13 Pull requests merged by 5 people

• Add redirect links to the contributor wiki

  #106863 merged Sep 22, 2023

• [Release-2.1]Add tensorboard to pip requirements

  #109823 merged Sep 21, 2023

• [CI] Add `torch.compile` works without numpy test (#109624)

  #109818 merged Sep 21, 2023

• [Release-2.1]Fix the parameter error in test_device_mesh.py

  #109826 merged Sep 21, 2023

• [Release-2.1]Add `finfo` properties for float8 dtypes

  #109808 merged Sep 21, 2023

• [Release/2.1][JIT] Fix typed enum handling in 3.11

  #109807 merged Sep 21, 2023

• [release only] Docker build - Setup release specific variables

  #109809 merged Sep 21, 2023

• [Release/2.1] [Docs] Fix `compiler.list_backends` invocation

  #109800 merged Sep 21, 2023

• [Release/2.1] [Docs] Fix typo in `torch.unflatten`

  #109801 merged Sep 21, 2023

• Remove torchtext from Build Official Docker images (#109799)

  #109803 merged Sep 21, 2023

• [release-2.1] Add docs for torch.compile(numpy)

  #109789 merged Sep 21, 2023

• Update dynamic shapes documentation

  #109787 merged Sep 21, 2023

• [release-2.1] Make numpy dependency optional for torch.compile

  #109608 merged Sep 19, 2023

160 Pull requests opened by 95 people

• [Decomposition] hann_window.periodic

  #109464 opened Sep 17, 2023

• Revert "[inductor] Fix inputs with existing offsets (#108168)"

  #109465 opened Sep 17, 2023

• Revert "[inductor] let codegen not rely on node order (#107320)"

  #109466 opened Sep 17, 2023

• Back out "[PyTorch] Add Expanded call stack to nodes (#108426)"

  #109467 opened Sep 17, 2023

• [vision hash update] update the pinned vision hash

  #109469 opened Sep 18, 2023

• [Docs][Distributed] Add migration notes for `--local-rank` option style change for `torchrun` in PyTorch 2.0

  #109480 opened Sep 18, 2023

• [xla hash update] update the pinned xla hash

  #109481 opened Sep 18, 2023

• [POC] Add caching for faketensor propagation

  #109485 opened Sep 18, 2023

• Fix access to unitialized memory in VSX vector functions for quantized values

  #109487 opened Sep 18, 2023

• Fix MultiProcess failure on nodes with 1 GPU

  #109491 opened Sep 18, 2023

• [inductor] Remove `is_big_gpu` check

  #109492 opened Sep 18, 2023

• Avoid cuda stubs libraries being RPATHed

  #109493 opened Sep 18, 2023

• Report NameError when name is not defined, rather than unimplemented

  #109503 opened Sep 18, 2023

• [torch.optim/C++] Add NAdam optimizer

  #109506 opened Sep 18, 2023

• [export] Verifier for exported program

  #109519 opened Sep 18, 2023

• DRAFT

  #109521 opened Sep 18, 2023

• [AOTInductor] Initial functionality for Inf and NaN checker

  #109526 opened Sep 18, 2023

• Support register_hook on intermediate tensors

  #109537 opened Sep 18, 2023

• [DO NOT LAND] Testing

  #109540 opened Sep 18, 2023

• Clean up fill_quant

  #109544 opened Sep 18, 2023

• Update torchbench pin

  #109545 opened Sep 18, 2023

• [foreach] check for empty tensors before dispatching to MTA

  #109550 opened Sep 18, 2023

• [wip]: fsspec remote code cache

  #109551 opened Sep 18, 2023

• [3/N][2D] Enable training with new 2D flow

  #109553 opened Sep 18, 2023

• [dynamo][`__torch_function__` 2/n] Refactor TensorWithTFOverrideVariable

  #109556 opened Sep 18, 2023

• [MPS] Fix mps to cpu copy with storage offset

  #109557 opened Sep 18, 2023

• Add requirement for input to AllGatherIntoTensor to be contiguous

  #109561 opened Sep 18, 2023

• Update triton pin to move past bugs affecting sparse kernels

  #109562 opened Sep 19, 2023

• [vision hash update] update the pinned vision hash

  #109563 opened Sep 19, 2023

• Adding T4 GPUs to inductor nightly benchmarks

  #109565 opened Sep 19, 2023

• [1/N] Dynamo skipfiles refactor

  #109567 opened Sep 19, 2023

• [WIP] fix: added check for convolution output shape wrt kernel_size and input length

  #109569 opened Sep 19, 2023

• [Not for merge][Repro] Unbacked symint in Inductor size_hint output

  #109573 opened Sep 19, 2023

• [MPS] Fix nll_loss with default ignore_index

  #109574 opened Sep 19, 2023

• Add Half support for AvgPool2d on CPU

  #109578 opened Sep 19, 2023

• Support FloatFunctional subclasses in eager mode

  #109580 opened Sep 19, 2023

• [MPS] Fix sort with empty tensor.

  #109584 opened Sep 19, 2023

• [WIP] Trace model attribute mutation

  #109590 opened Sep 19, 2023

• [dynamo][`__torch_function__` 3/n] TensorWithTFOverrideVariable inheritance from TensorVariable

  #109597 opened Sep 19, 2023

• ensure uint8 is honoured for cpu operations in dynamo

  #109598 opened Sep 19, 2023

• [inductor] Update triton pin

  #109601 opened Sep 19, 2023

• Fix send()/recv() to adhere to timeout

  #109611 opened Sep 19, 2023

• [Decomposition] glu

  #109618 opened Sep 19, 2023

• [Dynamo] Add native support for Triton Kernels to Dynamo

  #109623 opened Sep 19, 2023

• [ONNX] Enable more OpInfo tests in fx

  #109627 opened Sep 19, 2023

• [Inductor] Move fake_tensors to the same device as example_inputs

  #109630 opened Sep 19, 2023

• dynamo: break graph when "out" has complex dtype

  #109633 opened Sep 19, 2023

• Enable masked_scatter_backward for inductor

  #109642 opened Sep 19, 2023

• [pytorch] make IterableDataset of Iterable type

  #109645 opened Sep 19, 2023

• [DDP] Make _ReplicateState inherit from _State and make replicate eagerly initialized

  #109647 opened Sep 19, 2023

• Add `backend` concept to `torch.export` API

  #109649 opened Sep 19, 2023

• Add ``onnx`` backend to ``torch.export`` API

  #109650 opened Sep 19, 2023

• [FSDP][state_dict][dtensor][bugfix] Fix dtensor not picked up in load

  #109651 opened Sep 19, 2023

• [WIP] compiled autograd on inductor torchbench

  #109653 opened Sep 19, 2023

• WIP support for fp8 allgather FSDP

  #109654 opened Sep 19, 2023

• [aotinductor] Skip benchmarks with control flow

  #109661 opened Sep 19, 2023

• [vision hash update] update the pinned vision hash

  #109664 opened Sep 20, 2023

• Add Pass to move constructors from cpu to cuda

  #109665 opened Sep 20, 2023

• [inductor] Use _unsafe_view decompostion

  #109669 opened Sep 20, 2023

• [WIP] Dynamo CPU backend under Windows

  #109677 opened Sep 20, 2023

• [pytree] Make `optree` optional and populate members from `_pytree` when it not available

  #109684 opened Sep 20, 2023

• [MPS] add support for aten::nextafter

  #109685 opened Sep 20, 2023

• Higher order op for preserving leaf functions through trace, particularly for getting user defined hooks to compiled autograd

  #109690 opened Sep 20, 2023

• [3/N] Add -Wdeprecated and related fixes

  #109698 opened Sep 20, 2023

• Add torch.quint4x2 supoprt for mixed dtypes GEMM

  #109712 opened Sep 20, 2023

• [decomp] Fix baddbmm decomposition

  #109714 opened Sep 20, 2023

• Trace frames with `numpy.ndarray`.

  #109715 opened Sep 20, 2023

• [ONLY FOR TESTING] Revert "Reland 3rd try [finishing colesbury's PR 100642] Guard on nn.Module dicts and type (#109323)"

  #109726 opened Sep 20, 2023

• When doing typed typecheck, also check signature with symint removed

  #109727 opened Sep 20, 2023

• Histogram Fixes for QAT

  #109728 opened Sep 20, 2023

• [inductor] enable mypy checking in torch/_inductor/codegen/cpp.py

  #109729 opened Sep 20, 2023

• [pytorch][jit] allow passing in obj loader in unpickle api

  #109730 opened Sep 20, 2023

• [WIP][dynamo] make symbool input propagate guard to outer shape_env while dynamo tracing

  #109734 opened Sep 20, 2023

• [dynamo] lift the constraint that cannot make_fx a dynamo compiled function

  #109735 opened Sep 20, 2023

• Native c10d_functional ops

  #109738 opened Sep 20, 2023

• [C10D] Report detected failures when emitting collective end events.

  #109739 opened Sep 20, 2023

• Reland use std::make_unique after internal changes

  #109742 opened Sep 20, 2023

• [caffe2/torch] Package Importer with compatibility for Lazy Imports

  #109745 opened Sep 20, 2023

• rename nanogpt_generate to nanogpt to also support train

  #109746 opened Sep 20, 2023

• [AOTInductor] Switch ProxyExecutor to use AtenTensorHandle

  #109748 opened Sep 20, 2023

• [Quantization] Add "quantization_tag" as metadata to fx proxy

  #109749 opened Sep 20, 2023

• Extend storage access error api for untyped_storage()

  #109750 opened Sep 20, 2023

• [profiler] Show shapes for lists of tensors in chrome traces #109263

  #109751 opened Sep 20, 2023

• [vision hash update] update the pinned vision hash

  #109756 opened Sep 21, 2023

• Bump xla_base version tag to v1.1

  #109757 opened Sep 21, 2023

• ln + fp8 quant benchmark

  #109765 opened Sep 21, 2023

• [inductor] Add lowering for aten.take

  #109784 opened Sep 21, 2023

• Simplify the conditionals used for learning rate calculation for `ConstantLR` learning rate scheduler

