pytorch / pytorch

17 Pull requests merged by 14 people

• Update freezing API - changes from 52337

  #52392 merged Feb 18, 2021

• Add arm64 binary build (#52443)

  #52469 merged Feb 18, 2021

• onnx export of per channel fake quantize functions (#42835)

  #52430 merged Feb 18, 2021

• [v1.8.0] Fix upsample bicubic2d batching handling on CPU. (#52389)

  #52445 merged Feb 18, 2021

• update symeig backward note about similar eigenvalues (#52311)

  #52446 merged Feb 18, 2021

• [v1.8.0] .circleci: Downgrade CUDA 11.2 -> 11.1 for binaries (#52151)

  #52406 merged Feb 18, 2021

• [1.8][cmake] Add explicit cublas->cudart dependency

  #52404 merged Feb 18, 2021

• [1.8] Remove torch.vmap

  #52397 merged Feb 18, 2021

• [1.8] Fix libnvrtc discoverability in package patched by `auditwheel`

  #52365 merged Feb 18, 2021

• Skip OneDNN Convolution in case of groups = 24 #50042

  #52313 merged Feb 18, 2021

• [1.8 patch] Use side-stream in CPU to GPU copies in DDP

  #52270 merged Feb 17, 2021

• [ONNX] Update LayerNorm symbolic to handle autocasting

  #52199 merged Feb 17, 2021

• [ONNX] Update fuseLogSoftmaxNllLoss function to handle autocasting

  #51729 merged Feb 17, 2021

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

  #51795 merged Feb 16, 2021

• Ports fix for division documentation that removes "true" rounding mode

  #52242 merged Feb 14, 2021

• [v1.8 patch] [Resubmission] Add a documentation page for DDP communication hooks

  #52215 merged Feb 13, 2021

• [1.8] Do not print warning if CUDA driver not found (#51806)

  #52050 merged Feb 12, 2021

116 Pull requests opened by 72 people

• Support forward reference annotations

  #52210 opened Feb 12, 2021

• torch.Package extend PyTorchStreamWriter to track written records

  #52218 opened Feb 12, 2021

• [PyTorch] Try outlining error message construction in TORCH_{CHECK,INTERNAL_ASSERT}

  #52219 opened Feb 12, 2021

• [PyTorch] Make c10::str(const char*) return const char*

  #52222 opened Feb 12, 2021

• [c10] Take advantage of c10::str optis for simple CAFFE_ENFORCE

  #52223 opened Feb 12, 2021

• [PyTorch] Sync TORCH_INTERNAL_ASSERT optis with TORCH_CHECK

  #52226 opened Feb 12, 2021

• structured kernels - error check when structured_delegate is not marked structured

  #52227 opened Feb 12, 2021

• remove unreachable portion enable RuntimeError use

  #52228 opened Feb 12, 2021

• Make per-channel quantization the default for dynamic quantization

  #52231 opened Feb 12, 2021

• OpInfos: skip BFloat16 dtypes for JIT

  #52235 opened Feb 12, 2021

• Fix optim.lr_scheduler.CosineAnnealingWarmRestarts with last_epoch > 0

  #52238 opened Feb 12, 2021

• [NOT FOR SUBMIT] Unintrusive control flow experiments

  #52245 opened Feb 13, 2021

• [NOT FOR SUBMIT] Control flow ops

  #52247 opened Feb 13, 2021

• [TensorExpr] LoopNest: make API for creating a loopnest more explicit.

  #52259 opened Feb 14, 2021

• Adding PyTorch + DNNL + AMD BLIS path

  #52267 opened Feb 15, 2021

• Adding learning rate schedulers to C++ API

  #52268 opened Feb 15, 2021

• Remove hacky wrapper from a lot of unary operators.

  #52276 opened Feb 15, 2021

• Port sin to structured.

  #52277 opened Feb 15, 2021

• Included docs for `feature_dropout` method and classes

  #52282 opened Feb 15, 2021

• Add psutil to requirements.txt

  #52285 opened Feb 15, 2021

• fix zero_point rounding for fake_quant_per_channel_affine

  #52290 opened Feb 16, 2021

• temp

  #52292 opened Feb 16, 2021

• [special] add `torch.special` namespace

  #52296 opened Feb 16, 2021

• Add SoftplusTransform

  #52300 opened Feb 16, 2021

• [PyTorch Mobile] Create compile time string for passing in to the exception message instead of 4 arguments that will be concatenated at runtime

  #52303 opened Feb 16, 2021

• [quant] Reference option for conv module

  #52316 opened Feb 16, 2021

• add debug_flush_compilation_cache to `Method`

  #52317 opened Feb 16, 2021

• [package] rename ex/importer.py to package_ex/importer.py

  #52320 opened Feb 16, 2021

• fix the usage of is_alias_of in ddp reducer

  #52328 opened Feb 16, 2021

• Modernize for-loops

  #52330 opened Feb 16, 2021

• [Pytorch Mobile] Improve export_opnames Documentation

  #52333 opened Feb 17, 2021

• [JIT] Display an error message when with item is not an object

  #52335 opened Feb 17, 2021

• [JIT] Allow __exit__ to have a return value

  #52336 opened Feb 17, 2021

• Explicitly export submodules and variables from torch module

  #52339 opened Feb 17, 2021

• [StaticRuntime] Add function to check for memory leak

  #52342 opened Feb 17, 2021

• [static runtime] aten::to copy out variant

  #52343 opened Feb 17, 2021

• blas copy and axpy to aten

  #52345 opened Feb 17, 2021

• Optimize `setDebugName` time complexity

  #52346 opened Feb 17, 2021

• [ONNX] Fix for sequence of mutations in blocks (#51577)

  #52347 opened Feb 17, 2021

• [ONNX] Redesign onnx pass to enable shape type dependent pattern conversion - cont (#51795)

  #52348 opened Feb 17, 2021

• [ONNX] Update fuseLogSoftmaxNllLoss function to handle autocasting (#51729)

  #52349 opened Feb 17, 2021

• [ONNX] Update LayerNorm symbolic to handle autocasting (#52199)

  #52350 opened Feb 17, 2021

• Fix performance of CUDA trilinear interpolate backward

  #52351 opened Feb 17, 2021

• [WIP][optim] refactor asgd

  #52352 opened Feb 17, 2021

• torch.Package reverse dependencies printer

  #52355 opened Feb 17, 2021

• Support submodule manipulation in GraphModule

  #52358 opened Feb 17, 2021

• Reenables skipped tests for all CUDA versions except 11.2

  #52359 opened Feb 17, 2021

• Support any

  #52360 opened Feb 17, 2021

• Add guard to run on current thread

  #52361 opened Feb 17, 2021

• Ignore user annotated ignored attributes

  #52367 opened Feb 17, 2021

• ns for fx: allow graph matching of parents of cat

  #52368 opened Feb 17, 2021

• .github: Move anaconda prune to GHA

  #52372 opened Feb 17, 2021

• repro redefining forward error

  #52373 opened Feb 17, 2021

• Add support for pow

  #52374 opened Feb 17, 2021

• [glow aot] Support --onnxifi_min_ops in AOT flow

  #52380 opened Feb 17, 2021

• [Work in progress] Add AVX2_FAST dispatch

  #52382 opened Feb 17, 2021

• [alias analysis] Add schema based pure functions to functionalNonEscapingListUse

