented on Aug 17, 2021 • 3 new comments

• ONNX export fails for trivial torch.autograd.Function

  #61813 commented on Aug 17, 2021 • 3 new comments

• ONNX output mismatch with PyTorch. InstanceNormalization layer not using running_mean and running_var

  #61430 commented on Aug 19, 2021 • 3 new comments

• linalg matrix rank generates exception when calculating rank of a singular matrix

  #62166 commented on Aug 19, 2021 • 3 new comments

• Sparse CSR layout GPU backend tracking issue

  #60854 commented on Aug 20, 2021 • 3 new comments

• [feature request] conjugate transpose operator

  #45063 commented on Aug 20, 2021 • 3 new comments

• [special] Alias igamma, igammac to special.gammaninc, special.gammaincc

  #61902 commented on Aug 20, 2021 • 3 new comments

• Prototype checkpointing without re-entrant autograd

  #62964 commented on Aug 20, 2021 • 3 new comments

• LayerNorm

  #62980 commented on Aug 17, 2021 • 3 new comments

• Separating quantization test from distributed_test

  #63058 commented on Aug 21, 2021 • 3 new comments

• Add a common autograd TLS state

  #63114 commented on Aug 20, 2021 • 3 new comments

• clean up engine.cpp thread state

  #63115 commented on Aug 21, 2021 • 3 new comments

• remove special grad_mode tls handling

  #63116 commented on Aug 21, 2021 • 3 new comments

• New TLS to disable forward mode AD

  #63117 commented on Aug 21, 2021 • 3 new comments

• Allow shared functions to be inline to reduce binary size

  #63118 commented on Aug 16, 2021 • 3 new comments

• [PyTorch Edge][type] Add type check in compatibility api

  #63129 commented on Aug 17, 2021 • 3 new comments

• [fix] rename unique_dim to _unique_dim

  #63250 commented on Aug 16, 2021 • 3 new comments

• [docs][ao] add missing torch.choose_qparams_optimized documentation

  #63263 commented on Aug 16, 2021 • 3 new comments

• [ao][docs] Add description of qconfig and qengine to quantization page

  #63266 commented on Aug 19, 2021 • 3 new comments

• [dist] log node's hostname plus rank in the exception message

  #63174 commented on Aug 16, 2021 • 2 new comments

• `torch.disributed.GradBucket.is_last()` returns is first instead of is last

  #63210 commented on Aug 16, 2021 • 2 new comments

• cuda.is_available returns True in pycharms python console, False in code

  #63226 commented on Aug 16, 2021 • 2 new comments

• Create a boxed fallback / template recipe for Autograd that forwards, but errors on backwards

  #62032 commented on Aug 16, 2021 • 2 new comments

• DataParallel doesn't properly handle kwargs

  #31460 commented on Aug 17, 2021 • 2 new comments

• Seeing unit test failures for ProcessGroupShareTensorTest

  #62780 commented on Aug 17, 2021 • 2 new comments

• Split up test_nn.py

  #63085 commented on Aug 17, 2021 • 2 new comments

• mpiexec fails in GitHub Actions

  #60756 commented on Aug 17, 2021 • 2 new comments

• Notify dev infra team on PR failures that aren't the author's fault

  #56745 commented on Aug 17, 2021 • 2 new comments

• LazyTensor support for fuser integration investigation

  #62619 commented on Aug 18, 2021 • 2 new comments

• PyTorch(with mkldnn) in inference mode using non-optimal prop_kind for convolution

  #62788 commented on Aug 18, 2021 • 2 new comments

• Clarify sparse COO tensor coalesce behavior wrt overflow + how to binarize a sparse tensor

  #62162 commented on Aug 18, 2021 • 2 new comments

• RuntimeError - onnx export a model with transformer layer

  #61277 commented on Aug 19, 2021 • 2 new comments

• Exporting a quantized model to onnx with opset > 9 fails: nchw2nhwc is not supported

  #61512 commented on Aug 19, 2021 • 2 new comments

• [feature request] Implement CI testing of installation commands

  #62590 commented on Aug 19, 2021 • 2 new comments

• Support `torch.linalg.trace`

  #62255 commented on Aug 19, 2021 • 2 new comments

• [ONNX] Invalid ONNX graph when exporting prim::Constant of empty tensor data

  #61754 commented on Aug 19, 2021 • 2 new comments

• https://github.com/pytorch/pytorch/blob/master/CONTRIBUTING.md#pre-commit-tidylinting-hook seems out-of-date

  #62629 commented on Aug 20, 2021 • 2 new comments

• [feature request] BitTensors and bit packed tensors utilities

  #32867 commented on Aug 20, 2021 • 2 new comments

• Tracing models all the way down to basic functions

  #63076 commented on Aug 20, 2021 • 2 new comments

• RESOLVED: Retire component-wise closeness tests for complex tensors?