  #109785 opened Sep 21, 2023

• Fix tensor unpickling

  #109792 opened Sep 21, 2023

• Surround num-destroyed-communicators with spaces

  #109795 opened Sep 21, 2023

• Don't -Werror on cast-function-type

  #109796 opened Sep 21, 2023

• Add `pin_memory` to `torch.Tensor` type annotation args

  #109797 opened Sep 21, 2023

• Fix `functional::smooth_l1_loss` signatures to not override `beta`

  #109798 opened Sep 21, 2023

• [Not for Land] Add verbose all-gather info

  #109804 opened Sep 21, 2023

• [Inductor CUTLASS backend] Epilogue fusion codegen prototype

  #109810 opened Sep 21, 2023

• MAINT: pytorchify torch._numpy tests: core/ and fft/

  #109815 opened Sep 21, 2023

• [ROCM] enable skipped lazy and api cpp tests

  #109817 opened Sep 21, 2023

• [LTC] correct AdaptiveAvgPool3d channel dim index for shape inference

  #109822 opened Sep 21, 2023

• Added a UserWarning when using torch.{std,var,std_mean,std_var} with dof<=0

  #109824 opened Sep 21, 2023

• Update custom Function preserve torch function when inputs returned as-is

  #109825 opened Sep 21, 2023

• [quant] Enable quantization for wav2letter

  #109830 opened Sep 21, 2023

• Ensure fill_ works when value is a view of self

  #109835 opened Sep 21, 2023

• Fixed minor issues for bmm/mm decompositon

  #109836 opened Sep 21, 2023

• [vision hash update] update the pinned vision hash

  #109837 opened Sep 22, 2023

• [MPS] add support for heaviside

  #109838 opened Sep 22, 2023

• [Reland2] Update NVTX to NVTX3

  #109843 opened Sep 22, 2023

• Re-enable more Windows tests

  #109847 opened Sep 22, 2023

• Use _check_is_size for validate_dim_length

  #109849 opened Sep 22, 2023

• Fix S367052 to unblock ICVR MC3

  #109853 opened Sep 22, 2023

• Add default arguments to sym_constrain_range_for_size

  #109858 opened Sep 22, 2023

• Remove RangeConstraints in favor of ValueRanges

  #109859 opened Sep 22, 2023

• Move InputDim to torch.export instead of defining in a pass

  #109860 opened Sep 22, 2023

• Integrate xpu into torch.Generator and torch.seed

  #109866 opened Sep 22, 2023

• Don't link to libcpuinfo on s390x

  #109875 opened Sep 22, 2023

• [profiler] add PyCMethod_Type to assertion

  #109878 opened Sep 22, 2023

• Move at::{Refcounted,}MapAllocator to c10

  #109881 opened Sep 22, 2023

• [Easy] Fix "comparison of unsigned integer with zero" warning

  #109882 opened Sep 22, 2023

• Use wrapper instead of V.graph.wrapper_code

  #109883 opened Sep 22, 2023

• Fix CPU bitwise shifts for out-of-limit values in VSX-vec

  #109886 opened Sep 22, 2023

• Add torch.fx.experimental.recording to uninteresting_files()

  #109887 opened Sep 22, 2023

• Revert D49433268: Multisect successfully blamed "D49433268: [pytorch][PR] [Inductor] Extend Pattern Matcher to Match Equivalent Function Invocation" for test or build failures

  #109891 opened Sep 22, 2023

• Add support for item() and nonzero() codegen in Inductor

  #109893 opened Sep 22, 2023

• [Easy] Remove hook warning where source is always guaranteed

  #109898 opened Sep 22, 2023

• [core IR] Remove trunc decomp and add trunc to core

  #109902 opened Sep 22, 2023

• [core IR] Add decompositions for _assert_async to no-op

  #109905 opened Sep 22, 2023

• Reland "Update AOTAutograd to use FunctionalTensorMode instead of C++ functionalization (#106406)"

  #109906 opened Sep 22, 2023

• Add `endpoint` argument in `linspace` to match numpy behavior

  #109907 opened Sep 22, 2023

• [quant][pt2] Support cudnn_batch_norm in QAT fusion

  #109908 opened Sep 22, 2023

• Add torch.library.impl_abstract

  #109912 opened Sep 22, 2023

• [inductor] Avoid bool being upcast to int

  #109913 opened Sep 22, 2023

• Decompose to native_dropout in eval mode as well

  #109914 opened Sep 22, 2023

• [dynamo] fix reconstruct of ConvertSymintSource.

  #109916 opened Sep 22, 2023

• [BE]s/DEFINE_ENUM/DEFINE_ST_ENUM_VAL_/

  #109917 opened Sep 22, 2023

• [PTD][TP] Refactor the test and temporary disable one test case

  #109919 opened Sep 22, 2023

• [torch] Defer resolution of allowed/disallowed decorators

  #109921 opened Sep 22, 2023

• Define the public API for torch.distributed.fsdp

  #109922 opened Sep 22, 2023

• (pytorch) add List[float] type to get_lr

  #109927 opened Sep 22, 2023

• [vision hash update] update the pinned vision hash

  #109930 opened Sep 23, 2023

• [easy] Don't check uint32 >= 0

  #109932 opened Sep 23, 2023

• [WIP] Remove recordStream for callback that ends a profiler event

  #109933 opened Sep 23, 2023

• [pytorch vulkan] add tensor vulkan check for at::cat

  #109936 opened Sep 23, 2023

• Fix S367052 to unblock ICVR MC3

  #109937 opened Sep 23, 2023

• updated the tech doc

  #109939 opened Sep 23, 2023

• Rename `torch._C._TensorBase` to `TensorBase`

  #109940 opened Sep 23, 2023

• Add test for `ShapeEnv` recording fallback.

  #109944 opened Sep 23, 2023

• Add test for `ShapeEnv` state when not recording.

  #109945 opened Sep 23, 2023

• Enable typechecking for _inductor/fx_passes/split_cat.py

  #109951 opened Sep 23, 2023

• Enable typechecking for _inductor/fx_passes/pre_grad.py

  #109952 opened Sep 23, 2023

• 109605 dynamo scalar ndarray pow gen

  #109953 opened Sep 23, 2023

• Enable typechecking for _inductor/fx_passes/pad_mm.py

  #109954 opened Sep 23, 2023

• Enable typechecking for _inductor/fx_passes/joint_graph.py

  #109955 opened Sep 23, 2023

• [Dynamo] Support python class member_descriptor

  #109956 opened Sep 23, 2023

• [vision hash update] update the pinned vision hash

  #109957 opened Sep 24, 2023

• Clean up CMake target linking

  #109959 opened Sep 24, 2023

• [inductor] Do type promotion in pointless cumsum pattern replacement

  #109960 opened Sep 24, 2023

• [HigherOrderOp] wrap (and checkpoint) should accept pytree inputs

  #109962 opened Sep 24, 2023

• [2/N] Cleanup header inclusions in torch_cpu by iwyu

  #109964 opened Sep 24, 2023

• fix typo

  #109965 opened Sep 24, 2023

• [WIP] Make ONNX OpSchema function matcher more robust

  #109966 opened Sep 24, 2023

• [Dynamo] Lower builtin all for list iterators

  #109972 opened Sep 24, 2023

88 Issues closed by 35 people

• Can we do aggregate communication with dtensor through APIs like all_reduce?

  #109950 closed Sep 24, 2023

• "OutOfMemoryError" is not a valid exception class

  #109961 closed Sep 24, 2023

• `torch.overrides.get_overridable_functions()` throws a bunch of warnings

  #109842 closed Sep 23, 2023

• Failed running constrain_as_size: ShapeEnv not found

  #109844 closed Sep 23, 2023

• DISABLED test_non_standard_bool_values_masked_scatter_cuda_bool (__main__.TestCommonCUDA)

  #109006 closed Sep 23, 2023

• DISABLED test_nondeterministic_alert_MaxUnpool3d_cpu_float64 (__main__.TestTorchDeviceTypeCPU)

  #109007 closed Sep 23, 2023

• torch.mean() operation saves its input for backward (into _saved_self)

  #109876 closed Sep 23, 2023

• Fix torch.utils.benchmark.Timer.timeit while using privateuse1.

  #109776 closed Sep 23, 2023

• Error occurs when using focal loss in DaFormer

  #109867 closed Sep 23, 2023

• Track the accuracy regress for HF with max-autotune enabled

  #109736 closed Sep 23, 2023

• FSDP SHARDE_STATE_DICT Failing to load T5 model after saving on Nightlies

  #103627 closed Sep 22, 2023

• cm3leon_generate failing compilation

  #109926 closed Sep 22, 2023

• DISABLED test_nondeterministic_alert_MaxUnpool2d_cpu_float64 (__main__.TestTorchDeviceTypeCPU)

  #108887 closed Sep 22, 2023

• DISABLED test_jvp_ScaleGradGenVmapAutogradFunction_cpu_float32 (__main__.TestOperatorsCPU)

  #105129 closed Sep 22, 2023

• [FSDP] Problem with c10d pickle in distributed run

  #109396 closed Sep 22, 2023

• Patch release before v2.1.0?

  #107303 closed Sep 22, 2023

• DISABLED test_runtime_assert_equality_constraint (__main__.TestPasses)

  #107273 closed Sep 22, 2023

• Allow user to update metrics in Tensorboard SummaryWriter.add_hparam

  #37738 closed Sep 22, 2023

• Python Dispatcher Causes Infinite Recursion in to_copy

  #103532 closed Sep 22, 2023

• Multiple backward on multiple tensors returned from DDP with static_graph=True raises autograd assertion error

  #106953 closed Sep 22, 2023

• Torch Jit Script Running on CPU produces NAN values

  #108568 closed Sep 22, 2023

• Segmentation fault in 'flatbuffer_loader.cpp:298'

  #109793 closed Sep 21, 2023

• Symbolic size not traced for some NumPy tensors

  #106376 closed Sep 21, 2023

• Improve log1p(complex) speed/accuracy

  #107022 closed Sep 21, 2023

• ComplexHalf support

  #71680 closed Sep 21, 2023

• `index_put` with bool index errors out

  #93421 closed Sep 21, 2023

• Can not to install pytorch-2.0.0+rocm5.4.2

  #98257 closed Sep 21, 2023

• project c10_intrusive_ptr_benchmark link error

  #31315 closed Sep 21, 2023

• DISABLED test_all_gather_uneven (__main__.DeviceMeshCollectiveTest)

  #98049 closed Sep 21, 2023

• FSDP Dtensor for state_dict seems not picked up when loading

  #109648 closed Sep 21, 2023

• Torch Nightly with MPI giving RuntimeError: No backend type associated with device type cuda

  #109543 closed Sep 21, 2023

• share_memory does not work with 'meta' device

  #109702 closed Sep 21, 2023

• Move Inductor FX passes - pre grad pass to Dynamo

  #109357 closed Sep 21, 2023

• `torch.finfo(torch.float8_e4m3fn).max` crashes python runtime

  #109737 closed Sep 21, 2023

• DISABLED test_kineto_profiler_with_environment_variable (__main__.TestProfiler)

  #107383 closed Sep 21, 2023

• cudagraph tree segfault

  #105169 closed Sep 20, 2023

• [FX] Cannot trace calls with python Enum values

  #82135 closed Sep 20, 2023

• Enums + jit still not working in PyTorch 2.1 w/ Python 3.11

  #108933 closed Sep 20, 2023

• Tensor indexing error when using 'mps'

  #109716 closed Sep 20, 2023

• PT2 constant folding is using a lot of memory and caused GPU OOM

  #108388 closed Sep 20, 2023

• DISABLED test_sdpa_rewriter_11_cpu (__main__.SDPAPatternRewriterCpuTests)

  #109656 closed Sep 20, 2023

• `torch.compile` can not be used out of the box in nightly/2.1

  #109387 closed Sep 20, 2023

• aten::fused_moving_avg_obs_fake_quant' to ONNX opset version 14 is not supported

  #108834 closed Sep 20, 2023

• Test results for transformer change as the batch size changes

  #109676 closed Sep 20, 2023

• Exportability problems with `at::RecordFunction` due to ABI mismatch in layout of struct defined in header file

  #108566 closed Sep 20, 2023

• implement weight_norm on mps

  #104513 closed Sep 20, 2023

• Very poor performance using the unique function

  #107098 closed Sep 20, 2023

• [inductor][cpu] [amp] acc check fails

  #108333 closed Sep 20, 2023

• Dynamo pytree registration is distinct from user pytree registration, leading to behavior divergence

  #107315 closed Sep 20, 2023

• torch.var outputs inf for correction>N

  #109186 closed Sep 19, 2023

• SDPA Memory Efficient Attention is producing NaNs for certain mask patterns.

