1 Pull request merged by 1 person

• Add a docstring to build.sh

  #144566 merged Mar 5, 2025

201 Pull requests opened by 114 people

• Implement needs_exact_strides for mutable custom operators

  #148091 opened Feb 27, 2025

• Rename node.meta["arg_kwarg_vals"] to node.meta["eager_input_vals"]

  #148092 opened Feb 27, 2025

• [ROCm][Windows] Fix OpenMP Flags for clang-cl

  #148097 opened Feb 27, 2025

• [Experiment] meaure the effect of combining cpp-wrapper and cudagraphs

  #148100 opened Feb 27, 2025

• Inductor respects exact strides on custom ops by default

  #148104 opened Feb 27, 2025

• Use myst_nb in docs

  #148105 opened Feb 27, 2025

• stable torch library draft

  #148124 opened Feb 27, 2025

• Use correct boxed_forward_device_index when running `CompiledFxGraph.post_compile`

  #148130 opened Feb 27, 2025

• [WIP] [Inductor] Use real input to autotune user defined triton kernels

  #148131 opened Feb 27, 2025

• WIP enable aten convolution out in lowerings

  #148132 opened Feb 27, 2025

• Checks kv pair indexing in OrderedPreservingDictTest.test_range_insert

  #148136 opened Feb 27, 2025

• Disable cudnn to avoid creating guards that denies exporting

  #148140 opened Feb 28, 2025

• draft

  #148160 opened Feb 28, 2025

• [Don't merge]Upgrade submodule oneDNN to v3.7 (#147498)(Z7)

  #148163 opened Feb 28, 2025

• Improvement with comprehensive docstrings and implementation of class method for the code.

  #148170 opened Feb 28, 2025

• [Don't merge]Upgrade submodule oneDNN to v3.7 (#147498)(ZI)

  #148173 opened Feb 28, 2025

• Fix torch.matmul related out dtype check

  #148174 opened Feb 28, 2025

• Fix addbmm & addmv & baddbmm out dtype check

  #148176 opened Feb 28, 2025

• [Dynamo] Replace `unimplemented` with`unimplemented_v2` in `torch/_dynamo/variables/base.py`

  #148177 opened Feb 28, 2025

• [pytree] add another simplified pytree module `torch.pytree`

  #148180 opened Feb 28, 2025

• [BE][PYFMT] migrate PYFMT for `torch/ao/` to `ruff format`

  #148185 opened Feb 28, 2025

• [BE][PYFMT] migrate PYFMT for `test/inductor/` to `ruff format`

  #148186 opened Feb 28, 2025

• Support huggingface reading and writing for multi rank case

  #148189 opened Feb 28, 2025

• [BE][Ez]: Use itertools.chain.from_iterable when possible

  #148190 opened Feb 28, 2025

• Enable oneDNN dispatch for gemm bf16bf16->bf16

  #148197 opened Feb 28, 2025

• Move estimate runtime and pick loop order heuristics into choices.py

  #148202 opened Feb 28, 2025

• [debug] 'No available kernel' error for cudnn on A100

  #148204 opened Feb 28, 2025

• [cond] support output the same unbacked symbol from two branches

  #148206 opened Feb 28, 2025

• [inductor] support dilation in max_pool2d lowering

  #148209 opened Feb 28, 2025

• [inductor] Lowerings for max_pool3d

  #148210 opened Feb 28, 2025

• [not for merge] [AOTI] selectively build code at O1

  #148212 opened Feb 28, 2025

• Expose functions used in custom backend in torch_python dll

  #148213 opened Feb 28, 2025

• ROCm: Disable torch check for Multiplication of two Float8_e5m2 matrices

  #148228 opened Feb 28, 2025

• [cutlass backend] Expand addmm test to all 4 broadcastable shape bias

  #148234 opened Mar 1, 2025

• Make require_contiguous require exact strides instead of stride order

  #148235 opened Mar 1, 2025

• [cutlass backend] try reenable subproc add mm test

  #148236 opened Mar 1, 2025

• [ROCm][TunableOp] Add support for rowwise scaling on scaled GEMM.

