Insights: pytorch/pytorch

12 Pull requests merged by 8 people

• [v1.12.0] Fix non-reentrant hooks based checkpointing

  #79490 merged Jun 17, 2022

• Revert behavior of Dropout2d on 3D inputs to 1D channel-wise dropout behavior & warn

  #79611 merged Jun 17, 2022

• Add Dropout1d module

  #79610 merged Jun 17, 2022

• [Release 1.12][DataLoader] Fix the world_size when distributed sharding MapDataPipe (#79524)

  #79550 merged Jun 15, 2022

• Cherry pick tf32docs (#79537)

  #79539 merged Jun 15, 2022

• Link LazyLinalg with cusolver statically when needed (#79324)

  #79522 merged Jun 14, 2022

• Add docs for Python Registration

  #79481 merged Jun 14, 2022

• [CUDA graphs] Allows Adam and AdamW to be capture-safe (#77862)

  #79472 merged Jun 14, 2022

• cherry-pick

  #79455 merged Jun 13, 2022

• [CI] Turn flaky test signal to green (#79220) (#79220)

  #79416 merged Jun 13, 2022

• Removing cublas static linking (#79280)

  #79417 merged Jun 13, 2022

• Install NDK 21 after GitHub update (#79024) (#79024)

  #79429 merged Jun 13, 2022

202 Pull requests opened by 100 people

• [ONNX] Fix onnx logical functions' dtype

  #79339 opened Jun 11, 2022

• [meta] local_scalar_dense

  #79340 opened Jun 11, 2022

• fix some errors of clang-tidy

  #79342 opened Jun 11, 2022

• Add validation for mapper function in datapipes with `input_col`

  #79344 opened Jun 11, 2022

• [bazel] Add --config=shell for easier debugging

  #79350 opened Jun 11, 2022

• [bazel] Plumb all cc tests through rules.cc_test

  #79353 opened Jun 11, 2022

• compatible with glog-0.6.0

  #79360 opened Jun 12, 2022

• fix submodule imports by importing functions directly

  #79368 opened Jun 12, 2022

• [ONNX] Fix `any` and `all` outputs' shape

  #79371 opened Jun 12, 2022

• GenCode: Only gen lazy ts backend if enabled

  #79377 opened Jun 12, 2022

• Optimize GELU BFloat16 Impl in CPU path

  #79378 opened Jun 12, 2022

• GenCode: Set libtorch subset if python is off

  #79379 opened Jun 12, 2022

• Fix backward of binary_cross_entropy_with_logits

  #79381 opened Jun 12, 2022

• Validate Sparse Compressed tensor inputs

  #79385 opened Jun 12, 2022

• Add OpInfo for torch.equal and fix support for non-standard bools

  #79389 opened Jun 13, 2022

• Fix remaining CPU operators for non-standard bools

  #79390 opened Jun 13, 2022

• Support non-standard bools in masked_scatter CUDA

  #79391 opened Jun 13, 2022

• Support non-standard bools in CUDA unique

  #79392 opened Jun 13, 2022

• Support non-standard bools in mode CUDA kernels

  #79393 opened Jun 13, 2022

• [wip] add `script` and `env-variables` options to linux build

  #79396 opened Jun 13, 2022

• fix possible allocator issues

  #79398 opened Jun 13, 2022

• Clone tensor to write in ShardedTensor checkpoint

  #79400 opened Jun 13, 2022

• Optimize the copy of BFloat16 to Float and Float to BFloat16

  #79404 opened Jun 13, 2022

• Add new checks in CI system to verify the built linux pip wheel with cpu-cxx11-abi

  #79409 opened Jun 13, 2022

• add API for re-functionalization an aten op; other functionalization fixes

  #79420 opened Jun 13, 2022

• [ONNX] Clean up `onnx_supported_ops`

  #79424 opened Jun 13, 2022

• add saketh-are to torch.nn codeowners

  #79431 opened Jun 13, 2022

• ci: Remove pytorch- prefix from xla job name

  #79434 opened Jun 13, 2022

• ci: Remove pytorch- prefix from android jobs

  #79435 opened Jun 13, 2022

• ci: Move deploy prefix to job suffix

  #79436 opened Jun 13, 2022

• [Draft][No CI] FX graph partitioner and fuser

  #79439 opened Jun 13, 2022

• Added dynamic shape POC

  #79441 opened Jun 13, 2022

• [Static Runtime] Performance optimization for fork operation

  #79442 opened Jun 13, 2022

• [mergebot] post comment on finishing

  #79448 opened Jun 13, 2022

• test unused-local-typedef in internal CI

  #79458 opened Jun 13, 2022

• Add backward for spdiags (cuda)

  #79462 opened Jun 13, 2022

• Add deterministic impl of `scatter_add` CUDA for all input sizes

  #79466 opened Jun 13, 2022

• [caffe2] Don't copy Tensor dims during deserialization

  #79471 opened Jun 13, 2022

• Enable Conjugate and Negative Fallback for Meta Tensors

  #79475 opened Jun 13, 2022

• [CI] increase quota for linux.4xlarge.nvidia.gpu

  #79478 opened Jun 13, 2022

• [DataPipe] Simple graph fast-forwarding

  #79479 opened Jun 13, 2022

• Make linalg_det function structured

  #79486 opened Jun 14, 2022

• Add forward AD for linalg.det and simplify its backward

  #79487 opened Jun 14, 2022

• Migrate pull off linux-xenial-py3_7-clang7-onnx

  #79489 opened Jun 14, 2022

• switch computation to computationPtr

  #79491 opened Jun 14, 2022

• Retry - [JIT] Propagate profiled information to DifferentiableGraph outputs

  #79498 opened Jun 14, 2022

• [quant] Implement quantize APoT method

  #79499 opened Jun 14, 2022

• refactoring code for better vectorization

  #79500 opened Jun 14, 2022

• [ONNX] Fix `argmin` and `argmax` test cases

  #79503 opened Jun 14, 2022

• Adding codeowners to quantization, sparsity, ns, etc.

  #79505 opened Jun 14, 2022

• [ONNX] Fix linalg norm output's shapes and dtypes

  #79506 opened Jun 14, 2022

• [primTorch] Elementwise unary ops vi

  #79526 opened Jun 14, 2022

• [quant] Dequantize apot tensor

  #79530 opened Jun 14, 2022

• [MPS] Add test consistency from OpInfo based tests from PR 78504

  #79532 opened Jun 14, 2022

• test that OSS flags type-limits warning as error

  #79536 opened Jun 14, 2022

• [CUBLAS][TF32][CUDNN] Update numerical_accuracy.rst

  #79537 opened Jun 14, 2022

• [onnx] have all classes that call setUp inherit from common_utils.setUp

  #79543 opened Jun 14, 2022

• [ONNX] Ignore small tensors when deduplicating initializers

  #79552 opened Jun 14, 2022

• [ONNX] Enable test_fill script test

  #79555 opened Jun 14, 2022

• Profiler binding speed up

  #79558 opened Jun 14, 2022

• reinstate size and shape returning symints

  #79560 opened Jun 14, 2022

• Add linux-focal-py3.7-gcc7-no-ops to merge rules

  #79561 opened Jun 14, 2022

• Csl/pytest threads

  #79564 opened Jun 14, 2022

• unittest profile

  #79566 opened Jun 14, 2022

• Editing gen_jit_shape_functions to make lintrunner happy

  #79571 opened Jun 14, 2022

• [static runtime] Fixed argsort argument error in take_along_dim op test

  #79575 opened Jun 14, 2022

• Add numerical test comparing BetterDecoder and fairseq decoder

  #79576 opened Jun 14, 2022

• Add support for a couple ops

  #79581 opened Jun 14, 2022

• [c10d] Make allreduce as a custom op

  #79582 opened Jun 14, 2022

• [FSDP] Default to not checking exec order

  #79584 opened Jun 15, 2022

• [ONNX] Update onnx submodule to 1.12

  #79585 opened Jun 15, 2022

• [xla hash update] update the pinned xla hash

  #79587 opened Jun 15, 2022

• test1

  #79588 opened Jun 15, 2022

• test1.1

  #79589 opened Jun 15, 2022

• test1.2

  #79590 opened Jun 15, 2022

• test1.3

  #79591 opened Jun 15, 2022

• [ao] Added ModelReport class outline for Fx Graph Modules

  #79595 opened Jun 15, 2022

• Update ROCm base docker images to focal (ubuntu20.04)

  #79596 opened Jun 15, 2022

• [ao][sparsity] removing old sparsity API and updating tests

  #79597 opened Jun 15, 2022

• [ao][sparsity] adding type hints to sparsifier and utils

  #79598 opened Jun 15, 2022

• [quant][improvement][better-engineering] Refactored get_supported_device_types into common_quantization.py

  #79607 opened Jun 15, 2022

• Revert "Revert "Python Bindings for SymInts (#78135)""

  #79608 opened Jun 15, 2022

• add a new alias key for functional to non functional decompositions

  #79613 opened Jun 15, 2022

• [Profiler] Expose extra fields to Python

  #79623 opened Jun 15, 2022

• Improve small sort performance on CUDA

  #79627 opened Jun 15, 2022

• Use cub::BlockRadixSort to improve medium length sort performance

  #79628 opened Jun 15, 2022

• Add _reduce_scatter_base to ProcessGroupWrapper.

