pytorch / pytorch Public

10 Pull requests merged by 9 people

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

  #63170 merged Sep 16, 2021

• [LTC] Enable CI for building and testing

  #65008 merged Sep 16, 2021

• [JIT] Improve BatchMM mutability handling

  #64955 merged Sep 15, 2021

• remove git stamp from release documentation version

  #58849 merged Sep 15, 2021

• [ONNX] Update sum symbolic to handle dtypes

  #64289 merged Sep 15, 2021

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

  #63589 merged Sep 14, 2021

• Fix builder pinning

  #64971 merged Sep 14, 2021

• [ONNX] Add prim::PythonOp check back in export.cpp

  #64944 merged Sep 13, 2021

• [Operator] Adding support for Threshold operator

  #64790 merged Sep 13, 2021

• (torch.distributed) Add torch.distributed.is_torchelastic_launched() util method + make init_method=tcp:// compatible with torchelastic (#63910)

  #64826 merged Sep 12, 2021

145 Pull requests opened by 77 people

• Back out "[pytorch] deprecate default_op_deps.yaml"

  #64884 opened Sep 11, 2021

• [Testing] Combined PR for updating VS by Robot

  #64893 opened Sep 12, 2021

• [Testing] Update VS version by Robot

  #64894 opened Sep 12, 2021

• Revert D30728580

  #64898 opened Sep 12, 2021

• [TESTING] Test if any wobble in this file breaks CUDA

  #64900 opened Sep 12, 2021

• [WIP] per-RPC device mapping

  #64901 opened Sep 13, 2021

• [nn] Fold : no batch dim

  #64909 opened Sep 13, 2021

• [quant] Add imports to the torch/ao/quantization/__init__.py

  #64911 opened Sep 13, 2021

• [DO NOT MERGE] A temporary PR to upgrade oneDNN

  #64915 opened Sep 13, 2021

• [caffe2] [aten] Remove loose (unpaired) #pragma warning ( pop ) in Te…

  #64924 opened Sep 13, 2021

• Disable target determination

  #64925 opened Sep 13, 2021

• [Static Runtime] Support aten::to.prim_dtype overload

  #64928 opened Sep 13, 2021

• Logic to categorize all the commits

  #64929 opened Sep 13, 2021

• Skip failing tests when LAPACK and MAGMA are not available

  #64930 opened Sep 13, 2021

• [Static Runtime] Implement and enable variadic tuple unpack

  #64934 opened Sep 13, 2021

• .github: Specify names for all generated jobs

  #64940 opened Sep 13, 2021

• Update ProcessGroup collective C++ APIs to be non-pure virtual functions

  #64943 opened Sep 13, 2021

• [wip] Add manual target determination

  #64950 opened Sep 13, 2021

• Clean up `ListLiteral` and `ListComprehension` emission logic

  #64952 opened Sep 13, 2021

• Clean up `DictLiteral` and `DictComprehension` emission logic

  #64953 opened Sep 13, 2021

• TST Adds test for non-contiguous tensors

  #64954 opened Sep 13, 2021

• [do not merge] Test docs determination

  #64956 opened Sep 13, 2021

• .circleci: Remove Windows 10.1 workflow

  #64959 opened Sep 13, 2021

• Fix `ListType` variance rules to account for dynamic types

  #64960 opened Sep 13, 2021

• [NNC] Pass on tensor names to intermediate tensors

  #64961 opened Sep 13, 2021

• .github: Move windows back to default directory

  #64962 opened Sep 13, 2021

• log API usage for fsdp API in PyTorch

  #64964 opened Sep 13, 2021

• [wip] support interface call.

  #64966 opened Sep 13, 2021

• [Static Runtime] Add aten::remainder out variant

  #64967 opened Sep 14, 2021

• Creating a helper function to generate an unique name for an attr in a module

  #64970 opened Sep 14, 2021

• [vulkan] Use volk to load vulkan libraries

  #64973 opened Sep 14, 2021

• [nn] MarginRankingLoss : no batch dim

  #64975 opened Sep 14, 2021

• [BC-breaking] Fixed interpolation output size to match opencv if scale factor is specified

  #64983 opened Sep 14, 2021

• [Static Runtime] Add perf metrics for number of managed tensors & unmanaged values

  #64992 opened Sep 14, 2021

• [DataPipe] Unlimited buffer for Forker and Demultiplexer

  #64994 opened Sep 14, 2021

• Doc changes Release/1.9

  #64998 opened Sep 14, 2021

• [ONNX] Modify softplus symbolic to support beta!=1

  #65001 opened Sep 14, 2021

• Back out "[Pytorch Edge] Quantized Ops Dtype Selective"

  #65003 opened Sep 14, 2021

• [PT/ShardedTensor]Allow zero size local shard

  #65007 opened Sep 14, 2021

• [static runtime] release values to same storage values

  #65009 opened Sep 14, 2021

• [JIT] canonicalize aten::rsub

  #65014 opened Sep 14, 2021

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

  #65019 opened Sep 14, 2021

• [iOS][GPU] Consolidate array and non-array kernel for copy_nchw_to_metal & copy_metal_to_nchw

  #65020 opened Sep 14, 2021

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

  #65021 opened Sep 14, 2021

• Specify pytorch-dev-infra for CI related files

  #65025 opened Sep 14, 2021

• [ONNX] test ORT 1.9 RC

  #65029 opened Sep 15, 2021

• Reduce PyTorch warnings: Cast fix xplat/caffe2/c10/core/TensorOptions.h

  #65030 opened Sep 15, 2021

• Reduce PyTorch warnings: Cast fix xplat/caffe2/aten/src/ATen/core/DeprecatedTypeProperties.h

  #65031 opened Sep 15, 2021

• [Not for land] Chunked version of all_gather

  #65032 opened Sep 15, 2021

• [quant] AO migration of the `torch/quantization/quantize_fx.py` and `torch/quantization/fx/*`

  #65033 opened Sep 15, 2021

• Add support to lower acc_ops.transpose

  #65036 opened Sep 15, 2021

• Decouple forward AD checks from backward AD in OpInfo tests and gradcheck

  #65040 opened Sep 15, 2021

• Bump dlpack.h to latest version

  #65047 opened Sep 15, 2021

• `torch.linalg.lstsq`: forward/backward AD support

  #65054 opened Sep 15, 2021

• [quant] Add fp32/fp16 zero_point support for CPU fakeQuant

  #65055 opened Sep 15, 2021

• [TESTING ONLY]

  #65060 opened Sep 15, 2021

• [JIT] Initialize CUDA context before launching fused kernel

  #65064 opened Sep 15, 2021

• Repro for python_mode bug

  #65068 opened Sep 15, 2021

• Add dependencies installation in contributing

  #65074 opened Sep 15, 2021

• [CoreML][OSS] Add Simulator tests

  #65076 opened Sep 15, 2021

• [JIT] Support device as Dict key

  #65079 opened Sep 15, 2021

• Remove boolean dispatch for fractional_max_pool2d

  #65085 opened Sep 15, 2021

• WIP refactor version generation

  #65089 opened Sep 15, 2021

• Add force_on_cpu test to win cuda10.2 on GHA

  #65094 opened Sep 15, 2021

• .github: GHA retry docker pull

  #65103 opened Sep 15, 2021

• [quant][fx][graphmode] Fix a bug for sub

  #65109 opened Sep 15, 2021

• [LTC] Merge master

  #65112 opened Sep 15, 2021

• [Static Runtime] Fuse ListUnpack + gather_ranges_to_dense

  #65116 opened Sep 16, 2021

• [PyPer] copy-free freeze_module

  #65118 opened Sep 16, 2021

• Pass through allow-list from prepare_qat into propagate_qconfig_ to allow custom mapping and custom QAT module

  #65119 opened Sep 16, 2021

• [jit] Add cache for MemoryDag::collectAllContainedMemoryLocations

  #65122 opened Sep 16, 2021

• [Static Runtime] Mark StaticModule and StaticRuntime with C10_DISABLE_COPY_AND_ASSIGN

  #65126 opened Sep 16, 2021

• Test lazy source range

  #65127 opened Sep 16, 2021

• Matrix multiplication for CUDA batched and non-batched int32/int64 tensors.