  #52383 opened Feb 17, 2021

• [Pytorch Mobile] Error check that at least one input is provided

  #52386 opened Feb 17, 2021

• Modify tests to use assertWarnsOnceRegex instead of maybeWarnsRegex

  #52387 opened Feb 17, 2021

• [DDP] Separate error messages for unused params in forward and not all outputs used in loss computation

  #52391 opened Feb 17, 2021

• Test for distributed RL with RPC

  #52393 opened Feb 17, 2021

• ns for fx: support linear_relu for weight matching

  #52395 opened Feb 17, 2021

• use obj str when jit trace can't track the python class name

  #52398 opened Feb 17, 2021

• half precision fake quantization

  #52400 opened Feb 17, 2021

• Fake quantization support for half and double precision variables

  #52401 opened Feb 17, 2021

• ns for fx: update graph matching to not match nodes with equal types

  #52402 opened Feb 17, 2021

• Remove DepTracker from LoopNest

  #52405 opened Feb 18, 2021

• DO NOT MERGE, testing #52406

  #52407 opened Feb 18, 2021

• Port NumPy typing testing style to PyTorch

  #52408 opened Feb 18, 2021

• [ROCm] Enable test_jit_c10.py tests for ROCm

  #52410 opened Feb 18, 2021

• [quant][graphmode][fx] Enable test for non quantized input for add/mul

  #52412 opened Feb 18, 2021

• [quant][graphmode][fx] Guard the supported quantization type for add/mul

  #52413 opened Feb 18, 2021

• [qunat][graphmode][fx] Enable test for non quantized input for cat

  #52414 opened Feb 18, 2021

• [WIP][JIT] castToDevice pass

  #52418 opened Feb 18, 2021

• [PyTorch Mobile] Use real if constexpr behind macro in hot template (copy D26154964 in a different setting)

  #52420 opened Feb 18, 2021

• Numeric Suite: Enable QAT support

  #52421 opened Feb 18, 2021

• Support non-tensor inputs and outputs for checkpointed functions.

  #52422 opened Feb 18, 2021

• remove deprecated function

  #52426 opened Feb 18, 2021

• Implemented embedding_bag for SR

  #52429 opened Feb 18, 2021

• [testing] add macos_arm64 binary build

  #52441 opened Feb 18, 2021

• [WIP] Enable complex autograd for torch.unfold

  #52442 opened Feb 18, 2021

• [PyTorch Mobile] Do not create a static variable in Dispatcher::singleton()

  #52447 opened Feb 18, 2021

• Support broadcasting via slow path

  #52448 opened Feb 18, 2021

• Add type promotion to a fast path for foreach APIs

  #52449 opened Feb 18, 2021

• wip

  #52450 opened Feb 18, 2021

• Codegen python bindings to access attributes of grad_fn

  #52451 opened Feb 18, 2021

• Modernize for-loops in torch misc

  #52452 opened Feb 18, 2021

• [static runtime] register pow out variant

  #52454 opened Feb 18, 2021

• [DataLoader] Fix util ImportError

  #52459 opened Feb 18, 2021

• [pyper] fuse clip_ranges+gather_ranges

  #52461 opened Feb 18, 2021

• [FX acc]Store shape and dtype in serialized output node args

  #52462 opened Feb 18, 2021

• Add torch.linalg to generated annotated_args for test_overrides

  #52464 opened Feb 18, 2021

• [Take-2] [PyTorch Mobile] 15KiB size reduction by reducing MaxTypeIndex from 256 to 32

  #52466 opened Feb 18, 2021

• Allow Transformer accept output result that is not Proxy

  #52473 opened Feb 19, 2021

• Skip handle_r_to_c for dot & vdot

  #52474 opened Feb 19, 2021

• enable miopen for rnn f16

  #52475 opened Feb 19, 2021

• [PyTorch] update CMake to build libtorch lite

  #52476 opened Feb 19, 2021

• [BE] force cmake to always generate version.py

  #52477 opened Feb 19, 2021

• Add distributed debug mode func to python

  #52481 opened Feb 19, 2021

• [distributed launcher] auto find free port

  #52482 opened Feb 19, 2021

• [reland][complex] `masked_fill`: Complex Autograd support and update masked_scatter skips.

  #52483 opened Feb 19, 2021

• [Tessting] fix 52468

  #52484 opened Feb 19, 2021

• Enabled amax & amin for float16 & bfloat16 on CPU & CUDA

  #52487 opened Feb 19, 2021

• [complex] add autograd support for torch.polar

  #52488 opened Feb 19, 2021

• [WIP][JIT] Add default arguments for JIT CUDA Stream and Event

  #52489 opened Feb 19, 2021

• [DO NOT MERGE] Test azure pipeline

  #52490 opened Feb 19, 2021

• Added linalg.eig, linalg.eigvals

  #52491 opened Feb 19, 2021

• [NNC] Added reductions to NNC python bindings.

  #52492 opened Feb 19, 2021

• Adaptive ratio for L2 regularizer

  #52493 opened Feb 19, 2021

• [WIP][docs] Add updated TorchScript language reference

  #52494 opened Feb 19, 2021

• [PyTorch][CI] Fix typo to enable BUILD_TEST_LIBTORCH in macos

  #52496 opened Feb 19, 2021

• Restore fast path in OnnxifiOp::adjustOutputBatchSize

  #52498 opened Feb 19, 2021

• Fake Quantize for Double and Half precision variables

  #52499 opened Feb 19, 2021

• Finished GPU and backward for fake quantization support for half and double precision floats

  #52500 opened Feb 19, 2021

• Automatically set BUILD_SPLIT_CUDA for cpp exts

  #52503 opened Feb 19, 2021

• fetch git submodules during build

  #52505 opened Feb 19, 2021

64 Issues closed by 26 people

• Exporting model to onnx increases the model size

  #51343 closed Feb 19, 2021

• Implement torch.max for float16 and bfloat16 on CPU and CUDA

  #50790 closed Feb 19, 2021

• [jit] Python built-in function support

  #18627 closed Feb 19, 2021

• Scripting nonzero fails when as_tuple=True

  #52470 closed Feb 19, 2021

• all reduce hangs in certain multi-node configurations using the NCCL backend

  #52471 closed Feb 19, 2021

• Replace blacklist/whitelist throughout the PyTorch codebase

  #41443 closed Feb 19, 2021

• Replace blacklist/whitelist in torch/onnx/symbolic_opset7.py

  #41737 closed Feb 19, 2021

• Replace blacklist/whitelist in caffe2/opt/custom/fakefp16_transform.cc

  #41708 closed Feb 19, 2021

• Replace blacklist/whitelist in cmake/Modules_CUDA_fix/upstream/FindCUDA.cmake

  #41713 closed Feb 19, 2021

• Replace blacklist/whitelist in caffe2/opt/custom/glow_net_transform.cc

  #41732 closed Feb 19, 2021

• Replace blacklist/whitelist in torch/onnx/symbolic_helper.py

  #41736 closed Feb 19, 2021

• Replace blacklist/whitelist in torch/onnx/symbolic_opset8.py

  #41738 closed Feb 19, 2021

• Replace blacklist/whitelist in CMakeLists.txt

  #41739 closed Feb 19, 2021

• Replace blacklist/whitelist in aten/src/ATen/gen.py

  #41742 closed Feb 19, 2021

• Replace blacklist/whitelist in cmake/Codegen.cmake

  #41747 closed Feb 19, 2021

• Jit traces interpolate scale_factor as constant

  #36596 closed Feb 18, 2021

• Can we define a tensor with actually 1bit?