  #79633 opened Jun 15, 2022

• switched over to using faketensor in proxytensor

  #79634 opened Jun 15, 2022

• optimize `to_dense` for CSC

  #79635 opened Jun 15, 2022

• Updating miniz library from version 2.0.8 -> 2.1.0

  #79636 opened Jun 15, 2022

• Autogen native_batch_norm and native_batch_norm_backward

  #79637 opened Jun 15, 2022

• [ao] Moving the sparsity/experimental to sparsity/_experimental

  #79639 opened Jun 15, 2022

• [ONNX] Refactor `TestONNXRuntime` with parameterized class

  #79641 opened Jun 15, 2022

• fixing wrong example

  #79643 opened Jun 15, 2022

• finalized GHA for promoting to viable/strict

  #79649 opened Jun 15, 2022

• [FSDP] Introduce `FlatParamHandle`

  #79652 opened Jun 15, 2022

• [torch] Add more functions to __init__.pyi.in for torch._C for Node and Value

  #79654 opened Jun 15, 2022

• [DataPipe] Count number of yields for IterDataPipe

  #79657 opened Jun 15, 2022

• [DataPipe] Full graph fast-forwarding

  #79658 opened Jun 15, 2022

• add a new FunctionSchema kind called scratch

  #79659 opened Jun 15, 2022

• [JIT] Bind AliasInfo to decrease differences in interfaces across languages

  #79661 opened Jun 16, 2022

• Automated submodule update: kineto

  #79662 opened Jun 16, 2022

• [c10d] Make allgather as a custom op

  #79669 opened Jun 16, 2022

• [ao] Adding model report detector base class and implemented detectors

  #79671 opened Jun 16, 2022

• [Quant] Vectorize scalar remainder in quantized kernel for normalization

  #79673 opened Jun 16, 2022

• [BE] Add FSDP helper classes to public documentation

  #79674 opened Jun 16, 2022

• add MetricReport API

  #79678 opened Jun 16, 2022

• python bindings for create_metric_report

  #79679 opened Jun 16, 2022

• Revert "Revert "Put symint overloads on a different name""

  #79680 opened Jun 16, 2022

• Revert "Revert "Add support for multiply on direct SymInt""

  #79681 opened Jun 16, 2022

• [c10d] Make reduce_scatter as a custom op

  #79683 opened Jun 16, 2022

• Optimize the copy of BFloat16 to Float and Float to BFloat16

  #79685 opened Jun 16, 2022

• [c10d] Make reduce as a custom op

  #79686 opened Jun 16, 2022

• [c10d] Make gather as a custom op

  #79687 opened Jun 16, 2022

• [c10d] Make scatter as a custom op

  #79688 opened Jun 16, 2022

• [ONNX] Fix quantization outputs' dtype

  #79690 opened Jun 16, 2022

• [c10d] Make alltoall as a custom op

  #79691 opened Jun 16, 2022

• [complex] conv_tranpose1d

  #79694 opened Jun 16, 2022

• [ONNX] Fix `hardshrink` and `softshrink` output's shape

  #79695 opened Jun 16, 2022

• Back out "Back out "[const_fold] Set requires_grad based on the folded tensor; add device_for_folding option""

  #79696 opened Jun 16, 2022

• Avoid unnecessary copy of rvalue ref shared ptr in SharedPtrWrapper

  #79697 opened Jun 16, 2022

• [Vulkan] Fix mm op null pointer dereference on Android

  #79701 opened Jun 16, 2022

• [Vulkan] Optimize LSTM operator with pre-packing

  #79702 opened Jun 16, 2022

• [GH1] Make pull required for merge

  #79704 opened Jun 16, 2022

• [WIP] Testing using cuda rather then cudatoolkit

  #79706 opened Jun 16, 2022

• `permute` for COO sparse tensors

  #79707 opened Jun 16, 2022

• `__launch_bounds__` for `torch.mode` with CUDA 11.7

  #79710 opened Jun 16, 2022

• [vulkan] fixed calling of threshold op

  #79717 opened Jun 16, 2022

• pytest proc

  #79719 opened Jun 16, 2022

• [docs] Include `batch_isend_irecv` and `P2POp` of `torch.distributed`

  #79722 opened Jun 16, 2022

• [small changes] Add closing parentheses to CollectiveFingerprint output

  #79723 opened Jun 16, 2022

• [BE] Update ProcessGroupWrapper to add deserializer and improve logs

  #79724 opened Jun 16, 2022

• gen_aten output into aten/src/ATen/

  #79727 opened Jun 16, 2022

• switch Bazel to common aten codegen

  #79728 opened Jun 16, 2022

• Readability in fbcode/caffe2/aten/src/ATen/native/cuda/Loss.cu

  #79729 opened Jun 16, 2022

• switch Bazel to common aten codegen

  #79730 opened Jun 16, 2022

• switch Bazel to common aten codegen

  #79731 opened Jun 16, 2022

• [torchgen] Support multiple namespace in NativeFunctions.h

  #79733 opened Jun 16, 2022

• switch Bazel to common aten codegen

  #79736 opened Jun 16, 2022

• switch Bazel to common aten codegen

  #79737 opened Jun 16, 2022

• error when registering meta kernels to composite ops in core

  #79741 opened Jun 16, 2022

• Make slogdet, linalg.sloget and logdet support metatensors

  #79742 opened Jun 16, 2022

• Implement forward AD for slogdet-related operations and fix their complex derivative

  #79743 opened Jun 16, 2022

• [LT] fix lazy tensor tutorial and test for cpu only run

  #79744 opened Jun 16, 2022

• add check for ci sev

  #79745 opened Jun 16, 2022

• [Static Runtime] Added nested prim::fork and aten::wait test case

  #79746 opened Jun 16, 2022

• [JIT] fix continue + with

  #79749 opened Jun 16, 2022

• [ao] Added insertion of observer capability to ModelReport class

  #79752 opened Jun 16, 2022

• [PyTorch] Record Sequence Number to Match Forward and Backward Operators (2nd try)

  #79753 opened Jun 16, 2022

• [Not for Land][FSDP] Specify params to clone for state dict

  #79754 opened Jun 16, 2022

• Adding Refs Packet: Elementwise Unary Ops V

  #79755 opened Jun 16, 2022

• [AI Accelerators] softmax kernel for Nested Tensor (CPU)

  #79756 opened Jun 16, 2022

• [ROCm] fix for multiple runners and docker commands

  #79759 opened Jun 17, 2022

• [Profiler][Trivial] Update test_profiler_tree

  #79761 opened Jun 17, 2022

• [Profiler] Include ActivityType from Kineto

  #79763 opened Jun 17, 2022

• Revert nested tensors changes

  #79765 opened Jun 17, 2022

• (WIP) Fix retain grad behavior on in-place

  #79767 opened Jun 17, 2022

• Add ref for nn.functional.prelu

  #79768 opened Jun 17, 2022

• Small update to prelu

  #79769 opened Jun 17, 2022

• [typing] suppress errors in `caffe2/caffe2/python`

  #79770 opened Jun 17, 2022

• Upstream repeat shape function.