  #61906 commented on Aug 20, 2021 • 2 new comments

• Support `torch.linalg.diagonal`

  #62813 commented on Aug 20, 2021 • 2 new comments

• ReLU uses more VRAM for backward-pass than needed

  #63027 commented on Aug 20, 2021 • 2 new comments

• onnx export failed when output size are not factor of input size for adaptive_avg_pool2d

  #42653 commented on Aug 20, 2021 • 2 new comments

• Can we make torch.inverse FP16? - RuntimeError: "inverse_cuda" not implemented for 'Half'

  #40427 commented on Aug 20, 2021 • 2 new comments

• Ability to export pixel_unshuffle to ONNX

  #53539 commented on Aug 17, 2021 • 2 new comments

• Fix torch.onnx.export constant folding=False in doc

  #55451 commented on Aug 18, 2021 • 2 new comments

• add bf16 support for bucketize

  #55588 commented on Aug 20, 2021 • 2 new comments

• Add fast path for addmm when the inputs are conjugate

  #59380 commented on Aug 19, 2021 • 2 new comments

• NNAPI: Support const values in binary ops

  #59434 commented on Aug 20, 2021 • 2 new comments

• Enable BFloat16 LeakyReLU and RReLU in CPU path

  #61514 commented on Aug 20, 2021 • 2 new comments

• enable bf16 mkldnn path for gemm

  #61891 commented on Aug 20, 2021 • 2 new comments

• add BFloat16 operators on CPU: searchsorted, diag, fmod, cumsum, cumprod

  #61897 commented on Aug 20, 2021 • 2 new comments

• Introduce TestLogger

  #61910 commented on Aug 16, 2021 • 2 new comments

• [NNC] Add C++ codegen backend to NNC

  #62869 commented on Aug 19, 2021 • 2 new comments

• Automated submodule update: FBGEMM

  #62879 commented on Aug 17, 2021 • 2 new comments

• Improve performance of index_select by avoiding item

  #63008 commented on Aug 18, 2021 • 2 new comments

• test_neg_view: preseve sign of sample input

  #63010 commented on Aug 19, 2021 • 2 new comments

• [JIT] Add aten::slice optimization

  #63049 commented on Aug 17, 2021 • 2 new comments

• Add batch of unary functions

  #63050 commented on Aug 21, 2021 • 2 new comments

• [PyTorchEdge] promote prim ops v2

  #63093 commented on Aug 19, 2021 • 2 new comments

• ENH Adds label_smoothing to targets as class indices in cross_entropy

  #63122 commented on Aug 19, 2021 • 2 new comments

• Add BFloat16 support for logsigmoid, hardsigmoid, hardshrink, softshrink, hardswish, and softplus on CPU

  #63134 commented on Aug 19, 2021 • 2 new comments

• Adding alltoall_single collective to collective quantization API

  #63154 commented on Aug 19, 2021 • 2 new comments

• [DOC] Add doc for maybe_wrap_dim

  #63161 commented on Aug 16, 2021 • 2 new comments

• Cosine annealing warm restarts add gamma

  #63200 commented on Aug 16, 2021 • 2 new comments

• enable oneMKL&oneDNN on-demands verbose functinality

  #63212 commented on Aug 16, 2021 • 2 new comments

• Add BFloat16 support for multinomial and poisson on CPU

  #63215 commented on Aug 19, 2021 • 2 new comments

• [ONNX] Prepare for function extraction

  #63268 commented on Aug 21, 2021 • 2 new comments

• Ability to export `pixel_unshuffle` to ONNX

  #51181 commented on Aug 14, 2021 • 1 new comment

• Complex Numbers Support

  #33152 commented on Aug 15, 2021 • 1 new comment

• KeyError: 'max_unpool2d' pytorch to onnx

  #25088 commented on Aug 16, 2021 • 1 new comment

• [documentation] torch.distributed.elastic: illustrate how to write load_checkpoint and save_checkpoint in Train Script

  #63140 commented on Aug 16, 2021 • 1 new comment

• ARM Mac 16-core Neural Engine

  #47688 commented on Aug 16, 2021 • 1 new comment

• yolo5 convert to onnx failed.