  #52438 closed Feb 18, 2021

• nn.Conv1d non-deterministic behavior

  #52460 closed Feb 18, 2021

• training hanged with torch.distributed.launch

  #52433 closed Feb 18, 2021

• System stuck while training with 2 gpus

  #52294 closed Feb 18, 2021

• [nnc][perf] 50x performance regression from `_jit_override_can_fuse_on_cpu(True)`

  #52264 closed Feb 18, 2021

• torch.lerp does not broadcast all args

  #52254 closed Feb 18, 2021

• Access Forbidden for nightly aarch64 builds

  #52428 closed Feb 18, 2021

• The upsample_bicubic2d operator implemented by the CPU, when the input and output shapes are the same, There is a problem of data loss when copying data

  #49159 closed Feb 18, 2021

• Allow -1 in unflatten shape argument

  #51719 closed Feb 18, 2021

• Remove vmap from the 1.8 release

  #52396 closed Feb 18, 2021

• Data copy from CPU to GPU use default stream in nightly version.

  #49819 closed Feb 18, 2021

• DISABLED test_send_recv (__main__.TestDistBackendWithFork)

  #52283 closed Feb 18, 2021

• [jit] Make optional[T] sortable

  #45985 closed Feb 18, 2021

• JIT changes method kwarg argument names

  #43619 closed Feb 17, 2021

• Distributed Profiling does not work with DDP's all_reduce

  #52020 closed Feb 17, 2021

• Complex CPU torch.acos is wrong on some values

  #42952 closed Feb 17, 2021

• tutorials :: beginner :: audio_classifier_tutorial

  #52275 closed Feb 17, 2021

• Pytorch autograd sometimes fails with CUDNN_STATUS_INTERNAL_ERROR

  #52263 closed Feb 17, 2021

• [convenience] Make torch.nn.Module.eval() / torch.nn.Module.train() return self

  #52272 closed Feb 17, 2021

• Add custom types to default_collate

  #52353 closed Feb 17, 2021

• We need rtx A6000 sm_86 and cuda 11.2 support

  #52288 closed Feb 17, 2021

• Quantizable MHA error too high when bias=False

  #51662 closed Feb 17, 2021

• Documentation mention parameters `verbose` for torch.optim.lr_scheduler but it does not exist

  #44080 closed Feb 17, 2021

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

  #48174 closed Feb 16, 2021

• GNU and LLVM OpenMP conflict when using MKL and KINETO under conda build

  #51026 closed Feb 16, 2021

• CUDA error: too many resources requested for launch in Windows CI for CUDA 11.2

  #52002 closed Feb 16, 2021

• DataLoader for Video Loading that utilizes the GPU

  #51417 closed Feb 16, 2021

• Is SVD really GPU parallel?

  #52293 closed Feb 16, 2021

• [build system refactor] delete build_pytorch_libs.{sh,bat}, move functionality into cmake layer

  #10562 closed Feb 16, 2021

• problem: corrupted size vs. prev_size when using python3.6

  #10567 closed Feb 16, 2021

• grad_fn printout bug for torch::jit::(anonymous namespace)::ExecutionPlanAutogradFunction

  #10590 closed Feb 16, 2021

• Keep the requires_grad true after comparison of a matrix with a value

  #52250 closed Feb 16, 2021

• Compiler error in MultinomialKernel.cpp when building with gcc + libc++

  #52163 closed Feb 15, 2021

• CUDA Pytorch running on 1050ti and hangs on 2080ti running identical configurations

  #52019 closed Feb 15, 2021

• test_variant_consistency_jit tests fail on CPU for min & max when dtype is bfloat16 & dim argument is passed

  #51788 closed Feb 14, 2021

• linking error when using nightly pytorch prebuilt package

  #52075 closed Feb 14, 2021

• [proposal] Multi-dim torch.max + return_indices

  #51656 closed Feb 14, 2021

• CUDNN_STATUS_INTERNAL_ERROR and Illegal Memory Access on CUDA11.2 and CUDA10.2. Likely a faulty device

  #51382 closed Feb 13, 2021

• Keep the requires_grad true after matrix comparison

  #52249 closed Feb 13, 2021

• RuntimeError: Function not implemented at /pytorch/torch/lib/THD/process_group/General.cpp:17

  #10363 closed Feb 13, 2021

• Warn/error if importing torch from source directory

  #10391 closed Feb 13, 2021

• FP16 Grouped convolution is too slow on Titan V

  #10290 closed Feb 13, 2021

• torch.sum crash on Windows

  #10483 closed Feb 13, 2021

• [feature request] Add soversion for ATen

  #10431 closed Feb 13, 2021

• TensorOptions with sparse type does not work

  #10474 closed Feb 13, 2021

• [doc] specify `not allocate new memory` or `share underlying storage` at ops like `select()`

  #10493 closed Feb 13, 2021

• Optimize autograd engine init_to_execute() performance

  #34067 closed Feb 13, 2021

• Perf regression for YoloV3 CUDA eager train

  #51383 closed Feb 12, 2021

73 Issues opened by 54 people

• UnicodeDecodeError when parse kineto_event.name() for cuda112

  #52504 opened Feb 19, 2021

• F::interpolate get wrong result using libtorch

  #52502 opened Feb 19, 2021

• AMP + CUDA + tensor.sum op leads to dtype mismatch

  #52501 opened Feb 19, 2021

• Negative impact of Python versions on using Pytorch with Cuda11

  #52497 opened Feb 19, 2021

• link failed when using custom build pytorch on Android

  #52495 opened Feb 19, 2021

• OpInfo to support `sample_inputs_func` with just an Iterable output.

  #52486 opened Feb 19, 2021

• Dispatcher documentation needs update

  #52485 opened Feb 19, 2021

• Torchscript tutorial example is broken

  #52480 opened Feb 19, 2021

• torch.jit.is_scripting() example code is incorrect?

  #52479 opened Feb 19, 2021

• Could you please increase the wheel compiled Linux (aarch64) GPU, thank you very much!