  #79773 opened Jun 17, 2022

• [ao][sparsity] removing leading '.' from fqn in utils

  #79774 opened Jun 17, 2022

• [c10d] Make barrier as a custom op

  #79777 opened Jun 17, 2022

• [c10d] Make send/recv as custom ops

  #79779 opened Jun 17, 2022

• exclude jit_core_headers from hipification

  #79782 opened Jun 17, 2022

• switch Bazel to common aten codegen

  #79783 opened Jun 17, 2022

• Fix potencially unbound variable

  #79789 opened Jun 17, 2022

• [quant] Refactor nonuniform quantization mapping functions

  #79790 opened Jun 17, 2022

• [ao] Added generate report capability to ModelReport class

  #79792 opened Jun 17, 2022

• [ONNX] Support avgpool quantization

  #79793 opened Jun 17, 2022

• [3] move more pytorch buck targets to shared build

  #79794 opened Jun 17, 2022

• Wconstab/reland pysymint

  #79795 opened Jun 17, 2022

• ci: Move buck-build-test to periodic

  #79797 opened Jun 17, 2022

• Fix to preserve view relationships for Parameter.__deepcopy__()

  #79799 opened Jun 17, 2022

• [Expanded Weights] Fix instance norm

  #79800 opened Jun 17, 2022

• Csl/pytest asan

  #79801 opened Jun 17, 2022

• Make l1_loss composite

  #79804 opened Jun 17, 2022

• add heuristic and idle time computation

  #79805 opened Jun 17, 2022

• Relax the thread count assertion, that is modify EXPECT_EQ -> EXPECT_GE

  #79806 opened Jun 17, 2022

• Fix fx decomposition example

  #79807 opened Jun 17, 2022

• Add torch.nn.functional.threshold ref

  #79808 opened Jun 17, 2022

• Refine conditions under which index.Tensor has a dynamic shape

  #79809 opened Jun 17, 2022

• test commit

  #79810 opened Jun 17, 2022

• Fixing AO tests

  #79811 opened Jun 17, 2022

• [JIT] Handle aliasing cases where two arguments alias in SchemaCheckMode

  #79814 opened Jun 17, 2022

• [JIT] Nested fix (#79480)

  #79816 opened Jun 17, 2022

• [WIP][ao][sparsity] Base scheduler class for Data Schedulers

  #79817 opened Jun 17, 2022

• Reland #2 of "Added {logical_not, trace} refs, moved logical ops to use method overloads"

  #79819 opened Jun 17, 2022

• [primTorch] Adds contiguous and expand references

  #79820 opened Jun 17, 2022

• [primtorch] add reference for clamp_min/clamp_max

  #79821 opened Jun 17, 2022

• [WIP][ao][sparsity] Implemented the step() function for BaseDataScheduler class

  #79822 opened Jun 17, 2022

• [PrimTorch][Draft]

  #79823 opened Jun 17, 2022

• [nvFuser] Add real and imag to nvfuser and its python frontend

  #79824 opened Jun 17, 2022

• [_shard] make ShardedTensor be a Tensor and nn.Parameter

  #79825 opened Jun 18, 2022

• [ONNX] Inception

  #79826 opened Jun 18, 2022

• [WIP] Move addcmul from decompositions -> refs

  #79827 opened Jun 18, 2022

• Move shared_seed to cuda if NCCL is used

  #79829 opened Jun 18, 2022

• [CheckpointWrapper] Replace generic mod prefix

  #79830 opened Jun 18, 2022

• Support softmax(nested tensor)

  #79831 opened Jun 18, 2022

• automate hash updates - add new lint to appease lint

  #79832 opened Jun 18, 2022

• FSDP communication hook interface for NO_SHARD strategy

  #79833 opened Jun 18, 2022

• [WIP] Conv2d shape calculation for meta tensors

  #79834 opened Jun 18, 2022

69 Issues closed by 31 people

• ONNX tests intermittently ooming on master

  #79179 closed Jun 18, 2022

• Could not run aten::empty_strided with arguments from the 'PrivateUse1' backend

  #79812 closed Jun 18, 2022

• BytesWarning causes exception in torch.load() when running Python with -bb flag

  #74812 closed Jun 17, 2022

• Floating point exception in torch.nn.functional.conv1d/2d/3d/ and conv_transpose1d/2d/3d

  #49523 closed Jun 17, 2022

• ChainedScheduler documentation typoo

  #79720 closed Jun 17, 2022

• deit_small_patch16_224 weights not working

  #72654 closed Jun 17, 2022

• Cuda 11.3 Pull and Periodic Jobs are Failing

  #79738 closed Jun 16, 2022

• Update autograd engine to support nested tensors

  #79447 closed Jun 16, 2022

• Validate ONNX 1.12.0 release candidate

  #79461 closed Jun 16, 2022

• torch.nn.BCELoss throws 'invalid axes' error when using "mean" or "sum" reduction on torch.device("mps")

  #79527 closed Jun 16, 2022

• [FSDP] Investigate summon_full_params with CPU offloading

  #74166 closed Jun 16, 2022

• Error while loading model weights to mps device

  #78551 closed Jun 16, 2022

• RuntimeError: CUDA error: an illegal memory access was encountered on RTX 3080 with enough memory

  #79603 closed Jun 16, 2022

• Did Dropout2d change for Pytorch 1.11?

  #77081 closed Jun 16, 2022

• DISABLED test_comprehensive__masked_amin_cuda_uint8 (__main__.TestDecompCUDA)

  #79714 closed Jun 16, 2022

• test_dtypes__masked_amin_cuda is breaking with memory access issues

  #79711 closed Jun 16, 2022

• Failure to build with clang because of unknown warning option error

  #74805 closed Jun 16, 2022

• [ONNX] Add `__all__` to `torch.onnx`

  #78336 closed Jun 16, 2022

• [TorchData] Allow random_split to use percentages when applied to map-style datasets

  #78510 closed Jun 16, 2022

• Add forward prefetching option in FSDP API

  #78608 closed Jun 15, 2022

• Elementwise multiplication between MPS float32 and int Tensors result in LLVM error

  #78879 closed Jun 15, 2022

• `torch.gradient` need to check only dims specified by user and not all input tensor's dims

  #79346 closed Jun 15, 2022

• quantized linear dynamic should get reduce range from the observer rather than overwritting it in qnnpack/fbgemm

  #68278 closed Jun 15, 2022

• Pin PyTorch nightly used for comparison in BC test

  #79146 closed Jun 15, 2022

• torch.nn.CrossEntropyLoss output shape specification does not match behaviour

  #79531 closed Jun 15, 2022

• Performance drops after running tensor multiplication for 15 seconds on M1 MAX (Pytorch MPS).

  #79402 closed Jun 15, 2022

• Pytorch for PPC64LE

  #79419 closed Jun 15, 2022

• pytorch result not same as torchscript traced mod

  #79599 closed Jun 15, 2022

• Apply for Training Pytorch YOLOv5 by Apple M1 GPU

  #79594 closed Jun 15, 2022

• MPS backend support issue for int64

  #79200 closed Jun 15, 2022

• MPS device: Invalid type casting

  #79286 closed Jun 14, 2022

• DISABLED test_caching_pinned_memory_multi_gpu (__main__.TestCuda)

  #70875 closed Jun 14, 2022

• test_caching_pinned_memory_multi_gpu: Test has a race condition leading to flakiness

  #68299 closed Jun 14, 2022

• Random number generation yields different values on different devices, despite the same manual seed.

  #79496 closed Jun 14, 2022

• Different results between batched and one by one inference of nn.GRU

  #79501 closed Jun 14, 2022

• UnboundLocalError: local variable 'ws' referenced before assignment

  #79449 closed Jun 14, 2022

• MPSNDArray Error: buffer is not large enough

  #79181 closed Jun 14, 2022

• Cannot install Torch

  #79347 closed Jun 14, 2022

• Alexnet mode inference slows down after importing sklearn package

  #76970 closed Jun 14, 2022

• How to reimplement same behavior in AdaptiveAvgPooling2D

  #79332 closed Jun 14, 2022

• Fix warning: cast from type ‘const char*’ to type ‘char*’ casts away qualifiers

  #79519 closed Jun 14, 2022

• How to stacked RGB images

  #79495 closed Jun 14, 2022

• `libtorch.a` is not in pytorch `static` zip archives

  #70898 closed Jun 14, 2022

• `torch.load` does not propagate `map_location` arg to `torch.jit.load`

  #78331 closed Jun 14, 2022

• torch.load() fails on MPS backend ("don't know how to restore data location")

  #79384 closed Jun 14, 2022

• How to use torch.distributions.multivariate_normal.MultivariateNormal in multi-gpu mode

  #67013 closed Jun 14, 2022

• Move test/onnx/test_onnx_export.py into test_pytorch_onnx_no_runtime.py

  #78308 closed Jun 14, 2022

• nesterov in SGD

  #79405 closed Jun 14, 2022

• macos build time increased significantly after runner update to macos-12

  #78375 closed Jun 14, 2022

• ONNX export of torch.rand produces different data type

  #77845 closed Jun 14, 2022

• set data permits requires_grad=True on integer tensor

  #76008 closed Jun 14, 2022

• Move OptionalHasElement/If tests from no_runtime to runtime

  #79485 closed Jun 14, 2022

• Dynamic output of Upsample causes InstanceNorm2d error in onnx.

  #69346 closed Jun 13, 2022

• RocksDB 7 may have C++17 dependency

  #75496 closed Jun 13, 2022

• Remove the construction of unused tensors

  #79241 closed Jun 13, 2022

• How to find the code in '...'?