  #65133 opened Sep 16, 2021

• type annotate operator_support

  #65136 opened Sep 16, 2021

• Revert "Revert D30752939: [pytorch][PR] nvfuser update"

  #65137 opened Sep 16, 2021

• Enhancement/generator initialization [WIP]

  #65139 opened Sep 16, 2021

• Added antialias flag to interpolate (CPU only, bilinear)

  #65142 opened Sep 16, 2021

• [JIT][Easy] Shape cleanups

  #65148 opened Sep 16, 2021

• Fix logic to determine master vs PR

  #65155 opened Sep 16, 2021

• Track peak memory usage

  #65157 opened Sep 16, 2021

• [WIP] [JIT] Default value for Dict in SchemaParser

  #65158 opened Sep 16, 2021

• Remove orphan from cuDNN persistent note

  #65160 opened Sep 16, 2021

• [JIT] Delete obsolete message: or if you absolutely have to, use c10::impl::GenericDict(c10::impl::deprecatedUntypedDict())

  #65164 opened Sep 16, 2021

• Make new_empty/new_ones/new_zeros/new_full respect subclass

  #65169 opened Sep 16, 2021

• [PyTorch Edge] Revert copying of opname in Function::append_operator

  #65172 opened Sep 16, 2021

• [NCCL] Init dummy NCCL comms in constructor

  #65173 opened Sep 16, 2021

• Add register_module alias to nn.Module

  #65174 opened Sep 16, 2021

• [Static runtime] Micro-optimization pass on GetLivenessMap

  #65175 opened Sep 16, 2021

• [jit] Fix list copy in MemoryDAG

  #65176 opened Sep 16, 2021

• [jit] Implement one-argument AliasDb::mayContainAlias more efficiently

  #65177 opened Sep 16, 2021

• [jit] Implement one-element MemoryDAG::mayContainAlias more efficiently

  #65178 opened Sep 16, 2021

• [ddp] use named_params and named_buffers explicitly

  #65181 opened Sep 16, 2021

• Add __module__ to torch.dtype.__dict__

  #65182 opened Sep 16, 2021

• [pytorch] more informative error msg from fbgemm embedding spmdm call

  #65186 opened Sep 17, 2021

• [FX] Fix default tensor values

  #65187 opened Sep 17, 2021

• [Pipe] Add a `WithDevice` wrapper to specify device execution for a module.

  #65190 opened Sep 17, 2021

• feed model merge net lower benchmark

  #65191 opened Sep 17, 2021

• replace onlyOnCPUAndCUDA with onlyNativeDeviceTypes

  #65201 opened Sep 17, 2021

• [ROCm] update Magma for 4.3 release

  #65203 opened Sep 17, 2021

• [op_auth] Shape-specialization key for op caching

  #65204 opened Sep 17, 2021

• [op-auth] Helper to convert SpecializationKey to python object

  #65205 opened Sep 17, 2021

• [op-auth] Add call_with_numel interface for fast CUDA calls

  #65206 opened Sep 17, 2021

• [op-auth] Class for caching compilation results

  #65207 opened Sep 17, 2021

• [op-auth] Proxies for binding compilation results to python objects

  #65208 opened Sep 17, 2021

• [op-auth] Num arg-and-dim specialized cache for generated kernels

  #65209 opened Sep 17, 2021

• [op-auth] Num arg specialized cache

  #65210 opened Sep 17, 2021

• [op-auth] Complete compile cache, with in-out specialization

  #65211 opened Sep 17, 2021

• [op-auth] Python bindings for compilation cache

  #65212 opened Sep 17, 2021

• [nnc] Add BufHandle.store to python API

  #65213 opened Sep 17, 2021

• Remove device type suffix from parametrized test names

  #65217 opened Sep 17, 2021

• No-op version of general convolution_backward

  #65219 opened Sep 17, 2021

• [DataPipe] Fix deepcopy filehandle for Mapper and in-place modification for IterableWrapper

  #65220 opened Sep 17, 2021

• Use src size for memcpy in order to avoid fortify complaints

  #65222 opened Sep 17, 2021

• [fx] give warning instead of fatal the program when submod not found during adding get_attr

  #65225 opened Sep 17, 2021

• fix typo missing f string

  #65226 opened Sep 17, 2021

• Remove incorrect stride assert in Reduce.cuh

  #65227 opened Sep 17, 2021

• [DRAFT] New reduction kernel

  #65228 opened Sep 17, 2021

• OpInfo: nn.functional.conv2d

  #65233 opened Sep 17, 2021

• [PyTorchEdge] Compare stringstream while serializing

  #65234 opened Sep 17, 2021

• Fix autograd engine checks and update InputMetadata

  #65235 opened Sep 17, 2021

• [in progress] add op to mkldnn_convolution

  #65238 opened Sep 17, 2021

• Remove @balioglu from PyTorch Distributed code owners

  #65239 opened Sep 17, 2021

• Back out "Dispatch.h: Avoid including ivalue"

  #65240 opened Sep 17, 2021

• [quant] Add op benchmark for CPU FakeQuantizePerChannel with float zero_points

  #65241 opened Sep 17, 2021

• Avoid saving self for`softmax` and `log_softmax`

  #65242 opened Sep 17, 2021

• [XROS][ML] System specific adjustments for UTs to work.

  #65245 opened Sep 17, 2021

• [CUDA graphs] Beta, not prototype

  #65247 opened Sep 17, 2021

• [rfc] Add periodic auto-formatting job

  #65249 opened Sep 17, 2021

• Move Lazy Tensor into torch/csrc, build with CMake

  #65251 opened Sep 17, 2021

• [tmp]

  #65254 opened Sep 17, 2021

• [jit] Factor findAllNodes into one place.

  #65257 opened Sep 17, 2021

• [jit][edge] Export maybe-used interface methods from modules.

  #65258 opened Sep 17, 2021

• [jit][edge] Load interface methods to corresponding ClassTypes.

  #65259 opened Sep 17, 2021

• [jit][edge] Support INTERFACE_CALL to ClassType at runtime.

  #65260 opened Sep 17, 2021

• Revert "Revert D30558877: Ported std/var to ReductionOpInfo

  #65262 opened Sep 18, 2021

• [ONNX] publicize func for discovering unconvertible ops

  #65285 opened Sep 18, 2021

• [quant][fx2trt] Add quantize_per_channel in acc_ops and acc_ops_converter

  #65287 opened Sep 18, 2021

• [quant][fx2trt] Generate engine graph for explicit quant/implicit quant and fp16 graph

  #65289 opened Sep 18, 2021

• Adds keyword only args to gradcheck

  #65290 opened Sep 18, 2021

• [sparsity] Remove the pack_param from the sparsifier state_dict

  #65292 opened Sep 18, 2021

• [sparsity] Fix for accumulation bug in WeightNormSparsifier

  #65293 opened Sep 18, 2021

• [sparsity][doc] Docstring for WeightNormSparsifier

  #65294 opened Sep 18, 2021

• [sparsity] Add m-out-of-n support in the WeightNormSparsifier

  #65295 opened Sep 18, 2021

• [sparsity] Change API to take FQNs as configuration

  #65296 opened Sep 18, 2021

56 Issues closed by 32 people

• test_allgather_coalesced_async is flaky

  #65237 closed Sep 17, 2021

• [Apple M1] development build failure

  #64137 closed Sep 17, 2021

• How to register a Module as one custom OP when export to onnx

  #65199 closed Sep 17, 2021

• Any way to map the node in onnx and the module in pytorch modules?

  #65192 closed Sep 17, 2021

• DISABLED test_create_store_timeout_on_server (__main__.DistributedUtilTest)

  #64789 closed Sep 17, 2021

• Conda Package: cpuonly should be a mutex package

  #40213 closed Sep 17, 2021

• Passing an empty tensor to EmbeddingBag segfaults

  #50286 closed Sep 17, 2021

• Torchhub core tests are silently broken because of a Github API rate issue

  #64760 closed Sep 17, 2021

• can‘t build pytorch v1.9.0 error:`error: namespace "thrust" has no member "host_vector"`

  #64984 closed Sep 17, 2021

• torch.matmul does not correctly capture the shape mismatch error for CUDA tensors

  #65120 closed Sep 16, 2021

• importing torch.nn is giving error, but works well while importing torch i find it weird #10908

  #64877 closed Sep 16, 2021

• when custom op :RuntimeError: No Op registered for Qconv with domain_version of 9

  #65057 closed Sep 16, 2021

• [JIT] Improve BatchMM's handling of Mutability

  #64608 closed Sep 16, 2021

• Checking the dimensions of query, key, value in nn.MultiheadAttention

  #64888 closed Sep 16, 2021

• LSTM.weight_ih_l[k] dimensions with proj_size

  #65053 closed Sep 16, 2021

• distributed training not starting

  #65121 closed Sep 16, 2021

• Cannot export model using onnx - TypeError: elu() got an unexpected keyword argument 'scale' (occurred when translating elu)

  #19157 closed Sep 16, 2021

• RuntimeError when exporting model with reflectionpad into onnx

  #18323 closed Sep 16, 2021

• [TensorExpr] Var names are not unique in the Stmt (despite appearing unique when printed)

  #52727 closed Sep 16, 2021

• GHA failures require scanning the log

  #64614 closed Sep 15, 2021

• linalg.eig segfault if NaN in tensor with mkl

  #64671 closed Sep 15, 2021

• Why DataLoader need a separate thread for pin_memory?

  #65046 closed Sep 15, 2021

• Investigate warning of unclosed ArchiveStream in DataPipe

  #64281 closed Sep 15, 2021

• when custom op :RuntimeError: No Op registered for Qconv with domain_version of 9

  #65056 closed Sep 15, 2021

• Multiprocessing error with new NVIDIA driver

  #64800 closed Sep 15, 2021

• AttributeError: module 'torch' has no attribute 'unique_dim'

  #62793 closed Sep 15, 2021

• Why is there a GPU with a utilization rate of 0 when running pytorch's DDP task?