  #109517 closed Sep 19, 2023

• Backward memory usage of the embedded bag

  #109282 closed Sep 19, 2023

• Inputs to setup_context() of autograd.Function does not contain default values of forward()

  #108529 closed Sep 19, 2023

• InternalTorchDynamoError when using torch.compile with Huggingface Llama model and PEFT

  #98993 closed Sep 19, 2023

• `aten::fill_` causes H2D sync

  #109115 closed Sep 19, 2023

• DISABLED test_batchnorm_nhwc_cpu (__main__.TestNN)

  #106543 closed Sep 19, 2023

• Wrong example case in TORCH.UNFLATTEN

  #109559 closed Sep 19, 2023

• Executing `torch.cuda.is_available()` on a computer without an NVIDIA GPU will exit abnormally

  #101147 closed Sep 19, 2023

• RuntimeError: Unrecognized CachingAllocator option: C

  #101158 closed Sep 19, 2023

• [bug][docs]torch.compile.list_backends() should change to torch.compiler.list_backends()

  #109451 closed Sep 19, 2023

• bug at https://pytorch.org/get-started/previous-versions/

  #109570 closed Sep 19, 2023

• gdb debug pytorch but Undefined command: "torch-tensor-repr"

  #109472 closed Sep 19, 2023

• New PT2 Dashboard Tracker

  #96953 closed Sep 19, 2023

• `x.to(device="meta")` infinite loops when the python dispatcher is enabled

  #109558 closed Sep 18, 2023

• potential typo in torch.searchsorted documentation

  #109298 closed Sep 18, 2023

• Mismatch between PyTorch and onnxruntime when converting TransformerEncoder to onnx

  #109532 closed Sep 18, 2023

• hf_T5_generate: ERROR:common:call_function BuiltinVariable(str) [UserFunctionVariable()] {}

  #108699 closed Sep 18, 2023

• cm3leon_generate: ERROR:common:inline in skipfiles: _SpecialForm.getitem | inner /home/binbao/local/miniconda3/envs/pytorch-3.10/lib/python3.10/typing.py

  #108697 closed Sep 18, 2023

• [Dynamo] Graph break in Huggingface model Vilt

  #98045 closed Sep 18, 2023

• CrossEntropyLoss with label_smoothing not work in compiled model

  #98571 closed Sep 18, 2023

• _dynamo.skip doesn't actually add the function to skipfiles

  #98847 closed Sep 18, 2023

• 14k github models PT2 compilation time tracking

  #98997 closed Sep 18, 2023

• `torch.compile` failed in multi node distributed training

  #99067 closed Sep 18, 2023

• `torch.compile` failed with `pytorchddp`

  #99074 closed Sep 18, 2023

• [dynamo] `call_function BuiltinVariable` and `op returned non-Tensor` cause graph breaks

  #99752 closed Sep 18, 2023

• GPU Memory not being freed using PT 2.0, issue absent in earlier PT versions

  #99835 closed Sep 18, 2023

• Typo in annotation of DispatchKeySet in DispatchKeySet.h

  #108641 closed Sep 18, 2023

• [ONNX] cache_size_limit reached when batch exporting

  #108938 closed Sep 18, 2023

• mypy hints internal error while trying to check types, but lint runner hides it

  #104940 closed Sep 18, 2023

• DISABLED test_reduce_scatter_uneven (__main__.DeviceMeshCollectiveTest)

  #98024 closed Sep 18, 2023

• torch.compile Raises "NameError: name 'buf0' is not defined" for LeakyReLU with Empty Input Tensor

  #107995 closed Sep 18, 2023

• The different results from same MLP.

  #109371 closed Sep 18, 2023

• Build error in third_party/ideep

  #109397 closed Sep 18, 2023

• Errir building pytorch from source on Windows 10 Linking CXX shared library bin\torch_cuda.dll

  #102929 closed Sep 18, 2023

• FSDP + PEFT Prompt Tuning Issue

  #109077 closed Sep 18, 2023

• dynamo failed to handle "isinstance" builtin function with variable device

  #107856 closed Sep 18, 2023

• DISABLED test_vmapvjpvjp_linalg_lu_factor_ex_cuda_float32 (__main__.TestOperatorsCUDA)

  #104737 closed Sep 18, 2023

• TorchDynamo doesn't guard default dtype

  #109458 closed Sep 17, 2023

156 Issues opened by 82 people

• Resample in transforms.RandomRotation

  #109971 opened Sep 24, 2023

• torch-<version>.dist-info WHEEL file contains incorrect metadata for M1/M2 macOS platform

  #109970 opened Sep 24, 2023

• profiler doesn't work with multiple workers

  #109969 opened Sep 24, 2023

• Dtype hard-coded in DataLoader (for python floats).

  #109968 opened Sep 24, 2023

• WelfordReduction seems to have invalid/dead code when reduction_numel <= 1

  #109963 opened Sep 24, 2023

• How to compile torch 2.0.1 version from source?

  #109958 opened Sep 24, 2023

• Missing binary dependency for LibTorch (cpu) on Windows

  #109949 opened Sep 23, 2023

• Simple script segfaulting when grad is enabled

  #109948 opened Sep 23, 2023

• `self.profiler.kineto_results` always none which breaks assert

  #109947 opened Sep 23, 2023

• Indexed batch matrix multiplication to support MoEs and FFFs

  #109946 opened Sep 23, 2023

• Problems when loading PT files und Linux - Duda which are created under Mac Apple Silicon MPS

  #109943 opened Sep 23, 2023

• pytorch XLA document error

  #109942 opened Sep 23, 2023

• Need latest NCCL support to reduce GPU HBM consumption

  #109941 opened Sep 23, 2023

• Batching for is_in

  #109938 opened Sep 23, 2023

• test test_2d_fsdp_integration_fsdp_nested_param_groups failed

  #109934 opened Sep 23, 2023

• Memory access fault with AMD Rocm

  #109929 opened Sep 22, 2023

• [torch.compile] returns different value/type tensor with naive execution

  #109925 opened Sep 22, 2023

• Import order issue with torch and pybind11 Library Statically Linked to libstdc++

  #109923 opened Sep 22, 2023

• [AOTInductor] ERROR: undefined symbol: _binary_constants_bin_start

  #109920 opened Sep 22, 2023

• Dynamo error for autograd function

  #109910 opened Sep 22, 2023

• Large Discrepancies between PyTorch and ONNXRuntime Inference

  #109909 opened Sep 22, 2023

• Error using torch.onnx.dynamo_export

  #109903 opened Sep 22, 2023

• DISABLED test_tags_function (__main__.ActivationCheckpointingViaTagsTests)

  #109901 opened Sep 22, 2023

• Unbacked symint support in cpu inductor backend

  #109897 opened Sep 22, 2023

• moco: torch._dynamo.exc.Unsupported: hasattr: TensorVariable()

  #109895 opened Sep 22, 2023

• functorch_maml_omniglot, lennard_jones, maml_omniglot: SyntaxError: invalid decimal literal

  #109894 opened Sep 22, 2023

• Experiment moving torch.onnx.dynamo_export from torch._dynamo.export to torch.export

  #109889 opened Sep 22, 2023

• Abnormal GPU memory usage when `import torch.utils.cpp_extension`

  #109888 opened Sep 22, 2023

• DALLE2_pytorch: "torch._dynamo.exc.Unsupported: call_method NNModuleVariable() eval [] {}"

  #109885 opened Sep 22, 2023

• basic_gnn_gcn: ERROR:common:TypeError: object of type 'GreaterThan' has no len()

  #109884 opened Sep 22, 2023

• [FSDP ]How to convert sharded_state_dict files into full_state_dict offline without distributed process

  #109880 opened Sep 22, 2023

• DISABLED test_compile_selective_checkpoint_gemm_only (__main__.ActivationCheckpointingViaTagsTests)

  #109879 opened Sep 22, 2023

• [Profiler] RuntimeError: (((PyObject*)(arg))->ob_type) == &PyCFunction_Type INTERNAL ASSERT FAILED at "/pytorch/torch/csrc/autograd/profiler_python.cpp":807

  #109877 opened Sep 22, 2023

• [inductor][cpu] performance regression

  #109874 opened Sep 22, 2023

• Allow try except check for numpy bfloat16 representation

  #109873 opened Sep 22, 2023

• DISABLED test_tags_dropout (__main__.ActivationCheckpointingViaTagsTests)

  #109872 opened Sep 22, 2023

• masked_select for meta backend

  #109871 opened Sep 22, 2023

• Wrongly returns nan for vectorized complex numbers division on PPC/ZArch

  #109870 opened Sep 22, 2023

• If there functions（torch.autograd.Function） in the network can they be exported via jitscript?

  #109868 opened Sep 22, 2023

• DISABLED test_super1_dynamic_shapes (__main__.DynamicShapesNNModuleTests)

  #109865 opened Sep 22, 2023

• DISABLED test_tags_decomps (__main__.ActivationCheckpointingViaTagsTests)

  #109864 opened Sep 22, 2023

• [BUG?] Why Allocator use stream to manage Block?