  #79106 closed Jun 13, 2022

• Update Autograd to be more shape agnostic

  #79040 closed Jun 13, 2022

• Vectorized fmod on CPU computes wrong result

  #79009 closed Jun 13, 2022

• torch.nn.modules.pooling return_indices and ceil_mode reversed

  #79415 closed Jun 13, 2022

• Einsum producing inconsistent results to sum()

  #79394 closed Jun 13, 2022

• [ZeRO] `test_zero_model_parallel()` failing silently since not run in CI

  #79259 closed Jun 13, 2022

• GRU output dim issue

  #78711 closed Jun 13, 2022

• fails to build with: no matching function for call to max(int, long int)

  #70297 closed Jun 13, 2022

• Inconsistent behavior of `torch.Tensor.scatter_` for float `src`

  #61854 closed Jun 12, 2022

• float32 matrix dot product has different precision in CPU/GPU

  #79270 closed Jun 12, 2022

• Improve handling of precision in testing

  #31135 closed Jun 11, 2022

• pip install failure

  #79274 closed Jun 11, 2022

• `dim` argument not checked for any of the fixed dimensional `fft` functions

  #74380 closed Jun 11, 2022

• Adamw cannot set different lrs to different parameters

  #79338 closed Jun 11, 2022

91 Issues opened by 62 people

• Scalar add returns incorrect results for `int32` MPS array

  #79835 opened Jun 18, 2022

• Distributed shared seed in DataLoader should be moved to cuda with NCCL Process Group

  #79828 opened Jun 18, 2022

• [primTorch] `test_python_ref_executor` often fails due to concrete arg mismatch

  #79818 opened Jun 17, 2022

• torch.Tensor.to(memory_format=torch.contiguous_format) does not work properly.

  #79813 opened Jun 17, 2022

• [ONNX] Replace test inheritance for `test/onnx/test_models.py` with parameterizing

  #79802 opened Jun 17, 2022

• lite interpreted model performs differently in Python and Android

  #79791 opened Jun 17, 2022

• Parameter.__deepcopy__ doesn't preserve view relationships

  #79788 opened Jun 17, 2022

• Improve clarity by making sharding a static nightly update

  #79787 opened Jun 17, 2022

• android-tests is often flaky

  #79785 opened Jun 17, 2022

• .backward() on MSELoss fails with IndexError: Dimension out of range on MPS

  #79784 opened Jun 17, 2022

• onnx export with dict parameter problems

  #79781 opened Jun 17, 2022

• DISABLED test_binary_op_tensorlists_fastpath__foreach_add_cuda_bool (__main__.TestForeachCUDA)

  #79780 opened Jun 17, 2022

• iSTFT length mismatch when center=False

  #79778 opened Jun 17, 2022

• DISABLED test_inherit_tensor (__main__.TestMultiprocessing)

  #79776 opened Jun 17, 2022

• Linear not check shape

  #79771 opened Jun 17, 2022

• [FSDP] Test that module using mixed precision can be loaded into non-mp module

  #79766 opened Jun 17, 2022

• [JIT] failures with nested with blocks + loop continuation

  #79751 opened Jun 16, 2022

• DISABLED test_custom_stream (__main__.TensorPipeTensorPipeAgentCudaRpcTest)

  #79750 opened Jun 16, 2022

• [forward ad] squeeze_, unsqueeze_ : the shape of the tensor was modified directly without going through the PyTorch dispatcher

  #79748 opened Jun 16, 2022

• Compliance with PEP-0523

  #79740 opened Jun 16, 2022

• quantization: misleading backend config for linear_dynamic_fp16

  #79739 opened Jun 16, 2022

• Typo in nn.utils.parametrizations.orthogonal

  #79735 opened Jun 16, 2022

• Registering decompositions to all overloads is problematic for meta functions now.

  #79734 opened Jun 16, 2022

• DISABLED test_dispatch_meta__masked_amin_cuda_uint8 (__main__.TestMetaCUDA)

  #79716 opened Jun 16, 2022

• [FX] TypeError when tracing cat taking split's output as input

  #79715 opened Jun 16, 2022

• DISABLED test_comprehensive__masked_amin_cuda_uint8 (__main__.TestDecompCUDA)

  #79712 opened Jun 16, 2022

• ONEDNN testing is not done properly in quantization codebase

  #79709 opened Jun 16, 2022

• `backward(inputs=)` is not working as expected when inplace are involved (and grad() does work fine)

  #79708 opened Jun 16, 2022

• gradgradcheck fails for torch.native_layer_norm

  #79705 opened Jun 16, 2022

• Float and double tensors randomly initialized with the same seed get different values for size >= 16

  #79703 opened Jun 16, 2022

• Does torch.utils.checkpoint support custom cuda operators

  #79693 opened Jun 16, 2022

• Does Torch JIT Support Trace High-level Custom Op?

  #79684 opened Jun 16, 2022

• Tensor plus add result seems wrong [with Mac mps]

  #79672 opened Jun 16, 2022

• FakeTensor doesn't work with `logical_not_()`

  #79670 opened Jun 16, 2022

• tensorboard SummaryWriter.add_graph fails when model uses empty tuples

  #79647 opened Jun 15, 2022

• test_neg_conj_view_istft_cpu_complex128 fails on DEBUG=1

  #79646 opened Jun 15, 2022

• [FSDP] Progress of ParamExecOrderWrapPolicy

  #79644 opened Jun 15, 2022

• Eradicate TestCase.setUp() calls in the code for super().setUp()

  #79630 opened Jun 15, 2022

• Missing the time unit in duration time of DDP logging

  #79629 opened Jun 15, 2022

• [FSDP] Verify that FSDP-managed parameters are the same across ranks

  #79620 opened Jun 15, 2022

• binary_cross_entropy_with_logits forward-over-reverse rule is silently incorrect

  #79609 opened Jun 15, 2022

• PyTorch Preview (Nightly) version number does not comply with Conda conventions

  #79606 opened Jun 15, 2022

• [FSDP] RuntimeError when using FSDP with auto wrap for sequence-to-sequence language models such as T5, Pegasus

  #79605 opened Jun 15, 2022

• Some unit tests are failing

  #79604 opened Jun 15, 2022

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

  #79601 opened Jun 15, 2022

• [LTC] Introduce a `MetricsReport` python binding and allow backend to add their report as string

  #79592 opened Jun 15, 2022

• [LTC] `torch::lazy::Computation` to include a device

  #79586 opened Jun 15, 2022

• Torchscript Serialisation for numbers over 1000 include a comma

  #79583 opened Jun 15, 2022

• __torch__dispatch does not return new output in inplace function

  #79563 opened Jun 14, 2022

• Unable to use a parameter with torch.sparse_coo layout with DDP

  #79542 opened Jun 14, 2022

• TestModels_new_jit_API.test_inception is flaky

  #79540 opened Jun 14, 2022

• test_ops.py extremely slow on cuda11.3

  #79528 opened Jun 14, 2022

• Enable `dim=None` for all reduction functions

  #79525 opened Jun 14, 2022

• Display a "reference" link for ops that points to primTorch implementations

  #79518 opened Jun 14, 2022

• Torch FX return error "`__cuda_array_interface__` must be a dict"

  #79513 opened Jun 14, 2022

• Function to automatically calculate Conv shape

  #79512 opened Jun 14, 2022

• Suggest for a polynominal lr_scheduler

  #79511 opened Jun 14, 2022

• DISABLED test_caching_pinned_memory (__main__.TestCuda)

  #79508 opened Jun 14, 2022

• DISABLED test_checkpoint_wrapper_parity (__main__.CheckpointWrapperTest)

  #79510 opened Jun 14, 2022

• DISABLED test_noncontiguous_samples_nn_functional_conv_transpose3d_cuda_float32 (__main__.TestCommonCUDA)

  #79509 opened Jun 14, 2022

• DISABLED test_launch_user_script (__main__.TestDistributedLaunch)

  #79488 opened Jun 14, 2022

• Can not export pytorch model to onnx.

  #79483 opened Jun 14, 2022

• Implement NestedTensor size function

  #79477 opened Jun 13, 2022

• [META] Sign up to discuss significantly modifying CI

  #79476 opened Jun 13, 2022

• Add a new _broadcast_coalesced op for DDP

  #79469 opened Jun 13, 2022

• Ensure the guards in distributed_c10d.py wrappers get executed in the replay of the graph

  #79468 opened Jun 13, 2022

• Add autograd support for dispatch passable c10d ops

  #79467 opened Jun 13, 2022

• Iteration # 1-offset in DDP logging

  #79464 opened Jun 13, 2022

• Adding python bindings for `SymInt`

  #79460 opened Jun 13, 2022

• [BE][ZeRO] Enable multigpu unit tests

  #79459 opened Jun 13, 2022

• [LTC] Make `torch::lazy::BackendImplInterface::ExecuteComputation` takes `ComputationPtr` instead of `Computation`

  #79453 opened Jun 13, 2022

• Use c10d broadcast_object in Zero

  #79452 opened Jun 13, 2022

• API for accessing SymIntNode mandates refcount bump even when it is unnecessary

  #79425 opened Jun 13, 2022

• Add doc formatting check to lintrunner

  #79418 opened Jun 13, 2022

• BUG: Segmentation fault when calling pytorch function after np.exp (numpy 1.21.2)