  #148238 opened Mar 1, 2025

• handle jk for emulation runs

  #148240 opened Mar 1, 2025

• [fx] Move map_aggregate to C++

  #148243 opened Mar 1, 2025

• Set requires grad in TensorMaker::make_tensor()

  #148255 opened Mar 1, 2025

• [BE]: No include left behind - recursive glob setuptools support

  #148258 opened Mar 1, 2025

• [fx] Move Node._update_args_kwargs to C++

  #148260 opened Mar 1, 2025

• [fx] Move Node._prepend/Node._remove_from_list to C++

  #148261 opened Mar 1, 2025

• Typo Errors fixed in multiple files

  #148262 opened Mar 1, 2025

• Fix bug when Inductor include path contains spaces

  #148271 opened Mar 1, 2025

• [torch] Fix unsafe concurrent access to autocast_enabled

  #148281 opened Mar 2, 2025

• [fx] Optimizations for node name generation

  #148288 opened Mar 2, 2025

• [fx] Optimize TracerBase.create_arg and Graph._gen_python_code

  #148292 opened Mar 3, 2025

• Treat CUDA warnings as errors

  #148294 opened Mar 3, 2025

• ```torch.as_strided``` negative stride SIGSEV fix when using ```torch.compile```

  #148301 opened Mar 3, 2025

• Better log message to update pr_time_benchmarks/expected_results.csv

  #148303 opened Mar 3, 2025

• Reland: [inductor] Simplify grid handling

  #148305 opened Mar 3, 2025

• do not run `test_ck_blas_library` on cpu

  #148316 opened Mar 3, 2025

• [ATen][CUDA] Optimize 128 bit vectorization

  #148320 opened Mar 3, 2025

• Checking for cuda version to see if bf16 is natively supported or emulated

  #148322 opened Mar 3, 2025

• [Windows][Inductor][XPU] Unload triton pyd files to be able to remove them on Windows.

  #148323 opened Mar 3, 2025

• Flex unskip

  #148327 opened Mar 3, 2025

• [pytree] simplify public API exposition with `__module__`

  #148328 opened Mar 3, 2025

• Update CURL url for manywheel images

  #148343 opened Mar 3, 2025

• [test] cutlass

  #148351 opened Mar 3, 2025

• [DO NOT MERGE] Test new ROCm CI Navi31 nodes

  #148355 opened Mar 3, 2025

• test index_put

  #148357 opened Mar 3, 2025

• [inductor] Improve type annotations in _inductor/ir.py

  #148358 opened Mar 3, 2025

• [AOTI][dashboard] Skip torchbench models not supported by export

  #148359 opened Mar 3, 2025

• Enabling xpu in OffsetBasedRNGTracker .

  #148360 opened Mar 3, 2025

• [@no-merge] Enable process based async cp + caching

  #148373 opened Mar 3, 2025

• Documents torch.cuda.MemPool API

  #148374 opened Mar 3, 2025

• [Utilization] Add utilization monitor for linux build

  #148375 opened Mar 3, 2025

• [reland][ca] side-effect free inital trace: compiled_args

  #148376 opened Mar 3, 2025

• Throws error when using torch.cuda.MemPool with expandable segments

  #148378 opened Mar 3, 2025

• [Optimus][Auto-AC] Support activation quantization

  #148380 opened Mar 3, 2025

• [ca] remove compiled_autograd_tracing

  #148381 opened Mar 3, 2025

• [Docs][TunableOp] TunableOp documentation update

  #148384 opened Mar 4, 2025

• [dynamo] Remove dead code path around `functools.partial` objects

  #148386 opened Mar 4, 2025

• Add new GHA workflow to cache ROCm CI docker images on MI300 CI runners periodically

