pytorch / pytorch Public

25 Pull requests merged by 17 people

• [ONNX] Fix lstm reshape shape inference regression

  #72734 merged Feb 15, 2022

• [LTS] CherryPick: Fix failing test due to a bug in NumPy when using OpenBLAS

  #72820 merged Feb 15, 2022

• [LTS] CherryPick: Pin librosa

  #72668 merged Feb 15, 2022

• [1.11][DataPipe] Docs Improvement

  #72801 merged Feb 15, 2022

• [ONNX] Add symbolic support for torch.nn.cosinesimilarity

  #72128 merged Feb 15, 2022

• pad_sequence: fix regression - support tensor (#72436)

  #72697 merged Feb 14, 2022

• Cat shape analysis execution fix for -1 dim

  #72678 merged Feb 14, 2022

• Fix refcounting in access of saved for forward attribute (#72627)

  #72656 merged Feb 14, 2022

• Pin builder repo for GHA builds to release/1.11

  #72739 merged Feb 11, 2022

• Fix for builder repo not pinned in release branch (#72719)

  #72732 merged Feb 11, 2022

• Move debug_util and python_util to torch/csrc/lazy

  #72599 merged Feb 11, 2022

• [ONNX] Add Squeeze/Unsqueeze dynamic dimensions support when opset >= 13

  #71158 merged Feb 11, 2022

• [LT] Allow lazy_model.mark_step to specify a device

  #72683 merged Feb 10, 2022

• [PyTorch] Enable lightweight dispatch as an option in cmake build

  #71782 merged Feb 10, 2022

• Remove duplicate copy in lazy _copy_from()

  #72218 merged Feb 10, 2022

• Fix tagged build detection for binary builds (#72628)

  #72652 merged Feb 10, 2022

• Fix hardcoded `TsNode` in `lazy_ir.py`

  #71921 merged Feb 10, 2022

• [1.11] Fix SVD error code handling for OpenBLAS 0.3.15+ and MKL 2022+

  #72513 merged Feb 9, 2022

• Fix doc build for release branches (#72567)

  #72635 merged Feb 9, 2022

• Disable complex32

  #72604 merged Feb 9, 2022

• Add a flag to control whether to use thread pool for execution

  #72021 merged Feb 9, 2022

• fix shape cache hash for dynamic shape

  #72561 merged Feb 9, 2022

• [CircleCI] Remove `gpu.medium` class

  #72613 merged Feb 9, 2022

• [LT] Copy the device ordinal in MakeComputationDataFromTensor

  #72574 merged Feb 9, 2022

• [ONNX] Add torch.nan_to_num and torch.maximum/minimum symbolic

  #72090 merged Feb 9, 2022

127 Pull requests opened by 72 people

• Fix cusparse sync issue in bsrsv2 and bsrsm2

  #72568 opened Feb 9, 2022

• allow contiguous inputs run into qcat_nhwc_stub when dim is last dimension

  #72575 opened Feb 9, 2022

• [jiterator] stricter static_assert

  #72576 opened Feb 9, 2022

• Replacing tensor dtype query

  #72577 opened Feb 9, 2022

• [JIT][SR] Introduce prim::IfThenElse

  #72587 opened Feb 9, 2022

• [PTE] avoiding refcount bump

  #72590 opened Feb 9, 2022

• [Static Runtime] Remove ProcessedNode::num_outputs_

  #72592 opened Feb 9, 2022

• [test, do not merge] det_singular neg_view test for #72204

  #72593 opened Feb 9, 2022

• Make debug_pkl smaller by only emitting unique traces.

  #72596 opened Feb 9, 2022

• [FSDP] Implement apply()

  #72600 opened Feb 9, 2022

• [Static Runtime] Add a script to auto-generate out variant dispatchers

  #72602 opened Feb 9, 2022

• [Static Runtime] Add auto-generated out variant dispatchers

  #72603 opened Feb 9, 2022

• [LTS] CherryPick: Use latest ubuntu supported by CircleCI

  #72605 opened Feb 9, 2022

• Run example.py in CI

  #72606 opened Feb 9, 2022

• feat(profiler): allow use of backends to log traces

  #72608 opened Feb 9, 2022

• [LTS] Skip TestSerialization.test_lstm if avx512_vnni supported

  #72611 opened Feb 9, 2022

• [DO NOT MERGE] free one dispatch key (FPGA) for next PR to run

  #72621 opened Feb 9, 2022

• [Not For Merge][LT] Integrating with DDP with c10d comm ops falling back to eager

  #72631 opened Feb 9, 2022

• Support masked sum on CSR tensors [CPU]

  #72633 opened Feb 9, 2022

• [WIP] Data structures of a basic embedded runtime

  #72634 opened Feb 9, 2022

• Move torch::deploy tests to their own workflow job

  #72637 opened Feb 10, 2022

• Move dyn fusion api to jit/api/module/

  #72638 opened Feb 10, 2022

• Make fusion strategy api public

  #72639 opened Feb 10, 2022

• [HACK DO NOT MERGE] Testing OSS CI with local XNNPACK changes!!

  #72642 opened Feb 10, 2022

• [TEST] add back out-of-bounds bug and see if ASAN catches it with std::vector

  #72644 opened Feb 10, 2022

• [ao] bugfix for big tensors quantize to 0

  #72645 opened Feb 10, 2022

• run torchbench (DO NOT MERGE)

  #72646 opened Feb 10, 2022

• try to add more models to iOS test

  #72666 opened Feb 10, 2022

• Convert type comments to annotations across caffe2

  #72669 opened Feb 10, 2022

• Improve file system portability

  #72670 opened Feb 10, 2022

• [PyTorch] Handle non-vectorizable parameters for native MHA CUDA rescale kernel

  #72671 opened Feb 10, 2022

• Update __torch_dispatch__ to return op overload instead of the opoverload packet function

  #72673 opened Feb 10, 2022

• Pytorch test: What are these files?

  #72679 opened Feb 10, 2022

• qlinear: Remove legacy cpp_custom_type_hack support

  #72680 opened Feb 10, 2022

• remove torch asserts from deploy

  #72684 opened Feb 10, 2022

• add rocm ciflow/slow workflow

  #72686 opened Feb 10, 2022

• [PyTorch] RFC: existing MHA: can't we fuse the attn_mask addition easily?

  #72691 opened Feb 11, 2022

• add BFloat16 support for logcumsumexp on CPU

  #72694 opened Feb 11, 2022

• add BFloat16 operators on CPU: histc, atan2

  #72695 opened Feb 11, 2022

• [ao] bugfix for big tensors quantize to 0

  #72699 opened Feb 11, 2022

• align signature of make_tensor with other creation ops

  #72702 opened Feb 11, 2022

• noncontiguous -> non_contiguous in torch.testing.make_tensor

  #72703 opened Feb 11, 2022

• add layout parameter to torch.testing.make_tensor

  #72704 opened Feb 11, 2022

• deprecate torch.testing.make_non_contiguous

  #72705 opened Feb 11, 2022

• cleanup torch.testing namespace

  #72708 opened Feb 11, 2022

• [WIP] Faster `index_select` for sparse COO tensors.

  #72710 opened Feb 11, 2022

• [Quant][improvement] Rename ReferenceableQuantizedModule

  #72717 opened Feb 11, 2022

• masked std

  #72722 opened Feb 11, 2022

• masked argmin and argmax

  #72723 opened Feb 11, 2022

• masked cumsum and cumprod

  #72724 opened Feb 11, 2022

• masked median

  #72725 opened Feb 11, 2022

• masked logaddexp and logsumexp

  #72726 opened Feb 11, 2022

• [PyTorch] Add fused addmm path in linear for contiguous 3D input

  #72728 opened Feb 11, 2022

• Add access to inputs and outputs of op overloads

  #72729 opened Feb 11, 2022

• Update lazy_ir.py from lazy_tensor_staging

  #72730 opened Feb 11, 2022

• [PT-D][Sharded Tensor] new init api for local tensor and sharding spec auto inference