  #109863 opened Sep 22, 2023

• DISABLED test_symints_location (__main__.ActivationCheckpointingViaTagsTests)

  #109862 opened Sep 22, 2023

• Cannot use constrain_as_size from fake tensor implementations: RuntimeError: tried to get Int out of SymInt

  #109861 opened Sep 22, 2023

• Severe performance regression on deterministic algorithm in torch 2.0

  #109856 opened Sep 22, 2023

• Directly support assert on Scalar, instead of forcing Tensor

  #109854 opened Sep 22, 2023

• add_runtime_assertions_for_constraints_pass adds redundant asserts

  #109852 opened Sep 22, 2023

• torch._export has no logging

  #109850 opened Sep 22, 2023

• [dynamo][stream] Stream runtime operation in FX graph is ignored by remaining compiler

  #109848 opened Sep 22, 2023

• DISABLED test_sequential_with_duplicated_module2_dynamic_shapes (__main__.DynamicShapesNNModuleTests)

  #109846 opened Sep 22, 2023

• Implement Copy-on-write (COW) tensors

  #109833 opened Sep 21, 2023

• DISABLED test_kwargs (__main__.ActivationCheckpointingViaTagsTests)

  #109829 opened Sep 21, 2023

• PIN disabled tests for the release

  #109827 opened Sep 21, 2023

• LTC computes wrong shape for nn.AdapativeAvgPool3d

  #109821 opened Sep 21, 2023

• ValueError: only one element tensors can be converted to Python scalars

  #109819 opened Sep 21, 2023

• DISABLED test_named_children_dynamic_shapes (__main__.DynamicShapesNNModuleTests)

  #109814 opened Sep 21, 2023

• DISABLED test_submodules1_dynamic_shapes (__main__.DynamicShapesNNModuleTests)

  #109813 opened Sep 21, 2023

• Incompatible dimensions error for FusedMatMul

  #109806 opened Sep 21, 2023

• Custom autograd function that returns its input doesn't propagate tensor subclass

  #109805 opened Sep 21, 2023

• Bits types cannot be used under deterministic mode

  #109802 opened Sep 21, 2023

• Heap-buffer-overflow during tensor unpickling

  #109791 opened Sep 21, 2023

• DISABLED test_sequential_with_duplicated_module_dynamic_shapes (__main__.DynamicShapesNNModuleTests)

  #109783 opened Sep 21, 2023

• test/test_static_runtime.py: test_fork_wait_4 sometimes deadlocks

  #109782 opened Sep 21, 2023

• `torch.embedding`, `weight[indices]`, `torch.index_select` returns random data with indices on meta device

  #109781 opened Sep 21, 2023

• DISABLED test_modulemethod2_dynamic_shapes (__main__.DynamicShapesNNModuleTests)

  #109779 opened Sep 21, 2023

• an issue occurs while `loss.backward()`: You are trying to call the hook of a dead module

  #109778 opened Sep 21, 2023

• Wrong vector shift results on PowerPC

  #109777 opened Sep 21, 2023

• [DDP + Dynamo] Tracing DDP AllReduce

  #109774 opened Sep 21, 2023

• DISABLED test_seq_dynamic_shapes (__main__.DynamicShapesNNModuleTests)

  #109773 opened Sep 21, 2023

• DISABLED test_module_comparison_dynamic_shapes (__main__.DynamicShapesNNModuleTests)

  #109772 opened Sep 21, 2023

• Slow performance when running torch.jit traced model with Flash Attention using libtorch on Windows

  #109770 opened Sep 21, 2023

• LLaMA-2 70b model convert from PyTorch to ONNX format

  #109768 opened Sep 21, 2023

• DISABLED test_modulemethod1_dynamic_shapes (__main__.DynamicShapesNNModuleTests)

  #109766 opened Sep 21, 2023

• DTensor: summon full tensor API?

  #109762 opened Sep 21, 2023

• DISABLED test_module_class_method_dynamic_shapes (__main__.DynamicShapesNNModuleTests)

  #109761 opened Sep 21, 2023

• DISABLED test_parameters3_dynamic_shapes (__main__.DynamicShapesNNModuleTests)

  #109760 opened Sep 21, 2023

• fp16 parity issue with traced code on GPU

  #109753 opened Sep 20, 2023

• [RFC][TorchElastic] topology info in training apps/ranks

  #109747 opened Sep 20, 2023

• DISABLED test_modulelist_dynamic_shapes (__main__.DynamicShapesNNModuleTests)

  #109741 opened Sep 20, 2023

• DISABLED test_moduledict_dynamic_shapes (__main__.DynamicShapesNNModuleTests)

  #109733 opened Sep 20, 2023

• DISABLED test_module_static_method_dynamic_shapes (__main__.DynamicShapesNNModuleTests)

  #109732 opened Sep 20, 2023

• DISABLED test_layerlist_dynamic_shapes (__main__.DynamicShapesNNModuleTests)

  #109731 opened Sep 20, 2023

• Profiler should implicitly synchronize gpu devices

  #109725 opened Sep 20, 2023

• assert_is_valid_input_type is too weak

  #109724 opened Sep 20, 2023

• Make torch.cuda.graphs.is_current_stream_capturing() available in TorchScript

  #109719 opened Sep 20, 2023

• Regression on 2.1 RC RoCm: data parallel error on `torch._C._broadcast_coalesced`

  #109709 opened Sep 20, 2023

• Make standard container classes satisfy container Protocols.

  #109706 opened Sep 20, 2023

• DISABLED test_module_name_string_dynamic_shapes (__main__.DynamicShapesNNModuleTests)

  #109705 opened Sep 20, 2023

• DISABLED test_istraining2_dynamic_shapes (__main__.DynamicShapesNNModuleTests)

  #109704 opened Sep 20, 2023

• [inductor][cpu] [cppwrapper] performance regression

  #109703 opened Sep 20, 2023

• [inductor][cpu] [dynamic shapes][cppwrapper] performance regression

  #109701 opened Sep 20, 2023

• [inductor][cpu] performance regression

  #109700 opened Sep 20, 2023

• [TorchScript] Support ScriptFunction arguments in torch.jit.script calls.

  #109699 opened Sep 20, 2023

• [DDP + Dynamo] Traceable DDP hooks

  #109697 opened Sep 20, 2023

• Standardize `UserWarnings` for `torch.cov`, `torch.{std,var}` and `torch.{std,var}_mean`

  #109696 opened Sep 20, 2023

• DISABLED test_vmapvjp_NumpyExpMarkDirtyAutogradFunction_cpu_float32 (__main__.TestOperatorsCPU)

  #109694 opened Sep 20, 2023

• DISABLED test_isnonelayer_dynamic_shapes (__main__.DynamicShapesNNModuleTests)

  #109693 opened Sep 20, 2023

• DISABLED test_istraining1_dynamic_shapes (__main__.DynamicShapesNNModuleTests)

  #109692 opened Sep 20, 2023

• Extends the functionality of `nn.BatchNorm1d`.

  #109691 opened Sep 20, 2023

• [RFC]: Moving most torch.compile backends out of core by 12/1/23

  #109687 opened Sep 20, 2023

• DISABLED test_sdpa_rewriter_12_cpu (__main__.SDPAPatternRewriterCpuTests)

  #109682 opened Sep 20, 2023

• [FSDP] UnpicklingError when calling save_state_dict in distributed run

  #109675 opened Sep 20, 2023

• DISABLED test_intarg_dynamic_shapes (__main__.DynamicShapesNNModuleTests)

  #109672 opened Sep 20, 2023

• DISABLED test_iseval2_dynamic_shapes (__main__.DynamicShapesNNModuleTests)

  #109670 opened Sep 20, 2023

• DISABLED test_iseval1_dynamic_shapes (__main__.DynamicShapesNNModuleTests)

  #109671 opened Sep 20, 2023

• FSDP: ShardedStateDict support for world_size = 1

  #109666 opened Sep 20, 2023

• Inductor lowering error for aten fallbacks with multiple outputs

  #109658 opened Sep 19, 2023

• [inductor] AssertionError at torch/_inductor/codegen/wrapper.py `assert self.first_time`

  #109655 opened Sep 19, 2023

• InstanceNorm does not catch dim mismatch

  #109652 opened Sep 19, 2023

• DISABLED test_forward_directly_dynamic_shapes (__main__.DynamicShapesNNModuleTests)

  #109629 opened Sep 19, 2023

• DISABLED test_fnmembercmp2_dynamic_shapes (__main__.DynamicShapesNNModuleTests)

  #109628 opened Sep 19, 2023

• Layers being skipped during static quantization

  #109612 opened Sep 19, 2023

• AsyncCompile loses useful exception backtrace in __get_result

  #109610 opened Sep 19, 2023

• "RuntimeError: (*bias): last dimension must be contiguous" with F.scaled_dot_product_attention + torch.compile

  #109607 opened Sep 19, 2023

• [dynamo] We are not generating code for x**2 when x is an ndarray

  #109605 opened Sep 19, 2023

• Different graphbreaks may make that NumPy code is traced or not

  #109604 opened Sep 19, 2023

• ConvTranspose2d returns None with FakeTensor

  #109596 opened Sep 19, 2023

• DISABLED test_fnmembercmp1_dynamic_shapes (__main__.DynamicShapesNNModuleTests)

  #109595 opened Sep 19, 2023

• DISABLED test_fnmember_dynamic_shapes (__main__.DynamicShapesNNModuleTests)

  #109594 opened Sep 19, 2023

• test_memory_timeline fails on PPC due to extra temopraries

  #109592 opened Sep 19, 2023

• DISABLED test_cfgmod_dynamic_shapes (__main__.DynamicShapesNNModuleTests)

  #109589 opened Sep 19, 2023

• Max pool with negative integer inputs and channels_last memory layout gives the wrong values

  #109586 opened Sep 19, 2023

• [Torch-Onnx] Exporting the operator 'quantized::conv_transpose2d' to ONNX opset version 13 is not supported.