  #62090 commented on Aug 16, 2021 • 1 new comment

• [feature request] Type promotions for Boolean tensors with sub operation + Numpy compatability

  #63023 commented on Aug 16, 2021 • 1 new comment

• [TorchScript] Support for registering builtin operators with control flow

  #62775 commented on Aug 16, 2021 • 1 new comment

• Tests in CI are run from the test/ directory

  #56396 commented on Aug 16, 2021 • 1 new comment

• Create Github Action to create / register dummy node for self hosted runners

  #59303 commented on Aug 16, 2021 • 1 new comment

• Torch.nn.functional.interpolate() function memory release problem

  #63217 commented on Aug 16, 2021 • 1 new comment

• Replace Xenial with Bionic

  #61460 commented on Aug 16, 2021 • 1 new comment

• Some test reports are missing from the artifact

  #61151 commented on Aug 16, 2021 • 1 new comment

• IN_PULL_REQUEST env var might not be set correctly on CircleCI

  #60746 commented on Aug 16, 2021 • 1 new comment

• `periodic-pytorch-linux-xenial-cuda11.3-cudnn8-py3.6-gcc7-build` started to fail

  #62265 commented on Aug 16, 2021 • 1 new comment

• replace platform specific CI environment variables with generic ones

  #59478 commented on Aug 16, 2021 • 1 new comment

• [DDP/Reducer] Log reducer errors in standalone methods

  #59338 commented on Aug 17, 2021 • 1 new comment

• ONNX fails to export Fold

  #41423 commented on Aug 17, 2021 • 1 new comment

• Make axes selection keyword arguments in torch.diagonal and torch.transpose consistent

  #62955 commented on Aug 17, 2021 • 1 new comment

• Bring Python torch.slice bindings back or add step argument to torch.narrow

  #41625 commented on Aug 17, 2021 • 1 new comment

• Background threads not exiting after calling loss.backward() on Win7

  #48888 commented on Aug 17, 2021 • 1 new comment

• [jit]][script] handling of 'void' returns

  #62325 commented on Aug 17, 2021 • 1 new comment

• Forward/backward hooks for C++ torch::nn modules

  #25888 commented on Aug 17, 2021 • 1 new comment

• Add a new schema type for representing dimension lists

  #44409 commented on Aug 17, 2021 • 1 new comment

• `nn.Linear` allows 1d input tensors

  #16896 commented on Aug 17, 2021 • 1 new comment

• TestAdadelta.test_adadelta flaky on CI

  #12851 commented on Aug 18, 2021 • 1 new comment

• Using external memory allocator with PyTorch

  #43144 commented on Aug 18, 2021 • 1 new comment

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

  #22462 commented on Aug 18, 2021 • 1 new comment

• Synchronization problem in torch.distributed with MPI on CUDA

  #38642 commented on Aug 18, 2021 • 1 new comment

• ctx.save_for_backward doesn't save torch.Tensor subclasses fully

  #47117 commented on Aug 18, 2021 • 1 new comment

• Documentation for torch.finfo doesn't match implementation

  #59526 commented on Aug 18, 2021 • 1 new comment

• when call `torch.onnx.export()`, the graph is pruned by default ? how to cancel pruning

  #62951 commented on Aug 18, 2021 • 1 new comment

• RuntimeError: 0 INTERNAL ASSERT FAILED at "../torch/csrc/jit/ir/ir.cpp":518, please report a bug to PyTorch. 18118 not in scope

  #61543 commented on Aug 18, 2021 • 1 new comment

• [onnx] export: Exporting the operator cross to ONNX opset version 12 is not supported.

  #61445 commented on Aug 18, 2021 • 1 new comment

• ONNX export models with Map (and Sequence) input types

  #61517 commented on Aug 18, 2021 • 1 new comment

• copy.copy not working for ScriptModule

  #18106 commented on Aug 18, 2021 • 1 new comment

• ONNX and grid_sample layer

  #27212 commented on Aug 19, 2021 • 1 new comment

• test_python_dispatch fails under pytest

  #62501 commented on Aug 19, 2021 • 1 new comment

• Quantizable LSTMCell does not work correctly.

  #55945 commented on Aug 19, 2021 • 1 new comment

• Request for conjugate transposed triangular solve

  #45734 commented on Aug 19, 2021 • 1 new comment

• Document CI label and what they do

  #62585 commented on Aug 19, 2021 • 1 new comment

• [RFC] Should DDP support custom reduction logic for registered module buffers?