  #148394 opened Mar 4, 2025

• [Docs] update bucketize documentaion

  #148400 opened Mar 4, 2025

• [dynamo] show stack above dynamo in graph break user tracebacks

  #148401 opened Mar 4, 2025

• Use oneDNN v3.7.1 for Intel GPU

  #148403 opened Mar 4, 2025

• Add api info for torch._C._nn.pyi

  #148405 opened Mar 4, 2025

• ci: Move s390x builds with the rest

  #148406 opened Mar 4, 2025

•  Enable ASAN on inductor CUDA tests

  #148407 opened Mar 4, 2025

• [WIP] Add `device` arg to `_lazy_clone`

  #148408 opened Mar 4, 2025

• Add api info for torch._C._nn.pyi [1/N]

  #148410 opened Mar 4, 2025

• Disable flake8 advice C416

  #148412 opened Mar 4, 2025

• Automated perf_linter changes: generators

  #148413 opened Mar 4, 2025

• Automated perf_linter changes: list constructors

  #148414 opened Mar 4, 2025

• Automated perf_linter changes: x in (...)

  #148415 opened Mar 4, 2025

• Add perf_linter to auto-fix some anti-patterns

  #148416 opened Mar 4, 2025

• Add 'x in {...}' patterns to perf_linter

  #148417 opened Mar 4, 2025

• [2/N] Use Python 3.9 typing

  #148418 opened Mar 4, 2025

• ci: Add sccache to manylinux images

  #148419 opened Mar 4, 2025

• [set_linter] allow x in {...}

  #148422 opened Mar 4, 2025

• [Intel GPU][pt2e]: Collapse 3D input to 2D for matmul in qlinear_pointwise_binary fusion

  #148423 opened Mar 4, 2025

• Temp test

  #148424 opened Mar 4, 2025

• Optimize `torch.distributions` Score function

  #148429 opened Mar 4, 2025

• Introduce guard_or_true, guard_or_false and avoid guard_size_oblivious in decompositions.py

  #148430 opened Mar 4, 2025

• set non_blocking to true in torch._foreach_copy_ to improve performance

  #148431 opened Mar 4, 2025

• [ROCm] Add TF32 option for Flex Attention for gfx90a

  #148432 opened Mar 4, 2025

• Do not crash when compiling quantized LORA models

  #148435 opened Mar 4, 2025

• Expand docs for `nn.functional`, and make the wording consistent

  #148436 opened Mar 4, 2025

• [ROCm] Incorporate ROCm triton specific tuning parameters

  #148437 opened Mar 4, 2025

• Let `CUDAExtension` to find stub libs

  #148441 opened Mar 4, 2025

• Update s390x docker image

  #148444 opened Mar 4, 2025

• Fix test failures on non-x86 Linux

  #148445 opened Mar 4, 2025

• [inductor][triton] Block ptr analysis fix assert on matched index expression

  #148446 opened Mar 4, 2025

• Enable more nightly tests on s390x

  #148452 opened Mar 4, 2025

• Update docstring to match code.

  #148455 opened Mar 4, 2025

• [PP] RFC for fixing microbatch splitting for dim != 0

  #148458 opened Mar 4, 2025

• Remove `torch.testing` from `MOD_SKIPLIST`

  #148459 opened Mar 4, 2025

• meta registration for torch._scaled_mm with mxfp8

  #148461 opened Mar 4, 2025

• [aarch64] add libcufile for cu126 and cu128

  #148465 opened Mar 4, 2025

• [MPS] Introduce strides unary op

  #148468 opened Mar 4, 2025

• [dynamo] Properly account for non-list instances in list comparison

  #148470 opened Mar 4, 2025

• [BE][pytree] rename `NodeDef` member to match the type annotations: `*_fn -> *_func`

  #148474 opened Mar 4, 2025

• Remove warnings on non-buffer tensor constants

  #148483 opened Mar 4, 2025

• [BE][pytree] rename argument name in register function to match the type annotations: `*_fn -> *_func`