  #72733 opened Feb 11, 2022

• [qunat][fx][fix] Fix get_module_type for fusion

  #72735 opened Feb 11, 2022

• fix android test and update android/README

  #72736 opened Feb 11, 2022

• [LTS] CircleCI: Deprecate `gpu.medium` class

  #72742 opened Feb 11, 2022

• [LTS] CherryPick: Quote in setup-ci-env

  #72744 opened Feb 11, 2022

• Move shape inference code to master and only validate it in codegen

  #72752 opened Feb 11, 2022

• [Qunat] Refactor reference module mapping

  #72755 opened Feb 11, 2022

• Move Lazy Shape Inference functions to pytorch core

  #72756 opened Feb 11, 2022

• [JIT] document issues with eager autocast & scripting

  #72760 opened Feb 12, 2022

• [quant][core][gpu][eager] Improved quantized conv operator in cudnn

  #72770 opened Feb 13, 2022

• [fix] index_add support source broadcast in backward

  #72777 opened Feb 13, 2022

• Fix `nn.Module.state_dict()`

  #72780 opened Feb 13, 2022

• Add doctests for torch/_torch_docs.py

  #72781 opened Feb 13, 2022

• [LTS] CherryPick: Resolve attribute error in ONNX CI

  #72783 opened Feb 14, 2022

• DO NOT MERGE: Status of build workflow on lts branch after all active PRs

  #72785 opened Feb 14, 2022

• set SCCACHE_IGNORE_SERVER_IO_ERROR=1

  #72795 opened Feb 14, 2022

• eager quant: convert mapping for fused QAT Linear-Bn1d

  #72796 opened Feb 14, 2022

• Add optional encoding argument to fileopener

  #72800 opened Feb 14, 2022

• Rename undocumented storage types

  #72802 opened Feb 14, 2022

• [PTE] mobile call debug serializer/deserializer flatbuffer

  #72804 opened Feb 14, 2022

• Automated submodule update: FBGEMM

  #72805 opened Feb 14, 2022

• Change BLAS_LIBRARIES variable setting to MKL_LIBRARIES

  #72806 opened Feb 14, 2022

• [ROCM] Navi21 Enablement 3: Embedding kernels

  #72809 opened Feb 14, 2022

• DO NOT MERGE: Status of binary build workflow on lts branch after all active PRs

  #72817 opened Feb 14, 2022

• More update on the guidance

  #72818 opened Feb 14, 2022

• [LTS] Increase timeout for `TensorPipeDistAutogradTestWithSpawn.test_multiple_backward`

  #72819 opened Feb 14, 2022

• Test py-3.10 builds

  #72822 opened Feb 14, 2022

• [LTS] CherryPick: Skip broken svd tests

  #72823 opened Feb 14, 2022

• [LTS] CherryPick: Resolve jited `torch.isnan` crash for `torch.float16`

  #72825 opened Feb 14, 2022

• Implement Tanh Gelu Approximation

  #72826 opened Feb 14, 2022

• free up dispatch key space (in C++)

  #72827 opened Feb 15, 2022

• DispatchKeySet perf improvements

  #72828 opened Feb 15, 2022

• [DataPipe] Improve .pyi generation

  #72829 opened Feb 15, 2022

• Update stft tests to support latest librosa

  #72833 opened Feb 15, 2022

• Revert "Pin librosa"

  #72834 opened Feb 15, 2022

• Use irange in PyTorch

  #72836 opened Feb 15, 2022

• Fix undefined variable errors

  #72838 opened Feb 15, 2022

• [torchrec] remove __getstate__ from PyTorch base optimizer (#16)

  #72839 opened Feb 15, 2022

• 'typename Base' is checked repeatedly

  #72842 opened Feb 15, 2022

• [torch.distributed][DDP] Disable DDP bucketing for the first iteration

  #72843 opened Feb 15, 2022

• Adding details to kl.py

  #72845 opened Feb 15, 2022

• add BFloat16 sparse operators on CPU: copy, coalesce, sparse_mask, ad…

  #72846 opened Feb 15, 2022

• Fix command example

  #72847 opened Feb 15, 2022

• remove list of modules for py36

  #72849 opened Feb 15, 2022

• actions issue

  #72851 opened Feb 15, 2022

• [ROCm] Enable sort operator BF16 support

  #72854 opened Feb 15, 2022

• [Quant][fx] Add lowering for functional linear

  #72855 opened Feb 15, 2022

• Fix SequentialLR initialization

  #72856 opened Feb 15, 2022

• Back out "[pytorch][PR] Implement Tanh Gelu Approximation" (#5913)

  #72857 opened Feb 15, 2022

• Fix asserts in tests

  #72864 opened Feb 15, 2022

• Generate static docstrings for torch._masked functions.

  #72865 opened Feb 15, 2022

• fix model dump for the lowered module

  #72866 opened Feb 15, 2022

• [Hack] Allow true retracing of a scripted module

  #72867 opened Feb 15, 2022

• Move vectorized CPU codegen to after ATen codegen

  #72869 opened Feb 15, 2022

• [PyTorch] Hit fused addmm path in linear() for existing MHA

  #72871 opened Feb 15, 2022

• [nnc][aot_compiler] Memory formats args to aot_compiler

  #72873 opened Feb 15, 2022

• Use CType Union instead of manually listing everything out (#72853)

  #72875 opened Feb 15, 2022

• (2/2) Make TorchScript Preserve Fully Qualified Class Name for Python Exceptions: frontend change

  #72876 opened Feb 15, 2022

• [LTS] Skip large tensor tests that cause OOM kill on CircleCI

  #72877 opened Feb 15, 2022

• gitignore tools/bazel executable

  #72878 opened Feb 15, 2022

• [WIP][DRAFT] NestedTensor in core exploration

  #72881 opened Feb 15, 2022

• stft: move towards always returning complex

  #72882 opened Feb 15, 2022

• [ONNX] Capture predefined attributes for local function

  #72883 opened Feb 15, 2022

• [Join][BE] Fix typo; remove obsolete method

  #72886 opened Feb 15, 2022

• [easy][PT] Using toTupleRef insted of toTuple

  #72887 opened Feb 15, 2022

• [WIP][JIT] script & logging for extracting IR from logs

  #72889 opened Feb 15, 2022

• [LTS] Stick to conda-4.10 for 3.6 on Windows

  #72890 opened Feb 15, 2022

• Merge pytorch master into lazy_tensor_staging

  #72894 opened Feb 15, 2022

• Update nightly wheels to ROCm5.0 for GHA

  #72895 opened Feb 15, 2022

• [DataPipe] Fix issue with DataPipe serialization with `dill`

  #72896 opened Feb 15, 2022

• [AWS cluster] test debug and test PR sending

  #72898 opened Feb 16, 2022

• (2/2) Make TorchScript Preserve Fully Qualified Class Name for Python Exceptions: frontend change

  #72899 opened Feb 16, 2022

64 Issues closed by 35 people

• Allow params to be an iterable of dicts in ZeroRedundancyOptimizer

  #59973 closed Feb 16, 2022

• IterableDataset concrete subclasses are protocols which breaks isinstance checks

  #72863 closed Feb 15, 2022

• torch.linalg.det's gradcheck fails with Illegal Memory Access failure for singular matrices

  #72203 closed Feb 15, 2022

• [distributed] tell which address for RuntimeError: Address already in use

  #72816 closed Feb 15, 2022

• DISABLED test_dtypes_argsort_cuda (__main__.TestCommonCUDA)

  #72870 closed Feb 15, 2022

• symeig memory leak

  #42557 closed Feb 15, 2022

• request about perspective grid generator

  #72848 closed Feb 15, 2022

• [ZeRo] Parameter group support in constructor

  #71347 closed Feb 15, 2022

• Tests in test_constraints fail when compiled without LAPACK support

  #70670 closed Feb 15, 2022

• RuntimeError: ONNX export failed: Couldn't export Python operator Im2Col

  #22862 closed Feb 15, 2022

• Deprecating Python 3.6 support

  #66462 closed Feb 15, 2022

• Building ONNX (proto file not being generated)

  #31764 closed Feb 14, 2022

• RuntimeError: Exporting the operator roll to ONNX opset version 11 is not supported. Please open a bug to request ONNX export support for the missing operator.

  #68974 closed Feb 14, 2022

• Missing headers in ATen/cuda/DeviceUtils.cuh

  #72807 closed Feb 14, 2022

• Apparent different between `.to(dtype=torch.float32)` and `.float()` for TF32

  #63951 closed Feb 14, 2022

• FileOpenerIterDataPipe should support an encoding argument

  #72713 closed Feb 14, 2022

• Problem with typing for cudnn.allow_tf32

  #72753 closed Feb 14, 2022

• Gradient estimator methods like straight through estimators for gradient approximation of non-differentiable landscapes of a model's graph.