  #109585 opened Sep 19, 2023

• [dynamo][jagged tensor] Slow compilation time for a helper function of jagged tensor

  #109583 opened Sep 19, 2023

• Make Dropout take a dim=... argument

  #109582 opened Sep 19, 2023

• torch.optim.Adafactor

  #109581 opened Sep 19, 2023

• [Android: React Native] couldn't find DSO to load: libtorch-code-gen.so when loading model

  #109579 opened Sep 19, 2023

• ONNX Export error

  #109577 opened Sep 19, 2023

• DISABLED test_basicmodule2_dynamic_shapes (__main__.DynamicShapesNNModuleTests)

  #109576 opened Sep 19, 2023

• DISABLED test_constloop_dynamic_shapes (__main__.DynamicShapesNNModuleTests)

  #109575 opened Sep 19, 2023

• [fake/meta] Bad meta kernel for conv1d

  #109552 opened Sep 18, 2023

• DISABLED test_basicmodule1_dynamic_shapes (__main__.DynamicShapesNNModuleTests)

  #109547 opened Sep 18, 2023

• DISABLED test_children_dynamic_shapes (__main__.DynamicShapesNNModuleTests)

  #109546 opened Sep 18, 2023

• Torch FX SubgraphMatcher Any / Oneof Patterns

  #109539 opened Sep 18, 2023

• [dynamo] torch._dynamo.exc.Unsupported: call_function BuiltinVariable(float) [TensorVariable()] {}

  #109538 opened Sep 18, 2023

• attn_output_weights sometimes rerurn `None`

  #109528 opened Sep 18, 2023

• `TORCH_DISTRIBUTED_DEBUG=DETAIL` raises a RuntimeError on `_start_coalescing()`

  #109520 opened Sep 18, 2023

• _assert_bound_is_rational can fail

  #109514 opened Sep 18, 2023

• DISABLED test_rewrite_assert_noop_dynamic_shapes (__main__.DynamicShapesReproTests)

  #109511 opened Sep 18, 2023

• DISABLED test_maybe_multiply_symint_dynamic_shapes (__main__.DynamicShapesReproTests)

  #109512 opened Sep 18, 2023

• [dynamo] torch._dynamo.exc.Unsupported: torch.* op returned non-Tensor dtype call_function <built-in method promote_types of type object at 0x7f7d9c202160>

  #109508 opened Sep 18, 2023

• [dynamo] torch._dynamo.exc.Unsupported: call_function BuiltinVariable(setattr) [TensorVariable(), ConstantVariable(str), ConstantVariable(bool)] {}

  #109505 opened Sep 18, 2023

• [dynamo] torch._dynamo.exc.Unsupported: comparison SymNodeVariable() <built-in function is_> ListVariable()

  #109504 opened Sep 18, 2023

• [dynamo] torch._dynamo.exc.Unsupported: hasattr: ListVariable()

  #109502 opened Sep 18, 2023

• Very big differences in output of `torch.lobpcg` (values and run-time) compared to SciPy on a very ill-conditioned Laplacian matrix

  #109497 opened Sep 18, 2023

• DISABLED test_kwargs_out_list_variable_dynamic_shapes (__main__.DynamicShapesReproTests)

  #109495 opened Sep 18, 2023

• Performance degradation on AMD + A800 when computation is small

  #109494 opened Sep 18, 2023

• Investigate Strictness of torch.compile `is_big_gpu`

  #109489 opened Sep 18, 2023

• [bug] FALLBACK path has been taken inside: runCudaFusionGroup

  #109488 opened Sep 18, 2023

• [dynamo][symbolic shapes] Long compilation time for KJT helper function

  #109484 opened Sep 18, 2023

• DISABLED test_get_parameter_dtype_dynamic_shapes (__main__.DynamicShapesReproTests)

  #109479 opened Sep 18, 2023

• ProcessGroup is not automatically destroyed when the process exits

  #109478 opened Sep 18, 2023

• [DTensor] optimizer step performance is still too bad

  #109477 opened Sep 18, 2023

• DISABLED test_large_reduction_list (__main__.MiscTests)

  #109474 opened Sep 18, 2023

• DISABLED test_large_reduction_list_dynamic_shapes (__main__.DynamicShapesMiscTests)

  #109473 opened Sep 18, 2023

• Inconsistent behavior for in-place operations on coalesced sparse tensors

  #109462 opened Sep 17, 2023

317 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.

• [quant][pt2e] Enable constant folding for quantize ops

  #109343 commented on Sep 22, 2023 • 53 new comments

• [dynamo][guard refactor] C++ Guard data structure

  #108839 commented on Sep 23, 2023 • 33 new comments

• [Dynamo]`__torch_function__` tracing support

  #105167 commented on Sep 20, 2023 • 27 new comments

• [C10] PG observability hooks.

  #108815 commented on Sep 21, 2023 • 16 new comments

• Move negative index checking to common.py - Fix issue 97365

  #108690 commented on Sep 21, 2023 • 15 new comments

• [v.2.1.0] Release Tracker

  #108055 commented on Sep 21, 2023 • 14 new comments

• TorchInductor Opinfo fixes for rng ops

  #108170 commented on Sep 22, 2023 • 14 new comments

• [FSDP][optim_state_dict] Fuse allgather for optim_state_dict when use_orig_params is True

  #108298 commented on Sep 22, 2023 • 14 new comments

• Minor fixes in semi-structured sparse code

  #105595 commented on Sep 23, 2023 • 13 new comments

• Replace node.meta source_fn with source_fn_stack

  #108595 commented on Sep 21, 2023 • 13 new comments

• Python 3.10 Union operator | support for JIT

  #109293 commented on Sep 24, 2023 • 13 new comments

• Add PR number to metrics when available

  #109406 commented on Sep 22, 2023 • 13 new comments

• Add .item() and .tolist() support in Dynamo/Inductor without graph break

  #109262 commented on Sep 24, 2023 • 12 new comments

• Cannot install torchmetrics - ERROR 403

  #109014 commented on Sep 18, 2023 • 10 new comments

• TorchInductor CPU Performance Dashboard

  #93531 commented on Sep 22, 2023 • 10 new comments

• General MPS op coverage tracking issue

  #77764 commented on Sep 22, 2023 • 10 new comments

• AOTAutograd should put keep mutations in the graph during training

  #109240 commented on Sep 24, 2023 • 9 new comments

• [DONOTMERGE][ROCm]Test MI210 CI Nodes

  #105980 commented on Sep 21, 2023 • 9 new comments

• [foreach] Fix 0-size handling for real for real

  #109402 commented on Sep 22, 2023 • 9 new comments

• [Inductor] Extend Pattern Matcher to Match Equivalent Function Invocation

  #107832 commented on Sep 23, 2023 • 8 new comments

• Add meta kernel for ctc_loss.intList

  #107949 commented on Sep 21, 2023 • 8 new comments

• [Decomposition] unbind

  #108546 commented on Sep 20, 2023 • 8 new comments

• Release GIL in torch.cuda ops wherever possible.

  #109159 commented on Sep 24, 2023 • 8 new comments

• DISABLED test_conversions_all_patterns_backend_cutlass_cuda_bfloat16 (__main__.TestSparseSemiStructuredCUDA)

  #106793 commented on Sep 22, 2023 • 7 new comments

• DISABLED test_conversions_all_patterns_backend_cutlass_cuda_float16 (__main__.TestSparseSemiStructuredCUDA)

  #106796 commented on Sep 22, 2023 • 7 new comments

• Add support for float8_e4m3fnuz and _e5m2fnuz

  #107586 commented on Sep 21, 2023 • 7 new comments

• [inductor] benchmark fusion

  #108193 commented on Sep 22, 2023 • 7 new comments

• use reduced_precision_reduction flags in Triton matmul

  #108637 commented on Sep 20, 2023 • 7 new comments

• Allow reductions to write into pinned memory

  #109052 commented on Sep 20, 2023 • 6 new comments

• model.forward() get error with torch.compile() when using huggingface llama

  #107269 commented on Sep 22, 2023 • 6 new comments

• add fp16 support for native conv and deconv on CPU

  #99497 commented on Sep 24, 2023 • 6 new comments

• [Decomposition] split.Tensor

  #108543 commented on Sep 18, 2023 • 6 new comments

• [C10d] Cleanup collective sequence number.

  #109136 commented on Sep 22, 2023 • 6 new comments

• [feature request] Provide some sparse eigen solver(s) for PyTorch (maybe via `ARPACK` as in scipy) + SPD sparse / laplace linear system solver

  #108977 commented on Sep 18, 2023 • 5 new comments

• Support 0-sized batches in SyncBatchNorm cuda ops

  #97177 commented on Sep 19, 2023 • 5 new comments

• [MPS] Lerp tensor implementation

  #105547 commented on Sep 22, 2023 • 5 new comments

• [custom op] Use canonical API to constrain unbacked values

  #108372 commented on Sep 22, 2023 • 5 new comments

• Evict_first change and some other small stuff

  #108841 commented on Sep 23, 2023 • 5 new comments

• Fix the max pool kernel with channels_last memory layout

  #108852 commented on Sep 19, 2023 • 5 new comments

• ln + amax + fp8 quant inductor enablement

  #109301 commented on Sep 22, 2023 • 5 new comments

• [inductor] Fix triton compiler error in multilayer any

  #109325 commented on Sep 18, 2023 • 5 new comments

• [export] Support tracing constant attribute mutations

  #109412 commented on Sep 19, 2023 • 5 new comments

• Backward pass of inverse FFT is sometimes incorrect on GPU

  #109448 commented on Sep 19, 2023 • 4 new comments

• Improve IDE autocomplete for torch.Tensor methods

  #109438 commented on Sep 20, 2023 • 4 new comments

• Gradients across different ranks are not synchronized when using DDP

  #109191 commented on Sep 22, 2023 • 4 new comments

• Exporting the operator 'aten::_convolution_mode' to ONNX opset version 14 is not supported.

  #108975 commented on Sep 23, 2023 • 4 new comments

• [NCCL][CUDA][CUDA Graphs] Flush enqueued work before starting a graph capture

  #104487 commented on Sep 21, 2023 • 4 new comments

• Correcting error message for invalid output_size input in nn.AdaptiveAvgPool2d

  #104731 commented on Sep 20, 2023 • 4 new comments

• Reland "Make adding buffers more like adding parameters (#104069)" (take #2)

  #106981 commented on Sep 20, 2023 • 4 new comments

• [inductor] Fix bug in input mutation

  #107614 commented on Sep 18, 2023 • 4 new comments

• Generalize toAccumulateType()

  #108248 commented on Sep 20, 2023 • 4 new comments

• [dynamo][stream]support device-agnostic stream in dynamo and capture stream method in fx graph

  #108312 commented on Sep 23, 2023 • 4 new comments

• [Decomposition] bernoulli

  #108535 commented on Sep 17, 2023 • 4 new comments

• Make NestedTensor compilable with eager backend

  #109171 commented on Sep 21, 2023 • 4 new comments

• [dynamo]Scuba log some debug info about list of integer inputs

  #109452 commented on Sep 18, 2023 • 4 new comments

• Make Fx Generating Incorrect Graph For GPTQ model

  #109386 commented on Sep 18, 2023 • 3 new comments

• Encountering linker error starting nightly release 2023-07-20

  #109405 commented on Sep 19, 2023 • 3 new comments

• Inplace and out arguments for BatchNorm (and other norm layers: InstanceNorm / LayerNorm / GroupNorm ...)

  #26288 commented on Sep 20, 2023 • 3 new comments

• RNN Documentation is Confusing / Wrong

  #109443 commented on Sep 20, 2023 • 3 new comments

• torch.argmax fails for device='mps:0'

  #109067 commented on Sep 21, 2023 • 3 new comments

• DISABLED test_graph_breaks (__main__.LoggingTests)

  #104098 commented on Sep 21, 2023 • 3 new comments

• Exporting the operator 'aten::linalg_inv' to ONNX opset version 18 is not supported.