  #64989 closed Sep 15, 2021

• Exporting the Sigmoid operator svd to ONNX opset version 9 is not supported

  #44228 closed Sep 14, 2021

• JIT fails on creating bfloat16 fusions

  #53918 closed Sep 14, 2021

• [TensorExpr] Make `Placeholder` objects memory-managed

  #45705 closed Sep 14, 2021

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

  #61430 closed Sep 14, 2021

• TorchScript will report an error if my `lengths` is a CUDA tensor

  #41582 closed Sep 14, 2021

• isTensor() INTERNAL ASSERT FAILED at "C:\\w\\b\\windows\\pytorch\\aten\\src\\ATen/core/ivalue_inl.h":117

  #42160 closed Sep 14, 2021

• using Cuda, Libtorch. register_module(ModuleHolder/pimpl.h) error

  #64908 closed Sep 14, 2021

• Documentation for building from source not adequate

  #64875 closed Sep 14, 2021

• torch.unique_consecutive() is very slow when dim is specified even with 1-d tensors

  #62690 closed Sep 14, 2021

• PEP-503 simple repository API gives access denied

  #63098 closed Sep 14, 2021

• Segfault on setting gradient value to instance of user-defined class

  #64813 closed Sep 13, 2021

• RuntimeError: Exporting the operator linspace to ONNX opset version 11 is not supported.

  #58472 closed Sep 13, 2021

• `INTERNAL ASSERT FAILED` when profiling/timing on CUDA, not CPU

  #64946 closed Sep 13, 2021

• Run QAT model on TVM

  #64949 closed Sep 13, 2021

• Add custom quantization for saved for backwards tensors

  #64729 closed Sep 13, 2021

• Update torch.symeig doc to correctly reflect that it supports complex tensors too

  #53646 closed Sep 13, 2021

• Fix Doc Preview Links + add cppdocs build

  #64591 closed Sep 13, 2021

• torch.topk has trouble with inf's and nan's

  #16762 closed Sep 13, 2021

• Unhelpful error message for mismatched shapes for Linear layer on CUDA

  #64920 closed Sep 13, 2021

• torch.clamp fails with half precision args

  #64840 closed Sep 13, 2021

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

  #21987 closed Sep 13, 2021

• [BE] Clean up ProcessGroupAgent vestige

  #63169 closed Sep 13, 2021

• Dataloader halts when used with Keras TensorFlow Backend code

  #17871 closed Sep 13, 2021

• How to use python to implement _VF.lstm

  #64904 closed Sep 13, 2021

• About the C++ frontend API

  #17968 closed Sep 13, 2021

• Issue of "failed with initial frozen solve"

  #64906 closed Sep 13, 2021

• Implement improved DLPack support

  #55090 closed Sep 13, 2021

• How to get "finfo" in C++ torchlib like that in pytorch

  #64793 closed Sep 13, 2021

• test_python_dispatch fails under pytest

  #62501 closed Sep 13, 2021

103 Issues opened by 69 people

• torch.nn.CosineSimilarity() gives different results for same tensors with fp16 and fp32 datatypes

  #65298 opened Sep 18, 2021

• [JIT] TorchScript not works as expected when ModuleDict has modules with different signatures

  #65297 opened Sep 18, 2021

• Under what circumstances, DistributedDataParallel() doesn’t provide reduction in training time?

  #65288 opened Sep 18, 2021

• Add an `all_paths` method to PackageExporter

  #65261 opened Sep 18, 2021

• Add minimum image size to models page

  #65255 opened Sep 17, 2021

• TestCppExtensionJit's test_crash_handler fails for force_on_cpu job on GHA

  #65253 opened Sep 17, 2021

• `test_load_commit_from_forked_repo` started to fail with "Rate Limit Exceeded"

  #65252 opened Sep 17, 2021

• Rendered docs (in Dr. CI) failing to appear on recent PR

  #65250 opened Sep 17, 2021

• [Efficiency] Explicit prefetching data in FSDP

  #65248 opened Sep 17, 2021

• Links to PR CI runs on each commit missing from recent PR

  #65246 opened Sep 17, 2021

• Custom autograd sums output along batch dimension instead of failing

  #65244 opened Sep 17, 2021

• `empty_*` functions don't return subclasses properly

  #65243 opened Sep 17, 2021

• DISABLED test_allgather_coalesced_async (__main__.ProcessGroupGlooTest)

  #65231 opened Sep 17, 2021

• RuntimeError: Trying to create tensor with negative dimension -1741885395: [-1741885395]

  #65230 opened Sep 17, 2021

• at::parallel_for created max_threads for inputs larger than GRAIN_SIZE

  #65229 opened Sep 17, 2021

• [DataPipe] Mapper DataPipe should not deepcopy when index specified

  #65221 opened Sep 17, 2021

• indexing a tensor with a numpy array

  #65218 opened Sep 17, 2021

• Add function that retrieves a batch from a DataLoader

  #65216 opened Sep 17, 2021

• _disabled_torch_function_impl should not be necessary for __torch_dispatch__

  #65215 opened Sep 17, 2021

• Microsoft Visual C++ Redistributable is not installed, this may lead to the DLL load failure.

  #65214 opened Sep 17, 2021

• Upsample Module exist bug

  #65200 opened Sep 17, 2021

• Multi-processing leaking file descriptors

  #65198 opened Sep 17, 2021

• [Beta] Detect globally unused parameters in FSDP

  #65195 opened Sep 17, 2021

• [Beta] Avoid register hooks to AccumulateGrad in every FSDP iteration

  #65194 opened Sep 17, 2021

• [Better Engineering] Generalize DDPSink and consolidate it with FSDP's pre-backward hooks

  #65193 opened Sep 17, 2021

• [quant] Add support for Embedding/EmbeddingBag quantization via dynamic quant APIs

  #65185 opened Sep 17, 2021

• Review the Tensor and Operator Basics Onboarding page

  #65166 opened Sep 16, 2021

• Make gradcheck settings kwarg only

  #65165 opened Sep 16, 2021

• Parametrization cannot be parametrized

  #65163 opened Sep 16, 2021

• error: ‘_mm512_set_epi8’ was not declared in this scope when build from source on AMD 3600 CPU

  #65161 opened Sep 16, 2021

• [feature request] Jagged / padding version of torch.stack / torch.cat

  #65156 opened Sep 16, 2021

• Add simple backwards compatibility check for torch.package

  #65154 opened Sep 16, 2021

• Unification of model initialization methods / naming across domain libraries + support of skip_init

  #65153 opened Sep 16, 2021

• Registering a global fallback for all operators that defaults us to assuming that autograd is not implemented

  #65151 opened Sep 16, 2021

• jit.trace's returned module returns a tuple instead of a named tuple

  #65144 opened Sep 16, 2021

• C++ BatchNorm pretty_print() fails with optional momentum

  #65143 opened Sep 16, 2021

• profiler_legacy `_parse_legacy_records` deduplication logic has false-positive.

  #65140 opened Sep 16, 2021

• GNN (SAGE Convolutional) ONNX export

  #65138 opened Sep 16, 2021

• How to reference a tensor variable from a superclass of `torch.Tensor`?

  #65132 opened Sep 16, 2021

• Adding a device variable to a `torch.Tensor` subclass fails

  #65131 opened Sep 16, 2021

• User Experience Expectation Regarding `XLATensor` Design

  #65130 opened Sep 16, 2021

• [Comment] remove torch.cuda.synchronize() in state_dict()

  #65117 opened Sep 16, 2021

• [Comment] avoid to append a list of chunks in _get_shard()

  #65111 opened Sep 15, 2021

• [Efficiency] Avoid H2D copy and stream sync in clip_grad_norm_

  #65110 opened Sep 15, 2021

• [Comment] calc_grad_norm_ always compute the norm in full precision and avoid duplication code in clip_grad_norm_ of torch core

  #65108 opened Sep 15, 2021

• [Bug] [CUDA IPC] CUDA IPC memory cost

  #65100 opened Sep 15, 2021

• Bug: torch.jit.load cannot load from IOBuffer

  #65098 opened Sep 15, 2021

• [FX] `passes.split_module` doesn't have documentation

  #65096 opened Sep 15, 2021

• [AWS EC2 P3DN] torch sync.pipeline makes the tensor passing to the next stage in CPU device; it's working without using pipeline

  #65093 opened Sep 15, 2021

• Test reporting misses breakages in test framework

  #65091 opened Sep 15, 2021

• ShardedTensor.size() should be equivalent to torch.Tensor.size()

  #65087 opened Sep 15, 2021

• THPVariableCheck should check the return code for PyObject_IsInstance

  #65084 opened Sep 15, 2021

• THPVariable_subclass_dealloc (i.e. tp_dealloc for torch.Tensor) should handle Python error state gracefully

  #65083 opened Sep 15, 2021

• [Efficiency] Sometimes CudaMalloc is slow in FSDP

  #65078 opened Sep 15, 2021

• Pickling `torch.dtype` objects is unnecessarily slow

  #65077 opened Sep 15, 2021

• Invalid cuda context when running nvfuser with LTC

  #65073 opened Sep 15, 2021

• torch.linalg.cholesky does not raise a RuntimeError when the matrix is not symmetric

  #65069 opened Sep 15, 2021

• [DataPipe] Add Test Cases to Ensure Correct Behaviors When IterDataPipes Reset

  #65067 opened Sep 15, 2021

• Gloo and TensorPipe depend on different version of libuv

  #65062 opened Sep 15, 2021

• [Bug] `torch.linalg.lstsq` in backward + `gradcheck` = nondeterminism + Jacobian mismatch for TALL inputs

  #65061 opened Sep 15, 2021

• Add nnapi serialization for module components in Wav2Vec2Model

  #65058 opened Sep 15, 2021

• Use hyphens to separate long CLI parameters instead of underscores

  #65052 opened Sep 15, 2021

• deepcopy of Lazy modules returns an exception

  #65051 opened Sep 15, 2021

• torch.cross precision problem

  #65050 opened Sep 15, 2021

• Setting a random seed or passing a generator to Modules for deterministic sampling.