  #72772 closed Feb 14, 2022

• `sum` and `sparse.sum` have different behavior for scalar tensor

  #72793 closed Feb 14, 2022

• TorchVision Tries To Install Torch Again

  #72762 closed Feb 14, 2022

• How can I load and use a HDF5 format model in C ++ using libtorch

  #72774 closed Feb 14, 2022

• support for modern python

  #72779 closed Feb 13, 2022

• Determinism for MaxPool3d and AvgPool3d

  #72776 closed Feb 13, 2022

• devices or gpus in pytorch_lightning.Trainer

  #72771 closed Feb 13, 2022

• kernel assertion with torch 1.10.2 and CUDA 11.3

  #72696 closed Feb 12, 2022

• Using Autograd before and after forking should be possible

  #69839 closed Feb 12, 2022

• Builder repo is not pinned in release branch

  #72655 closed Feb 11, 2022

• [ROCm] test_linalg.py failures after hipMAGMA integration

  #51303 closed Feb 11, 2022

• Pytorch should not log to python root logger

  #72648 closed Feb 11, 2022

• jit.script mis-optimizes code causing cpu fallbacks

  #72720 closed Feb 11, 2022

• PyTorch 1.10.2 failed to create workable onnx for sequential LSTMs

  #72399 closed Feb 11, 2022

• torch.complex32 is missing from the PyTorch 1.11 RC

  #72721 closed Feb 11, 2022

• TCPStoreTest.testWatchKeyCallbackWithPrefix Stuck in CI

  #59797 closed Feb 11, 2022

• Add retries to ecr get-login call: https://github.com/pytorch/pytorch/runs/3948026530

  #67699 closed Feb 11, 2022

• RuntimeError: Caffe2 -The tensor has a non-zero number of elements, but its data is not allocated yet

  #71914 closed Feb 11, 2022

• Functional version of `MultiheadAttention`, `torch.nn.functional.multi_head_attention_forward` has no documentation

  #72597 closed Feb 11, 2022

• torch.no_grad() can cause nan and random results

  #72594 closed Feb 10, 2022

• {TypeError}pad_sequence(): argument 'sequences' (position 1) must be tuple of Tensors, not Tensor

  #71365 closed Feb 10, 2022

• Forward-over-reverse gradgradcheck fails on ROCm for `cumulative_trapezoid` due to memory exception

  #69897 closed Feb 10, 2022

• Forward-over-reverse gradgradcheck fails for `cumulative_trapezoid` on CUDA

  #69866 closed Feb 10, 2022

• [structured] Preserve computed values to implementation kernels

  #55526 closed Feb 10, 2022

• Calculation is different from decimal point.

  #72640 closed Feb 10, 2022

• `StreamWrapper` does not work with `open`

  #71221 closed Feb 10, 2022

• CUDA initialization makes DataLoader slower

  #68383 closed Feb 10, 2022

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

  #36035 closed Feb 10, 2022

• discuss.pytorch.org is down

  #72651 closed Feb 10, 2022

• [C++] Inference Pytorch and TensorRT got errors on Windows

  #71079 closed Feb 10, 2022

• DeepLung pretrained weights gets changed while importing weights from Pytorch to ONNX.

  #72392 closed Feb 10, 2022

• Custom function recent change introduces bad refcounting

  #72612 closed Feb 10, 2022

• forward-mode AD testing does not test inputs with requires_grad=False

  #72113 closed Feb 10, 2022

• Installing torch_nighly using pip shows error

  #72571 closed Feb 10, 2022

• pip install does not work for LTS (1.8) - wheel index page might be broken

  #72579 closed Feb 10, 2022

• Seeing several addmm_sparse_csr_cuda float16/bf16 failures

  #72234 closed Feb 10, 2022

• DOCS, CI: pushing a tag is not triggering a docs build

  #72519 closed Feb 9, 2022

• DISABLED test_success (__main__.SpawnTest)

  #72298 closed Feb 9, 2022

• binary_linux_manywheel_3_7m_cu102_devtoolset7_test is broken

  #72610 closed Feb 9, 2022

• BLAS options: OpenBLAS vs Accelerate

  #71712 closed Feb 9, 2022

• RuntimeError: set_sizes_and_strides is not allowed on a Tensor created from .data or .detach().

  #72569 closed Feb 9, 2022

• In windows, DataLoader with num_workers > 0 is extremely slow (50 times slower)

  #12831 closed Feb 9, 2022

• ONNX export on nightly with opset13 started to fail Unsqueeze

  #51208 closed Feb 9, 2022

• [FSDP] Root frees params in `_post_backward_hook()` though `reshard_after_forward=False`

  #72541 closed Feb 9, 2022

• dbr quantization: add support for functional convolution variants

  #70620 closed Feb 9, 2022

• libtorch with Cuda 11.3 not linked properly on Windows using Visual Studio 2022

  #72396 closed Feb 9, 2022

• Inconsistent behavior between torch.nn.functional.pad and torchvision.transforms.functional.pad when setting padding to be an eight-element lists

  #72394 closed Feb 9, 2022

71 Issues opened by 56 people

• Hanging Validation

  #72897 opened Feb 16, 2022

• DISABLED test_Sparse_to_Sparse_copy__cuda_bfloat16 (__main__.TestSparseCUDA)

  #72892 opened Feb 15, 2022

• DISABLED test_coalesce_cuda_bfloat16 (__main__.TestSparseCUDA)

  #72893 opened Feb 15, 2022

• DISABLED test_ddp_comparison_uneven_inputs (__main__.TensorPipeDdpComparisonTest)

  #72891 opened Feb 15, 2022

• Support c10::IntArrayRef in c10::IValue

  #72885 opened Feb 15, 2022

• Bazel fails in an obscure way if submodules are not initialized

  #72880 opened Feb 15, 2022

• SequentialLR scheduler incorrect initialization

  #72874 opened Feb 15, 2022

• DISABLED test_data_parallel_module_cuda_float64 (__main__.TestDataParallelDeviceTypeCUDA)

  #72868 opened Feb 15, 2022

• Doc issue in `nn.Module`

  #72862 opened Feb 15, 2022

• torchscripted model fails to load on triton server

  #72861 opened Feb 15, 2022

• Some tests misusing assertTrue for comparisons

  #72860 opened Feb 15, 2022

• Profiler crashes with ProfilerActivity.CUDA on AWS p4d.24xlarge with A100-SXM4-40GB

  #72859 opened Feb 15, 2022

• Profiler crashes in export_chrome_trace with seg fault if any of record_shapes=True or with_flops=True

  #72858 opened Feb 15, 2022

• Type signature for tools.codegen.api.lazy.isValueType is a bit suspect

  #72852 opened Feb 15, 2022

• N-dimensional Convolutions

  #72850 opened Feb 15, 2022

• torch.distributed hangs at barrier()

  #72835 opened Feb 15, 2022

• Toggling deterministic mode for individual autograd backward functions

  #72831 opened Feb 15, 2022

• Some loss functions support `dtype` broadcast but some do not

  #72824 opened Feb 14, 2022

• `{Batch,Instance}Norm{1,2,3}d` works when `num_features != C`!

  #72821 opened Feb 14, 2022

• Allow `nn.Sequential` to take a normal dict (along with OrderedDict)

  #72815 opened Feb 14, 2022

• Standardize Naming for Workflows/Jobs

  #72814 opened Feb 14, 2022

• DISABLED test_conv_noncontig_weights_and_bias_cuda (main.TestNNDeviceTypeCUDA)

  #72813 opened Feb 14, 2022

• DISABLED test_fast_tasks (main.TestFunctionalAutogradBenchmark)

  #72812 opened Feb 14, 2022

• FX graph mode quant: support fusing linear -> view -> BN

  #72797 opened Feb 14, 2022

• why ram memory surges while loading model, with change in torch load device from CPU to GPU

  #72791 opened Feb 14, 2022

• Export torch.maximum() op in ONNX format

  #72790 opened Feb 14, 2022

• Docs bug: type annotations for linspace (and logspace) start and end arguments is wrong

  #72788 opened Feb 14, 2022

• grad strides do not match bucket view strides

  #72787 opened Feb 14, 2022

• Torch version in docker container does not match tag

  #72784 opened Feb 14, 2022

• Error during training: falseINTERNAL ASSERT FAILED

  #72782 opened Feb 13, 2022

• BUG (potential crash) with `state_dict()` implementation and overload

  #72778 opened Feb 13, 2022

• [vulkan] Vulkan backend fails creating tensor on x86_64 Linux

  #72775 opened Feb 13, 2022

• Add NCCL and MPI version printing to torch.utils.collect_env

  #72768 opened Feb 12, 2022

• Feature Request: Deterministic MaxPool3d and AvgPool3d

  #72766 opened Feb 12, 2022

• [KL divergence] Adding details in error when KL divergence registered between two distributions and improving the doctstring.