  #63041 commented on Aug 19, 2021 • 1 new comment

• Helping test example code blocks in the docs

  #41499 commented on Aug 19, 2021 • 1 new comment

• RuntimeError: Unsupported: ONNX export of operator adaptive_max_pool1d

  #61172 commented on Aug 20, 2021 • 1 new comment

• "Unsupported: ONNX export of transpose for tensor of unknown rank." with transpose option when mix trace-based and script-based exporter

  #56663 commented on Aug 20, 2021 • 1 new comment

• Python/C++ API Parity: torch.nn modules and functional

  #25883 commented on Aug 20, 2021 • 1 new comment

• ONNX export doubles RAM usage

  #61263 commented on Aug 20, 2021 • 1 new comment

• NN Module functional API

  #49171 commented on Aug 20, 2021 • 1 new comment

• Unexpected Behavior when Pointwise Operations Write to Expanded Tensors

  #10756 commented on Aug 21, 2021 • 1 new comment

• Make deprecation warnings use stacklevel

  #60548 commented on Aug 21, 2021 • 1 new comment

• [feature request] Expand a given dim

  #62352 commented on Aug 21, 2021 • 1 new comment

• [cuda] faster kernelTransformReduceInnermostDimIndex using cub

  #21295 commented on Aug 18, 2021 • 1 new comment

• [DO NOT DELETE OR MERGE] Nightly release

  #26921 commented on Aug 21, 2021 • 1 new comment

• [DO NOT DELETE OR MERGE] Postnightly release

  #27167 commented on Aug 21, 2021 • 1 new comment

• [JIT] Fix type annotations of pooling modules

  #47888 commented on Aug 16, 2021 • 1 new comment

• add channels last for GroupNorm

  #49821 commented on Aug 18, 2021 • 1 new comment

• add channel last support for MaxUnpool2d

  #49984 commented on Aug 20, 2021 • 1 new comment

• Enable Union typing

  #53180 commented on Aug 19, 2021 • 1 new comment

• optimize BFloat16 elemwise operators CPU: sigmoid, sigmoid_backward, tanh_backward, addcmul, addcdiv

  #55221 commented on Aug 18, 2021 • 1 new comment

• Support `expand_dims`

  #57116 commented on Aug 17, 2021 • 1 new comment

• [DO NOT MERGE] [BENCHMARK] Get scores for `optimize_for_inference`

  #60066 commented on Aug 15, 2021 • 1 new comment

• [ONNX] Support tensorList when Infers shape and type uninitialized output

  #60534 commented on Aug 20, 2021 • 1 new comment

• move kernel launch checks from `torch.testing` to `torch.testing._internal.check_kernel_launches`

  #60862 commented on Aug 19, 2021 • 1 new comment

• fix resize bug

  #61166 commented on Aug 18, 2021 • 1 new comment

• Enable `bfloat16` in NCCL

  #61799 commented on Aug 18, 2021 • 1 new comment

• add BFloat16 support for GLU, randperm, bernoulli, and nan_to_num operators on CPU

  #61944 commented on Aug 19, 2021 • 1 new comment

• Clean up related to type refinements

  #62444 commented on Aug 20, 2021 • 1 new comment

• Simplify ccache instructions in CONTRIBUTING.md

  #62549 commented on Aug 21, 2021 • 1 new comment

• ENH Adds no_batch_dim for NLLLoss

  #62651 commented on Aug 21, 2021 • 1 new comment

• [WIP] Support `torch.linalg.trace`

  #62714 commented on Aug 17, 2021 • 1 new comment

• [OpInfo] Make SkipInfo assert failure by default

  #62725 commented on Aug 21, 2021 • 1 new comment

• Detect out argument in the schema

  #62755 commented on Aug 21, 2021 • 1 new comment

• add BFloat16 support for logcumsumexp on CPU

  #62803 commented on Aug 18, 2021 • 1 new comment

• Update CONTRIBUTING.md

  #62832 commented on Aug 16, 2021 • 1 new comment

• Catch exception when registering a fallback

  #63003 commented on Aug 21, 2021 • 1 new comment

• Make torch.floor_divide error on negative results

  #63011 commented on Aug 19, 2021 • 1 new comment

• Profiler as a service

  #63021 commented on Aug 18, 2021 • 1 new comment

• Derivatives of relu, leaky_relu (#63027)