  #107948 commented on Sep 21, 2023 • 3 new comments

• DISABLED test_make_fx_symbolic_exhaustive_special_airy_ai_cpu_float32 (__main__.TestProxyTensorOpInfoCPU)

  #107739 commented on Sep 22, 2023 • 3 new comments

• DISABLED test_make_fx_symbolic_exhaustive_special_bessel_j1_cpu_float32 (__main__.TestProxyTensorOpInfoCPU)

  #107762 commented on Sep 22, 2023 • 3 new comments

• DISABLED test_make_fx_symbolic_exhaustive_special_bessel_y0_cpu_float32 (__main__.TestProxyTensorOpInfoCPU)

  #107767 commented on Sep 22, 2023 • 3 new comments

• Python Crashes When Importing Torch With C API

  #104128 commented on Sep 24, 2023 • 3 new comments

• [triton hash update] update the pinned triton hash

  #98728 commented on Sep 18, 2023 • 3 new comments

• add a few overarching definitions; revise a few definitions

  #100906 commented on Sep 20, 2023 • 3 new comments

• Fix Clang compilation error with Lib ATen for ppc64le

  #106446 commented on Sep 21, 2023 • 3 new comments

• Enable function declaration check in Vulkan and Metal backends

  #106762 commented on Sep 21, 2023 • 3 new comments

• [WIP] Fix static lib

  #107673 commented on Sep 18, 2023 • 3 new comments

• Fix python decomps for OpOverloadPackets and add tests

  #107707 commented on Sep 20, 2023 • 3 new comments

• Add CUTLASS-based support for mixed dtypes matrix multiplication

  #107782 commented on Sep 23, 2023 • 3 new comments

• Intra-graph reordering pass on Inductor scheduler IR (based on #100762)

  #108091 commented on Sep 22, 2023 • 3 new comments

• [Decomposition] full_like

  #108537 commented on Sep 18, 2023 • 3 new comments

• [Decomposition] randint

  #108538 commented on Sep 17, 2023 • 3 new comments

• [RFC] Allow "spawn" start method for torchinductor workers.

  #108850 commented on Sep 23, 2023 • 3 new comments

• Apply clang-tidy to c10/test/*cpp

  #109278 commented on Sep 19, 2023 • 3 new comments

• FSDP do not support `ignored_parameters` when `auto_wrap_policy` is specified

  #109100 commented on Sep 18, 2023 • 2 new comments

• DISABLED test_inplace_grad_fmod_cuda_float64 (__main__.TestBwdGradientsCUDA)

  #95148 commented on Sep 19, 2023 • 2 new comments

• CPU and MPS floating point math is different (in a significant way)

  #84936 commented on Sep 19, 2023 • 2 new comments

• [inductor][cpu][dynamic shapes] performance regression

  #108754 commented on Sep 19, 2023 • 2 new comments

• Adding a method called `T` in native_functions causes undefined behavior on Windows

  #21176 commented on Sep 19, 2023 • 2 new comments

• TransformerEncoder truncates output when some token positions are masked by `src_key_padding_mask` across batch

  #97111 commented on Sep 19, 2023 • 2 new comments

• DISABLED test_conversions_all_patterns_backend_cutlass_cuda_int8 (__main__.TestSparseSemiStructuredCUDA)

  #106795 commented on Sep 19, 2023 • 2 new comments

• torch.onnx.export does not support divisor_override in AvgPool2d

  #105349 commented on Sep 19, 2023 • 2 new comments

• NestedTensor cannot be pickled

  #104198 commented on Sep 20, 2023 • 2 new comments

• DISABLED test_make_fx_symbolic_exhaustive_special_bessel_y1_cpu_float32 (__main__.TestProxyTensorOpInfoCPU)

  #107114 commented on Sep 21, 2023 • 2 new comments

• DISABLED test_make_fx_symbolic_exhaustive_special_bessel_j0_cpu_float32 (__main__.TestProxyTensorOpInfoCPU)

  #102775 commented on Sep 21, 2023 • 2 new comments

• Support CUDA 12.2

  #104417 commented on Sep 21, 2023 • 2 new comments

• DISABLED test_backend_match_guard_multi_threads (__main__.MiscTests)

  #109321 commented on Sep 21, 2023 • 2 new comments

• IPEX as TorchDynamo Backend Performance Dashboard

  #101273 commented on Sep 22, 2023 • 2 new comments

• DISABLED test_encoder_padding_and_src_mask_bool_cpu (__main__.TestTransformersCPU)

  #101785 commented on Sep 22, 2023 • 2 new comments

• Inconsistent Behavior of `ConvTranspose2d` on CPU and CUDA

  #109453 commented on Sep 22, 2023 • 2 new comments

• DISABLED test_compute_local_shape_and_global_offset_1D (__main__.UtilTest)

  #109379 commented on Sep 22, 2023 • 2 new comments

• cuda streams run sequentially, expected to run parallel

  #59692 commented on Sep 22, 2023 • 2 new comments

• Inconsistent, platform-dependent torch.ones_like behavior on metatensors

  #108870 commented on Sep 23, 2023 • 2 new comments

• Memory usage steadily increasing when using back propagation with sparse CSR parameter matrices on CPU

  #109445 commented on Sep 23, 2023 • 2 new comments

• [ndk] Clean up LLVM and libc++ 12 and 13

  #107326 commented on Sep 22, 2023 • 2 new comments

• quantized module serialization through prepack function registration

  #108400 commented on Sep 21, 2023 • 2 new comments

• [Decomposition] randn_like

  #108541 commented on Sep 17, 2023 • 2 new comments

• [Quantization] Add "quantization_tag" as metadata to fx proxy

  #108764 commented on Sep 20, 2023 • 2 new comments

• Add function to port FX minified graph to HLO via StableHLO

  #109084 commented on Sep 22, 2023 • 2 new comments

• Re-land: Break graph on `manual_seed`.

  #109109 commented on Sep 23, 2023 • 2 new comments

• Register decomposition for empty.memory_format into empty_strided

  #109197 commented on Sep 24, 2023 • 2 new comments

• Add a multiprocess CI job to torchbench dynamo runner

  #109311 commented on Sep 21, 2023 • 2 new comments

• Use weakref in fast tracebacks

  #109422 commented on Sep 24, 2023 • 2 new comments

• [Dynamo] Match closures by code ID

  #109427 commented on Sep 24, 2023 • 2 new comments

• Python 3.10 Union operator `|` is not supported in jit script

  #101777 commented on Sep 17, 2023 • 1 new comment

• TCPStore() RuntimeError: unmatched '}' in format string

  #109374 commented on Sep 18, 2023 • 1 new comment

• SSL Certificate Error when downloading pretrained weights

  #109080 commented on Sep 18, 2023 • 1 new comment

• [onnx]Unsupported: ONNX export of convolution for kernel of unknown shape

  #98497 commented on Sep 18, 2023 • 1 new comment

• Libtorch consumes too much memory as 16225

  #99225 commented on Sep 18, 2023 • 1 new comment

• [Poll] Support higher-order gradient computation in DDP

  #63929 commented on Sep 18, 2023 • 1 new comment

• MultiheadAttention set(-inf) cause 'Nan' in loss computing

  #40932 commented on Sep 18, 2023 • 1 new comment

• Inference ran on new thread leak memory on Android

  #59690 commented on Sep 18, 2023 • 1 new comment

• [Dynamo] Avoid id test on closures

  #107866 commented on Sep 18, 2023 • 1 new comment

• Fuse softmax and masking in MultiheadAttention

  #44945 commented on Sep 18, 2023 • 1 new comment

• nn.MultiheadAttention causes gradients to become NaN under some use cases

  #41508 commented on Sep 18, 2023 • 1 new comment

• cuda rng state for 2.0.1 cannot be used for 2.1.0

  #109383 commented on Sep 18, 2023 • 1 new comment

• Cannot access data pointer of Tensor that doesn't have storage when using `torch.func.jvp` with `torch.compile`

  #96041 commented on Sep 18, 2023 • 1 new comment

• Allow `low` and `high` to be tensors in `torch.randint`

  #89438 commented on Sep 18, 2023 • 1 new comment

• Enhanced torch.chunk and torch.split

  #60531 commented on Sep 18, 2023 • 1 new comment

• gfx906 ROCM print black images all ai torch: 2.0.1+rocm5.4.2/rocm5.5 only works with torch=1.13.0+rocm5.2

  #103973 commented on Sep 18, 2023 • 1 new comment

• torch.jit.script does not custom state_dicts

  #74112 commented on Sep 18, 2023 • 1 new comment

• Using DDP with num_workers > 0 hangs before entering the first training epoch loop

  #82077 commented on Sep 18, 2023 • 1 new comment

• FSDP: enhanced shared parameter support

  #77724 commented on Sep 18, 2023 • 1 new comment

• Can't construct a tensor from List[SymFloat]

  #107188 commented on Sep 18, 2023 • 1 new comment

• Support `cond` branches that reference variables defined in an outer scope

  #99401 commented on Sep 18, 2023 • 1 new comment

• opacus_cifar10 fails in dynamo due to hooks

  #99201 commented on Sep 18, 2023 • 1 new comment

• TORCH_COMPILE_ABLATE envvar

  #99155 commented on Sep 18, 2023 • 1 new comment

• DISABLED test_redistribute_sub_mesh (__main__.DTensorMeshTest)

  #103345 commented on Sep 18, 2023 • 1 new comment

• [dynamo] hf_Reformer's graph break has increased

  #98087 commented on Sep 18, 2023 • 1 new comment

• Pytorch 2.0.1 pypi wheel does not install dependent cuda libraries

  #100974 commented on Sep 18, 2023 • 1 new comment

• Training results from using MPS backend are poor compared to CPU and CUDA

  #109457 commented on Sep 19, 2023 • 1 new comment

• torch.onnx.errors.OnnxExporterError: Unsupported: ONNX export of operator unsafe_chunk, unknown dimension size.