  #72765 opened Feb 12, 2022

• RuntimeError: input_values.size() == param_count_list.size()INTERNAL ASSERT FAILED at "..\\torch\\csrc\\jit\\python\\script_init.cpp":480

  #72763 opened Feb 12, 2022

• Add softplus inverse

  #72759 opened Feb 12, 2022

• pytorchmergebot doesn't react to comments left from "files" tab

  #72749 opened Feb 11, 2022

• [ROCm] ERROR test_fast_tasks and test_conv_noncontig_weights_and_bias_cuda

  #72747 opened Feb 11, 2022

• CPU execution/dispatch time dominates and slows down small TorchScript GPU models

  #72746 opened Feb 11, 2022

• [JIT] torch.jit.fuser("fuser1") should enable cpu fusion

  #72743 opened Feb 11, 2022

• Pattern Matching with Tensors

  #72737 opened Feb 11, 2022

• [ZeRO] Investigate ROCm test flakiness

  #72718 opened Feb 11, 2022

• Investigate support for parameters(), named_parameters() in FSDP

  #72716 opened Feb 11, 2022

• Add `pct_end` parameter to `OneCycleLR`

  #72714 opened Feb 11, 2022

• Better support for pypip packages implementing torch cuda extentions

  #72712 opened Feb 11, 2022

• Feature request: Implement `gelsd` backend for `linalg.lstsq` via `linalg.svd`

  #72711 opened Feb 11, 2022

• Feature: Use iterative refinement algorithm from cuSOLVER for the least-squares solver on CUDA

  #72709 opened Feb 11, 2022

• Torch typing: `load_state_dict` overconstrained

  #72707 opened Feb 11, 2022

• GeneralTensorShapeOpQuantizeHandler skip unless reference mode

  #72698 opened Feb 11, 2022

• Create a CI workflow for XLA testing using the XLA test image

  #72693 opened Feb 11, 2022

• `test_qconv_transpose3d` test run-time failure

  #72681 opened Feb 10, 2022

• [JIT] Add pass that compares outputs of nvfuser & non-fused implementations

  #72674 opened Feb 10, 2022

• nn.functional No-batch-dim support should have OpInfo examples

  #72672 opened Feb 10, 2022

• Tests should be disabled in forks when they are expected to fail

  #72661 opened Feb 10, 2022

• Add a unit test which uses an extension module + ordered importer

  #72659 opened Feb 10, 2022

• deit_small_patch16_224 weights not working

  #72654 opened Feb 10, 2022

• C++ linking to Pytorch on Linux (CPU) fails with libtorch 1.10.1 - cannot find MKL

  #72653 opened Feb 10, 2022

• Some inplace ops don't raise on incompatible shapes and meta device

  #72650 opened Feb 10, 2022

• upstream `apex.normalization.FusedRMSNorm`

  #72643 opened Feb 10, 2022

• Code in torch/_masked is getting executed during `import torch`

  #72636 opened Feb 9, 2022

• `torch.kaiser_window` fails for meta tensors and `window_length >= 2`

  #72630 opened Feb 9, 2022

• DISABLED test_terminate_exit (__main__.SpawnTest)

  #72624 opened Feb 9, 2022

• DISABLED test_success_first_then_exception (__main__.SpawnTest)

  #72625 opened Feb 9, 2022

• TorchScript assertion failure for a `List[...]` inside a `NamedTuple`

  #72617 opened Feb 9, 2022

• Can't forward pass conv2d with kernel_size=1, and padding=1

  #72591 opened Feb 9, 2022

• [DDP][FSDP][BE] Refactor `no_sync()` context

  #72589 opened Feb 9, 2022

• Update torch.testing.make_tensor "device" and "dtype" arguments to be kwarg-only

  #72588 opened Feb 9, 2022

• Avoid using thrust:: directly, use THRUST_NS_QUALIFIER:: instead

  #72582 opened Feb 9, 2022

• Zero-copy on shared memory of NVIDIA Jetson devices

  #72581 opened Feb 9, 2022

• Setting a list of Modules as an attribute does not work like setting a Module as an attribute, and throws no warning

  #72580 opened Feb 9, 2022

347 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.

• [FSDP] Implement local_state_dict and load_local_state_dict

  #72469 commented on Feb 16, 2022 • 38 new comments

• expanded weights without fast rules

  #70140 commented on Feb 15, 2022 • 27 new comments

• [PyTorch] Add codegen unboxing ability

  #69881 commented on Feb 16, 2022 • 20 new comments

• Check if the iterator is valid before dereferencing it

  #72405 commented on Feb 15, 2022 • 20 new comments

• [ONNX] Relax sequence tensor dim_param serialization

  #70651 commented on Feb 15, 2022 • 18 new comments

• [tensorexp] ExternalCall2 without memcpy

  #72225 commented on Feb 15, 2022 • 16 new comments

• Create a CI workflow for XLA tests using the XLA test image

  #72496 commented on Feb 15, 2022 • 16 new comments

• [v.1.11.0] Release Tracker

  #72267 commented on Feb 16, 2022 • 13 new comments

• Improve numerical stability of `torch.distributions.wishart.Wishart`

  #72059 commented on Feb 16, 2022 • 13 new comments

• Reduce boilerplate for the `_out` overload of structured kernels that require specific strides.

  #69813 commented on Feb 14, 2022 • 12 new comments

• Split the allow list and increase the warning level for forever BC breaking changes

  #72474 commented on Feb 11, 2022 • 12 new comments

• [ONNX] ONNX Exporter logging

  #71342 commented on Feb 16, 2022 • 11 new comments

• [Quant] Add qint32 quantization support

  #72472 commented on Feb 15, 2022 • 10 new comments

• only compare attributes for meta tensors

  #72508 commented on Feb 16, 2022 • 10 new comments

• Error in SVD cusolver on Linux

  #69203 commented on Feb 15, 2022 • 9 new comments

• Add `unique_indices` arg and update nondeterministic alerts for `scatter` and `scatter_add`

  #70873 commented on Feb 15, 2022 • 9 new comments

• Port `margin_ranking_loss` to `TensorIterator`

  #72321 commented on Feb 11, 2022 • 9 new comments

• Enable Half/BFloat16 support for to_dense and coalesce methods.

  #72397 commented on Feb 15, 2022 • 9 new comments

• Port `index.Tensor` to structured kernels.

  #69607 commented on Feb 10, 2022 • 7 new comments

• fix: onnx PReLU unidirectional broadcasting

  #70571 commented on Feb 11, 2022 • 7 new comments

• Fix `functional.max_poolNd` warning spam in the CI

  #71258 commented on Feb 10, 2022 • 7 new comments

• [ONNX] Make Non-Float Op Exportation Compatible to Avoid Invalid ONNX Models

  #72401 commented on Feb 14, 2022 • 7 new comments

• numpy-compat: Add an offset to `torch.eye`

  #72443 commented on Feb 14, 2022 • 7 new comments

• macOS Monterey 12.2 PyTorch-1.7.1 SGD Optimizer Segfaults with libomp.dylib`__kmp_suspend_initialize_thread

  #72400 commented on Feb 16, 2022 • 6 new comments

• Micro-optimisations for matmul

  #64387 commented on Feb 15, 2022 • 6 new comments

• ignore sccache server error

  #71758 commented on Feb 12, 2022 • 6 new comments

• [pkg] add generic ZipFile Reader/Writer

  #72237 commented on Feb 11, 2022 • 6 new comments

• quant: add QAT fused Linear-Bn1d [1/x]: prepared module

  #72431 commented on Feb 14, 2022 • 6 new comments

• [RFC] enable oneMKL&oneDNN on-demands verbose functinality

  #63212 commented on Feb 15, 2022 • 5 new comments

• Added check for unsupported dispatch key in codegen

  #67961 commented on Feb 11, 2022 • 5 new comments

• Automated submodule update: kineto

  #68177 commented on Feb 15, 2022 • 5 new comments

• add native quantization support for pixel_shuffle

  #68328 commented on Feb 12, 2022 • 5 new comments

• Update persons of interest for ONNX

  #72072 commented on Feb 15, 2022 • 5 new comments

• derivative for channel_shuffle is not implemented

  #67240 commented on Feb 10, 2022 • 4 new comments

• Idiom for Tensor subclasses that are also Parameters

  #65018 commented on Feb 11, 2022 • 4 new comments

• FP32 depthwise convolution is slow in GPU

  #18631 commented on Feb 13, 2022 • 4 new comments

• [Feature request] Let DistributedSampler take a Sampler as input

  #23430 commented on Feb 14, 2022 • 4 new comments

• `torch.sspaddmm` should broadcast the input tensor

  #69348 commented on Feb 14, 2022 • 4 new comments

• Write explicit meta-kernels for `normal`

  #70089 commented on Feb 11, 2022 • 4 new comments

• Fix empty tensor handling in RReLU

  #70496 commented on Feb 15, 2022 • 4 new comments

• Fix for norm() return `nan` when computing for a tensor of zeros.

  #71372 commented on Feb 16, 2022 • 4 new comments

• masked_softmax_backward

  #71502 commented on Feb 10, 2022 • 4 new comments

• Introduce an environment variable to change c10 log level

  #71746 commented on Feb 12, 2022 • 4 new comments

• [tensorexpre] Add typecast when src and dest buf types are different in PlacementAllocate

  #71934 commented on Feb 11, 2022 • 4 new comments

• release notes script changes

  #72154 commented on Feb 10, 2022 • 4 new comments

• saving progress so far on categorizing

  #72178 commented on Feb 10, 2022 • 4 new comments

• [ONNX] List of files to consider for mergebot onnx rule

  #72297 commented on Feb 15, 2022 • 4 new comments

• Add BUILD_LAZY_CUDA_LINALG option

  #72306 commented on Feb 14, 2022 • 4 new comments

• Change API type for `Tensor[]`: `ITensorList`.