  #63089 commented on Aug 19, 2021 • 1 new comment

• move some codegen utilities into utils.py

  #63094 commented on Aug 16, 2021 • 1 new comment

• Add squeeze, unsqueeze, transpose shape functins

  #63099 commented on Aug 21, 2021 • 1 new comment

• Introduce the torch.distributed.ops module

  #63100 commented on Aug 17, 2021 • 1 new comment

• [BE][exp] experiment distributed run_test options

  #63147 commented on Aug 19, 2021 • 1 new comment

• preserve memory format in `cudnn_convolution_add_relu_out`

  #63151 commented on Aug 16, 2021 • 1 new comment

• Add crow_/col_indices to view types

  #63176 commented on Aug 19, 2021 • 1 new comment

• [pruner] modify base pruner to prune bias by default

  #63202 commented on Aug 21, 2021 • 1 new comment

• OpInfo for `nn.functional.batch_norm`

  #63218 commented on Aug 20, 2021 • 1 new comment

• Array API: Add `torch.linalg.matmul` alias to `torch.matmul`

  #63227 commented on Aug 20, 2021 • 1 new comment

• Enhance shape (#60335)

  #63249 commented on Aug 20, 2021 • 1 new comment

• [LTC] Fix some typos in QUICKSTART.md

  #63261 commented on Aug 18, 2021 • 1 new comment

• Windows Builds on GHA have a lot of sccache misses

  #63036 commented on Aug 15, 2021 • 0 new comments

• document our compatibility with upstream packages (e.g. languages, compilers, tools)

  #63207 commented on Aug 15, 2021 • 0 new comments

• Memory leak in `ddp_zero_hook.py` hooks

  #63211 commented on Aug 15, 2021 • 0 new comments

• Check global rank and process group usage in `ZeroRedundancyOptimizer`

  #63213 commented on Aug 15, 2021 • 0 new comments

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

  #43948 commented on Aug 15, 2021 • 0 new comments

• SVD error message is not correct

  #63220 commented on Aug 15, 2021 • 0 new comments

• AdaHessian Implementation for PyTorch

  #63221 commented on Aug 15, 2021 • 0 new comments

• `help(torch.finfo)` doesn't give useful information even though class has documentation

  #63257 commented on Aug 15, 2021 • 0 new comments

• torch.cuda.nccl.version() is incorrect if NCCL_MINOR >= 10

  #62295 commented on Aug 16, 2021 • 0 new comments

• RFC: CIFlow for PyTorch: a flexible CI dispatcher that enables better PR experience with configurable testing validation set

  #61888 commented on Aug 16, 2021 • 0 new comments

• Update _symbolic_trace.py for proxy_placeholder function

  #63020 commented on Aug 16, 2021 • 0 new comments

• [quant] Add support for floating point zero_point in fake_quantize ops

  #61866 commented on Aug 16, 2021 • 0 new comments

• Make --help print all CLI options for test scripts

  #45945 commented on Aug 16, 2021 • 0 new comments

• Migrate build/test jobs from CircleCI to GitHub Actions

  #57686 commented on Aug 16, 2021 • 0 new comments

• Consolidate and pin Python dependencies in CI

  #61375 commented on Aug 16, 2021 • 0 new comments

• Use pytorch-probot for PyTorch-specific stuff + MOTD in Dr. CI comment

  #63038 commented on Aug 16, 2021 • 0 new comments

• Audit use of `C10_UNUSED`

  #62973 commented on Aug 16, 2021 • 0 new comments

• Race condition in docker image autogeneration

  #62738 commented on Aug 16, 2021 • 0 new comments

• Automate bumping of `clang-tidy` docker image tag on CI

  #60983 commented on Aug 16, 2021 • 0 new comments

• ci: file_diff_from_base does not work for ghstack PRs on GHA

  #60111 commented on Aug 16, 2021 • 0 new comments

• [BE] optimize hard-coded branch name in util scripts that might not work on release branch

  #56556 commented on Aug 16, 2021 • 0 new comments

• Move `tensordot` into `torch.linalg`

  #61649 commented on Aug 17, 2021 • 0 new comments

• Conv2d layers should accept 4-tuple for padding argument

  #17150 commented on Aug 17, 2021 • 0 new comments

• Sum issues with FP32

  #59863 commented on Aug 17, 2021 • 0 new comments

• max-sum operation

  #59545 commented on Aug 17, 2021 • 0 new comments

• [perf] 10x improvement when doing `x.sum(-1)` manually

  #57610 commented on Aug 17, 2021 • 0 new comments

• bool_tensor.sum(dtype=torch.int32) creates int32-copy of the original int8 tensor