  #79410 opened Jun 13, 2022

• Conda enviroment

  #79407 opened Jun 13, 2022

• SymInt equality tests are unsound

  #79395 opened Jun 13, 2022

• Init connect timeout when use torch.distributed.run

  #79388 opened Jun 13, 2022

• caffe2_nvrtc is produced even when it won't be used

  #79387 opened Jun 13, 2022

• Incorrect image upscaling on MPS backend

  #79383 opened Jun 12, 2022

• torch failure to open libcuda.so.1 on macOS

  #79382 opened Jun 12, 2022

• TorchScript bidirectional lnlstm from example doesn't work

  #79375 opened Jun 12, 2022

• [ONNX Export] Interpolation likely should be exported with `half_pixel` instead of `pytorch_half_pixel`

  #79361 opened Jun 12, 2022

• [build] No documented way to install C++ binaries for pure-python development of pytorch

  #79359 opened Jun 12, 2022

• MSVC: vec_base size template fails with divide by sizeof(scalar_t)

  #79358 opened Jun 12, 2022

• [bazel] build spams warnings

  #79355 opened Jun 11, 2022

• [CI] [bazel] run GPU-requiring tests on a GPU machine

  #79354 opened Jun 11, 2022

• Adam not optimally implemented: unnecessary torch.div

  #79352 opened Jun 11, 2022

• [bazel] ability to run gpu tests on gpu machines in RBE

  #79351 opened Jun 11, 2022

• PyTorch get positive log_prob of a multivariate normal distribution

  #79349 opened Jun 11, 2022

• [bazel] build doesn't use sccache

  #79348 opened Jun 11, 2022

278 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] First implementation of ParamExecOrderWrapPolicy (non-recursive wrap policy)

  #79238 commented on Jun 18, 2022 • 39 new comments

• Add prims::{real, imag}, refs::{real, imag, isposinf, isneginf}, remove prims::is_infinite

  #78655 commented on Jun 18, 2022 • 28 new comments

• [BE][flatbuffer] Remove code duplications and refactor

  #79184 commented on Jun 17, 2022 • 26 new comments

• Add opset16 onnx support for `torch.scatter_add`

  #79103 commented on Jun 16, 2022 • 18 new comments

• General MPS op coverage tracking issue

  #77764 commented on Jun 17, 2022 • 17 new comments

• Less strict size check in scatter

  #79084 commented on Jun 17, 2022 • 17 new comments

• [v.1.12.0] Release Tracker

  #78005 commented on Jun 17, 2022 • 16 new comments

• fix kl_div for negative targets

  #69212 commented on Jun 14, 2022 • 15 new comments

• `kl_div`: fix for grads wrt `target`, double backward, forward-over-reverse AD support.

  #79007 commented on Jun 16, 2022 • 12 new comments

• torch.elastic fails to shutdown despite crashed processes

  #76287 commented on Jun 17, 2022 • 10 new comments

• [FSDP] ssd_offload first commit importing basic building blocks and u…

  #77136 commented on Jun 14, 2022 • 10 new comments

• Generate non-structured dispatch-less composite kernels.

  #77484 commented on Jun 17, 2022 • 10 new comments

• [vision hash update] update the pinned vision hash

  #79256 commented on Jun 18, 2022 • 9 new comments

• Add Div reference

  #77936 commented on Jun 17, 2022 • 8 new comments

• [static runtime] Add JIT prim ops: aten::cpu, aten::list, aten::numel, aten::__range_length

  #79111 commented on Jun 18, 2022 • 8 new comments

• [2] move more pytorch buck targets to shared build

  #79330 commented on Jun 17, 2022 • 8 new comments

• [ONNX] Internal assert error during export

  #79333 commented on Jun 17, 2022 • 7 new comments

• [ONNX] Enable data propagation from ONNX to simplify current shape inference

  #75307 commented on Jun 15, 2022 • 7 new comments

• integrate functionalization <> LTC torchscript backend

  #75527 commented on Jun 17, 2022 • 7 new comments

• Fixing JIT compatibility

  #79012 commented on Jun 15, 2022 • 7 new comments

• PyTorch fails to calculate the log_prob of a multivariate normal distribution when the number of data samples is larger than 524280

  #79191 commented on Jun 15, 2022 • 6 new comments

• Fix SequentialLR initialization

  #72856 commented on Jun 17, 2022 • 6 new comments

• [WIP] Added support for `expand` in LazyTensor shape inference

  #77830 commented on Jun 17, 2022 • 6 new comments

• Add spdiags sparse matrix initialization

  #78439 commented on Jun 13, 2022 • 6 new comments

• [primTorch] Implements refs for gcd, lcm and remainder

  #78747 commented on Jun 14, 2022 • 6 new comments

• [torchdynamo hash update] update the pinned torchdynamo hash

  #79149 commented on Jun 18, 2022 • 6 new comments

• Modify nccl_dependency to take dev mode

  #79169 commented on Jun 18, 2022 • 6 new comments

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

  #43259 commented on Jun 16, 2022 • 5 new comments

• More stable computation of KL between two Bernoulli distributions

  #20164 commented on Jun 14, 2022 • 5 new comments

• [quant][fx] Support keyword arguments for functional linear

  #79095 commented on Jun 17, 2022 • 5 new comments

• Corner cases of ShardedTensor checkpoint when using TorchRec

  #79016 commented on Jun 15, 2022 • 4 new comments

• CrossEntropyLoss computes SoftMax always across the second dimension

  #77256 commented on Jun 17, 2022 • 4 new comments

• Add prelu op and module for quantized CPU backend

  #73491 commented on Jun 17, 2022 • 4 new comments

• To add hipify_torch as a submodule in pytorch/third_party

  #74704 commented on Jun 14, 2022 • 4 new comments

• [ONNX] Inline prim::PythonOp for Autograd Function Export

  #74765 commented on Jun 14, 2022 • 4 new comments

• Fix: memory cross-border access on the ROCM platform

  #76100 commented on Jun 17, 2022 • 4 new comments

• Added decomposition for nll_loss_forward

  #78491 commented on Jun 16, 2022 • 4 new comments

• [composite compliance] frobenius norm

  #78515 commented on Jun 17, 2022 • 4 new comments

• DistributedDataParallel `static_graph=True` fails to handle unused parameters

  #79171 commented on Jun 13, 2022 • 3 new comments

• CI workflow creates too many tags in RSS feed

  #79083 commented on Jun 13, 2022 • 3 new comments

• OOM Issues despite plenty of resources on M1 Mac

  #79283 commented on Jun 15, 2022 • 3 new comments

• Add fast path of qmean/qstd for quantized CPU

  #70172 commented on Jun 17, 2022 • 3 new comments

• [static-runtime] add nnc codegen for aten::div

  #76903 commented on Jun 18, 2022 • 3 new comments

• [NNC] enable fusion of conv with elementwise OP

  #77157 commented on Jun 17, 2022 • 3 new comments

• [FSDP] Fix a small bug of pre_backward_hook params prefetch

  #78851 commented on Jun 17, 2022 • 3 new comments

• [ROCm] More Sparse UTs enablement and more hipification mappings.

  #78939 commented on Jun 16, 2022 • 3 new comments

• Add shape function for split like operations.

  #79194 commented on Jun 17, 2022 • 3 new comments

• ROCm runlogs are not uplodaed to the hud

  #79074 commented on Jun 13, 2022 • 2 new comments

• Convolution weight gradient computation in nn.grad is wrong with groups > 1

  #51430 commented on Jun 13, 2022 • 2 new comments

• Significant perf reduction on Python GIL contention with dataloader pinning thread.

  #77139 commented on Jun 13, 2022 • 2 new comments

• Batch norm doc mentions using biased variance estimator, but it actually uses the unbiased variance estimator

  #77427 commented on Jun 13, 2022 • 2 new comments

• Support float tensors as float inputs

  #79279 commented on Jun 14, 2022 • 2 new comments

• RuntimeError on ONNX export of `fake_quantize_per_tensor_affine`

  #77455 commented on Jun 14, 2022 • 2 new comments

• Supper `DimensionNode::isDynamic()` API via `lazy::Shape::is_symbolic()`

  #77909 commented on Jun 14, 2022 • 2 new comments

• [Tracer] RuntimeError: _Map_base::at when tracing fake quantization

  #76486 commented on Jun 15, 2022 • 2 new comments

• python test/run_test.py

  #79022 commented on Jun 15, 2022 • 2 new comments

• Mismatch in clang toolchain lead to binary incompatibilities on M1 between torch and torchvision

  #79421 commented on Jun 15, 2022 • 2 new comments

• [Device MPS] Error: buffer is not large enough. Must be 19200 bytes

  #78916 commented on Jun 15, 2022 • 2 new comments

• torch.nn.Sequential can use `+` (add operator) to concatenate

  #78512 commented on Jun 16, 2022 • 2 new comments

• RFC: multi-part `torch.load`/`torch.save` to support huge models and/or low CPU memory

  #64327 commented on Jun 16, 2022 • 2 new comments

• Observing a strange behavior - Row parallelism

  #76232 commented on Jun 17, 2022 • 2 new comments

• [WIP] Integrate xdoctest

  #15648 commented on Jun 11, 2022 • 2 new comments

• fix `torch.nn.functional.l1_loss` for complex inputs

  #65681 commented on Jun 12, 2022 • 2 new comments

• Fixes bias not working when not qkv has diff size

  #66544 commented on Jun 16, 2022 • 2 new comments

• Remove set_default_dtype call in jit_metaprogramming_utils.py

  #69624 commented on Jun 13, 2022 • 2 new comments

• Make language std configurable.