  #72414 commented on Feb 10, 2022 • 4 new comments

• Enable win-arm64

  #72424 commented on Feb 15, 2022 • 4 new comments

• CUSOLVER_STATUS_EXECUTION_FAILED when using `torch.linalg.solve`

  #70122 commented on Feb 10, 2022 • 3 new comments

• Support matrix operations between complex and real tensors

  #61909 commented on Feb 10, 2022 • 3 new comments

• ComplexHalf support

  #71680 commented on Feb 10, 2022 • 3 new comments

• Implement Truly Parallel Ensemble Layers

  #54147 commented on Feb 13, 2022 • 3 new comments

• RFC: torch.vmap

  #42368 commented on Feb 15, 2022 • 3 new comments

• `test_neg_view_linalg_det_singular_cuda_float64` fails with Illegal Memory Access failure for singular matrices

  #72204 commented on Feb 15, 2022 • 3 new comments

• `TestProfilerCUDA. test_mem_leak` failing for CUDA 11.5 on Linux

  #69023 commented on Feb 15, 2022 • 3 new comments

• [LTC] add shape info to types in TS lowering

  #62377 commented on Feb 14, 2022 • 3 new comments

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

  #63230 commented on Feb 11, 2022 • 3 new comments

• Add JIT graph fuser for oneDNN Graph API (Preview4)

  #68111 commented on Feb 11, 2022 • 3 new comments

• [ONNX] Support optional type

  #68793 commented on Feb 16, 2022 • 3 new comments

• Fix a typo: add a missing space

  #70086 commented on Feb 11, 2022 • 3 new comments

• Unify checks for `normal`

  #70087 commented on Feb 11, 2022 • 3 new comments

• Disable torch.complex32 dtype

  #70606 commented on Feb 15, 2022 • 3 new comments

• Support masked sum on sparse COO tensors.

  #71239 commented on Feb 15, 2022 • 3 new comments

• Extrapolated on equiv between linalg @ and solve

  #71769 commented on Feb 15, 2022 • 3 new comments

• Add scalar support for `logsumexp`.

  #71870 commented on Feb 11, 2022 • 3 new comments

• Import packaging.version in torch_version, if available

  #71902 commented on Feb 12, 2022 • 3 new comments

• Add Einsum to Scalar type analysis ONNX pass

  #71916 commented on Feb 11, 2022 • 3 new comments

• Fix a few aliasing schemas

  #72097 commented on Feb 10, 2022 • 3 new comments

• Port `amax` to structured kernel

  #72124 commented on Feb 14, 2022 • 3 new comments

• [shard] Extensible ShardingSpec

  #72130 commented on Feb 11, 2022 • 3 new comments

• [pytorch] use cublas lt interface for bias fusion (#71200)

  #72148 commented on Feb 16, 2022 • 3 new comments

• Testing of masked reductions on mixed layout inputs.

  #72398 commented on Feb 15, 2022 • 3 new comments

• Enable FC tests

  #72479 commented on Feb 11, 2022 • 3 new comments

• Pillow v9.0.0 conda-forge package causes "OSError: [WinError 182] [...] error loading ‘caffe2_detectron_ops.dll'"

  #72293 commented on Feb 9, 2022 • 2 new comments

• NotImplementedError in torch.distributions

  #72258 commented on Feb 9, 2022 • 2 new comments

• Make `torch.Tensor.view` support autograd for appropriate cases

  #68208 commented on Feb 9, 2022 • 2 new comments

• DISABLED test_linalg_lstsq_cpu_float64 (__main__.TestLinalgCPU)

  #72281 commented on Feb 9, 2022 • 2 new comments

• DISABLED test_ind_worker_queue (__main__.TestIndividualWorkerQueue)

  #68643 commented on Feb 9, 2022 • 2 new comments

• Better (optional) tensor list handling in native kernels

  #66328 commented on Feb 10, 2022 • 2 new comments

• DataLoader tests are quite flaky

  #71187 commented on Feb 10, 2022 • 2 new comments

• NCCL Backend does not support ComplexFloat data type

  #71613 commented on Feb 10, 2022 • 2 new comments

• Forward-over-reverse gradgradcheck failing for `trapz` and `trapezoid` on `meta` device

  #69868 commented on Feb 10, 2022 • 2 new comments

• c10d Python memory allocated without holding the GIL

  #26475 commented on Feb 11, 2022 • 2 new comments

• upstream `apex.optimizers.FusedAdam` to replace `torch.optim.AdamW`

  #71274 commented on Feb 12, 2022 • 2 new comments

• download speed

  #17023 commented on Feb 12, 2022 • 2 new comments

• `torch._sources.normalize_source_lines()` and consequently `torch.jit.script` fail with lambda functions

  #71912 commented on Feb 14, 2022 • 2 new comments

• `torch.broadcast_shapes` should not handle shape with negative dimension

  #68957 commented on Feb 14, 2022 • 2 new comments

• Pruning off NaN values in the gradient graph still produces NaN gradients.

  #23156 commented on Feb 14, 2022 • 2 new comments

• [Github] Exporting “torch.stft“ and” torch.istft” function to ONNX is not supported.

  #65666 commented on Feb 14, 2022 • 2 new comments

• TestSpectralOps tests incompatible with updated librosa 0.9.0

  #72550 commented on Feb 15, 2022 • 2 new comments

• No dtype check for zero sparse tensor!

  #69665 commented on Feb 15, 2022 • 2 new comments

• einsum "jk,ijkl->il" is ~16x slower than numpy

  #57121 commented on Feb 15, 2022 • 2 new comments

• ImportError: cannot import name '_VF' from partially initialized module 'torch'

  #72105 commented on Feb 15, 2022 • 2 new comments

• torch.meshgrid is divergent from np.meshgrid

  #50276 commented on Feb 15, 2022 • 2 new comments

• CompositeImplicitAutograd operators should not perform operations that do not dispatch

  #61669 commented on Feb 15, 2022 • 2 new comments

• add channels last support for PixelShuffle and PixelUnshuffle

  #50573 commented on Feb 10, 2022 • 2 new comments

• Enable fp32/bf16 PRelu forward and backward in MkldnnCPU path

  #60427 commented on Feb 14, 2022 • 2 new comments

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

  #61897 commented on Feb 15, 2022 • 2 new comments

• [Draft][ONNX] STFT and ISTFT

  #62412 commented on Feb 14, 2022 • 2 new comments

• Support `torch.linalg.trace`

  #62714 commented on Feb 15, 2022 • 2 new comments

• initial implementation of PartitionedDataset

  #65995 commented on Feb 15, 2022 • 2 new comments

• [tensorexpr] Memory planning: add pre-allocation for reused intermediate buffers

  #66967 commented on Feb 10, 2022 • 2 new comments

• Virtualize `<type>Storage` classes

  #66970 commented on Feb 15, 2022 • 2 new comments

• [caffe2] Remove OperatorBase::newstyle_outputs_

  #67093 commented on Feb 14, 2022 • 2 new comments

• Remove native_functions.yaml dependency from CUDA distributions

  #67875 commented on Feb 15, 2022 • 2 new comments

• [WIP] Dynamic shapes for LTC POC

  #68078 commented on Feb 11, 2022 • 2 new comments

• Use cub::DeviceSelect::UniqueByKey for EmbeddingBackward

  #68376 commented on Feb 10, 2022 • 2 new comments

• improve multi-core performance of qavg_pool2d

  #69517 commented on Feb 15, 2022 • 2 new comments

• improve multi-core performance of qmax_pool2d

  #69598 commented on Feb 15, 2022 • 2 new comments

• improve multi-core performance of qbatch_norm2d

  #69599 commented on Feb 15, 2022 • 2 new comments

• Add ONEDNN quantization backend

  #69820 commented on Feb 15, 2022 • 2 new comments

• expanded weights faster rules

  #70141 commented on Feb 14, 2022 • 2 new comments

• [shard] use scatter in shard_parameter API

  #72160 commented on Feb 10, 2022 • 2 new comments

• remove torch from package exporter

  #72242 commented on Feb 11, 2022 • 2 new comments

• primTorch aliases

  #72289 commented on Feb 11, 2022 • 2 new comments

• [PyTorch] MHA: add test for transform_bias_rescale_qkv

  #72464 commented on Feb 15, 2022 • 2 new comments

• Support `eye` for sparse layouts (CSR and COO)

  #72497 commented on Feb 10, 2022 • 2 new comments

• Using external memory allocator with PyTorch

  #43144 commented on Feb 9, 2022 • 1 new comment

• INTERNAL ASSERT FAILED at "../aten/src/ATen/MapAllocator.cpp":263

  #67864 commented on Feb 9, 2022 • 1 new comment

• Python version is 3.7.11 in latest pytorch docker image

  #68901 commented on Feb 9, 2022 • 1 new comment

• Output shape of ONNX exported "index_select" function is wrong on pytorch 1.10.0