  #55366 commented on Aug 17, 2021 • 0 new comments

• Reductions tracking issue

  #61417 commented on Aug 17, 2021 • 0 new comments

• [discussion] Support other index dtypes for scatter and other indexing functions in addition to int64

  #61819 commented on Aug 17, 2021 • 0 new comments

• [discussion] Remove the need of mandatory super() module call

  #61686 commented on Aug 17, 2021 • 0 new comments

• Port TH operators to Aten (umbrella issue)

  #24507 commented on Aug 17, 2021 • 0 new comments

• log10 is not one of supported operators in torch.onnx

  #61332 commented on Aug 17, 2021 • 0 new comments

• Conversion issue

  #17202 commented on Aug 18, 2021 • 0 new comments

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

  #24592 commented on Aug 18, 2021 • 0 new comments

• Sign in slogdet is set to requires_grad = False even when using complex numbers.

  #63138 commented on Aug 18, 2021 • 0 new comments

• void onnxruntime::BroadcastIterator::Init(ptrdiff_t, ptrdiff_t) axis == 1 || axis == largest was false. Attempting to broadcast an axis by a dimension other than 1. 3 by 224

  #61345 commented on Aug 18, 2021 • 0 new comments

• LazyTensor recomputes intermediate tensor in fragmented computation

  #62614 commented on Aug 18, 2021 • 0 new comments

• `torch.triangular_solve`'s documentation is misleading when `transpose` is explicitly specified but `upper` is not.

  #54258 commented on Aug 19, 2021 • 0 new comments

• Update docs for torch.triangular_solve to describe rectangular matrix behavior

  #56327 commented on Aug 19, 2021 • 0 new comments

• torch.triangular_solve on CUDA returns nan if triangular matrix is not full rank

  #56326 commented on Aug 19, 2021 • 0 new comments

• Addition of sparse CSR tensors is not yet implemented

  #59060 commented on Aug 19, 2021 • 0 new comments

• CUDA graphs user exposure for 1.10

  #61880 commented on Aug 19, 2021 • 0 new comments

• CompositeImplicitAutograd operators should not perform operations that do not dispatch

  #61669 commented on Aug 19, 2021 • 0 new comments

• Checkpointing without re-entrant autograd

  #62654 commented on Aug 20, 2021 • 0 new comments

• Use fast gpuAtomicAdd for FP16 data type in CUDA kernels

  #62695 commented on Aug 20, 2021 • 0 new comments

• Move `lu`, `lu_solve`, and `lu_unpack` into `torch.linalg`

  #61657 commented on Aug 20, 2021 • 0 new comments

• OpInfo Tracker

  #54261 commented on Aug 20, 2021 • 0 new comments

• Label tracking meta-issue (edit me to get automatically CC'ed on issues! cc bot)

  #24422 commented on Aug 20, 2021 • 0 new comments

• Support more ops in LazyTensor to TorchScript lowering

  #62431 commented on Aug 20, 2021 • 0 new comments

• Enable benchmark scribe reporting for GHA

  #61809 commented on Aug 20, 2021 • 0 new comments

• torch.package: TypeError: 'NoneType' object is not iterable

  #60089 commented on Aug 20, 2021 • 0 new comments

• add channels last support for ChannelShuffle

  #50247 commented on Aug 18, 2021 • 0 new comments

• add channels last support for PixelShuffle and PixelUnshuffle

  #50573 commented on Aug 18, 2021 • 0 new comments

• add channels last support for ConvTranspose2d

  #51185 commented on Aug 18, 2021 • 0 new comments

• Add onnx test for batched_nms

  #53175 commented on Aug 20, 2021 • 0 new comments

• [WIP][ONNX] Add support for max_unpool op in export

  #55411 commented on Aug 20, 2021 • 0 new comments

• [ONNX] Add support for torch.mv

  #55472 commented on Aug 20, 2021 • 0 new comments

• [WIP][ONNX] Update lower tuple pass

  #55473 commented on Aug 20, 2021 • 0 new comments

• add channels last (2d) support for mkldnn_convolution

  #55584 commented on Aug 18, 2021 • 0 new comments

• [WIP][ONNX] Extend support for args/kwargs in optional inputs API

  #55792 commented on Aug 20, 2021 • 0 new comments

• enable BFloat16 mkldnn_convolution on both contiguous and channels last memory format