  #75519 commented on Jun 12, 2022 • 2 new comments

• Respect the global std option if set.

  #75729 commented on Jun 12, 2022 • 2 new comments

• BUG: Evade segfault by throwing a RuntimeError for `nn.ChannelShuffle` and empty input tensors

  #77029 commented on Jun 13, 2022 • 2 new comments

• Experimental module perf changes

  #77098 commented on Jun 17, 2022 • 2 new comments

• simple c10 implementation for std::call_once

  #78051 commented on Jun 15, 2022 • 2 new comments

• [ao][sparsity] add option for tensor_fqn to sparsity API

  #79113 commented on Jun 18, 2022 • 2 new comments

• Version 1.3 no longer supporting Tesla K40m?

  #30532 commented on Jun 11, 2022 • 1 new comment

• Hash mismatch for METADATA file

  #47916 commented on Jun 11, 2022 • 1 new comment

• Automate cleanup of header includes

  #77165 commented on Jun 11, 2022 • 1 new comment

• [bazel] quantization::_Bfloat16QuantizedToFloat kernel de-registration fails

  #79236 commented on Jun 11, 2022 • 1 new comment

• PyTorch fails to compile on gcc 11.2 due to breakpad

  #73113 commented on Jun 12, 2022 • 1 new comment

• Execute C++ based GTest unit tests from a python wrapper

  #64067 commented on Jun 12, 2022 • 1 new comment

• Custon C++ classes are not traceable

  #47162 commented on Jun 13, 2022 • 1 new comment

• Feature Request: CUDA torch.histogram (and histogramdd)

  #69519 commented on Jun 13, 2022 • 1 new comment

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

  #72763 commented on Jun 13, 2022 • 1 new comment

• Feature request: reverse_padded_sequence

  #1794 commented on Jun 13, 2022 • 1 new comment

• Convolution for complex tensors

  #71108 commented on Jun 13, 2022 • 1 new comment

• Incosistency with args for `nn.functional.max_poolNd` vs `nn.MaxPoolNd` functions

  #62545 commented on Jun 13, 2022 • 1 new comment

• [bug] `TensorIterator` segfaults when inputs broadcast over output, and there is a 0-dim input.

  #79218 commented on Jun 13, 2022 • 1 new comment

• RFC: [primTorch] Stride-agnostic Operator Semantics

  #78050 commented on Jun 13, 2022 • 1 new comment

• Potential race conditions between multiple workers trying to download and cache the same file

  #68320 commented on Jun 13, 2022 • 1 new comment

• DDP Freezes w/ No Output for PyTorch Geometric GNN Multi-GPU Node Classification

  #79049 commented on Jun 13, 2022 • 1 new comment

• Memory allocation errors when attempting to initialize a large number of small feed-forward networks in RAM with shared memory despite having enough memory

  #78274 commented on Jun 13, 2022 • 1 new comment

• Move Optional tests from test_pytorch_onnx_no_runtime.py to test_pytorch_onnxruntime.py

  #78119 commented on Jun 14, 2022 • 1 new comment

• destroy_process_group() has a certain probability of hangs

  #75097 commented on Jun 14, 2022 • 1 new comment

• PyTorch installation doesn't work with Python 3.10.0

  #68982 commented on Jun 14, 2022 • 1 new comment

• [FSDP] `ignored_modules` follow-ups

  #77141 commented on Jun 14, 2022 • 1 new comment

• DDP window TCP bug [socket.cpp:558] [c10d] The client socket has failed to

  #77523 commented on Jun 14, 2022 • 1 new comment

• `cumsum` op: pytorch failed to run GPT-2 model in M1's MPS device

  #79112 commented on Jun 14, 2022 • 1 new comment

• [Poll] Support DistributedDataParallel (DDP) in PyTorch C++ API (libtorch)

  #48959 commented on Jun 15, 2022 • 1 new comment

• Unable to install Preview (Nightly) on M1 macOS: "Symbol not found"

  #78681 commented on Jun 15, 2022 • 1 new comment

• Integrate run_android_tests.yml into trunk workflow

  #79182 commented on Jun 15, 2022 • 1 new comment

• Pytorch very slow to convert list of numpy arrays into tensors

  #13918 commented on Jun 15, 2022 • 1 new comment

• Support Positional-only Arguments in JIT

  #76774 commented on Jun 15, 2022 • 1 new comment

• SEGFAULT on "import torch"

  #70245 commented on Jun 15, 2022 • 1 new comment

• Split up and reorganize RPC tests

  #78367 commented on Jun 15, 2022 • 1 new comment

• fatal_signal_asan_no_sig_test in current master hang.

  #78880 commented on Jun 15, 2022 • 1 new comment

• Slicing with step does not work correctly with MPS backend

  #78886 commented on Jun 15, 2022 • 1 new comment

• Dropout2d doesn't drop channels for (C, H, W)

  #69801 commented on Jun 15, 2022 • 1 new comment

• JIT fuser throws compilation error (1.10 regression)

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

• Support torch::jit::script::Module converting to torch::nn::Module.

  #51328 commented on Jun 16, 2022 • 1 new comment

• Support more granular test decoration

  #79161 commented on Jun 16, 2022 • 1 new comment

• ninja: build stopped: subcommand failed.

  #32694 commented on Jun 16, 2022 • 1 new comment

• ReduceLROnPlateau doesn't have the get_last_lr()

  #50715 commented on Jun 16, 2022 • 1 new comment

• can't load model on cuda after call cudaDeviceReset functions

  #27970 commented on Jun 16, 2022 • 1 new comment

• Export torch.maximum() op in ONNX format

  #72790 commented on Jun 16, 2022 • 1 new comment

• DISABLED test_dtypes__masked_amin_cuda (__main__.TestCommonCUDA)

  #71248 commented on Jun 16, 2022 • 1 new comment

• C10d Elastic Training master_addr ERROR

  #74824 commented on Jun 16, 2022 • 1 new comment

• DISABLED test_fs_sharing (__main__.TestMultiprocessing)

  #67002 commented on Jun 16, 2022 • 1 new comment

• permute returning unexpected results on MPS devices

  #78445 commented on Jun 16, 2022 • 1 new comment

• DISABLED test_multiprocessing_contexts (__main__.TestDataLoader)

  #71229 commented on Jun 17, 2022 • 1 new comment

• ONNX export failed on ATen operator _thnn_fused_gru_cell

  #51553 commented on Jun 17, 2022 • 1 new comment

• PyTorch CI Stability Action Tracker

  #65439 commented on Jun 17, 2022 • 1 new comment

• Segmentation Fault when importing Torch, 2021 version

  #66782 commented on Jun 17, 2022 • 1 new comment

• [Meta] CI Revert Tracker

  #66178 commented on Jun 17, 2022 • 1 new comment

• Batches are being duplicated from go http call

  #79336 commented on Jun 17, 2022 • 1 new comment

• Massive initial memory overhead GPU

  #12873 commented on Jun 18, 2022 • 1 new comment

• Fix documentation for atan2

  #10878 commented on Jun 13, 2022 • 1 new comment

• [WIP] Add Beta CDF

  #26141 commented on Jun 13, 2022 • 1 new comment

• Update the autograd notes in the doc

  #30253 commented on Jun 13, 2022 • 1 new comment

• Punctuation updated the tech docs

  #45806 commented on Jun 11, 2022 • 1 new comment

• std/var: Deprecate default correction value

  #55679 commented on Jun 15, 2022 • 1 new comment

• std/var: Require correction or unbiased parameter be given

  #55680 commented on Jun 15, 2022 • 1 new comment

• std/var: Deprecate overloads with "unbiased" argument

  #55681 commented on Jun 15, 2022 • 1 new comment

• Give std/var correction overloads proper defaults

  #56398 commented on Jun 14, 2022 • 1 new comment

• Support `torch.linalg.trace`

  #62714 commented on Jun 12, 2022 • 1 new comment

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

  #63289 commented on Jun 16, 2022 • 1 new comment

• [torchelastic] Set correct default rdzv_id

  #66280 commented on Jun 17, 2022 • 1 new comment

• error in codegen if vendor provides kernel for a composite op with no…

  #67090 commented on Jun 17, 2022 • 1 new comment

• add native quantization support for pixel_shuffle

  #68328 commented on Jun 13, 2022 • 1 new comment

• remove kl_div_backward from aten namespace

  #68419 commented on Jun 12, 2022 • 1 new comment

• Use 64bit type in multiplications to avoid integer overflow in max_pool2d

  #68682 commented on Jun 14, 2022 • 1 new comment

• [torchelastic][2/n] Fix agent, timer tests

  #69035 commented on Jun 17, 2022 • 1 new comment

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

  #70873 commented on Jun 13, 2022 • 1 new comment

• [torchelastic][etcd] Make default `last_call_timeout` as two times `timeout`

  #72033 commented on Jun 17, 2022 • 1 new comment

• [torchelastic] Add warning message if `last_call_timeout` for `etcd` rendezvous is less than `2*monitor_interval`