  #65049 opened Sep 15, 2021

• Difference in git behaviour in CentOS vs Ubuntu base dockers for ROCm

  #65048 opened Sep 15, 2021

• RuntimeError: Exporting the operator index_add to ONNX opset version 12 is not supported.

  #65039 opened Sep 15, 2021

• [pytorch v1.9.0] for torch sync.pipeline in a single GPU, passing tensor layer by layer is better or using the skip-connection API is better?

  #65038 opened Sep 15, 2021

• CUDA out of memory

  #65027 opened Sep 14, 2021

• Implement .gather API for ShardedTensor

  #65026 opened Sep 14, 2021

• [AWS EC2 P3DN, EFA is enabled] Torch RPC tensorpipe/common/ibv.h:172 "": Operation not supported

  #65022 opened Sep 14, 2021

• Idiom for Tensor subclasses that are also Parameters

  #65018 opened Sep 14, 2021

• Fix autograd engine checks

  #65016 opened Sep 14, 2021

• Some tests are running multiple times

  #65012 opened Sep 14, 2021

• Traceback tensor mode

  #65002 opened Sep 14, 2021

• Error on MacOS related to custom class registration in torchscript

  #65000 opened Sep 14, 2021

• [feature request] OpInfo: possibility to test forward AD without backward AD implemented

  #64999 opened Sep 14, 2021

• FX failed to normalize op `block_diag` and `broadcast_tensors`

  #64997 opened Sep 14, 2021

• Bump DLPack dependency

  #64995 opened Sep 14, 2021

• [DataPipe] Forker and Demultiplexer should be able to take unlimited buffer

  #64991 opened Sep 14, 2021

• Distributed Elastic Training Failed on Two Amazon g4dn.xlarge Instances

  #64990 opened Sep 14, 2021

• RandomSampler / DistributedSampler does not seem really random

  #64986 opened Sep 14, 2021

• Feature request: convert .pth to ONNX

  #64980 opened Sep 14, 2021

• torch.nn.functional.linear function doesn't validate weight dimension == 2

  #64978 opened Sep 14, 2021

• `grid_sample` backward pass performance scales poorly with `input` size

  #64977 opened Sep 14, 2021

• NNC should rely on schema rather than on node kind in lowerings

  #64963 opened Sep 13, 2021

• `INTERNAL ASSERT FAILED` after `torch.jit.freeze`

  #64957 opened Sep 13, 2021

• Quantile is limited to 16 million elements and have poor performance.

  #64947 opened Sep 13, 2021

• `with-ssh` session are not terminated when job is completed/cancelled

  #64941 opened Sep 13, 2021

• BC CI error message should link to some information about how to squash the warning

  #64939 opened Sep 13, 2021

• [proposal] Let's unify the various gather/scatter ops.

  #64936 opened Sep 13, 2021

• [RFC] TorchStore - A Shared-Memory Tensor Store

  #64932 opened Sep 13, 2021

• native functions should not be allowed to take in a `grad` argument

  #64931 opened Sep 13, 2021

• Distributed Elastic Training Got Stuck on Rendezvous Stage With Multiple NICs

  #64926 opened Sep 13, 2021

• TorchScript error loading module with float literal default argument

  #64918 opened Sep 13, 2021

• [nn] im2col and col2im mutate input if it has no batch dim

  #64907 opened Sep 13, 2021

• [docs] [feature request] Tutorial on implementing correctly distributed dataset evaluators

  #64905 opened Sep 13, 2021

• Test determination is incorrectly skipping tests

  #64902 opened Sep 13, 2021

• Either support torch.mean on BoolTensors or fix the error message

  #64897 opened Sep 12, 2021

• pytorch1.9.0 can not use torch.jit.load() to load model which save in pytorch1.3

  #64896 opened Sep 12, 2021

• AttributeError: '_MultiProcessingDataLoaderIter' object has no attribute '_workers_status'

  #64895 opened Sep 12, 2021

• "RuntimeError: CUDA error: out of memory" for no reason

  #64889 opened Sep 12, 2021

• Set warning or error with `use_deterministic_algorithms`

  #64883 opened Sep 11, 2021

362 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.

• [JIT] memory planning base with naive strategy

  #64347 commented on Sep 15, 2021 • 75 new comments

• upstreaming fairscale fsdp to PyTorch

  #63881 commented on Sep 18, 2021 • 74 new comments

• Add basic ltc codegen to staging branch

  #64104 commented on Sep 17, 2021 • 56 new comments

• [Structured Kernels] Port for `baddbmm` and `bmm`

  #64805 commented on Sep 18, 2021 • 40 new comments

• Add functional api for `nn.Module`

  #61447 commented on Sep 17, 2021 • 27 new comments

• add slice/select/diagonal_scatter variants as primitive ops

  #64430 commented on Sep 16, 2021 • 19 new comments

• Add a Functionalization pass in core

  #64432 commented on Sep 18, 2021 • 16 new comments

• enable convolution relu fusion path for cpu path

  #64639 commented on Sep 15, 2021 • 15 new comments

• [ShapProp] Refactor TensorMeta to use qparams for more general quantization support

  #64787 commented on Sep 18, 2021 • 15 new comments

• OpInfo for `nn.functional.batch_norm`

  #63218 commented on Sep 17, 2021 • 13 new comments

• Cleaning up DDP SPMD in reducer.cpp

  #64113 commented on Sep 17, 2021 • 11 new comments

• Basic implementation of ShardedLinear using ShardedTensor.

  #64128 commented on Sep 18, 2021 • 11 new comments

• [ROCm] Changes not to rely on CUDA_VERSION or HIP_VERSION

  #64597 commented on Sep 17, 2021 • 11 new comments

• Automated submodule update: FBGEMM

  #64640 commented on Sep 17, 2021 • 11 new comments

• [nn] TripletMarginLoss and PairwiseDistance : no batch dim

  #64882 commented on Sep 17, 2021 • 11 new comments

• enable signature forwarding in `boolean_dispatch`

  #64420 commented on Sep 16, 2021 • 10 new comments

• Include shape of the tensor

  #64421 commented on Sep 16, 2021 • 10 new comments

• Added nearest-exact interpolation mode

  #64501 commented on Sep 17, 2021 • 9 new comments

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

  #62084 commented on Sep 17, 2021 • 8 new comments

• [RELAND] Enable ncclAvg for reductions

  #62835 commented on Sep 17, 2021 • 8 new comments

• Micro-optimisations for matmul

  #64387 commented on Sep 15, 2021 • 8 new comments

• torch.unique acting up for a binary tensor

  #64041 commented on Sep 14, 2021 • 7 new comments

• Installation of torch and torchvision not happening with Poetry

  #64520 commented on Sep 15, 2021 • 7 new comments

• RuntimeError - onnx export a model with transformer layer

  #61277 commented on Sep 18, 2021 • 7 new comments

• sync c10d NCCL version parsing with formatting

  #63155 commented on Sep 15, 2021 • 7 new comments

• support using gradients named for outputs in derivatives

  #63947 commented on Sep 17, 2021 • 7 new comments

• [ONNX] Enable export of __xor_ (#64042)

  #64581 commented on Sep 17, 2021 • 7 new comments

• TST Adds module test comparing output of cpu and gpu

  #64694 commented on Sep 14, 2021 • 7 new comments

• Incorporate Posit Support

  #57724 commented on Sep 17, 2021 • 6 new comments

• _disabled_torch_function_impl is unimplementable (must be special cased)

  #64690 commented on Sep 17, 2021 • 6 new comments

• optimize sort

  #64808 commented on Sep 16, 2021 • 6 new comments

• I am unable to detect torch.Tensor._version change during __torch_dispatch__ call

  #64874 commented on Sep 14, 2021 • 5 new comments

• Add support to profile all RPC worker threads that are run on a process

  #64799 commented on Sep 14, 2021 • 5 new comments

• Java NVIDIA GPU Inference Support

  #63581 commented on Sep 15, 2021 • 5 new comments

• CompositeImplicitAutograd operators should not perform operations that do not dispatch

  #61669 commented on Sep 16, 2021 • 5 new comments

• Ability to explicitly close/dispose a DataLoader

  #64766 commented on Sep 18, 2021 • 5 new comments

• Support `torch.linalg.trace`

  #62714 commented on Sep 16, 2021 • 5 new comments

• simplify `torch.meshgrid`'s shape computation

  #62905 commented on Sep 17, 2021 • 5 new comments

• Fix batch_isend_irecv tests for err case

  #63112 commented on Sep 15, 2021 • 5 new comments

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

  #63621 commented on Sep 17, 2021 • 5 new comments

• Fixed custom op sizes not appearing in profiler trace file or execution graph.