  #55864 commented on Aug 18, 2021 • 0 new comments

• add BFloat16 support for bernoulli and Dropout on CPU

  #56372 commented on Aug 18, 2021 • 0 new comments

• add BFloat16 support for AdaptiveAvgPool2d on CPU

  #56902 commented on Aug 18, 2021 • 0 new comments

• add BFloat16 support for MaxPool2d on CPU

  #56903 commented on Aug 18, 2021 • 0 new comments

• optimize transpose copy for float32 and bfloat16 on CPU

  #56904 commented on Aug 18, 2021 • 0 new comments

• Add type refinement for Union

  #57011 commented on Aug 19, 2021 • 0 new comments

• Allow for Union-typed memory aliasing

  #57012 commented on Aug 19, 2021 • 0 new comments

• Preserve Union type annotations during serialization

  #57013 commented on Aug 19, 2021 • 0 new comments

• [BE] Improve docs and fix grammar/spelling errors

  #57014 commented on Aug 19, 2021 • 0 new comments

• add BFloat16 support for UpSample on CPU

  #58297 commented on Aug 18, 2021 • 0 new comments

• opitimze ConvTransposedND with mkldnn float32 and bfloat16 on CPU

  #58348 commented on Aug 18, 2021 • 0 new comments

• Canonicalize Optional[T] as Union[T, None]

  #58673 commented on Aug 19, 2021 • 0 new comments

• Preserve types during empty container assignment

  #58911 commented on Aug 20, 2021 • 0 new comments

• [PyTorch] Make arithmetic ops etc. use_const_ref_for_mutable_tensors

  #58935 commented on Aug 19, 2021 • 0 new comments

• [PyTorch] use borrow_from_optional_tensor in CPU matmul

  #59062 commented on Aug 19, 2021 • 0 new comments

• [PyTorch] use_const_ref_for_mutable_tensors in matmul_out and linalg_inv_out

  #59091 commented on Aug 19, 2021 • 0 new comments

• [ONNX][do not review][do not merge] Test ORT bump

  #59329 commented on Aug 20, 2021 • 0 new comments

• [cuDNN v8 API] cuDNN benchmark, convolution bwd / transposed convolution fwd, `bfloat16`, conv-bias-activation fusion

  #60755 commented on Aug 20, 2021 • 0 new comments

• Add cuSPARSE descriptors and update CSR addmm

  #60838 commented on Aug 19, 2021 • 0 new comments

• add BFloat16 support for BatchNorm2d on CPU

  #61015 commented on Aug 18, 2021 • 0 new comments

• Sparse CSR CUDA: add `addmv_out`

  #61407 commented on Aug 19, 2021 • 0 new comments

• Sparse CSR: add test parametrization for enabling sparse_csr

  #61408 commented on Aug 19, 2021 • 0 new comments

• [PyTorch] Reduce code size of register_prim_ops.cpp

  #61494 commented on Aug 19, 2021 • 0 new comments

• [PyTorch] Move c10::toString(ScalarType) to .cpp file

  #61501 commented on Aug 20, 2021 • 0 new comments

• Sparse CSR CPU: add `addmv_out`

  #61536 commented on Aug 19, 2021 • 0 new comments

• [ROCm] Refactor how to specify AMD gpu targets

  #61706 commented on Aug 18, 2021 • 0 new comments

• add out variant of linear

  #61801 commented on Aug 19, 2021 • 0 new comments

• out variant for cudnn batch norm

  #61824 commented on Aug 19, 2021 • 0 new comments

• Port triangular_solve to structured kernel

  #61857 commented on Aug 19, 2021 • 0 new comments

• Sparse CSR CUDA: add `triangular_solve_out`

  #61858 commented on Aug 19, 2021 • 0 new comments

• [ONNX] Adding optional type in ONNX exporter

  #61938 commented on Aug 20, 2021 • 0 new comments

• Add export_graph in profiler

  #61948 commented on Aug 16, 2021 • 0 new comments

• Enable CUPTI for kineto by default on windows.