  #148484 opened Mar 4, 2025

• Demote logger of runtime_asserts_frozen to be fired only on debug mode

  #148485 opened Mar 4, 2025

• Suppress more warnings

  #148488 opened Mar 4, 2025

• [Ads] Slice convertIdScoreListFeaturesToIValue in half

  #148489 opened Mar 4, 2025

• [aot cache][ca] remove restriction on caching ca's aot inference graph

  #148491 opened Mar 4, 2025

• [triton hash update] update the pinned triton hash

  #148492 opened Mar 4, 2025

• Implement fast access to individual elements of jagged nested tensors

  #148497 opened Mar 4, 2025

• [codemod] Remove unused-variable in caffe2/caffe2/core/context_gpu.cu +2

  #148501 opened Mar 4, 2025

• [Inductor-CPU] Fix perf regression for templated int8 WoQ GEMM for small M dimension

  #148502 opened Mar 4, 2025

• [triton] Warp specialization support in torchinductor

  #148503 opened Mar 4, 2025

• Change constexpr annotation to specific initialization (test: triton_kernel_constants)

  #148505 opened Mar 4, 2025

• Support basic TorchBind in aot_compile and aoti_compile_and_package

  #148506 opened Mar 4, 2025

• WIP: record how many paramaters we're parsing

  #148508 opened Mar 4, 2025

• Enable autocast on MAIA device

  #148511 opened Mar 5, 2025

• Add sparsity

  #148513 opened Mar 5, 2025

• [ca][aot] maybe mark activations as dynamic

  #148516 opened Mar 5, 2025

• [PT2] Port use_triton_dot_compress to PT2 pre_grad passes

  #148517 opened Mar 5, 2025

• [cutlass backend] Forward fix for less aligned gemm shapes

  #148521 opened Mar 5, 2025

• [Intel GPU][pt2e] Enable quantized grouped convolution at XPU

  #148522 opened Mar 5, 2025

• Implement gradient for the `residuals` of `torch.linalg.lstsq`

  #148526 opened Mar 5, 2025

• Fix clang-tidy bugprone* warnings

  #148529 opened Mar 5, 2025

• [WIP] Initial implementation of Grouped Gemm API

  #148531 opened Mar 5, 2025

• Skip buffer in dense update

  #148533 opened Mar 5, 2025

• [dtensor] add CuDNN SDPA op support to DTensor

  #148537 opened Mar 5, 2025

• [XPU] Add test/kernel.errors.txt to .gitignore.

  #148538 opened Mar 5, 2025

• Enable Direct Use of Arm Compute Library (ACL) in ATen

  #148542 opened Mar 5, 2025

• remove TORCH_NCCL_AVOID_RECORD_STREAMS,use stashed_for_allocator_safety_ to save the input ref

  #148553 opened Mar 5, 2025

• [BE] format `test/inductor/s429861_repro.py`

  #148554 opened Mar 5, 2025

• [DEBUG] Custom ops perf

  #148555 opened Mar 5, 2025

• [BE] Remove `onlyCPU` decorator from test_local_scalar_dense

  #148559 opened Mar 5, 2025

• [ROCm][Windows] Fix ROCm/HIP version header

  #148560 opened Mar 5, 2025

• [WIP] First version of StaticCudaLauncher

  #148561 opened Mar 5, 2025

• [ROCm][Windows] Enable hipblaslt for Windows

  #148563 opened Mar 5, 2025

• [BE][pytree] cleanup parameterized pytree tests

  #148569 opened Mar 5, 2025

• [DCP] Save Plan Caching: Fix the missing all_plans update in the cache.