  #64325 commented on Sep 18, 2021 • 5 new comments

• [CUDA][Linalg] Add gesvd as SVD fallback; optimize SVD gesvdj performance

  #64533 commented on Sep 14, 2021 • 5 new comments

• Port `min` and `max` full reductions to structured kernels

  #64735 commented on Sep 13, 2021 • 5 new comments

• `torch.lu_unpack`: forward AD support

  #64810 commented on Sep 15, 2021 • 5 new comments

• profiler crashes the program when trying to stop on different threads without context manager

  #64795 commented on Sep 13, 2021 • 4 new comments

• [JIT] Serialization of Sparse Tensors doesnt work

  #63987 commented on Sep 15, 2021 • 4 new comments

• Add option for Convolutions to operate over NTC and NHWC tensors

  #50962 commented on Sep 15, 2021 • 4 new comments

• [Fixed] Enable Half, BFloat16, and Complex dtypes for coo-coo sparse matmul [CUDA]

  #59980 commented on Sep 16, 2021 • 4 new comments

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

  #61168 commented on Sep 17, 2021 • 4 new comments

• Port `sort` kernel to structured kernels.

  #62391 commented on Sep 16, 2021 • 4 new comments

• Add relative and absolute tolerances for matrix_rank, pinv

  #63102 commented on Sep 17, 2021 • 4 new comments

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

  #63364 commented on Sep 16, 2021 • 4 new comments

• Remove native_functions.yaml dependency from Activation.cu

  #64499 commented on Sep 15, 2021 • 4 new comments

• [JIT] Make output of prim::TupleConstruct alias only with its inputs

  #64879 commented on Sep 16, 2021 • 4 new comments

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

  #61686 commented on Sep 13, 2021 • 3 new comments

• [v.1.9.1] Release Tracker

  #62586 commented on Sep 14, 2021 • 3 new comments

• Key already registered with the same priority: uv

  #62876 commented on Sep 15, 2021 • 3 new comments

• [Meta] PyTorch GitHub Action migration issues

  #61604 commented on Sep 15, 2021 • 3 new comments

• Improve F.interpolate error message

  #63845 commented on Sep 16, 2021 • 3 new comments

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

  #64327 commented on Sep 17, 2021 • 3 new comments

• Full Backward Hooks get applied even when grad is disabled

  #59901 commented on Sep 18, 2021 • 3 new comments

• DistributedStreamSampler class: support stream sampler in distributed setting

  #32420 commented on Sep 12, 2021 • 3 new comments

• Document torch.quantile interpolation kwarg

  #59397 commented on Sep 17, 2021 • 3 new comments

• Port `min` kernel to structured kernels.

  #61450 commented on Sep 16, 2021 • 3 new comments

• [pytorch][PR] Remove dtype from torch.Storage and use only torch.ByteStorage

  #62030 commented on Sep 18, 2021 • 3 new comments

• Enable CUPTI for kineto by default on windows.

  #62175 commented on Sep 18, 2021 • 3 new comments

• clarify implementation of check_grad_usage

  #64439 commented on Sep 17, 2021 • 3 new comments

• TST Adds gradcheck and gradgradcheck to module info

  #64444 commented on Sep 14, 2021 • 3 new comments

• [DDP] Custom buffer reduction

  #64513 commented on Sep 18, 2021 • 3 new comments

• Fix nll_backward for negative weights

  #64572 commented on Sep 14, 2021 • 3 new comments

• Bug in CosineAnnealingWarmRestarts in optim/lr_scheduler.py

  #64758 commented on Sep 15, 2021 • 3 new comments

• `__torch_function__` docs should use @classmethod

  #63767 commented on Sep 13, 2021 • 2 new comments

• Improve Docs builds to not have to push to a repository to update

  #64711 commented on Sep 13, 2021 • 2 new comments

• Opset warning keeps popping up even if opset version is set to 11 for torch.onn.export

  #64485 commented on Sep 14, 2021 • 2 new comments

• Distributed: RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation: [torch.cuda.FloatTensor [2048]] is at version 4; expected version 3 instead

  #62474 commented on Sep 14, 2021 • 2 new comments

• PyTorch 1.6 DataParallel causes CUDNN_STATUS_BAD_PARAM in backward pass

  #44551 commented on Sep 14, 2021 • 2 new comments

• Tracing models all the way down to basic functions

  #63076 commented on Sep 14, 2021 • 2 new comments

• DataParallel doesn't properly handle kwargs

  #31460 commented on Sep 14, 2021 • 2 new comments

• Detect OpenMP Loop and this application may hang warnings

  #52047 commented on Sep 15, 2021 • 2 new comments

• Named Tensors to support custom unify functions

  #64743 commented on Sep 15, 2021 • 2 new comments

• Simplification of pruned models

  #58846 commented on Sep 15, 2021 • 2 new comments

• Exporting the operator inverse to ONNX is not supported

  #64577 commented on Sep 16, 2021 • 2 new comments

• CUDA error (cublasSgemm / cuda runtime error (719) / etc)

  #29795 commented on Sep 16, 2021 • 2 new comments

• Conversion from bool to float sometimes produces tensor with values 255.0 for True instead of 1.0

  #54789 commented on Sep 16, 2021 • 2 new comments

• torch.onnx.export crash while export quantized model

  #64477 commented on Sep 17, 2021 • 2 new comments

• FSDP followup improvements

  #64618 commented on Sep 17, 2021 • 2 new comments

• torchhub for private repository

  #44174 commented on Sep 14, 2021 • 2 new comments

• [JIT] Display error message when NamedTuples have user-defined functions

  #49115 commented on Sep 17, 2021 • 2 new comments

• Automated submodule update: kineto

  #59674 commented on Sep 16, 2021 • 2 new comments

• Added optimize kwarg to torch.einsum

  #60191 commented on Sep 17, 2021 • 2 new comments

• Implements allreduce_coalesced for ProcessGroupNCCL

  #62140 commented on Sep 15, 2021 • 2 new comments

• avoid re-allocation of view_shape for every tensor in `torch.meshgrid`

  #62908 commented on Sep 17, 2021 • 2 new comments

• Updated documentation for all reduction ops

  #63334 commented on Sep 17, 2021 • 2 new comments

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

  #63393 commented on Sep 16, 2021 • 2 new comments

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

  #63634 commented on Sep 17, 2021 • 2 new comments

• Add InplaceOrView boxed kernel

  #63878 commented on Sep 14, 2021 • 2 new comments

• Remove duplicated AVX512 detection

  #63924 commented on Sep 15, 2021 • 2 new comments

• Fix NCCL version parsing for 2.10.3

  #64028 commented on Sep 14, 2021 • 2 new comments

• Add tensor.{adjoint(),H,mT,mH} methods and properties

  #64179 commented on Sep 16, 2021 • 2 new comments

• [ONNX] Add log10 symbolic (#63418)

  #64374 commented on Sep 17, 2021 • 2 new comments

• [ONNX] Update instance_norm implementation and support training (#60538)

  #64375 commented on Sep 17, 2021 • 2 new comments

• Fix empty size constant creation (#63607)

  #64376 commented on Sep 17, 2021 • 2 new comments

• [ONNX] Fix input sequence for pad op (#60554)

  #64377 commented on Sep 17, 2021 • 2 new comments

• [ONNX] Fix cuda test case (#63597)

  #64378 commented on Sep 17, 2021 • 2 new comments

• [ONNX] Fix gather squeeze axis in constant folding (#63588)

  #64379 commented on Sep 17, 2021 • 2 new comments

• Add onnx test for batched_nms (#53175)

  #64381 commented on Sep 17, 2021 • 2 new comments

• [codegen] teach C++ reference types how to make copies of themselves

  #64431 commented on Sep 16, 2021 • 2 new comments

• [ONNX] Add diagonal symbolic

  #64454 commented on Sep 17, 2021 • 2 new comments

• change default grain size in gather scatter kernel to improve CPU performance

  #64478 commented on Sep 15, 2021 • 2 new comments

• [DDP] Allow await of custom buffer reduction in backward

  #64515 commented on Sep 18, 2021 • 2 new comments

• [ONNX] Fix remainder export (#64230)

  #64578 commented on Sep 17, 2021 • 2 new comments

• [ONNX] Update doc and error message for indexing export (#64290)

  #64579 commented on Sep 17, 2021 • 2 new comments

• [ONNX] Fix inplace fill_ dtype export mismatch (#64233)

  #64580 commented on Sep 17, 2021 • 2 new comments

• Make it so you cannot call THPModule_disable_torch_function directly

  #64693 commented on Sep 14, 2021 • 2 new comments

• [tensorexpr] Add Op handling for mobilenetv3 large

  #64741 commented on Sep 17, 2021 • 2 new comments

• [tensorexpr] Handle linear with addmm

  #64751 commented on Sep 17, 2021 • 2 new comments

• [PyTorch] Store Argument::alias_info_ on the heap

  #64824 commented on Sep 17, 2021 • 2 new comments

• [WIP] Add efficient zero tensors

  #64837 commented on Sep 13, 2021 • 2 new comments

• Inconsistent NaN handling by cholesky between CPU and CUDA

  #64818 commented on Sep 11, 2021 • 1 new comment

• How to view VLOG information

  #57118 commented on Sep 12, 2021 • 1 new comment

• creation of a tensor from a numba.cuda array

  #23067 commented on Sep 13, 2021 • 1 new comment

• PyTorch inference on tensors of a particular size cause Illegal Instruction (core dumped) on Jetson Nano

  #64812 commented on Sep 13, 2021 • 1 new comment

• ONNX Slice of dynamic tensors

  #64732 commented on Sep 13, 2021 • 1 new comment

• Building without LAPACK causes tests to fail

  #59662 commented on Sep 13, 2021 • 1 new comment

• Add support for GPU based data loading and transformation

  #64881 commented on Sep 13, 2021 • 1 new comment

• Unreasonable memory usage for hello world example

  #64720 commented on Sep 13, 2021 • 1 new comment

• linalg.lstsq out variant fails internal assert because it uses non-inplace view op for some inputs

  #64794 commented on Sep 13, 2021 • 1 new comment

• support grid_sample with batch=1 but supprting batch affine parameters

  #2732 commented on Sep 13, 2021 • 1 new comment

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

  #48279 commented on Sep 13, 2021 • 1 new comment

• Grad strides do not match bucket view strides.