  #72098 commented on Jun 17, 2022 • 1 new comment

• [Quant][core][improvement] Separated implementations for quantized & non-quantized tensors in item

  #72333 commented on Jun 14, 2022 • 1 new comment

• [Quant][core][improvements] Combined dispatch registration for max_pool2d & quantized_max_pool2d and implemented max_pool2d_with_indices_out_quantized_cpu

  #72353 commented on Jun 14, 2022 • 1 new comment

• Add xpu path for distributed data parallel

  #74821 commented on Jun 16, 2022 • 1 new comment

• fix torch.config can't respect USE_MKLDNN flag issue

  #75001 commented on Jun 16, 2022 • 1 new comment

• [Quant][core][improvements] Combined dispatch registration for max_pool1d & quantized_max_pool1d

  #75261 commented on Jun 14, 2022 • 1 new comment

• [quant][bc-breaking][improvements] Removed quantized_max_pool2d registration

  #75304 commented on Jun 14, 2022 • 1 new comment

• [quant][bc-breaking][improvements] Removed quantized_max_pool1d registration

  #75305 commented on Jun 14, 2022 • 1 new comment

• [quant][ns] Rename the weight tensor name to match fx NS tool

  #75530 commented on Jun 12, 2022 • 1 new comment

• [FSDP] `scatter_full_optim_state_dict()`: async broadcast

  #75603 commented on Jun 14, 2022 • 1 new comment

• [PyTorch][3/4 Static dispatch with multiple backends] Move Tensor method definitions to TensorMethod.cpp

  #75606 commented on Jun 14, 2022 • 1 new comment

• [PyTorch][4/4 Static dispatch with multiple backends]

  #75607 commented on Jun 14, 2022 • 1 new comment

• Sets rtol>0 for some tests that use tf32

  #75612 commented on Jun 17, 2022 • 1 new comment

• [NOT WORKING][quant][gpu][core][improvement] Added broadcasting by channels addition support for quantized cudnn conv operator

  #75668 commented on Jun 13, 2022 • 1 new comment

• [PyTorchEdge] Version bump and make flatbuffer deafult for targets that enable faltbuffer

  #75742 commented on Jun 12, 2022 • 1 new comment

• ci: Add 5 minute sleep to allow user logins

  #75745 commented on Jun 12, 2022 • 1 new comment

• [testing] -D_GLIBCXX_ASSERTIONS

  #75766 commented on Jun 13, 2022 • 1 new comment

• use context manager for path extension in torch.hub

  #75786 commented on Jun 13, 2022 • 1 new comment

• Init commit

  #75793 commented on Jun 13, 2022 • 1 new comment

• Add warning message when model is out of SLA

  #75812 commented on Jun 13, 2022 • 1 new comment

• Add softmax/log_softmax out variants

  #75813 commented on Jun 13, 2022 • 1 new comment

• temporarily duplicating the stack from #75819

  #75835 commented on Jun 17, 2022 • 1 new comment

• temporarily duplicating the stack from #75819

  #75839 commented on Jun 14, 2022 • 1 new comment

• improve package error message

  #75842 commented on Jun 14, 2022 • 1 new comment

• POC sparse structured kernels

  #75871 commented on Jun 14, 2022 • 1 new comment

• [torch package] show where broken dependency comes from in packaging error message

  #75873 commented on Jun 14, 2022 • 1 new comment

• Add alpha_dropout opinfo

  #75882 commented on Jun 14, 2022 • 1 new comment

• [quant][gpu][core] Removed TORCH_API from quantized cudnn linear and conv packed param classes

  #75934 commented on Jun 15, 2022 • 1 new comment

• free up 3 dispatch keys: make autocast + SparseCSR per-backend keys

  #76112 commented on Jun 13, 2022 • 1 new comment

• Initial integration of ZenDNN as backend into PyTorch

  #76242 commented on Jun 17, 2022 • 1 new comment

• Add Linear-(BN)-LeakyReLU fusion

  #76424 commented on Jun 12, 2022 • 1 new comment

• [torch] print out gemm params when we get errors from cublasLtMatmul

  #76653 commented on Jun 16, 2022 • 1 new comment

• [JIT] handle branching tuple outputs

  #77120 commented on Jun 17, 2022 • 1 new comment

• Provide a default implementation of NVFuser callback

  #77672 commented on Jun 17, 2022 • 1 new comment

• Add support for BSR <-> Strided conversion with any number of Batch dims

  #78303 commented on Jun 17, 2022 • 1 new comment

• [JIT] autodiff - don't create groups for subgraphs without requires_grad

  #78593 commented on Jun 14, 2022 • 1 new comment

• Symbolic Shape Analaysis: Add Generalized List of Tensor Shape Support

  #78679 commented on Jun 15, 2022 • 1 new comment

• [primTorch] Elementwise binary ops I tests

  #78837 commented on Jun 13, 2022 • 1 new comment

• Remove const from function return type if returning const value

  #78986 commented on Jun 17, 2022 • 1 new comment

• [distributed] Make DDP work with python process group

  #79176 commented on Jun 16, 2022 • 1 new comment

• Do not do an inplace clone with `freeze_module`

  #79227 commented on Jun 14, 2022 • 1 new comment

• [1.9.1] [collect_env] collect_env does not collect actual runtime-loaded cudnn version

  #78489 commented on Jun 11, 2022 • 0 new comments

• Function request: logerfc, logerfcx special functions

  #31945 commented on Jun 11, 2022 • 0 new comments

• PyTorch master failed to build with glog 0.5.0.

  #58054 commented on Jun 12, 2022 • 0 new comments

• AlBert quantization

  #78722 commented on Jun 13, 2022 • 0 new comments

• `with torch.backends.cudnn.flags(deterministic=True)` doesn't give an exception for ctc_loss backward on CUDA

  #78475 commented on Jun 13, 2022 • 0 new comments

• If large enough tensor is being cloned, parallel dataloading hangs on M1 Mac

  #78924 commented on Jun 13, 2022 • 0 new comments

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

  #24422 commented on Jun 13, 2022 • 0 new comments

• Using a Subclassed Tensor Results in Significant Decrease in Training GPU Throughput

  #79321 commented on Jun 13, 2022 • 0 new comments

• [NVFuser] hitting fallbacks on demucs (from torchbench + lazy tensor)

  #79325 commented on Jun 13, 2022 • 0 new comments

• RuntimeError: cuDNN error: CUDNN_STATUS_INTERNAL_ERROR

  #79014 commented on Jun 13, 2022 • 0 new comments

• [NVFuser] bad performance on pyhpc_equation_of_state

  #79244 commented on Jun 13, 2022 • 0 new comments

• Conda install from pytorch-nightly channel does not install the expected version on macOS

  #79337 commented on Jun 13, 2022 • 0 new comments

• Three memory copies of every dataloader cpu tensor

  #78018 commented on Jun 13, 2022 • 0 new comments

• scripted fft Convolutions are faster than nn.Conv1d with large kernels

  #79222 commented on Jun 13, 2022 • 0 new comments

• [ONNX] Make it obvious when a test failure is scripting vs tracing in test_pytorch_onnx_onnxruntime.py

  #75630 commented on Jun 14, 2022 • 0 new comments

• `adapative_{avg, max}_pool` will crash in the backward pass on cuda on empty output size

  #78868 commented on Jun 14, 2022 • 0 new comments

• `kl_div` will crash in the backward pass on cuda

  #78867 commented on Jun 14, 2022 • 0 new comments

• PyTorch gets stuck when using an NVLink/A6000 and more than two GPUs

  #79299 commented on Jun 14, 2022 • 0 new comments

• `prepare_qat_fx` docstring doesn't run

  #79307 commented on Jun 14, 2022 • 0 new comments

• [ONNX] Support unit tests in scripting that we already support in tracing

  #75625 commented on Jun 14, 2022 • 0 new comments

• Add Autograd Support for Nested Tensor

  #79039 commented on Jun 15, 2022 • 0 new comments

• [build] push containers used for PyTorch build in CI to GitHub container registry