  #52478 opened Feb 19, 2021

• Windows libtorch nightly builds for CUDA 11.1 failing binary smoke tests with unresolved symbols

  #52468 opened Feb 18, 2021

• Windows CXX11-ABI libtorch debug builds for CUDA 11.1 fail with unresolved external symbol

  #52467 opened Feb 18, 2021

• [nnc][perf] Performance decrease with CPU fusion on `freeze(script(pytorch_mobilenet_v3))`

  #52465 opened Feb 18, 2021

• optimize_for_mobile segfaults during forward of mobilenet_v3

  #52463 opened Feb 18, 2021

• Add autograd tests to verify correctness for R -> C cases

  #52457 opened Feb 18, 2021

• [JIT] Context Manager to Ignore Source Ranges in IR Dumps

  #52456 opened Feb 18, 2021

• Skip handle_r_to_c for torch.dot and torch.vdot in derivatives.yaml

  #52455 opened Feb 18, 2021

• torch::jit::load crashes with "memmove_avx_unaligned_erms" error

  #52453 opened Feb 18, 2021

• [proposal] Way to export/display the convolution algorithm found with cudnn.benchmark = True

  #52439 opened Feb 18, 2021

• [windows] test_variant_consistency_eager_masked_fill_cuda_complex128: Misaligned Address CUDA

  #52437 opened Feb 18, 2021

• Incorrect .ONNX representation for models that contains integer constants.

  #52436 opened Feb 18, 2021

• [JIT] pytorch greater operation will introduce a cast to double node in onnx

  #52435 opened Feb 18, 2021

• [NNC] NNC usability list

  #52434 opened Feb 18, 2021

• Problem exporting to onnx with registered buffer

  #52432 opened Feb 18, 2021

• [bug] fractional_max_pool2d backward: CUDA doesn't support input without batch dimenion, CPU does

  #52427 opened Feb 18, 2021

• Result of torch.Tensor.__mod__ is different from torch.fmod() function.

  #52425 opened Feb 18, 2021

• [JIT] Regression? in accessing NamedTuple named field

  #52419 opened Feb 18, 2021

• TypeError: _export_onnx(): incompatible function arguments. The following argument types are supported:

  #52416 opened Feb 18, 2021

• torch.autograd.Function doesn't support non-Tensor outputs

  #52415 opened Feb 18, 2021

• autograd.functional.vjp with function that mutates inputs in-place can cause confusion

  #52390 opened Feb 17, 2021

• torch.empty_like example seems wrong

  #52375 opened Feb 17, 2021

• RuntimeError: Can't redefine method: forward on class: __torch__.torch.nn.quantized.modules.conv.Conv1d (of Python compilation unit at: 0x5571a223a770)

  #52363 opened Feb 17, 2021

• Normal_like operator

  #52334 opened Feb 17, 2021

• Reciprocals of complex tensors with infinities are different from NumPy.

  #52332 opened Feb 16, 2021

• Build failure with "TORCH_CUDA_API" is undefined and more

  #52331 opened Feb 16, 2021

• Errors in scripting when using `_pair_from_first`

  #52329 opened Feb 16, 2021

• TorchConfig.cmake entries of if(ON) are not accepted by CMake as "true", to be replaced by if(1)

  #52318 opened Feb 16, 2021

• TorchScript can't wrap Python constants in current module

  #52312 opened Feb 16, 2021

• Implementation of many complex functions is fast but inaccurate in libc++

  #52310 opened Feb 16, 2021

• [JIT] __getitem__ on scripted ModuleDict returns unscripted module

  #52309 opened Feb 16, 2021

• Model produces same outputs for different inputs when using GPU

  #52307 opened Feb 16, 2021

• [TorchScript usability] Should report error when instance attributes are annotated outside `__init__`

  #52306 opened Feb 16, 2021

• [TorchScript Usability] Problems with `Any` type

  #52301 opened Feb 16, 2021

• `torch.acos` on Windows returns different values between CUDA and CPU

  #52299 opened Feb 16, 2021

• [FR} Implement Skellam distribution

  #52298 opened Feb 16, 2021

• Printing in backward implementations can trigger a Vmap-related error.

  #52295 opened Feb 16, 2021

• "LayerNormKernelImpl" not implemented for 'Half' - CPU

  #52291 opened Feb 16, 2021

• DataParallel module fails to handle data with size not divisible by number of GPUs

  #52289 opened Feb 16, 2021

• JIT makes model run x14 times slower in pytorch nightly

  #52286 opened Feb 15, 2021

• DISABLED test_send_recv_with_tag (__main__.TestDistBackendWithFork)

  #52284 opened Feb 15, 2021

• [docs] Official apex -> torch.cuda.amp migration guide

  #52279 opened Feb 15, 2021

• Distributed tests fail on A100 GPUs

  #52278 opened Feb 15, 2021

• Port unary elementwise ops to structured kernels

  #52274 opened Feb 15, 2021

• torchscript should work with type: ignore comments from mypy even if they are not at the end of the line

  #52271 opened Feb 15, 2021

• GPU (+CPU/RAM?) self-check script or instructions in core

  #52266 opened Feb 15, 2021

• 'TypeError: expected CPU (got CUDA)' when placing a CUDA tensor on a class that is inheriting torch.Tensor

  #52265 opened Feb 14, 2021

• tolist called on torch scalar does not return a list

  #52262 opened Feb 14, 2021

• Impossible to export ONNX as RAW

  #52261 opened Feb 14, 2021

• Unlucky batch->device split by DataParallel causes an exception

  #52260 opened Feb 14, 2021

• MultiHeadAttention bias=False regression

  #52257 opened Feb 14, 2021

• [docs] Bad docs SEO

  #52256 opened Feb 13, 2021

• 🐛 memory leaks at backward algorithm for sparse_sum and sparse_matmul

  #52253 opened Feb 13, 2021

• ONNX exporter sets invalid output type for casted tensor

  #52252 opened Feb 13, 2021

• torch.where leads to unexpected gradients

  #52248 opened Feb 13, 2021

• ProcessGroupNCCL::all_reduce is not profiled correctly

  #52246 opened Feb 13, 2021

• CTCLoss gradient is incorrect

  #52241 opened Feb 12, 2021

• CosineAnnealingWarmRestarts LR scheduler fails when lash_epoch != -1

  #52236 opened Feb 12, 2021

• torch.cat different Windows vs Linux behavior

  #52233 opened Feb 12, 2021

• Removing mutation on block input should print any meaning message.

  #52230 opened Feb 12, 2021

• [JIT] get_overload_annotations ignores __jit_ignored_attributes__

  #52217 opened Feb 12, 2021

• `hasattr` check not supported on objects other than `self`

  #52214 opened Feb 12, 2021

• logcumsumexp returns the wrong result in GPU

  #52213 opened Feb 12, 2021

• Conv3D error : CUDNN_STATUS_INTERNAL_ERROR

  #52211 opened Feb 12, 2021

258 Unresolved conversations

Sometimes conversations happen on old items that aren’t yet closed. Here is a list of all the Issues and Pull Requests with unresolved conversations.

• Add property binding in torchbind

  #50670 commented on Feb 19, 2021 • 48 new comments

• [PyTorch] update CMake to build libtorch lite

  #51419 commented on Feb 19, 2021 • 30 new comments

• Sparse compressed row layout

  #50937 commented on Feb 19, 2021 • 20 new comments

• [ONNX] Symbolic shape inference

  #51481 commented on Feb 19, 2021 • 20 new comments

• Migrate `mode` from TH to ATen

  #52043 commented on Feb 19, 2021 • 20 new comments

• Change TCPStore world_size and is_master to be optional

  #51809 commented on Feb 19, 2021 • 17 new comments

• Optimized bilinear interpolation using TensorIterator

  #51653 commented on Feb 18, 2021 • 16 new comments

• [v.1.8.0] Release Tracker

  #51886 commented on Feb 18, 2021 • 13 new comments

• [package] Introduce BaseImporter to manage module namespace collisions.