  #97861 commented on Sep 19, 2023 • 1 new comment

• `ignore_index` isn't used for MPS backend in CrossEntropyLoss / F.cross_entropy

  #108148 commented on Sep 19, 2023 • 1 new comment

• torch.fx.symbolic_trace fails on torch.arange with input-dependent size

  #51803 commented on Sep 19, 2023 • 1 new comment

• uninformative OOM error

  #108645 commented on Sep 19, 2023 • 1 new comment

• NO building and NO module

  #19757 commented on Sep 19, 2023 • 1 new comment

• isTuple()INTERNAL ASSERT FAILED at \"C:\\\\w\\\\b\\\\windows\\\\pytorch\\\\aten\\\\src\\\\ATen/core/ivalue_inl.h\":1400, please report a bug to PyTorch. Expected Tuple but got String

  #92560 commented on Sep 19, 2023 • 1 new comment

• aten::_linalg_solve_ex.result' is not currently implemented for the MPS

  #98222 commented on Sep 19, 2023 • 1 new comment

• CPU memory cannot get released after `torch.compile` (caused by importing `AsyncCompile`)

  #109442 commented on Sep 19, 2023 • 1 new comment

• PyTorch 2.1 smoke test requirements

  #109289 commented on Sep 19, 2023 • 1 new comment

• C++ functional::smooth_l1_loss overrides beta in options with default

  #70163 commented on Sep 19, 2023 • 1 new comment

• Passing dict in datapipe/dataset will have memory leak problem

  #103581 commented on Sep 20, 2023 • 1 new comment

• Training got stuck due to timeout from dataloader

  #33296 commented on Sep 20, 2023 • 1 new comment

• `torch.jit.is_scripting()` not set when scripting a Module

  #67146 commented on Sep 20, 2023 • 1 new comment

• Implement missing torch.nan* operators

  #61474 commented on Sep 20, 2023 • 1 new comment

• Unnecessary cuda synchronizations that we should remove in PyTorch

  #108968 commented on Sep 21, 2023 • 1 new comment

• [dtensor] Add debug tool to visualize sharding

  #108746 commented on Sep 21, 2023 • 1 new comment

• caffe does not respect CUDNN_LIB_DIR when building from source (cmake)

  #107389 commented on Sep 21, 2023 • 1 new comment

• Enable AMP for MPS devices

  #88415 commented on Sep 21, 2023 • 1 new comment

• Multiple dimensions support for `torch.max`

  #105053 commented on Sep 21, 2023 • 1 new comment

• torch.compile graph breaks should be independent of DDP buckets

  #108966 commented on Sep 21, 2023 • 1 new comment

• rocm support for windows

  #106161 commented on Sep 21, 2023 • 1 new comment

• Undefined Symobl: pybind11::detail::type_caster<at::Tensor, void>::load(pybind11::handle, bool)

  #108041 commented on Sep 21, 2023 • 1 new comment

• [PT2] Some errors with `cond` and `torch.compile`

  #98844 commented on Sep 21, 2023 • 1 new comment

• [CUDA][CUDA 12] CUDA 12 Support Tracking Issue

  #91122 commented on Sep 21, 2023 • 1 new comment

• [FX] Ability to wrap functions in other modules for symbolic tracing

  #53534 commented on Sep 21, 2023 • 1 new comment

• [discussion] Consolidation of audio-visual I/O in a new package

  #81102 commented on Sep 21, 2023 • 1 new comment

• torch pollutes libgomp symbols when import _C

  #109446 commented on Sep 21, 2023 • 1 new comment

• DISABLED test_cond_nested_dynamic_shapes (__main__.DynamicShapesMiscTests)

  #101512 commented on Sep 21, 2023 • 1 new comment

• `ray` multiprocessing interference by torch import

  #106989 commented on Sep 21, 2023 • 1 new comment

• convit_base: AssertionError: Mutating module attribute rel_indices during export.

  #105530 commented on Sep 22, 2023 • 1 new comment

• Add Lambert W function as torch.special.lambertw

  #108948 commented on Sep 22, 2023 • 1 new comment

• Can dtensor flexibly modify the layout via devicemesh?

  #109437 commented on Sep 22, 2023 • 1 new comment

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

  #64947 commented on Sep 22, 2023 • 1 new comment

• Remove parameter `self` in `typeConvertIndices`

  #107946 commented on Sep 22, 2023 • 1 new comment

• Get errors after compiling and running PyTorch MINIMAL EXAMPLE for c++ Mac M1 with make

  #104502 commented on Sep 22, 2023 • 1 new comment

• Torch C++ API can't initiate from ATen c10::getCustomClassTypeMap() in ivalue.h

  #60824 commented on Sep 22, 2023 • 1 new comment

• Windres error while configuring cmake

  #99188 commented on Sep 22, 2023 • 1 new comment

• pr build failures in inductor dynamic shape test for operation tests with simple tensors. Side effect of current test framework

  #109016 commented on Sep 22, 2023 • 1 new comment

• c10::CUDAError

  #67978 commented on Sep 22, 2023 • 1 new comment

• Supporting Block_Ptrs in inductor code gen

  #109420 commented on Sep 22, 2023 • 1 new comment

• [CPP API] Add Adadelta, Adamax, ASGD, NAdam, RAdam and Rprop

  #107224 commented on Sep 22, 2023 • 1 new comment

• [Dynamo] Unable to Trace AdamW Optimizer when there is LR Scheduler

  #107076 commented on Sep 22, 2023 • 1 new comment

• CUBLAS_WORKSPACE_CONFIG can not be parsed

  #103499 commented on Sep 22, 2023 • 1 new comment

• 14k github models on PyTorch 2.0 pass rates dashboard

  #93667 commented on Sep 23, 2023 • 1 new comment

• Difference between allocated and reserved CUDA memory

  #40989 commented on Sep 23, 2023 • 1 new comment

• [MPS] Add support for aten::sgn.out for MPS backend

  #86805 commented on Sep 23, 2023 • 1 new comment

• RuntimeError: "slow_conv2d_cpu" not implemented for 'Half'

  #84348 commented on Sep 23, 2023 • 1 new comment

• [FSDP] Ignored modules on meta device seem to be initialized on CUDA device

  #108023 commented on Sep 23, 2023 • 1 new comment

• DISABLED test_complex_half_reference_testing_pow_cuda_complex32 (__main__.TestCommonCUDA)

  #108993 commented on Sep 23, 2023 • 1 new comment

• [discussion] move-semantics for tensors

  #106596 commented on Sep 23, 2023 • 1 new comment

• [numpy compat] torch.stack and torch.tensor doesn't support nested list+tensors (NumPy does support) - at least document the difference in the error message

  #71479 commented on Sep 23, 2023 • 1 new comment

• Unexpected modification to CPU affinity of Dataloader workers

  #101850 commented on Sep 23, 2023 • 1 new comment

• Recommendations for Grid Sample/Affine Grid/Displacement Fields/Optical Flow

  #24870 commented on Sep 24, 2023 • 1 new comment

• torch._dynamo.exc.Unsupported: call_function BuiltinVariable(all) [ListIteratorVariable()] {} in DynamicShapesReproTests.test_chunk_reformer_ff_dynamic_shapes

  #103620 commented on Sep 24, 2023 • 1 new comment

• [mta] Implement fused SGD

  #94791 commented on Sep 18, 2023 • 1 new comment

• Cleanup redundant CMake code

  #95680 commented on Sep 23, 2023 • 1 new comment

• change stacksize_analysis to worklist algorithm for better result

  #95767 commented on Sep 23, 2023 • 1 new comment

• [MPS] Add support for autocast in MPS

  #99272 commented on Sep 18, 2023 • 1 new comment

• (WIP; DO NOT REVIEW) Use python tensor subclass version of NT for PT2

  #99753 commented on Sep 18, 2023 • 1 new comment

• [functorch] hard error on direct use with torch.compile

  #101384 commented on Sep 18, 2023 • 1 new comment

• [BE]: pyupgrade Python to 3.8 - remove extraneous parentheses only

  #101606 commented on Sep 22, 2023 • 1 new comment

• add Half support for AdaptiveAvgPool2d and AdaptiveMaxPool2d on CPU

  #102079 commented on Sep 22, 2023 • 1 new comment

• [MPS] Add support for native binary ops

  #102103 commented on Sep 17, 2023 • 1 new comment

• [TEST] use py3.11 for benchmark image

  #102295 commented on Sep 19, 2023 • 1 new comment

• [dynamo] Update Unsupported to raise from fake tensor exceptions

  #103987 commented on Sep 22, 2023 • 1 new comment

• Enable quantization dispatch for backend QuantizedPrivateUse1

  #104241 commented on Sep 24, 2023 • 1 new comment

• Deprecated the device usage without device_type

  #104457 commented on Sep 22, 2023 • 1 new comment

• recompile fx.GraphModule lazily

  #105257 commented on Sep 20, 2023 • 1 new comment

• Expose Some FunctionsManual Api

  #105441 commented on Sep 18, 2023 • 1 new comment

• Change default autograd fallback mode to "Warn"

  #105592 commented on Sep 24, 2023 • 1 new comment

• Enable Mypy checking for scheduler.py

  #105600 commented on Sep 18, 2023 • 1 new comment

• Add z3-solver as dependency to dynamo tests

  #105640 commented on Sep 23, 2023 • 1 new comment

• [TEST ONLY] lstm ideep API change

  #105647 commented on Sep 18, 2023 • 1 new comment

• [LTC] Fix type inference for native_layer_norm_backward

  #105664 commented on Sep 22, 2023 • 1 new comment

• [pytorch] Add bernoulli.p_out

  #105704 commented on Sep 19, 2023 • 1 new comment

• Update Core Aten Doc

  #105722 commented on Sep 19, 2023 • 1 new comment

• [pytorch] replace __FILE__ with __FILE_NAME__ for exceptions

  #105725 commented on Sep 22, 2023 • 1 new comment

• [FAILING] Make guard after freeze a hard error

  #105734 commented on Sep 22, 2023 • 1 new comment

• [WIP] Fix Prims as_strided_scatter

  #105781 commented on Sep 23, 2023 • 1 new comment

• Remove TORCH_API from OpaqueTensorImpl

  #105835 commented on Sep 22, 2023 • 1 new comment

• [WIP][Experiment] Avoid real computation for dynamo export

  #105858 commented on Sep 22, 2023 • 1 new comment

• [vision hash update] update the pinned vision hash

  #105884 commented on Sep 23, 2023 • 1 new comment

• [state_dict][1/N] Implement the basic functions of distributed.checkpoint._state_dict

  #105902 commented on Sep 22, 2023 • 1 new comment

• Enable xpu backend in totchdynamo benchmarks

  #105907 commented on Sep 23, 2023 • 1 new comment

• enabling fused A16W8 mm through prologue fusion WIP

  #105926 commented on Sep 24, 2023 • 1 new comment

• [FSDPxMTPG] Migrate one more test

  #105999 commented on Sep 24, 2023 • 1 new comment

• [ComposablexMTPG] Migrate some composable tests to MTPG

  #106000 commented on Sep 24, 2023 • 1 new comment

• [FSDPExecOrder] Migrate one test to MTPG

  #106002 commented on Sep 24, 2023 • 1 new comment

• [FSDPxMTPG] Migrate TestFSDPTraversal

  #106003 commented on Sep 24, 2023 • 1 new comment

• Enable thp(transparent huge pages) for buffer sizes >=2MB

  #107697 commented on Sep 21, 2023 • 1 new comment

• [PyTorch] torch.empty_permuted: rename param name from 'physical_layout' to 'dim_order'

  #107967 commented on Sep 22, 2023 • 1 new comment

• fix: adam(w) ignore stride mismatch when dim is size 1

  #107984 commented on Sep 22, 2023 • 1 new comment

• Update to newest CUTLASS version 3.2.0

  #108070 commented on Sep 19, 2023 • 1 new comment

• [fx] Show original user stack trace on GraphModule exception if it's available

  #108101 commented on Sep 20, 2023 • 1 new comment

• [pytorch] Test key ET models export to core aten ir

  #108614 commented on Sep 20, 2023 • 1 new comment

• [Decomposition] all.dim

  #108714 commented on Sep 17, 2023 • 1 new comment

• [Decomposition] rand_like

  #108727 commented on Sep 17, 2023 • 1 new comment

• Validate that storage have enough memory allocated

  #108872 commented on Sep 21, 2023 • 1 new comment

• [dynamo] fix functools.wraps on nested functions

  #108885 commented on Sep 18, 2023 • 1 new comment

• Adding T4 GPUs to inductor nightly benchmarks

  #109345 commented on Sep 19, 2023 • 1 new comment

• [HigherOrderOp] Flatten outputs of `wrap`.