  #62175 commented on Aug 18, 2021 • 0 new comments

• Sparse CSR CPU: add `triangular_solve_out`

  #62180 commented on Aug 19, 2021 • 0 new comments

• [DO NOT REVIEW YET] Allow NCCL pg to send and recv CPU tensors with UCC

  #62270 commented on Aug 17, 2021 • 0 new comments

• add tensorboard plugin tip in console output

  #62384 commented on Aug 16, 2021 • 0 new comments

• [Draft][ONNX] STFT and ISTFT

  #62412 commented on Aug 20, 2021 • 0 new comments

• Update CMake and use native CUDA language support

  #62445 commented on Aug 21, 2021 • 0 new comments

• Add workaround for nvcc header dependecies bug

  #62550 commented on Aug 21, 2021 • 0 new comments

• Add BFloat16 support for smooth_l1_loss on CPU

  #62558 commented on Aug 16, 2021 • 0 new comments

• [PyTorch Edge][type] Add type support for NamedTuple custom class (export)

  #62612 commented on Aug 17, 2021 • 0 new comments

• Implement forward AD for linalg.matrix_exp

  #62716 commented on Aug 20, 2021 • 0 new comments

• DO NOT MERGE, testing conda pytorch-mutex

  #62718 commented on Aug 17, 2021 • 0 new comments

• implement "xy" indexing for torch.meshgrid

  #62724 commented on Aug 21, 2021 • 0 new comments

• Better and more consistent error messages in torch.linalg

  #62734 commented on Aug 20, 2021 • 0 new comments

• remove unnecessary torch.meshgrid overloads

  #62743 commented on Aug 21, 2021 • 0 new comments

• require that clients provide `indexing` to torch.meshgrid

  #62749 commented on Aug 21, 2021 • 0 new comments

• change `torch.meshgrid` indexing default to "xy"

  #62752 commented on Aug 21, 2021 • 0 new comments

• Add note on ifdefing based on CUDA_VERSION for ROCm path

  #62850 commented on Aug 18, 2021 • 0 new comments

• [WIP][ONNX] exporter changes for torch hub models

  #62856 commented on Aug 21, 2021 • 0 new comments

• [CI] move distributed test into its own CI job

  #62896 commented on Aug 16, 2021 • 0 new comments

• Add OpInfo for `nn.functional.cosine_similarity`

  #62959 commented on Aug 15, 2021 • 0 new comments

• update rpc tensorpipe logic for sparse tensors

  #62960 commented on Aug 21, 2021 • 0 new comments

• [Static runtime] refactor static runtime to expose useful liveness methods

  #62995 commented on Aug 21, 2021 • 0 new comments

• Update test_aten and remove IN_WHEEL_TEST flag

  #63066 commented on Aug 17, 2021 • 0 new comments

• Update test_vec256

  #63067 commented on Aug 17, 2021 • 0 new comments

• [PyTorch Edge][type] Add type support for NamedTuple custom class (import)

  #63130 commented on Aug 17, 2021 • 0 new comments

• [pruner] refactor `ActivationReconstruction` forward hooks

  #63157 commented on Aug 18, 2021 • 0 new comments

• [pruner] refactor `ActivationReconstruction` forward hooks

  #63158 commented on Aug 21, 2021 • 0 new comments

• [DOC] fix torch.randint doc

  #63179 commented on Aug 14, 2021 • 0 new comments

• Working on concatting multiple linear layers together

  #63198 commented on Aug 20, 2021 • 0 new comments

• document policy for upstream package compatibility

  #63208 commented on Aug 16, 2021 • 0 new comments

• fix(torch.distributed.elastic): Failed to print the first info statement

  #63222 commented on Aug 15, 2021 • 0 new comments

• [doc][hackathon] To add Adam Optimizer to the documentation

  #63251 commented on Aug 16, 2021 • 0 new comments

• [doc][hackathon] To add AdamW Optimizer to the documentation

  #63252 commented on Aug 16, 2021 • 0 new comments

• [doc][hackathon] To add Adagrad Optimizer to the documentation

  #63254 commented on Aug 15, 2021 • 0 new comments

• [doc][hackathon] To add Adadelta Optimizer to the documentation

  #63255 commented on Aug 16, 2021 • 0 new comments

• Docathon: Add docs for nn.functional.*d_max_pool

  #63264 commented on Aug 17, 2021 • 0 new comments

• Initial set of changes to replacing

  #63267 commented on Aug 18, 2021 • 0 new comments

• [ao][docs] Clarify operator support for quantization

  #63270 commented on Aug 20, 2021 • 0 new comments

• [DO NOT COMMIT] Build pip binaries with llvm

  #63273 commented on Aug 15, 2021 • 0 new comments

• [DO NOT COMMIT] Build conda binaries with llvm

  #63274 commented on Aug 15, 2021 • 0 new comments