  #68003 commented on Feb 9, 2022 • 1 new comment

• To have single cuda context across multiple processes

  #42080 commented on Feb 9, 2022 • 1 new comment

• Can't Export Pytorch Model to ONNX

  #72329 commented on Feb 9, 2022 • 1 new comment

• DistributedDataParallel creates too many threads

  #72107 commented on Feb 9, 2022 • 1 new comment

• Add dim to view_as_real and view_as_complex

  #71066 commented on Feb 9, 2022 • 1 new comment

• [typing] new_ones has wrong signature

  #63291 commented on Feb 9, 2022 • 1 new comment

• make pytorch support different hardware acceleratioin configuration

  #71364 commented on Feb 9, 2022 • 1 new comment

• DataLoader: Segmentation Fault (core dumped)

  #31758 commented on Feb 9, 2022 • 1 new comment

• Enforce quotas on CI users

  #72555 commented on Feb 9, 2022 • 1 new comment

• Incorrect ONNX Export for Unidirectional Broadcasting in PReLU

  #70570 commented on Feb 10, 2022 • 1 new comment

• Update weight initialisations to current best practices

  #18182 commented on Feb 10, 2022 • 1 new comment

• `pip==22.0` breaks installation process

  #72045 commented on Feb 10, 2022 • 1 new comment

• When modified the model python file, the pytorch will raise the KeyError of this file

  #43120 commented on Feb 10, 2022 • 1 new comment

• Make to_dense() a no-op for non-sparse Tensors

  #59958 commented on Feb 10, 2022 • 1 new comment

• torch.fft.rfft2 doesn't support half dtype

  #70664 commented on Feb 10, 2022 • 1 new comment

• Well documented, safe method to deserialize model parameters from untrusted sources

  #6107 commented on Feb 10, 2022 • 1 new comment

• Transpose of a sparse tensor is not a view operation

  #68699 commented on Feb 10, 2022 • 1 new comment

• Add support for comparing meta tensors

  #68592 commented on Feb 10, 2022 • 1 new comment

• [TorchScript] Can we support type refinement by Python type annotation?

  #55211 commented on Feb 10, 2022 • 1 new comment

• Decouple `TensorIteratorBase` output from structured kernel outputs.

  #70678 commented on Feb 10, 2022 • 1 new comment

• Mechanism for Tensor subclasses to "disable autograd"

  #72179 commented on Feb 10, 2022 • 1 new comment

• Integrate Libtorch into Unreal Engine 4: _ivalue_INTERNAL ASSERT FAILED

  #69425 commented on Feb 11, 2022 • 1 new comment

• [RFC] Model Sharding for distributed training

  #55207 commented on Feb 11, 2022 • 1 new comment

• [Perf request] Make index_select on sparse COO tensors as fast as that from rusty1s/pytorch_sparse (1000x)

  #72212 commented on Feb 11, 2022 • 1 new comment

• Windows "smoke tests" are not very "smoky"

  #62298 commented on Feb 11, 2022 • 1 new comment

• halt or exit function implementation

  #72099 commented on Feb 13, 2022 • 1 new comment

• Error building docker image: No module named 'typing_extensions'

  #55437 commented on Feb 13, 2022 • 1 new comment

• Lack of type check in `nn.functional` APIs

  #71211 commented on Feb 13, 2022 • 1 new comment

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

  #43259 commented on Feb 13, 2022 • 1 new comment

• SEGFAULT on "import torch"

  #70245 commented on Feb 13, 2022 • 1 new comment

• finfo operator not bound into JIT

  #25661 commented on Feb 13, 2022 • 1 new comment

• Helping test example code blocks in the docs

  #41499 commented on Feb 13, 2022 • 1 new comment

• Bug about distributed launch

  #72034 commented on Feb 14, 2022 • 1 new comment

• nn.Bilinear cannot be used inside nn.Sequential

  #37092 commented on Feb 14, 2022 • 1 new comment

• `torch.hstack` should raise an error when tensor is 0 dimensional

  #69408 commented on Feb 14, 2022 • 1 new comment

• Advanced indexing gradient is extremely slow when there are many duplicate indices

  #41162 commented on Feb 14, 2022 • 1 new comment

• Support for Multi-Categorical in torch.distributions

  #43250 commented on Feb 14, 2022 • 1 new comment

• Keys of a `ModuleDict` cannot have the same name as existing `ModuleDict` class attributes.

  #71203 commented on Feb 14, 2022 • 1 new comment

• Add ONNX model generator with IR 0.0.3 for TensorRT

  #31249 commented on Feb 14, 2022 • 1 new comment

• the results when using opset10 and when using opset11 are different.

  #32267 commented on Feb 14, 2022 • 1 new comment

• Runtime Error when multipling complex32 with float32

  #69209 commented on Feb 15, 2022 • 1 new comment

• `torch.scatter` will return random value when `input` is empty tensor

  #71059 commented on Feb 15, 2022 • 1 new comment

• `torch.nn.{Constant,Zero}Pad` unexpectedly fail

  #71078 commented on Feb 15, 2022 • 1 new comment

• `torch.sparse.addmm` does not check the dtype of zero tensor

  #69518 commented on Feb 15, 2022 • 1 new comment

• `torch.median` will return -2147483648 when input is an empty tensor

  #71636 commented on Feb 15, 2022 • 1 new comment

• RuntimeError: Global alloc not supported yet in TorchScript

  #69078 commented on Feb 15, 2022 • 1 new comment

• Unknown builtin op: torchvision::nms when loading scripted FasterRCNN

  #48932 commented on Feb 15, 2022 • 1 new comment

• [Feature Request] Sparse-Dense elementwise Multiplication

  #3158 commented on Feb 15, 2022 • 1 new comment

• Add Python-3.10 support

  #66424 commented on Feb 15, 2022 • 1 new comment

• Huggingface model not being torch scriptable

  #51376 commented on Feb 15, 2022 • 1 new comment

• CPU Capability is being reported as AVX512 even if PyTorch is built without AVX512 ATen kernels

  #66712 commented on Feb 15, 2022 • 1 new comment

• `torch.set_default_dtype` will crash with `complex` dtype

  #71674 commented on Feb 15, 2022 • 1 new comment

• JIT fuser throws compilation error (1.10 regression)

  #67362 commented on Feb 15, 2022 • 1 new comment

• Summing over large tensors in InferenceMode is throwing a RuntimeError for tensors with more than 2**15 elements.

  #60539 commented on Feb 15, 2022 • 1 new comment

• Follow Up on the usage for cudatoolkit across pytorch projects

  #69691 commented on Feb 15, 2022 • 1 new comment

• CUDNN 8.3.0 static library names have changed

  #67905 commented on Feb 15, 2022 • 1 new comment

• Compilation of <torch/extension.h> error on Windows CUDA 11.5

  #69460 commented on Feb 15, 2022 • 1 new comment

• PyTorch master failed to build with glog 0.5.0.

  #58054 commented on Feb 15, 2022 • 1 new comment

• undefined symbol curandCreateGenerator for torch extensions

  #69666 commented on Feb 15, 2022 • 1 new comment

• Errors when coercing complex numbers of various sizes

  #50098 commented on Feb 15, 2022 • 1 new comment

• DISABLED test_fd_sharing (__main__.TestMultiprocessing)