  #47163 commented on Sep 14, 2021 • 1 new comment

• Mixed precision causes NaN loss

  #40497 commented on Sep 14, 2021 • 1 new comment

• Fix the non-converging issue of SVD on GPU for large matrices

  #64237 commented on Sep 14, 2021 • 1 new comment

• RuntimeError: Unsupported: ONNX export of operator adaptive_max_pool1d

  #61172 commented on Sep 14, 2021 • 1 new comment

• [feature request] Simplified torch.gather with not specifying full index (naming: index_left? batched index_select?)

  #64208 commented on Sep 14, 2021 • 1 new comment

• [ONNX] scatter_add may result in wrong ONNX

  #32960 commented on Sep 14, 2021 • 1 new comment

• CUDA `linalg.norm` matrix order +/-2 and nuclear norm for extreme values may be incorrect

  #52633 commented on Sep 14, 2021 • 1 new comment

• Onnx export of softplus with beta!=1

  #64811 commented on Sep 14, 2021 • 1 new comment

• Make sure `data_ptr` for non-zero-size input tensors stays the same after the VariableType dispatch

  #16589 commented on Sep 14, 2021 • 1 new comment

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

  #63445 commented on Sep 14, 2021 • 1 new comment

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

  #63174 commented on Sep 14, 2021 • 1 new comment

• Per channel weight observer for ConvTranspose

  #54816 commented on Sep 15, 2021 • 1 new comment

• torch.onnx.export cannot work with torch.unfold when dynamic_axes is set.

  #57086 commented on Sep 15, 2021 • 1 new comment

• Nightly MacOS M1 builds are broken

  #64649 commented on Sep 15, 2021 • 1 new comment

• [iOS][Metal] Cannot construct tensor directly on GPU: METAL device type not enabled

  #64736 commented on Sep 15, 2021 • 1 new comment

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

  #23509 commented on Sep 15, 2021 • 1 new comment

• ONNX export doubles RAM usage

  #61263 commented on Sep 15, 2021 • 1 new comment

• improve jit error message for legacy constructor

  #18434 commented on Sep 16, 2021 • 1 new comment

• i want to parsing waveglow model to onnx but i have a error ㅜㅜ

  #19399 commented on Sep 16, 2021 • 1 new comment

• TestTorchDeviceTypeCPU.test_float_to_int_conversion_finite_cpu_uint8 is broken on PowerPC

  #39671 commented on Sep 16, 2021 • 1 new comment

• test_float_to_int_conversion_finite_cpu_int16 is failing on MacOS

  #38752 commented on Sep 16, 2021 • 1 new comment

• add torch.unravel_index

  #35674 commented on Sep 16, 2021 • 1 new comment

• Keyword-only function not allowed with TorchScript

  #47565 commented on Sep 16, 2021 • 1 new comment

• BoxedKernelWrapper triggers MSVC bug in internal Windows build

  #48763 commented on Sep 16, 2021 • 1 new comment

• torch.package: re-export docs suggestions don't work + other sharp edges

  #64462 commented on Sep 16, 2021 • 1 new comment

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

  #61345 commented on Sep 16, 2021 • 1 new comment

• Reuse tensors from state_dict in load_state_dict

  #64601 commented on Sep 16, 2021 • 1 new comment

• Wrapper-based torch_dispatch objects report maybe incorrect device

  #62730 commented on Sep 17, 2021 • 1 new comment

• Migrate build/test jobs from CircleCI to GitHub Actions

  #57686 commented on Sep 17, 2021 • 1 new comment

• [JIT] Fusion of Dropout without constant is_training parameter is unsuccessful

  #24032 commented on Sep 17, 2021 • 1 new comment

• [Better Engineering] Refactor FSDP unit tests to use common distributed helpers

  #64664 commented on Sep 17, 2021 • 1 new comment

• Add a deterministic version of scatter_add_cuda_kernel

  #50469 commented on Sep 18, 2021 • 1 new comment

• Add padding as an option to torch.cat

  #11494 commented on Sep 17, 2021 • 1 new comment

• Reduce scenarios that plan executor issues SIGABRT

  #13084 commented on Sep 17, 2021 • 1 new comment

• Refactor C2 pybind global workspace into a library

  #16993 commented on Sep 17, 2021 • 1 new comment

• update CPU version 32bit-Libtorch guide

  #25149 commented on Sep 17, 2021 • 1 new comment

• [JIT] Support already scripted submodule as module member

  #39284 commented on Sep 17, 2021 • 1 new comment

• [JIT] Remove dead store in loop_unrolling.cpp

  #40612 commented on Sep 17, 2021 • 1 new comment

• [JIT] Fix clang-tidy issues in JIT

  #41150 commented on Sep 17, 2021 • 1 new comment

• [WIP][JIT] Allow DAG module hierarchies and module container attributes

  #43311 commented on Sep 17, 2021 • 1 new comment

• ONNX export: propagate node metadata across passes

  #45256 commented on Sep 15, 2021 • 1 new comment

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

  #47950 commented on Sep 17, 2021 • 1 new comment

• [JIT] Throw error when no_grad is used as a decorator

  #50129 commented on Sep 17, 2021 • 1 new comment

• [JIT][WIP] Enable Python-language hooks for ScriptModule

  #50776 commented on Sep 17, 2021 • 1 new comment

• [WIP][JIT] Make and/or return the value of their operands

  #51684 commented on Sep 17, 2021 • 1 new comment

• [WIP][JIT] castToDevice pass

  #52418 commented on Sep 17, 2021 • 1 new comment

• [WIP][package] Add package dependency visualizer

  #54242 commented on Sep 17, 2021 • 1 new comment

• [WIP] Add torch.package shell

  #55110 commented on Sep 17, 2021 • 1 new comment

• [WIP][package] Factor out dependency resolution

  #55625 commented on Sep 17, 2021 • 1 new comment

• [WIP] Migrate _fft_c2c_mkl to structured kernel

  #55730 commented on Sep 17, 2021 • 1 new comment

• [package] Add raise_packaging_errors flag to PackageExporter

  #56234 commented on Sep 17, 2021 • 1 new comment

• [fx] Make Tracer._autowrap_search a list of module names

  #57220 commented on Sep 17, 2021 • 1 new comment

• [package] (Un)package GraphModules using corresponding Tracer

  #57221 commented on Sep 17, 2021 • 1 new comment

• [torch][package] Make load_library a no-op inside a package

  #58901 commented on Sep 17, 2021 • 1 new comment

• [docs] Add explanation section to torch.package docs

  #59504 commented on Sep 17, 2021 • 1 new comment

• Python Operator Authoring w/ NNC Prototype

  #60684 commented on Sep 17, 2021 • 1 new comment

• Sparse CSR: add test parametrization for enabling sparse_csr

  #61408 commented on Sep 16, 2021 • 1 new comment

• Port `max` kernel to structured kernels.

  #61449 commented on Sep 13, 2021 • 1 new comment

• implementing _out for trapezoid and cumulative_trapezoid

  #61616 commented on Sep 17, 2021 • 1 new comment

• [special] special alias for softmax

  #62251 commented on Sep 17, 2021 • 1 new comment

• Fix rsub with alpha != 1

  #62397 commented on Sep 17, 2021 • 1 new comment

• Replace calls to _aminmax and add deprecation warning

  #62432 commented on Sep 17, 2021 • 1 new comment

• [WIP] Port `native_batch_norm` to structured

  #62452 commented on Sep 17, 2021 • 1 new comment

• [WIP] Port `native_batch_norm_backward` to structured

  #62513 commented on Sep 17, 2021 • 1 new comment

• Enable BFloat16 ELU, SELU and CELU in CPU path

  #62546 commented on Sep 17, 2021 • 1 new comment

• Refactor RecordFunction to use CustomClass

  #62710 commented on Sep 17, 2021 • 1 new comment

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

  #63129 commented on Sep 15, 2021 • 1 new comment

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

  #63230 commented on Sep 14, 2021 • 1 new comment

• Add ONNX export support for argsort Fixes #33412

  #63283 commented on Sep 14, 2021 • 1 new comment

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

  #63461 commented on Sep 17, 2021 • 1 new comment

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

  #63506 commented on Sep 17, 2021 • 1 new comment

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

  #63575 commented on Sep 13, 2021 • 1 new comment

• [quant] Making isPlaceholderObserver work post migration

  #63599 commented on Sep 14, 2021 • 1 new comment

• Implement n-dimensional hermitian FFTs

  #63890 commented on Sep 13, 2021 • 1 new comment

• OBSOLETE (put tools tests in reporoot/tools/test/): add mechanism to allow for tools imports in our tests