  #51975 commented on Feb 19, 2021 • 13 new comments

• Ge v1

  #52136 commented on Feb 18, 2021 • 13 new comments

• Add torch.linalg.vector_norm function

  #51099 commented on Feb 18, 2021 • 12 new comments

• torch.Package zipfile debugging printer

  #52176 commented on Feb 19, 2021 • 12 new comments

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

  #7313 commented on Feb 19, 2021 • 10 new comments

• [wip] [Pipe] supporting None and non-Tensors in forward's input/output

  #50693 commented on Feb 18, 2021 • 10 new comments

• migrate `sort` from the TH to Aten (CUDA)

  #50887 commented on Feb 16, 2021 • 10 new comments

• fractional_max_pool{2/3}d : Fix segfaults for incorrect kernel_size and output_size

  #51626 commented on Feb 18, 2021 • 10 new comments

• matmul performance benchmarks

  #51647 commented on Feb 19, 2021 • 10 new comments

• Example LSTMCell

  #51983 commented on Feb 19, 2021 • 10 new comments

• OpInfo does not work well with operators that take Tensor[] as input.

  #51996 commented on Feb 18, 2021 • 9 new comments

• Implements cpu_kernel_multiple_outputs and torch.frexp

  #51097 commented on Feb 17, 2021 • 9 new comments

• A ~17x speed optimization of creating tensor from multiple numpy arrays using torch.tensor and torch.Tensor.

  #51731 commented on Feb 18, 2021 • 8 new comments

• [fx] Separate globals assignment from code generation.

  #51974 commented on Feb 19, 2021 • 8 new comments

• [ONNX] Support inplace operations on inplace indexing

  #52063 commented on Feb 19, 2021 • 8 new comments

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

  #47027 commented on Feb 17, 2021 • 7 new comments

• Added CUDA support for torch.orgqr

  #51348 commented on Feb 17, 2021 • 7 new comments

• Change cmake to allow building with MLC kick-off build

  #51326 commented on Feb 19, 2021 • 6 new comments

• Update vec_mergee operand specifiers (_vecb)

  #52091 commented on Feb 17, 2021 • 6 new comments

• [quant][graphmode][api] debug --> output_a_reference_model

  #52179 commented on Feb 18, 2021 • 6 new comments

• Skip some nodes during discovery using sequence number

  #52180 commented on Feb 17, 2021 • 6 new comments

• Analytical metric package

  #51777 commented on Feb 15, 2021 • 5 new comments

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

  #47993 commented on Feb 17, 2021 • 5 new comments

• MacOs build from source error

  #44449 commented on Feb 18, 2021 • 5 new comments

• Add conj OpInfo and fix out inconsistency

  #52059 commented on Feb 18, 2021 • 5 new comments

• Add support for builtin sum

  #52188 commented on Feb 17, 2021 • 5 new comments

• third_party subdirectories missing from source releases.

  #13586 commented on Feb 18, 2021 • 4 new comments

• Update foreach binary ops with a scalar list

  #49249 commented on Feb 19, 2021 • 4 new comments

• Modernize for-loops

  #50797 commented on Feb 17, 2021 • 4 new comments

• Add support for sparse complex tensors for CPU/CUDA

  #50984 commented on Feb 18, 2021 • 4 new comments

• Update foreach APIs to use scalar lists

  #51893 commented on Feb 19, 2021 • 4 new comments

• [pytorch] reintroduce static dispatch

  #51957 commented on Feb 19, 2021 • 4 new comments

• Kernel launch checks for aten/src/ATen

  #52185 commented on Feb 18, 2021 • 4 new comments

• remove index_fill side-effect for scalar tensors

  #52209 commented on Feb 19, 2021 • 4 new comments

• Bazel Support

  #35316 commented on Feb 15, 2021 • 3 new comments

• Status of pip wheels with _GLIBCXX_USE_CXX11_ABI=1

  #51039 commented on Feb 15, 2021 • 3 new comments

• [Quantization] Output tensor type is lost after serializing and loading back a quantized model

  #39690 commented on Feb 15, 2021 • 3 new comments

• tensor operations return incorrect data with libtorch v1.7.1-rc3 on ARM64 on QNX

  #52159 commented on Feb 16, 2021 • 3 new comments

• [TorchScript usability] More intuitive error message for type annotated `self`

  #52053 commented on Feb 16, 2021 • 3 new comments

• jit internal python bindings all get added to the `torch.` namespace.

  #51691 commented on Feb 17, 2021 • 3 new comments

• [feature request] Singular values and spectral norm for convolutional layers

  #13505 commented on Feb 17, 2021 • 3 new comments

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

  #51430 commented on Feb 19, 2021 • 3 new comments

• Stable sort for the CPU take 2.

  #51790 commented on Feb 18, 2021 • 3 new comments

• Some simple optimizations

  #51831 commented on Feb 18, 2021 • 3 new comments

• port upsample_nearest3d and upsample_trilinear3d to structured

  #52065 commented on Feb 13, 2021 • 3 new comments

• [JIT] Create torch._C._jit submodule for JIT bindings

  #52158 commented on Feb 18, 2021 • 3 new comments

• [JIT] [Remove Mutation] Add handling of normal_

  #52175 commented on Feb 17, 2021 • 3 new comments

• Port index_copy from TH to ATen

  #52203 commented on Feb 17, 2021 • 3 new comments

• [doc] functionalities not documented

  #9886 commented on Feb 12, 2021 • 2 new comments

• rorch.fliplr and torch.flip have bugs with 3-D tensor (CxHxW)

  #52201 commented on Feb 12, 2021 • 2 new comments

• Unreachable Code in /tools/codegen/api/python.py

  #52189 commented on Feb 12, 2021 • 2 new comments

• onnx trace and JIT broken where export works OK in pytorch1.6

  #50687 commented on Feb 15, 2021 • 2 new comments

• Reduce code duplication in interpolate and make it more generic

  #10482 commented on Feb 16, 2021 • 2 new comments

• Function request: np.corrcoef

  #1254 commented on Feb 16, 2021 • 2 new comments

• Sending CUDA tensor to process, and then back, does not work

  #10375 commented on Feb 16, 2021 • 2 new comments

• [TensorExpr] Loop transformations: IR Simplifier fails to fold index expression after loop flattening

  #51173 commented on Feb 16, 2021 • 2 new comments

• torch.special tracking issue

  #50345 commented on Feb 17, 2021 • 2 new comments

• RFC: Private CUDA memory pools

  #51075 commented on Feb 17, 2021 • 2 new comments

• Cannot manually assign a tensor to .grad from TorchScript

  #40561 commented on Feb 17, 2021 • 2 new comments

• Changing order of field assignments breaks TorchScript classes

  #46946 commented on Feb 17, 2021 • 2 new comments

• [ONNX] training parameter of batch_norm is ignored

  #49229 commented on Feb 18, 2021 • 2 new comments

• [Feature Request] Support tensor creation from objects that implement the __array__ interface

  #51156 commented on Feb 18, 2021 • 2 new comments

• [Feature Request] Implement "same" padding for convolution operations?