  #68831 commented on Feb 15, 2022 • 1 new comment

• Need to fix nn.quantized.Linear API

  #36703 commented on Feb 16, 2022 • 1 new comment

• Add SoftplusTransform

  #52300 commented on Feb 15, 2022 • 1 new comment

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

  #60755 commented on Feb 10, 2022 • 1 new comment

• small grammar/syntax fixes

  #61423 commented on Feb 12, 2022 • 1 new comment

• add out variant of cudnn fused conv

  #61802 commented on Feb 12, 2022 • 1 new comment

• refactor layer_norm and expose out variant

  #61821 commented on Feb 12, 2022 • 1 new comment

• implement out variant for softmax

  #61823 commented on Feb 12, 2022 • 1 new comment

• out variant for cudnn batch norm

  #61824 commented on Feb 12, 2022 • 1 new comment

• hacky conditional recompile depending on generated code

  #61915 commented on Feb 12, 2022 • 1 new comment

• start iwyu project

  #62010 commented on Feb 12, 2022 • 1 new comment

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

  #62084 commented on Feb 14, 2022 • 1 new comment

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

  #62995 commented on Feb 12, 2022 • 1 new comment

• Optimize PReLU (float32) and enable PReLU BFloat16 support in CPU path

  #63634 commented on Feb 9, 2022 • 1 new comment

• [JIT][WIP] memorization memory planning

  #63873 commented on Feb 12, 2022 • 1 new comment

• Simplify OptionalArrayRef

  #64092 commented on Feb 14, 2022 • 1 new comment

• [JIT] memory planning base with naive strategy

  #64347 commented on Feb 12, 2022 • 1 new comment

• [JIT] linear scan memory planning strategy

  #64348 commented on Feb 12, 2022 • 1 new comment

• [JIT] greedy by size memory planning strategy

  #64349 commented on Feb 12, 2022 • 1 new comment

• [JIT] greedy by breadth memory planning strategy

  #64350 commented on Feb 12, 2022 • 1 new comment

• [Core] Update Exception.h

  #64553 commented on Feb 12, 2022 • 1 new comment

• [DRAFT] New reduction kernel

  #65228 commented on Feb 14, 2022 • 1 new comment

• get aten memory ops in the graph

  #65445 commented on Feb 12, 2022 • 1 new comment

• ufunc codegen

  #65851 commented on Feb 15, 2022 • 1 new comment

• Mild refactor of native_functions.yaml dispatch parsing

  #66109 commented on Feb 15, 2022 • 1 new comment

• When loading model on a big-endian platform, do byteswap

  #66613 commented on Feb 10, 2022 • 1 new comment

• [caffe2] Remove IValue include from operator.h

  #67094 commented on Feb 14, 2022 • 1 new comment

• Create BoxedKernel as a subset of KernelFunction

  #67095 commented on Feb 14, 2022 • 1 new comment

• [caffe2] Export operators to c10 without including ATen/Tensor.h

  #67096 commented on Feb 14, 2022 • 1 new comment

• Fix crash on unload torch cpu dll

  #67632 commented on Feb 10, 2022 • 1 new comment

• Don't build nvfuser benchmarks by default

  #67857 commented on Feb 9, 2022 • 1 new comment

• [nnc][quantization] Quantized.add to native_functions yaml

  #68313 commented on Feb 9, 2022 • 1 new comment

• [JIT][Symbolic Shape Analysis] expose op shape functions

  #68564 commented on Feb 12, 2022 • 1 new comment

• [JIT] jemalloc experiments

  #68904 commented on Feb 12, 2022 • 1 new comment

• codegen: do not generate code for dispatch_namespaced_definitions

  #69074 commented on Feb 16, 2022 • 1 new comment

• improve multi-core performance of qupsample_nearest2d

  #69600 commented on Feb 9, 2022 • 1 new comment

• improve multi-core performance of qupsample_bilinear2d

  #69601 commented on Feb 9, 2022 • 1 new comment

• improve qcat_nhwc performance on both multi-core and single-core

  #69667 commented on Feb 9, 2022 • 1 new comment

• Jetson tx2 launchbound fixes

  #69728 commented on Feb 14, 2022 • 1 new comment

• Add example of incorrect results from lazy tensor, to debug/fix

  #69762 commented on Feb 9, 2022 • 1 new comment

• [not for land] dbr quant: prototyping script execution order annotation

  #69850 commented on Feb 14, 2022 • 1 new comment

• Refactor ATen node on ONNX converter for consistency and debugability

  #69906 commented on Feb 9, 2022 • 1 new comment

• [PyTorch][Static Runtime] Fix to() w/dtype bool

  #69930 commented on Feb 14, 2022 • 1 new comment

• Add mkl implementation for exponential on CPU

  #69967 commented on Feb 14, 2022 • 1 new comment

• make DispatchKeySets static singletons

  #69986 commented on Feb 14, 2022 • 1 new comment

• Expand composite compliance tests

  #69990 commented on Feb 14, 2022 • 1 new comment

• lazy bench in subprocess

  #70026 commented on Feb 14, 2022 • 1 new comment

• [WIP] throw when sparse csr shallow_copy.* is called

  #70077 commented on Feb 15, 2022 • 1 new comment

• Bug fix: allow std 0 in the meta definition of `normal_`

  #70085 commented on Feb 11, 2022 • 1 new comment

• Changes to support input sequence ID tracking

  #70264 commented on Feb 11, 2022 • 1 new comment

• [WIP] Implementation of Inverse Wishart distribution

  #70275 commented on Feb 16, 2022 • 1 new comment

• Docs: build with Sphinx 4

  #70309 commented on Feb 14, 2022 • 1 new comment

• CPU Kernel: Use per-operator headers

  #71137 commented on Feb 14, 2022 • 1 new comment

• Remove `SHARED` for `caffe2_detectron_ops`

  #71273 commented on Feb 14, 2022 • 1 new comment

• [Profiler] Optimize `reportMemoryUsage`

  #71538 commented on Feb 15, 2022 • 1 new comment

• DOC: Initialize numpy compatibility note

  #71688 commented on Feb 11, 2022 • 1 new comment

• Performance and memory improvements to batched torch.linalg.solve (2nd attempt)

  #71756 commented on Feb 15, 2022 • 1 new comment

• [fix] Contiguity of `torch.ravel`!

  #71771 commented on Feb 9, 2022 • 1 new comment

• Implement numpy-like function `torch.block`

  #71859 commented on Feb 15, 2022 • 1 new comment

• qcat: use direct memcpy when all the inputs and output share the same scale and zero_point

  #71903 commented on Feb 9, 2022 • 1 new comment

• [ROCm] Enable topk operator for bfloat16 dtype

  #71913 commented on Feb 14, 2022 • 1 new comment

• remove torch from package importer and add fbcode shims for BC

  #72246 commented on Feb 11, 2022 • 1 new comment

• #71806 Remove references to master branch

  #72363 commented on Feb 11, 2022 • 1 new comment

• Exit early if the type pointer is null

  #72406 commented on Feb 11, 2022 • 1 new comment

• Strip GCC5 stuff from PyTorch

  #72438 commented on Feb 10, 2022 • 1 new comment

• [ONNX] Add pixel_unshuffle support in opset 9

  #72449 commented on Feb 14, 2022 • 1 new comment

• Quantized LSTM/GRU: Remove legacy API support

  #72522 commented on Feb 9, 2022 • 1 new comment

• Small typing improvements

  #72526 commented on Feb 13, 2022 • 1 new comment

• [Do not merge] Dummy commit

  #72557 commented on Feb 11, 2022 • 1 new comment

• add reduce package function to storage.py in order to deperecate presistent_id in OSS

  #72563 commented on Feb 11, 2022 • 1 new comment

• [DDP] Static graph should print out unused parameters when they are detected.

  #68833 commented on Feb 9, 2022 • 0 new comments

• `Only 2D, 3D, 4D, 5D padding with non-constant padding are supported` on 3D array

  #72521 commented on Feb 9, 2022 • 0 new comments

• Add _reduce_scatter_base and _allgather_base to processGroupWrapper

  #66329 commented on Feb 9, 2022 • 0 new comments

• Memory leak in distributions.multivariate_normal.MultivariateNormal

  #71149 commented on Feb 10, 2022 • 0 new comments

• Feature Request: Implement `torch.sparse.spdiags` and `torch.sparse.diags`

  #70926 commented on Feb 10, 2022 • 0 new comments

• `make_tensor` tracking issue

  #61758 commented on Feb 10, 2022 • 0 new comments

• Support sparse inputs for torch.block_diag

  #31942 commented on Feb 10, 2022 • 0 new comments

• KL divergence between two Continuous Bernoulli is negative

  #72525 commented on Feb 10, 2022 • 0 new comments

• Define and Put in place process to regularly update Grafana instance in order to avoid outages

  #70250 commented on Feb 10, 2022 • 0 new comments

• Support gradient accumulation without no_sync context manager in FSDP

  #72185 commented on Feb 10, 2022 • 0 new comments

• `eye` should support other diagonals than the main one

  #70910 commented on Feb 10, 2022 • 0 new comments

• [JIT][Autocasting] Add autocasting & constant propagation as part of freezing

  #72295 commented on Feb 10, 2022 • 0 new comments

• Revise `gather()` method of `ShardedTensor`

  #66187 commented on Feb 10, 2022 • 0 new comments

• JIT / TorchScript docs missing any mention of typing.cast

  #69814 commented on Feb 11, 2022 • 0 new comments

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

  #47953 commented on Feb 11, 2022 • 0 new comments

• last_epoch parameter of CyclicLR and OneCycleLR is not the number of epochs

  #69753 commented on Feb 11, 2022 • 0 new comments

• PyTorch CI Stability Action Tracker

  #65439 commented on Feb 11, 2022 • 0 new comments

• JIT: Support for `torch.autograd.functional.jacobian` in TorchScript

  #69741 commented on Feb 11, 2022 • 0 new comments

• Incomplete error message at tensor indexing (when indexing with set)

  #69800 commented on Feb 12, 2022 • 0 new comments

• [FSDP] Enable tests for Gloo backend

  #69891 commented on Feb 12, 2022 • 0 new comments

• [INSTANCENORM] Instance Normalization ignores track_running_stats=True when exporting to ONNX.

  #72057 commented on Feb 12, 2022 • 0 new comments

• Conversion error from pytorch model to libtorch model

  #69830 commented on Feb 13, 2022 • 0 new comments

• The DataLoader can't work in Apple Silicon.