  #109433 commented on Sep 23, 2023 • 1 new comment

• Improve IDE Type Hinting for torch.Tensor class methods

  #109441 commented on Sep 20, 2023 • 1 new comment

• test_torchinductor_opinfo tracker

  #105534 commented on Sep 18, 2023 • 0 new comments

• FSDP vs. MiCS

  #109180 commented on Sep 18, 2023 • 0 new comments

• [FSDP] supports QLora finetuning

  #109440 commented on Sep 18, 2023 • 0 new comments

• Libtorch C++ model forward crashed on windows10, CUDA 11.2, Qt ,RTX 3070, but libtorch C++ works with cpu successfully

  #66853 commented on Sep 18, 2023 • 0 new comments

• Quantizable LSTMCell does not work correctly.

  #55945 commented on Sep 18, 2023 • 0 new comments

• [Quant] Framework observes weight in convert, changing numerics

  #73941 commented on Sep 18, 2023 • 0 new comments

• torch.ao.quantization.ReuseInputObserver should also reuse the dtype of the input

  #73738 commented on Sep 18, 2023 • 0 new comments

• Feature request: INT4 format support

  #74627 commented on Sep 18, 2023 • 0 new comments

• MultiHeadAttention in quantizable seems incorrect with batch_first=True

  #67651 commented on Sep 18, 2023 • 0 new comments

• fusion in fx graph mode did not take care of direct attribute access

  #68892 commented on Sep 18, 2023 • 0 new comments

• Skip LSTM quantization by default in get_default_qconfig_dict and get_default_qat_qconfig_dict

  #71028 commented on Sep 18, 2023 • 0 new comments

• torch.compile not work in WSL

  #97501 commented on Sep 18, 2023 • 0 new comments

• RuntimeError: Cannot call sizes() on tensor with symbolic sizes/strides w/ `dynamo.export`, `make_fx` and `functionalize`

  #99774 commented on Sep 18, 2023 • 0 new comments

• Changing module attributes doesn't retrigger compilation

  #98928 commented on Sep 18, 2023 • 0 new comments

• Target indices are ignored when writing to a CPU tensor from an MPS tensor

  #108978 commented on Sep 18, 2023 • 0 new comments

• Cannot export a quantized model that permutes a quantized tensor to ONNX

  #109425 commented on Sep 18, 2023 • 0 new comments

• RaisesTypeError: 'torch._C._TensorMeta' object is not iterable when import torch

  #104496 commented on Sep 19, 2023 • 0 new comments

• pytorch1.12.0 building erros on win10

  #107203 commented on Sep 19, 2023 • 0 new comments

• torch.cuda.is_available() closes the Python command prompt

  #96998 commented on Sep 19, 2023 • 0 new comments

• Support negative step sizes for slicing

  #59786 commented on Sep 19, 2023 • 0 new comments

• DISABLED test_out_randn_cuda_float32 (__main__.TestCommonCUDA)

  #109070 commented on Sep 19, 2023 • 0 new comments

• Introduce 'backend' concept to torch.export.export API

  #109131 commented on Sep 19, 2023 • 0 new comments

• Efficient and robust calculation of diag(sparse @ diag @ sparse)

  #108483 commented on Sep 20, 2023 • 0 new comments

• [BUG][pytree] treespec serialization for locally defined classes and namedtuple types

  #109460 commented on Sep 20, 2023 • 0 new comments

• [profiler] Show shapes for lists of tensors in chrome traces

  #109263 commented on Sep 20, 2023 • 0 new comments

• skipIfTorchInductor Tracking Issue

  #102207 commented on Sep 21, 2023 • 0 new comments

• [Optimizer Perf] Improve speed of _init_group to c++

  #108090 commented on Sep 21, 2023 • 0 new comments

• Add a unittest for ModuleWrapPolicy callable

  #109266 commented on Sep 22, 2023 • 0 new comments

• Cross Entropy doesn't work with the specific batch, but works with each sample from this batch

  #108345 commented on Sep 22, 2023 • 0 new comments

• SGD does not converge properly

  #109104 commented on Sep 22, 2023 • 0 new comments

• RWKV + Adam exp_avg_sq will change from positive to negative after loss.backward()

  #108493 commented on Sep 22, 2023 • 0 new comments

• Suppport Fused AdamW on CPU

  #108491 commented on Sep 22, 2023 • 0 new comments

• _foreach_copy_ with scalar second arg

  #108445 commented on Sep 22, 2023 • 0 new comments

• stride of gradient is not same as the corresponding tensor

  #106951 commented on Sep 22, 2023 • 0 new comments

• Other overloads of `_foreach_clamp`

  #106931 commented on Sep 22, 2023 • 0 new comments

• Optimizers should use learning rates passed as tensors directly

  #106802 commented on Sep 22, 2023 • 0 new comments

• Got error when train models with more than one param_group in torch2.0

  #106121 commented on Sep 22, 2023 • 0 new comments

• Export+AOTInductor issue tracker

  #105217 commented on Sep 22, 2023 • 0 new comments

• OneCycleLR's state_dict includes a full reference to the optimizer

  #106243 commented on Sep 22, 2023 • 0 new comments

• ReduceLROnPlateau will throw IndexError: list index out of range with modified optimizer's param_groups.

  #104361 commented on Sep 22, 2023 • 0 new comments

• Asynchronous CUDA AveragedModel

  #103449 commented on Sep 22, 2023 • 0 new comments

• Torch's `LayerNorm` and Adam optimizer vs those in tensorflow

  #104857 commented on Sep 22, 2023 • 0 new comments

• torch compile error with SyncBatchNorm

  #107372 commented on Sep 22, 2023 • 0 new comments

• [FSDP] Implement additional check for turn on 2D TP + FSDP extension

  #109392 commented on Sep 22, 2023 • 0 new comments

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

  #24422 commented on Sep 23, 2023 • 0 new comments

• Enable Mypy Checking in torch/_inductor

  #105230 commented on Sep 24, 2023 • 0 new comments

• PyTorch 2.0: AttributeError: __torch__.torch.classes.c10d.ProcessGroup (of Python compilation unit at: 0) does not have a field with name 'shape'

  #94599 commented on Sep 24, 2023 • 0 new comments

• add fp16 support for gemm

  #99498 commented on Sep 24, 2023 • 0 new comments

• Add pin_memory and is_pinned to NT

  #102445 commented on Sep 19, 2023 • 0 new comments

• [inductor] Memory planning

  #102625 commented on Sep 19, 2023 • 0 new comments

• Add Half support for softmax and log_softmax on CPU

  #103315 commented on Sep 22, 2023 • 0 new comments

• Insert nvtx markers into generated triton kernels

  #103644 commented on Sep 24, 2023 • 0 new comments

• [ONNX] Move out onnx bench bash scripts

  #103983 commented on Sep 23, 2023 • 0 new comments

• Add mutliple CUDA streams support to TorchInductor

  #104072 commented on Sep 20, 2023 • 0 new comments

• [test-only] Tensor load endianness default value

  #104470 commented on Sep 23, 2023 • 0 new comments

• [not ready for review yet], AOTDispatch subclass

  #104483 commented on Sep 22, 2023 • 0 new comments

• [not ready for review yet] torch.compile support for parseSemiStructuredTensor

  #104974 commented on Sep 21, 2023 • 0 new comments

• [not ready for review yet] torch.compile DTensor E2E

  #105236 commented on Sep 22, 2023 • 0 new comments

• [MPS] Add mps support for max unpool2d

  #105637 commented on Sep 19, 2023 • 0 new comments

• Automated submodule update: kineto

  #106149 commented on Sep 19, 2023 • 0 new comments

• Automated submodule update: FBGEMM

  #106168 commented on Sep 24, 2023 • 0 new comments

• [inductor] Add ir.Scan and lower aten.cumsum on CUDA

  #106581 commented on Sep 21, 2023 • 0 new comments

• [torch.optim/C++] Add Adagrad state initialization

  #106896 commented on Sep 19, 2023 • 0 new comments

• Add drop_remainder & redistribute to torch.chunk and drop_remainder for torch.split

  #107484 commented on Sep 19, 2023 • 0 new comments

• [Dynamic Shapes] Pre-autograd FX transform that replaces boolean mask updates with torch.where calls

  #107844 commented on Sep 22, 2023 • 0 new comments

• [FSDP] New rate limiter

  #108234 commented on Sep 18, 2023 • 0 new comments

• Benchmark to include slowdowns when computing gmean of speedups over eager

  #108375 commented on Sep 23, 2023 • 0 new comments

• [ONNX] dort to inline onnx model before running ort

  #108376 commented on Sep 23, 2023 • 0 new comments

• [inductor] Add CI jobs to test AOTInductor

  #108419 commented on Sep 22, 2023 • 0 new comments

• [FSDP][optim_state_dict] Enable cpu_offload config for optimzer state_dict

  #108434 commented on Sep 21, 2023 • 0 new comments

• Eliminate calls of c10::guts::conjunction,c10::guts::disjunction,c10::guts::negation,c10::guts::void_t, c10::invoke and c10::guts::apply

  #108510 commented on Sep 19, 2023 • 0 new comments

• [state_dict][2/N] Implement the flattening and unflattening of optimizer state_dict feature

  #108790 commented on Sep 19, 2023 • 0 new comments

• Allow public access for imports

  #108914 commented on Sep 19, 2023 • 0 new comments

• feat(inductor): Accumulate fp16 for bmm and mm

  #108982 commented on Sep 18, 2023 • 0 new comments

• [TEST] Release only changes

  #109210 commented on Sep 19, 2023 • 0 new comments

• metric table

  #109245 commented on Sep 19, 2023 • 0 new comments

• Support inference_mode decorator

  #109274 commented on Sep 18, 2023 • 0 new comments

• Unblock QAT multiprocessing

  #109288 commented on Sep 22, 2023 • 0 new comments

• Move eval_frame global variables into module state

  #109419 commented on Sep 20, 2023 • 0 new comments

• gh-108197 Update AdaptiveMaxPooling2d.cpp

  #109432 commented on Sep 22, 2023 • 0 new comments