  #3867 commented on Feb 19, 2021 • 2 new comments

• optim.Optimizer should copy "params" before modifying them

  #51794 commented on Feb 19, 2021 • 2 new comments

• Add padding='same' mode to conv{1,2,3}d

  #45667 commented on Feb 18, 2021 • 2 new comments

• Expose nbins and ratio

  #50398 commented on Feb 17, 2021 • 2 new comments

• LayerNorm Support in autodiff:

  #50467 commented on Feb 19, 2021 • 2 new comments

• Fix loop type

  #50484 commented on Feb 13, 2021 • 2 new comments

• [package] implement `get_resource_reader` API

  #51674 commented on Feb 16, 2021 • 2 new comments

• move rnn cell size check to cpp

  #51964 commented on Feb 16, 2021 • 2 new comments

• fix nccl version detection

  #52128 commented on Feb 14, 2021 • 2 new comments

• [CI][IOS] Add lite interpreter ios build job

  #52200 commented on Feb 12, 2021 • 2 new comments

• Support CUDA 11.2

  #50232 commented on Feb 12, 2021 • 1 new comment

• Asynchronous Execution on CPU

  #44343 commented on Feb 12, 2021 • 1 new comment

• Learning rate scheduler in C++ API

  #50577 commented on Feb 12, 2021 • 1 new comment

• download speed

  #17023 commented on Feb 12, 2021 • 1 new comment

• Implementation of Normalized Euclidean Distance (NED) in pytorch

  #52005 commented on Feb 12, 2021 • 1 new comment

• Batched matrix multiplication copying the input data(CUDA)

  #52111 commented on Feb 13, 2021 • 1 new comment

• [feature request] Runtime warning for inappropriate labels (among others)

  #10298 commented on Feb 13, 2021 • 1 new comment

• MKLDNN doesnt work and is slower than normal cpu mode

  #32826 commented on Feb 13, 2021 • 1 new comment

• CPU eval BatchNorm2d is not threaded

  #52011 commented on Feb 13, 2021 • 1 new comment

• RPC hangs in callback when there is an error

  #52132 commented on Feb 14, 2021 • 1 new comment

• [RFC] DataLoader architecture updates and TarDataset implementation

  #49440 commented on Feb 14, 2021 • 1 new comment

• torch.nn.functional.cosine_similarity outputs NaN

  #51912 commented on Feb 15, 2021 • 1 new comment

• Using pretrained inception network in inference mode for evaluating GANs while training the main model in DistributedDataParallel (DDP) mode

  #51449 commented on Feb 15, 2021 • 1 new comment

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

  #48279 commented on Feb 15, 2021 • 1 new comment

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

  #48251 commented on Feb 15, 2021 • 1 new comment

• [jit] support for generators and `yield`

  #41416 commented on Feb 15, 2021 • 1 new comment

• torch.triangular_solver doesn't work on batched inputs

  #36734 commented on Feb 15, 2021 • 1 new comment

• Optimize CUDA svd and eig

  #9083 commented on Feb 16, 2021 • 1 new comment

• torch.clamp_max clamp_min shouldn't be there

  #37443 commented on Feb 16, 2021 • 1 new comment

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

  #12831 commented on Feb 16, 2021 • 1 new comment

• Could not run 'aten::mm' with arguments from the 'Metal' backend

  #50877 commented on Feb 16, 2021 • 1 new comment

• Conflict with cudatoolkit 11.0.221

  #51080 commented on Feb 16, 2021 • 1 new comment

• `fractional_max_pool{2, 3}d` inconsistent between CUDA and CPU

  #51622 commented on Feb 16, 2021 • 1 new comment

• JIT engine does not fuse elementwise on 1.7.1 while 1.6.0 is fine

  #52169 commented on Feb 16, 2021 • 1 new comment

• Deadlock in a single machine multi-gpu using dataparlel when cpu is AMD

  #52142 commented on Feb 17, 2021 • 1 new comment

• Complex backward returns NaN values

  #46155 commented on Feb 17, 2021 • 1 new comment

• Why zero_point in _fake_quantize_learnable_per_tensor is different from _fake_quantize_learnable_per_channel_affine?

  #52038 commented on Feb 17, 2021 • 1 new comment

• Failed to cross compile: '/bin/mkalias', needed by 'sleef/src/libm/include/alias_advsimd.h', missing and no known rule to make it