  #70344 commented on Feb 13, 2022 • 0 new comments

• Implement aten::equal for sparse tensors

  #69786 commented on Feb 13, 2022 • 0 new comments

• torchscript does not work with `SyncBatchNorm` layers

  #69972 commented on Feb 13, 2022 • 0 new comments

• torch.nn.DataParallel caused inference failure with cpu set as device on NV machine

  #69966 commented on Feb 13, 2022 • 0 new comments

• [JIT] torch.exp roughly 20 times slower in TorchScript vs. PyTorch

  #69931 commented on Feb 13, 2022 • 0 new comments

• torch.nn.functional.ctc_loss with invalid input produce NaN or infinity gradient, while the batch entries are fine

  #69912 commented on Feb 13, 2022 • 0 new comments

• FeatureAlphaDropout doesn't drop channels for (C, D, H, W)

  #69960 commented on Feb 13, 2022 • 0 new comments

• Re-raise the exception when the `forward` of a parametrization raises

  #72452 commented on Feb 14, 2022 • 0 new comments

• Question about collect tensor in distributed dataparallel?

  #70037 commented on Feb 14, 2022 • 0 new comments

• [JIT] Cannot `jit.export` a `@staticmethod`

  #70047 commented on Feb 14, 2022 • 0 new comments

• Docs for torch.nn.MSELoss are confusing

  #69984 commented on Feb 14, 2022 • 0 new comments

• [shard] determine if there's device overlap in pg

  #70053 commented on Feb 14, 2022 • 0 new comments

• Numerical-reproducibility issue in torch.matmul

  #60234 commented on Feb 14, 2022 • 0 new comments

• input.dim() == 4 INTERNAL ASSERT FAILED mkldnn/Pooling.cpp:201

  #72418 commented on Feb 14, 2022 • 0 new comments

• [RFC] Activation Checkpoint API improvements

  #70060 commented on Feb 14, 2022 • 0 new comments

• [feature request] [onnx] Export torch.stft as Conv1d till ONNX supports stft op

  #31317 commented on Feb 14, 2022 • 0 new comments

• Tool for detecting inefficent striding for nn.Conv2d

  #70065 commented on Feb 14, 2022 • 0 new comments

• torch jit script segm fault

  #70102 commented on Feb 15, 2022 • 0 new comments

• Question: what is "Parameter indices"?

  #70099 commented on Feb 15, 2022 • 0 new comments

• Unknown builtin op: spconv::get_indice_pairs

  #70058 commented on Feb 15, 2022 • 0 new comments

• Tensor transfer between gpus doesnt work

  #70048 commented on Feb 15, 2022 • 0 new comments

• Add support for a `default` arg in `ModuleDict.pop`

  #70095 commented on Feb 15, 2022 • 0 new comments

• Performance improvement in Autograd Forward AD using ZeroTensors

  #69687 commented on Feb 15, 2022 • 0 new comments

• Clarify test dependencies (e.g., into a test-requirements.txt file)

  #72556 commented on Feb 15, 2022 • 0 new comments

• DISABLED test_data_parallel_module_cuda_float16 (__main__.TestDataParallelDeviceTypeCUDA)

  #71770 commented on Feb 15, 2022 • 0 new comments

• channels_last/channels_last_3d memory format not supported for some modules on ROCm that should be supported on CUDA

  #70125 commented on Feb 15, 2022 • 0 new comments

• Feature Request: dim parameter of torch.nn.functional.normalize should accept tuples

  #70100 commented on Feb 15, 2022 • 0 new comments

• Bazel build documentation

  #56940 commented on Feb 15, 2022 • 0 new comments

• error in bazel build //...

  #46631 commented on Feb 15, 2022 • 0 new comments

• Bazel target all_tests improperly reports failures on CPU-only (non-CUDA) build

  #66565 commented on Feb 15, 2022 • 0 new comments

• bazel build warning: Artifact 'torch/csrc/api/include/torch/version.h' is duplicated

  #63972 commented on Feb 15, 2022 • 0 new comments

• PyTorch Profiler built with Bazel doesn't produce GPU trace

  #67349 commented on Feb 15, 2022 • 0 new comments

• scatter_reduce

  #63780 commented on Feb 15, 2022 • 0 new comments

• Make quantized::prepack_fp16 op just do prepacking

  #44676 commented on Feb 16, 2022 • 0 new comments

• [DO NOT DELETE OR MERGE] Nightly release

  #26921 commented on Feb 15, 2022 • 0 new comments

• [DO NOT DELETE OR MERGE] Postnightly release

  #27167 commented on Feb 15, 2022 • 0 new comments

• [distributed] add flock-based wrapper around built-in print

  #61803 commented on Feb 15, 2022 • 0 new comments

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

  #63289 commented on Feb 13, 2022 • 0 new comments

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

  #63932 commented on Feb 11, 2022 • 0 new comments

• [ROCm] Update the handling of hipRuntimeGetVersion()

  #64471 commented on Feb 15, 2022 • 0 new comments

• optimize sort

  #64808 commented on Feb 15, 2022 • 0 new comments

• Adds cudaMallocAsync as an alternative backend for the CUDA allocator

  #65365 commented on Feb 13, 2022 • 0 new comments

• [Foreach Reduction] Use `OpMathType` tensor for intermediate results

  #68107 commented on Feb 15, 2022 • 0 new comments

• no longer coalesce sparse COO tensors before comparison

  #69751 commented on Feb 16, 2022 • 0 new comments

• subtests to skip on jetson CI

  #70648 commented on Feb 15, 2022 • 0 new comments

• remove references to dataclasses

  #71086 commented on Feb 15, 2022 • 0 new comments

• CUDA Kernels: Use per-operator headers (1/4)

  #71212 commented on Feb 14, 2022 • 0 new comments

• CUDA Kernels: Use per-operator headers (2/4)

  #71213 commented on Feb 14, 2022 • 0 new comments

• CUDA Kernels: Use per-operator headers (3/4)

  #71214 commented on Feb 14, 2022 • 0 new comments

• CUDA Kernels: Use per-operator headers (4/4)

  #71215 commented on Feb 14, 2022 • 0 new comments

• cuDNN/miopen: Use per-operator headers

  #71216 commented on Feb 14, 2022 • 0 new comments

• QuantizedCPU: Use per-operator headers

  #71217 commented on Feb 14, 2022 • 0 new comments

• [WIP][JIT] OpInfo tests for nvfuser

  #71299 commented on Feb 15, 2022 • 0 new comments

• Replace `get_all_` type macros with the ATen dispatch macros.

  #71561 commented on Feb 15, 2022 • 0 new comments

• Add debug_name attribute to JIT

  #71684 commented on Feb 13, 2022 • 0 new comments

• Optimize grid sample 3d

  #71759 commented on Feb 14, 2022 • 0 new comments

• Testing circleci builds

  #71882 commented on Feb 9, 2022 • 0 new comments

• DO NOT MERGE, testing minimal macOS builds

  #71893 commented on Feb 10, 2022 • 0 new comments

• Set pull_request checkout to head sha

  #71974 commented on Feb 16, 2022 • 0 new comments

• torch.hub security improvement

  #72060 commented on Feb 14, 2022 • 0 new comments

• DO NOT MERGE: Status of lts branch with binary build

  #72224 commented on Feb 10, 2022 • 0 new comments

• DO NOT MERGE: Status of lts branch with build

  #72228 commented on Feb 11, 2022 • 0 new comments

• changed tests to reflect generic zip file reader/writer

  #72247 commented on Feb 9, 2022 • 0 new comments

• [Quant][bc-breaking][devs] Combined dispatch registration for max_pool2d & quantized_max_pool2d

  #72353 commented on Feb 15, 2022 • 0 new comments

• DO NOT MERGE: test_ltsm issue fix

  #72453 commented on Feb 15, 2022 • 0 new comments

• [PyTorch] MHA: fix contiguity assumption in transform_bias_rescale_qkv

  #72465 commented on Feb 15, 2022 • 0 new comments

• [Quant][core][devs] Removed support for non-quantized tensors in reflection_pad1d_cpu

  #72485 commented on Feb 15, 2022 • 0 new comments

• [Perf] Reduce unnecessary ref count bumps

  #72523 commented on Feb 15, 2022 • 0 new comments

• update onnx to 1.11.0

  #72531 commented on Feb 16, 2022 • 0 new comments

• [PTE][Flatbuffer] Save JIT files

  #72535 commented on Feb 15, 2022 • 0 new comments

• [PyTorch] IValue(const c10::Scalar&) improvements

  #72551 commented on Feb 10, 2022 • 0 new comments

• Add naive native FFN in aten

  #72564 commented on Feb 9, 2022 • 0 new comments

• Add BatchNormalization to ONNX Scalar type analysis pass to support fp16

  #72565 commented on Feb 9, 2022 • 0 new comments