  #63946 commented on Sep 16, 2021 • 1 new comment

• Cleanup and modernize Optional.h

  #64020 commented on Sep 17, 2021 • 1 new comment

• Added OptionalArrayRef template class

  #64084 commented on Sep 17, 2021 • 1 new comment

• Simplify OptionalArrayRef

  #64092 commented on Sep 17, 2021 • 1 new comment

• [WIP] Add depth tracking

  #64309 commented on Sep 13, 2021 • 1 new comment

• [ONNX] Remove argument _retain_param_name from torch.onnx.export() function. (#61702)

  #64370 commented on Sep 13, 2021 • 1 new comment

• [ONNX] Remove strip_doc_string param from torch.onnx.export() function. (#61712)

  #64371 commented on Sep 13, 2021 • 1 new comment

• [ONNX] Add supplementary tests and description for custom_opsets param from torch.onnx.export() function. (#62085)

  #64372 commented on Sep 13, 2021 • 1 new comment

• [ONNX] minor doc improvements and cleanup (#62514)

  #64373 commented on Sep 13, 2021 • 1 new comment

• [ONNX] Deprecated the example_outputs param from torch.onnx.export() function. (#62815)

  #64380 commented on Sep 13, 2021 • 1 new comment

• [ONNX] Deprecate use_external_data_format param from torch.onnx.export() function. (#62257)

  #64382 commented on Sep 13, 2021 • 1 new comment

• [sharded_tensor] simplify init_from_local_shards API

  #64481 commented on Sep 16, 2021 • 1 new comment

• [WIP] working POC of quantization via dynamic tracing

  #64676 commented on Sep 17, 2021 • 1 new comment

• [Static Runtime] Add record functions

  #64698 commented on Sep 14, 2021 • 1 new comment

• [tensorexpr] conv2d handle optional bias

  #64750 commented on Sep 17, 2021 • 1 new comment

• [ONNX] Support torch.isfinite export

  #64759 commented on Sep 16, 2021 • 1 new comment

• Masked reductions: sum, prod, mean, amax, amin

  #64809 commented on Sep 16, 2021 • 1 new comment

• [PyTorch] add fastToString for infer_schema

  #64823 commented on Sep 17, 2021 • 1 new comment

• [quant] Member variable access in quantizable LSTM

  #64833 commented on Sep 17, 2021 • 1 new comment

• [feature request] New copyseq_ method

  #64254 commented on Sep 12, 2021 • 0 new comments

• Python Array API Compatibility Tracker

  #58743 commented on Sep 13, 2021 • 0 new comments

• torch.cuda.is_available() returns misleading value

  #17966 commented on Sep 13, 2021 • 0 new comments

• Empty channel support in convolution

  #56998 commented on Sep 13, 2021 • 0 new comments

• Empty batch support for SyncBatchNorm

  #36530 commented on Sep 13, 2021 • 0 new comments

• Softmax/logsoftmax use more memory for backward than needed

  #64000 commented on Sep 13, 2021 • 0 new comments

• CPU MaxPool2d is very slow

  #28733 commented on Sep 13, 2021 • 0 new comments

• Generalized CPU vector reductions

  #40570 commented on Sep 13, 2021 • 0 new comments

• max_pool2d always compute indices even when it's not required

  #41099 commented on Sep 13, 2021 • 0 new comments

• torch.topk documentation is incorrect about the meaning of returned indices

  #50331 commented on Sep 13, 2021 • 0 new comments

• Interpolation tracking issue

  #50341 commented on Sep 13, 2021 • 0 new comments

• torch.nn.functional.kl_div and torch.nn.functional.nll_loss outputs NaN

  #51910 commented on Sep 13, 2021 • 0 new comments

• MaxPool1d Inconsistent Shape Checking

  #51954 commented on Sep 13, 2021 • 0 new comments

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

  #57121 commented on Sep 13, 2021 • 0 new comments

• MaxPool2D Returns Wrong Shape With Ceil_Mode

  #57178 commented on Sep 13, 2021 • 0 new comments

• Clarity of error message in einsum regressed in performance improvements

  #58380 commented on Sep 13, 2021 • 0 new comments

• PyTorch public API cleanup

  #64185 commented on Sep 13, 2021 • 0 new comments

• Memory leak in multi-thread inference

  #64412 commented on Sep 13, 2021 • 0 new comments

• Improving ease of contribution for beginner student contributors

  #64503 commented on Sep 13, 2021 • 0 new comments

• Support more ops in LazyTensor to TorchScript lowering

  #62431 commented on Sep 13, 2021 • 0 new comments

• RFC: torch.vmap

  #42368 commented on Sep 13, 2021 • 0 new comments

• [CI] Add GPU+AVX512 capable runners to the CI config

  #64767 commented on Sep 13, 2021 • 0 new comments

• Error converting to onnx: forward function contains for loop

  #58289 commented on Sep 13, 2021 • 0 new comments

• Make interpolation output size compatible with opencv, scikit-image and scipy for floating scale factor

  #62396 commented on Sep 14, 2021 • 0 new comments

• [feature request] Core API for invertible/inplace and flow-like ops + memory-saving sequential container for RevNets to allow for much larger batch-sizes in academic setting

  #23756 commented on Sep 14, 2021 • 0 new comments

• Reduce or eliminate thrust usage in pytorch

  #57505 commented on Sep 14, 2021 • 0 new comments

• Complex Number support for distributed

  #45760 commented on Sep 14, 2021 • 0 new comments

• [jit] Optional type refinement on non-named expressions

  #22155 commented on Sep 14, 2021 • 0 new comments

• Tracing doesn't work with spectral normalization

  #49643 commented on Sep 14, 2021 • 0 new comments

• Cannot manually assign a tensor to .grad from TorchScript

  #40561 commented on Sep 14, 2021 • 0 new comments

• JIT, nn.utils.weight_norm and {save,load}_state_dict produce wrong results

  #30754 commented on Sep 14, 2021 • 0 new comments

• SyncBatchNorm for JIT and a list of not supported operations

  #40507 commented on Sep 14, 2021 • 0 new comments

• TracedModule 'to' attribute doesn't work for tensors created on forward.

  #17655 commented on Sep 14, 2021 • 0 new comments

• JIT does not batch linear layers in an ensemble

  #18157 commented on Sep 14, 2021 • 0 new comments

• Jit fail with TracingCheckError with tracing model with layers created after init.

  #18520 commented on Sep 14, 2021 • 0 new comments

• [JIT] traced model with optimization shows no performance improvement

  #19978 commented on Sep 14, 2021 • 0 new comments

• [jit] Better type refinement for class attributes which are class types

  #44077 commented on Sep 14, 2021 • 0 new comments

• [ONNX] Can't export LSTM

  #24235 commented on Sep 14, 2021 • 0 new comments

• Add ONNX support for the new linalg.norm, linalg.matrix_norm and linalg.vector_norm functions.

  #58081 commented on Sep 14, 2021 • 0 new comments

• Rollup: No-batch-dim support for torch.nn modules

  #60585 commented on Sep 14, 2021 • 0 new comments

• Consider adding meta:1, meta:2 devices

  #55192 commented on Sep 14, 2021 • 0 new comments

• Python 3.10 support

  #64424 commented on Sep 15, 2021 • 0 new comments

• OpInfo Tracker

  #54261 commented on Sep 15, 2021 • 0 new comments

• LazyTensor support for fuser integration investigation

  #62619 commented on Sep 15, 2021 • 0 new comments

• [Meta] CI infra flakiness tracker and status update

  #64390 commented on Sep 15, 2021 • 0 new comments

• Feature request: Use LinAlgError instead of raising RuntimeError for linalg operations

  #64785 commented on Sep 16, 2021 • 0 new comments

• The latest version of onnx-caffe2 does not support “pow” ？

  #18475 commented on Sep 16, 2021 • 0 new comments

• torch.any and torch.all map uint8 -> uint8 but should map uint8 -> bool

  #50342 commented on Sep 16, 2021 • 0 new comments

• Reductions tracking issue

  #61417 commented on Sep 16, 2021 • 0 new comments

• Port kernels to be structured [tracker]

  #55070 commented on Sep 16, 2021 • 0 new comments

• [feature request] Forward-mode automatic differentiation

  #10223 commented on Sep 16, 2021 • 0 new comments

• Export decoder to onnx failed: RuntimeError: 0 INTERNAL ASSERT FAILED at "..\\torch\\csrc\\jit\\ir\\ir.cpp":518, please report a bug to PyTorch. 114 not in scope

  #59621 commented on Sep 16, 2021 • 0 new comments

• Support custom tensor-like objects as module parameters

  #62787 commented on Sep 17, 2021 • 0 new comments

• Forward mode AD for linear algebra functions

  #64545 commented on Sep 17, 2021 • 0 new comments

• aminmax increases dimension from 0-dim to 1-dim when given scalar inputs and keepdim.