  #51710 commented on Feb 17, 2021 • 1 new comment

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

  #47980 commented on Feb 17, 2021 • 1 new comment

• Simple functions shouldn't go through dispatcher

  #44809 commented on Feb 17, 2021 • 1 new comment

• Build libtorch with AMD support

  #31318 commented on Feb 17, 2021 • 1 new comment

• torch.save has a maximum size regardless of RAM

  #47201 commented on Feb 17, 2021 • 1 new comment

• Build Fail with vulkan on ARM64 and wayland drivers

  #50206 commented on Feb 17, 2021 • 1 new comment

• [BUG] Can't Deepcopy module with weightnorm

  #28594 commented on Feb 18, 2021 • 1 new comment

• Make it possible to use mypy to typecheck code that uses PyTorch

  #16574 commented on Feb 18, 2021 • 1 new comment

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

  #45687 commented on Feb 18, 2021 • 1 new comment

• Measure communicated data volume used in DDP communication hooks

  #52033 commented on Feb 18, 2021 • 1 new comment

• N dimensional histogram function in pytorch as in numpy.histogramdd

  #29209 commented on Feb 18, 2021 • 1 new comment

• Code coverage tool not capturing ONNX tests

  #44120 commented on Feb 18, 2021 • 1 new comment

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

  #42218 commented on Feb 19, 2021 • 1 new comment

• Pytorch can't be installed on Python3.8+

  #47849 commented on Feb 19, 2021 • 1 new comment

• numa.h conflicts with system numa.h

  #9376 commented on Feb 16, 2021 • 1 new comment

• Accelerated 3D depthwise convolution implementation

  #40801 commented on Feb 19, 2021 • 1 new comment

• [caffe2] get rid of Python 3 deprecation warnings about getargspec

  #44186 commented on Feb 14, 2021 • 1 new comment

• fix typo errors in quantization-support.rst Line320

  #44447 commented on Feb 18, 2021 • 1 new comment

• try to solve duplication of Config.h

  #44458 commented on Feb 14, 2021 • 1 new comment

• Add torch utils optimize for vulkan

  #44915 commented on Feb 15, 2021 • 1 new comment

• Use asymmetric mkldnn convolutions for padding='same' convolutions

  #45668 commented on Feb 17, 2021 • 1 new comment

• Autograd functional API should enable_grad

  #47543 commented on Feb 19, 2021 • 1 new comment

• [JIT] Fix type annotations of pooling modules

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

• [WIP] [cuDNN] Enable Conv3d channels-last

  #48430 commented on Feb 19, 2021 • 1 new comment

• compare_model_stub_fx API

  #48701 commented on Feb 15, 2021 • 1 new comment

• optimize channels last for BatchNorm2d on CPU

  #48919 commented on Feb 19, 2021 • 1 new comment

• [NNC] add fusion support for nan_to_num

  #48975 commented on Feb 15, 2021 • 1 new comment

• Dec14 cuda stream

  #49342 commented on Feb 12, 2021 • 1 new comment

• add OneDNN pooling backward

  #49454 commented on Feb 19, 2021 • 1 new comment

• [static runtime] move memonger to new ownership model

  #50060 commented on Feb 17, 2021 • 1 new comment

• [WIP] np.cov

  #50466 commented on Feb 17, 2021 • 1 new comment

• Introduce `mlc` device (ML Compute device) to PyTorch's device list

  #50634 commented on Feb 19, 2021 • 1 new comment

• Added pow() on CPU for float16 & bfloat16

  #50999 commented on Feb 15, 2021 • 1 new comment

• codegen'd API for redispatching

  #51474 commented on Feb 12, 2021 • 1 new comment

• moved nccl module to latest stable: v2.8.3-1

  #51814 commented on Feb 17, 2021 • 1 new comment

• Fix DistributedSampler mem usage on large datasets

  #51841 commented on Feb 18, 2021 • 1 new comment

• Changed documentation and added warning message for writing a tensor created with from_numpy()

  #51870 commented on Feb 18, 2021 • 1 new comment

• Refactor ForeachUnaryOp.cu

  #51894 commented on Feb 19, 2021 • 1 new comment

• Trace linear as aten::linear

  #51897 commented on Feb 18, 2021 • 1 new comment

• var(single observation) should be 0 not NaN

  #51953 commented on Feb 13, 2021 • 1 new comment

• [fx/package] make GraphModules packageable

  #51976 commented on Feb 19, 2021 • 1 new comment

• Automated submodule update: tensorpipe

  #52007 commented on Feb 18, 2021 • 1 new comment

• require non-empty inputs to grad() calls in the API

  #52016 commented on Feb 19, 2021 • 1 new comment

• [PyTorch] Try outlining error message construction in TORCH_{CHECK,INTERNAL_ASSERT}

  #52097 commented on Feb 12, 2021 • 1 new comment

• [PyTorch] Try outlining error message construction in TORCH_{CHECK,INTERNAL_ASSERT}

  #52098 commented on Feb 12, 2021 • 1 new comment

• [PyTorch] Try outlining error message construction in TORCH_{CHECK,INTERNAL_ASSERT}

  #52099 commented on Feb 12, 2021 • 1 new comment

• [PyTorch][easy] Make shared empty string static instead of thread_local

  #52100 commented on Feb 12, 2021 • 1 new comment

• [rfc][wip] Introduce a new way of expressing reductions

  #52121 commented on Feb 12, 2021 • 1 new comment

• [DataLoader] Move BufferedShuffle from Dataset to DataPipe

  #52141 commented on Feb 18, 2021 • 1 new comment

• [RPC] Allow rref.rpc_async() to be non blocking

  #52166 commented on Feb 16, 2021 • 1 new comment

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

  #13918 commented on Feb 13, 2021 • 0 new comments

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

  #7455 commented on Feb 14, 2021 • 0 new comments

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

  #24523 commented on Feb 15, 2021 • 0 new comments

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

  #24422 commented on Feb 16, 2021 • 0 new comments

• [feature request] Adding Pre and Post padding functionalities to pad_sequence function

  #10536 commented on Feb 16, 2021 • 0 new comments

• [TorchScript] Rough edges discovered during type system rewrite

  #50444 commented on Feb 16, 2021 • 0 new comments

• Safe way of loading only weights from *.pt file

  #52181 commented on Feb 16, 2021 • 0 new comments

• [group conv backward] wrong value calculated

  #50042 commented on Feb 16, 2021 • 0 new comments

• [JIT/Futures] Future callbacks/then() APIs should throw the correct exception

  #51673 commented on Feb 16, 2021 • 0 new comments

• Assignment target is transposed when using jit.script and avanced indexing

  #51856 commented on Feb 16, 2021 • 0 new comments

• Allow Custom Classes to register a handler for .to operations

  #51994 commented on Feb 16, 2021 • 0 new comments

• Detect OpenMP Loop and this application may hang warnings

  #52047 commented on Feb 16, 2021 • 0 new comments

• common_jit.check_against_reference() should check gradients w.r.t. weights

  #52052 commented on Feb 16, 2021 • 0 new comments

• quantization - document get_default_qconfig

  #48106 commented on Feb 16, 2021 • 0 new comments

• investigate reported issues with quantized BatchNorm2d

  #43774 commented on Feb 17, 2021 • 0 new comments

• Audit quantization functions to ensure proper argument sizes

  #42336 commented on Feb 17, 2021 • 0 new comments

• Inconsistency loading quantized models when state_dict with Version==None

  #41684 commented on Feb 17, 2021 • 0 new comments

• whitelist keyword to quantization.prepare is implemented incorrectly

  #41633 commented on Feb 17, 2021 • 0 new comments

• Observer in Quantization throws Warning

  #40278 commented on Feb 17, 2021 • 0 new comments

• Pytorch 1.5.0 (installed from conda) errors with complaints about incompatibility between MKL and libgomp when using Pytorch's multiprocessing

  #37377 commented on Feb 17, 2021 • 0 new comments

• Add a minimum compression rate threshold for PowerSGD communication hook

  #52034 commented on Feb 17, 2021 • 0 new comments

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

  #24751 commented on Feb 17, 2021 • 0 new comments

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

  #24614 commented on Feb 17, 2021 • 0 new comments

• Add distributed examples into PyTorch CI tests

  #46321 commented on Feb 17, 2021 • 0 new comments

• Allow ProcessGroup specific options in init_process_group and new_group

  #50295 commented on Feb 17, 2021 • 0 new comments

• [PyTorch Mobile] Can't use Vulkan backend

  #50723 commented on Feb 17, 2021 • 0 new comments

• FP32 depthwise convolution is slow in GPU

  #18631 commented on Feb 17, 2021 • 0 new comments

• [JIT] Support for CUDA streams in TorchScript

  #41355 commented on Feb 17, 2021 • 0 new comments

• [JIT] Rough edges found while exploring expressions supported by TorchScript

  #51154 commented on Feb 17, 2021 • 0 new comments

• Logspace broken for complex dtypes

  #38779 commented on Feb 18, 2021 • 0 new comments

• [feature request] Set limit on GPU memory use

  #18626 commented on Feb 18, 2021 • 0 new comments

• [Feature request] modified Cholesky decomposition

  #25785 commented on Feb 18, 2021 • 0 new comments

• Complex Numbers Support

  #33152 commented on Feb 18, 2021 • 0 new comments

• [feature request] convtbc with group convolution

  #10685 commented on Feb 19, 2021 • 0 new comments

• Build system doesn't prevent ATen/core from including non-core files

  #10691 commented on Feb 19, 2021 • 0 new comments

• Tensor.register_hook is not passing the tensor object to the hook function

  #10683 commented on Feb 19, 2021 • 0 new comments

• [Not to Land] Add Verify Shape analysis pass

  #18813 commented on Feb 12, 2021 • 0 new comments

• [DO NOT DELETE OR MERGE] Nightly release

  #26921 commented on Feb 19, 2021 • 0 new comments

• [DO NOT DELETE OR MERGE] Postnightly release

  #27167 commented on Feb 19, 2021 • 0 new comments

• fixes #38775 #38779: complex support for linspace and logspace

  #38875 commented on Feb 19, 2021 • 0 new comments

• `torch.sign` support output of different dtype than input

  #45374 commented on Feb 17, 2021 • 0 new comments

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

  #47715 commented on Feb 17, 2021 • 0 new comments

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

  #47950 commented on Feb 13, 2021 • 0 new comments

• [wip] Pushed unsupported scenarios to slow path

  #48224 commented on Feb 19, 2021 • 0 new comments

• add channels last for MaxPool2d

  #48917 commented on Feb 19, 2021 • 0 new comments

• add channels last support for AvgPool2d on CPU

  #48918 commented on Feb 19, 2021 • 0 new comments

• add channels last for AdapativeMaxPool2d

  #48920 commented on Feb 19, 2021 • 0 new comments

• [wip] Replace optimizers in torch.optim with the ones from torch.optim._multi_tensor