  #79323 commented on Jun 15, 2022 • 0 new comments

• [bazel] basic cuda tensor creation is getting program stuck

  #79304 commented on Jun 15, 2022 • 0 new comments

• Add type() support for mps backend

  #78929 commented on Jun 15, 2022 • 0 new comments

• Random functions should infer device from user-specified Generator

  #79018 commented on Jun 16, 2022 • 0 new comments

• MPS: Add support for TopK (k>16) on M1 GPU

  #78915 commented on Jun 16, 2022 • 0 new comments

• [distributed_test.py] Improve `test_barrier`

  #78885 commented on Jun 16, 2022 • 0 new comments

• torch.linalg.inv gives incorrect results for any index besides zero on M1/mps GPU device

  #78363 commented on Jun 16, 2022 • 0 new comments

• Composite Compliance Problems Tracker

  #69991 commented on Jun 17, 2022 • 0 new comments

• Installing conda package from pytorch-nighly fails with `libcupti.so.X.Y` not found

  #74473 commented on Jun 17, 2022 • 0 new comments

• RReLU and PReLU do not propagate input strides to outputs

  #60028 commented on Jun 17, 2022 • 0 new comments

• Add operator and derivative coverage for nestedtensors

  #79044 commented on Jun 17, 2022 • 0 new comments

• [ONNX] Enable more operators to support data propagation

  #79208 commented on Jun 17, 2022 • 0 new comments

• Enhance add_out_dense_sparse_cpu for hybrid sparse tensor

  #23057 commented on Jun 15, 2022 • 0 new comments

• [DO NOT DELETE OR MERGE] Nightly release

  #26921 commented on Jun 17, 2022 • 0 new comments

• [DO NOT DELETE OR MERGE] Postnightly release

  #27167 commented on Jun 17, 2022 • 0 new comments

• Add the release testing workflow for 1.9.0 and future releases.

  #59260 commented on Jun 17, 2022 • 0 new comments

• optimize sort

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

• Add mkl implementation for exponential on CPU

  #69967 commented on Jun 15, 2022 • 0 new comments

• [DataPipe] Snapshotting prototype

  #70216 commented on Jun 18, 2022 • 0 new comments

• [DataPipe] Basic snapshotting with IterableWrapper

  #70373 commented on Jun 18, 2022 • 0 new comments

• [quant] Add QuantizedLSTM class

  #71189 commented on Jun 14, 2022 • 0 new comments

• [quant] Add QuantizedMHA class

  #71190 commented on Jun 14, 2022 • 0 new comments

• [quant] Quantizable documentation

  #71191 commented on Jun 14, 2022 • 0 new comments

• [quant] Create default custom modules for LSTM and MHA

  #71192 commented on Jun 14, 2022 • 0 new comments

• Automated submodule update: FBGEMM

  #74729 commented on Jun 18, 2022 • 0 new comments

• improve sort cpu bfloat16 perf by directly comparing on acc_type

  #74898 commented on Jun 15, 2022 • 0 new comments

• [jiterator] logical_{or, xor} : complex

  #75947 commented on Jun 16, 2022 • 0 new comments

• add a new alias key for functional to non functional decompositions

  #76358 commented on Jun 15, 2022 • 0 new comments

• [WIP] Optionally avoid recordStream calls in ProcessGroupNCCL

  #76861 commented on Jun 15, 2022 • 0 new comments

• improve index_select performance on CPU

  #76868 commented on Jun 15, 2022 • 0 new comments

• test mkldnn conv2d channels last when weight is nchw format

  #77348 commented on Jun 15, 2022 • 0 new comments

• Allow external pytorch libraries to see kineto

  #77746 commented on Jun 14, 2022 • 0 new comments

• [PyTorch Edge] Remove Original Weight Tensor from QNNPack Sparse Quantized Linear Packed Params

  #77757 commented on Jun 13, 2022 • 0 new comments

• [PyTorch Edge] Add serialization/deserialization of Sparse Quantize Linear Packed Params

  #77758 commented on Jun 13, 2022 • 0 new comments

• gen_aten uses full paths to aten files

  #77788 commented on Jun 17, 2022 • 0 new comments

• eliminate generated_cpu_cpp and generated_cuda_cpp from Bazel

  #77790 commented on Jun 17, 2022 • 0 new comments

• move fbcode ATEN_CORE_H to shared build structure

  #77863 commented on Jun 17, 2022 • 0 new comments

• introduce aten-core-headers to Bazel

  #77864 commented on Jun 17, 2022 • 0 new comments

• switch aten_headers to use strip_include_prefix instead of includes

  #77865 commented on Jun 17, 2022 • 0 new comments

• [quant] Adding tests for qconv output shapes

  #77927 commented on Jun 17, 2022 • 0 new comments

• [quant] Enable 'same' and 'valid' padding in static qconv

  #77928 commented on Jun 17, 2022 • 0 new comments

• [quant] Enable 'same' and 'valid' padding in dynamic qconv

  #77929 commented on Jun 17, 2022 • 0 new comments

• Landtime

  #77943 commented on Jun 15, 2022 • 0 new comments

• More forward AD formulas

  #77975 commented on Jun 17, 2022 • 0 new comments

• [ROCm] enable jiterator

  #77982 commented on Jun 16, 2022 • 0 new comments

• Enable SyncBatchNorm in CPU backend

  #78081 commented on Jun 16, 2022 • 0 new comments

• move aten_core_headers to common build structure

  #78093 commented on Jun 17, 2022 • 0 new comments

• move the GENERATED_CORE_H headers into aten_core_headers

  #78097 commented on Jun 17, 2022 • 0 new comments

• use full paths to files to generate-code genrule

  #78104 commented on Jun 17, 2022 • 0 new comments

• [quant] Removing unnecessary arguments from F.conv

  #78386 commented on Jun 17, 2022 • 0 new comments

• extract aten_core_srcs function into shared build structure

  #78535 commented on Jun 17, 2022 • 0 new comments

• [quant] Signatures check for F.convXd

  #78543 commented on Jun 17, 2022 • 0 new comments

• [WIP] Dynamic shape poc

  #78587 commented on Jun 13, 2022 • 0 new comments

• Add HPU Activity Type

  #78645 commented on Jun 13, 2022 • 0 new comments

• [quant][ao_migration] `torch.nn.quantized.functional` → `torch.ao.nn.quantized.functional`

  #78712 commented on Jun 14, 2022 • 0 new comments

• [quant][ao_migration] `torch.nn.quantized.modules` → `torch.ao.nn.quantized.modules`

  #78713 commented on Jun 14, 2022 • 0 new comments

• [quant][ao_migration] `torch.nn.quantized.dynamic` → `torch.ao.nn.quantized.dynamic`

  #78714 commented on Jun 14, 2022 • 0 new comments

• [quant][ao_migration] `torch.nn.quantized._reference` → `torch.ao.nn.quantized._reference`

  #78715 commented on Jun 14, 2022 • 0 new comments

• [quant][ao_migration] `torch.nn.qat` → `torch.ao.nn.qat`

  #78716 commented on Jun 14, 2022 • 0 new comments

• [quant][ao_migration] `torch.nn.quantizable` → `torch.ao.nn.quantizable`.

  #78717 commented on Jun 14, 2022 • 0 new comments

• JIT SSA Add support for Stack

  #78769 commented on Jun 15, 2022 • 0 new comments

• torch.special.airy_ai

  #78902 commented on Jun 13, 2022 • 0 new comments

• torch.special.gamma

  #78904 commented on Jun 13, 2022 • 0 new comments

• Add backward for spdiags (cpu) and enable autograd

  #78965 commented on Jun 13, 2022 • 0 new comments

• test ops auto

  #79059 commented on Jun 15, 2022 • 0 new comments

• [WIP] switch InputMetadata to using `SymInt`s

  #79070 commented on Jun 14, 2022 • 0 new comments

• Rocm enable tensorexpr

  #79075 commented on Jun 13, 2022 • 0 new comments

• Exponential integrals

  #79080 commented on Jun 13, 2022 • 0 new comments

• [buck2][caffe2] stable paths in generated file comments

  #79133 commented on Jun 13, 2022 • 0 new comments

• Add assertions to fix torch::jit::load bugs

  #79192 commented on Jun 15, 2022 • 0 new comments

• Implement cuda kernel for torch.sparse.spdiags

  #79228 commented on Jun 14, 2022 • 0 new comments

• proc

  #79234 commented on Jun 16, 2022 • 0 new comments

• [ONNX] Add dtype check in onnx verification

  #79263 commented on Jun 16, 2022 • 0 new comments

• ONNX: Convert aten numpy_T to ONNX transpose

  #79269 commented on Jun 16, 2022 • 0 new comments

• Port cholesky to structured kernels

  #79300 commented on Jun 11, 2022 • 0 new comments

• [WIP] Support float tensor as args when tracing

  #79319 commented on Jun 12, 2022 • 0 new comments