  #64008 commented on Sep 17, 2021 • 0 new comments

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

  #63645 commented on Sep 17, 2021 • 0 new comments

• Extend @ops decorator with kwarg to apply dynamic decorators

  #63494 commented on Sep 17, 2021 • 0 new comments

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

  #62354 commented on Sep 17, 2021 • 0 new comments

• Implement missing torch.nan* operators

  #61474 commented on Sep 17, 2021 • 0 new comments

• Optimize torch.einsum

  #60295 commented on Sep 17, 2021 • 0 new comments

• `test_reductions` ignoring some tests

  #53704 commented on Sep 17, 2021 • 0 new comments

• Link to `torch.einsum` in `torch.tensordot`

  #50802 commented on Sep 17, 2021 • 0 new comments

• [Meta] Change default branch name to `main` for repos in `pytorch` project

  #62540 commented on Sep 17, 2021 • 0 new comments

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

  #24422 commented on Sep 17, 2021 • 0 new comments

• GHA checkout or workspace permission randomly failed

  #61684 commented on Sep 17, 2021 • 0 new comments

• [rfc] Simplify the PyTorch CI

  #62370 commented on Sep 17, 2021 • 0 new comments

• ciflow tracking meta issues

  #64124 commented on Sep 17, 2021 • 0 new comments

• How to use CUDA Dynamic Parallelism in PyTorch CPP extension?

  #44279 commented on Sep 18, 2021 • 0 new comments

• [Beta] Refactoring and documentation enhancements

  #64685 commented on Sep 18, 2021 • 0 new comments

• [DO NOT DELETE OR MERGE] Nightly release

  #26921 commented on Sep 18, 2021 • 0 new comments

• [DO NOT DELETE OR MERGE] Postnightly release

  #27167 commented on Sep 18, 2021 • 0 new comments

• onnx: Use Reciprocal operator

  #38708 commented on Sep 14, 2021 • 0 new comments

• Vectorize fmax and fmin for float and double on CPU

  #55867 commented on Sep 16, 2021 • 0 new comments

• Give std/var correction overloads proper defaults

  #56398 commented on Sep 16, 2021 • 0 new comments

• Add detailed error message to the test test_maximum_minimum_float_nan_and_inf

  #57702 commented on Sep 16, 2021 • 0 new comments

• Deprecate bytecode version 3

  #57775 commented on Sep 17, 2021 • 0 new comments

• [v1.9.0] Getting signal for `release/1.9`

  #58519 commented on Sep 15, 2021 • 0 new comments

• [PyTorch] Make arithmetic ops etc. use_const_ref_for_mutable_tensors

  #58935 commented on Sep 15, 2021 • 0 new comments

• [PyTorch] use borrow_from_optional_tensor in CPU matmul

  #59062 commented on Sep 15, 2021 • 0 new comments

• Fix `searchsorted` with non-contiguous out

  #60508 commented on Sep 16, 2021 • 0 new comments

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

  #60755 commented on Sep 13, 2021 • 0 new comments

• [ROCm] Refactor how to specify AMD gpu targets

  #61706 commented on Sep 14, 2021 • 0 new comments

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

  #62612 commented on Sep 15, 2021 • 0 new comments

• [Static Runtime] Disable ListUnpack fusion if its output is reachable from graph output

  #62697 commented on Sep 17, 2021 • 0 new comments

• DO NOT MERGE, testing conda pytorch-mutex

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

• DO NOT MERGE until after 30/sep/2021: remove unnecessary torch.meshgrid overloads

  #62743 commented on Sep 17, 2021 • 0 new comments

• DO NOT MERGE until after 1.10 is cut: require that clients provide `indexing` to torch.meshgrid

  #62749 commented on Sep 17, 2021 • 0 new comments

• DO NOT MERGE until after 1.10 is cut: change `torch.meshgrid` indexing default to "xy"

  #62752 commented on Sep 17, 2021 • 0 new comments

• Add BFloat16 support for grid_sample on CPU

  #62875 commented on Sep 18, 2021 • 0 new comments

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

  #62995 commented on Sep 13, 2021 • 0 new comments

• move some codegen utilities into utils.py

  #63094 commented on Sep 15, 2021 • 0 new comments

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

  #63130 commented on Sep 15, 2021 • 0 new comments

• Add crow_/col_indices to view types

  #63176 commented on Sep 17, 2021 • 0 new comments

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

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

• Add low level torch.profiler.kineto_profiler base class

  #63302 commented on Sep 17, 2021 • 0 new comments

• less strict size check in scatter

  #63452 commented on Sep 13, 2021 • 0 new comments

• [nvFuser] [WIP] Normalization support

  #63542 commented on Sep 18, 2021 • 0 new comments

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

  #63651 commented on Sep 17, 2021 • 0 new comments

• [PyTorch] Outline destructor of CppFunction

  #63688 commented on Sep 14, 2021 • 0 new comments

• Create test for builtin tensorrt module in torch deploy

  #63819 commented on Sep 17, 2021 • 0 new comments

• [JIT] memorization memory planning

  #63873 commented on Sep 13, 2021 • 0 new comments

• [Do not merge] Flatbuffer prototype

  #63875 commented on Sep 15, 2021 • 0 new comments

• plumb c10::nullopt Device through python binding layer, match TensorOptions defaulting logic with python binding defaulting logic

  #63906 commented on Sep 13, 2021 • 0 new comments

• Add native_dropout

  #63937 commented on Sep 17, 2021 • 0 new comments

• Autodiff update for Add operator

  #63938 commented on Sep 17, 2021 • 0 new comments

• Add fp16/fp32 autocasting to JIT/TorchScript

  #63939 commented on Sep 17, 2021 • 0 new comments

• Add softplus support to autodiff

  #63942 commented on Sep 17, 2021 • 0 new comments

• [PyTorch] Fix many Tuple::elements() callsites

  #64065 commented on Sep 17, 2021 • 0 new comments

• [PyTorch] RFC: Add tuple inline storage

  #64066 commented on Sep 17, 2021 • 0 new comments

• Enable NCCL_BLOCKING_WAIT if TORCH_DIST_DEBUG=detail

  #64073 commented on Sep 18, 2021 • 0 new comments

• Move intraop_launch_future from Parallel.h

  #64166 commented on Sep 14, 2021 • 0 new comments

• Remove Tensor.h include from TensorIterator.h

  #64167 commented on Sep 14, 2021 • 0 new comments

• Remove cuda/Loops.cuh dependency on native_functions.yaml

  #64168 commented on Sep 14, 2021 • 0 new comments

• Remove native_functions.yaml dependency from binary ops

  #64169 commented on Sep 14, 2021 • 0 new comments

• Remove native_functions.yaml dependency from unary ops

  #64170 commented on Sep 14, 2021 • 0 new comments

• Remove native_functions.yaml dependency from PowKernel

  #64171 commented on Sep 14, 2021 • 0 new comments

• Remove native_functions.yaml dependency from PointwiseOps

  #64172 commented on Sep 14, 2021 • 0 new comments

• Remove native_functions.yaml dependency from some reduction operators

  #64173 commented on Sep 14, 2021 • 0 new comments

• [PyTorch] Model loading: cache operators by string address

  #64200 commented on Sep 14, 2021 • 0 new comments

• Moving the constant parameter check to a more common file

  #64251 commented on Sep 13, 2021 • 0 new comments

• [PyTorch Edge] Avoid string copying in TypeParser

  #64278 commented on Sep 17, 2021 • 0 new comments

• Support bool input tensors for argmax and other functions

  #64343 commented on Sep 15, 2021 • 0 new comments

• [JIT] linear scan memory planning strategy

  #64348 commented on Sep 13, 2021 • 0 new comments

• [JIT] greedy by size memory planning strategy

  #64349 commented on Sep 13, 2021 • 0 new comments

• [JIT] greedy by breadth memory planning strategy

  #64350 commented on Sep 13, 2021 • 0 new comments

• [Structured Kernels] Port `cross` to structured kernels

  #64355 commented on Sep 16, 2021 • 0 new comments

• Update distributed debug doc

  #64466 commented on Sep 18, 2021 • 0 new comments

• ONNX: Delete or document skipped ORT tests

  #64470 commented on Sep 18, 2021 • 0 new comments

• optimize dim reduce performance on norm, argmax and argmin

  #64479 commented on Sep 14, 2021 • 0 new comments

• [DDP] Refactor and remove sync_params

  #64514 commented on Sep 18, 2021 • 0 new comments

• add automatic type promotion tests for binary ufunc ops

  #64634 commented on Sep 14, 2021 • 0 new comments

• [TEST ONLY] Replacing codegen with boxed for testing

  #64654 commented on Sep 14, 2021 • 0 new comments

• [PyTorch] Use SmallVector for FunctionSchema Arguments

  #64691 commented on Sep 17, 2021 • 0 new comments

• improve CPU performance for log_softmax when dim != -1 on both float32 and bfloat16

  #64726 commented on Sep 14, 2021 • 0 new comments

• [jit] Shape propagation batch_norm, dropout, quantize, hardswidh

  #64740 commented on Sep 17, 2021 • 0 new comments

• [quant] Correct name for quantized LSTM

  #64834 commented on Sep 16, 2021 • 0 new comments

• [PyTorch] Fix missing moves in SchemaParser::parseArgument

  #64839 commented on Sep 17, 2021 • 0 new comments

• [iOS][OSS][BE] Add simulator tests for Metal

  #64852 commented on Sep 14, 2021 • 0 new comments

• [LTC][Not for Merge] Try out the new TorchBench on-demand PR CI

  #64863 commented on Sep 17, 2021 • 0 new comments

• [wip] Test Linux CPU shards 2 -> 6

  #64867 commented on Sep 18, 2021 • 0 new comments