  #49039 commented on Feb 19, 2021 • 0 new comments

• Add Sparse support for asinh

  #49160 commented on Feb 19, 2021 • 0 new comments

• Update foreach binary ops with a single scalar and list of tensors

  #49250 commented on Feb 18, 2021 • 0 new comments

• Fix crash for topk(k=0)

  #49311 commented on Feb 19, 2021 • 0 new comments

• add OneDNN relu backward and reshape backward

  #49455 commented on Feb 19, 2021 • 0 new comments

• add channels last support for thnn_conv2d (non-dilated)

  #49582 commented on Feb 19, 2021 • 0 new comments

• Refactor foreach unary ops tests to use OpInfo

  #49712 commented on Feb 18, 2021 • 0 new comments

• Update foreach max/min

  #49714 commented on Feb 18, 2021 • 0 new comments

• add channels last for GroupNorm

  #49821 commented on Feb 19, 2021 • 0 new comments

• add channel last support for MaxUnpool2d

  #49984 commented on Feb 19, 2021 • 0 new comments

• [quantization] Add some support for 3d operations

  #50003 commented on Feb 17, 2021 • 0 new comments

• Stable sort for CPU

  #50052 commented on Feb 17, 2021 • 0 new comments

• changed TE 'Allocate/Free' API to take argument 'Buf' instead of arguments 'Var', 'dtype', 'dims'.

  #50167 commented on Feb 19, 2021 • 0 new comments

• add channels last support for ChannelShuffle

  #50247 commented on Feb 19, 2021 • 0 new comments

• add OneDNN batch_norm backward

  #50460 commented on Feb 19, 2021 • 0 new comments

• Huber loss

  #50553 commented on Feb 19, 2021 • 0 new comments

• add channels last support for PixelShuffle and PixelUnshuffle

  #50573 commented on Feb 19, 2021 • 0 new comments

• Refactor foreach binary ops tests with scalars to use OpInfo

  #51058 commented on Feb 19, 2021 • 0 new comments

• Refactor foreach binary ops tests with tensor lists to use OpInfo

  #51059 commented on Feb 19, 2021 • 0 new comments

• Refactor foreach pointwise ops tests to use OpInfo

  #51060 commented on Feb 19, 2021 • 0 new comments

• Refactor ForeachUtils.h

  #51131 commented on Feb 18, 2021 • 0 new comments

• [quant][refactor] Factor out MHA code dup from nn and nn.quantizable

  #51169 commented on Feb 17, 2021 • 0 new comments

• Updated foreach pointwise ops to support complex and bloat16 on CUDA

  #51170 commented on Feb 19, 2021 • 0 new comments

• Upgrade onednn to v.1.8.1

  #51184 commented on Feb 19, 2021 • 0 new comments

• add channels last support for ConvTranspose2d

  #51185 commented on Feb 19, 2021 • 0 new comments

• Second prototype version: Ban inference tensor in autograd

  #51274 commented on Feb 13, 2021 • 0 new comments

• Raise error in `torch.normal` if `std` Tensor <= 0

  #51317 commented on Feb 12, 2021 • 0 new comments

• [ONNX] Update ONNX documentation

  #51362 commented on Feb 18, 2021 • 0 new comments

• update autograd kernels to use redispatch

  #51363 commented on Feb 12, 2021 • 0 new comments

• [CUDA graphs] Private mempools for CUDA graphs

  #51436 commented on Feb 19, 2021 • 0 new comments

• Improve test_reference_numerics

  #51604 commented on Feb 19, 2021 • 0 new comments

• [quant][graphmode][fx] Add support for packed params in state_dict

  #51639 commented on Feb 18, 2021 • 0 new comments

• [package] make importer a little more obscure

  #51676 commented on Feb 16, 2021 • 0 new comments

• update sphinx to 3.5.1

  #51774 commented on Feb 19, 2021 • 0 new comments

• Implemented torch.linalg.multi_dot

  #51807 commented on Feb 19, 2021 • 0 new comments

• remove unnecessary/dead code in upsample_nearest1d cuda kernel

  #51916 commented on Feb 12, 2021 • 0 new comments

• port upsample_linear1d to structured

  #51917 commented on Feb 12, 2021 • 0 new comments

• [ONNX] Fix export of copy_ operator

  #51938 commented on Feb 18, 2021 • 0 new comments

• Add discontiguous kwarg to make_tensor

  #51985 commented on Feb 19, 2021 • 0 new comments

• [v1.8.0] Getting signal for `release/1.8`

  #51995 commented on Feb 19, 2021 • 0 new comments

• codegen'd API for redispatching

  #52008 commented on Feb 12, 2021 • 0 new comments

• update tracing codegen to use redispatch API

  #52009 commented on Feb 12, 2021 • 0 new comments

• port upsample_bilinear2d and upsample_bicubic2d to structured

  #52012 commented on Feb 12, 2021 • 0 new comments

• Adding infos= variant for torch.linalg.inv

  #52046 commented on Feb 12, 2021 • 0 new comments

• Move is_complex() directly onto Tensor

  #52060 commented on Feb 17, 2021 • 0 new comments

• [docs] Add deprecation note to torch.chain_matmul

  #52088 commented on Feb 19, 2021 • 0 new comments

• [JIT] Add CUDA Conv-Relu fusion for Freezing Model Optmization

  #52102 commented on Feb 18, 2021 • 0 new comments

• Add scatter_add to amp promote list

  #52133 commented on Feb 13, 2021 • 0 new comments

• Remove rounding_mode="true" option

  #52139 commented on Feb 12, 2021 • 0 new comments

• DO NOT MERGE, testing CUDA 11.1 builds

  #52152 commented on Feb 17, 2021 • 0 new comments

• [RPC] Move rref proxy timeout tests to general RpcTest

  #52167 commented on Feb 17, 2021 • 0 new comments

• [doc] FX Graph Mode Quantization - fix preamble

  #52192 commented on Feb 13, 2021 • 0 new comments

• [testing] support op with args/kwargs in test_unary_ufunc

  #52194 commented on Feb 12, 2021 • 0 new comments

• add OpInfo entry for signbit

  #52198 commented on Feb 13, 2021 • 0 new comments