  #148577 opened Mar 5, 2025

• [CI][CUDA][Distributed]Update test_composability.py

  #148578 opened Mar 5, 2025

• [wip][inductor]lowering scan to while_loop

  #148580 opened Mar 5, 2025

• Enable Direct Use of Arm Compute Library (ACL) in ATen

  #148581 opened Mar 5, 2025

• Enable Direct Use of Arm Compute Library (ACL) in ATen

  #148584 opened Mar 5, 2025

• Enable fast qlinear static/dynamic path for AArch64 through ACL directly

  #148585 opened Mar 5, 2025

• [PGNCCL] Launch kernel on current stream & remove `record_stream` entirely

  #148590 opened Mar 5, 2025

• [AOTI] Swith to local cpp compile for fbcode

  #148592 opened Mar 5, 2025

• Add XPU device to nested_layer_norm

  #148593 opened Mar 5, 2025

• [CUDA Graphs][NCCL] Set event queries to happen under thread-local mode in `ProcessGroupNCCL.cpp`

  #148594 opened Mar 5, 2025

• [c10d] Make getDefaultBackend more fault tolerant

  #148596 opened Mar 5, 2025

• [pytorch] Update flexattention bwd config generation

  #148600 opened Mar 5, 2025

• Fix for AOTI + CUDAGraphs when calling from Python

  #148601 opened Mar 5, 2025

• [CI][CUDA] Move away from cuda12.4, Add cuda12.6 eager CI tests

  #148602 opened Mar 5, 2025

• [ONNX] Expose verification utilities

  #148603 opened Mar 5, 2025

• [cuda] Add new faster gammabeta backward kernel

  #148605 opened Mar 5, 2025

• Remove warnings on non-buffer tensor constants (#148483)

  #148611 opened Mar 5, 2025

• [CI] [inductor] Add cu126 inductor jobs and move away cu124

  #148612 opened Mar 5, 2025

• [For Discussion][Dynamo] Avoiding skipping module.py inner() frame, to keep forward hooks and forward in the same graph

  #148613 opened Mar 5, 2025

• Bump to AOTriton 0.9.2 to fix version strings

  #148615 opened Mar 5, 2025

• Optimize AOTInductor: Caching, Reduced Decompositions, and Improved JSON Handling

  #148616 opened Mar 5, 2025

• [dynamo] Don't affect stack traces under TORCHDYNAMO_DISABLE

  #148618 opened Mar 5, 2025

• fix 142457 , fixes double free corruption by adding TORCH_CHECK to ensure weights have the proper size

  #148620 opened Mar 5, 2025

• update get_default_device to also respect torch.device ctx manager

  #148621 opened Mar 5, 2025

• stage 2 of depreate silent fallback of tuning gemm

  #148622 opened Mar 5, 2025

• [mm_logs] follow up to add count info based on shape for inductor `aten.mm`s

  #148623 opened Mar 6, 2025

• Remove Cuda 12.4 from nightly Binaries

  #148625 opened Mar 6, 2025

• [triton 3.3] test_triton_kernel_constants fix

  #148626 opened Mar 6, 2025

• [ONNX] Use torch export to get dynamic shapes for JIT convert strategy

  #148627 opened Mar 6, 2025

• Remove CAFFE2_USE_EIGEN_FOR_BLAS

  #148628 opened Mar 6, 2025

• [inductor] lowering for fractional_max_pool3d

  #148630 opened Mar 6, 2025

• [Window][Inductor UT] Fix for tempfile.NamedTemporaryFile(delete=True) not work on Windows.

  #148632 opened Mar 6, 2025

• update torch.nn.RelicationPad{1,2,3}d deternimistic documentation

  #148633 opened Mar 6, 2025

• Subprocess compile (attempt 2)

  #148635 opened Mar 6, 2025

• [cutlass backend] switch host optimizer to O3

  #148637 opened Mar 6, 2025

• Remove cppcoreguidelines-pro-type-member-init_fix suppression

  #148638 opened Mar 6, 2025

• [Intel GPU][quant] Refine zero-point memory creation

  #148640 opened Mar 6, 2025

• Fix torch.utils.checkpoint import error

  #148641 opened Mar 6, 2025

• [XPU] Add an implict conversion from XPUStream to sycl::queue*

  #148646 opened Mar 6, 2025

• [SGD] Add SGD capturable API and tests

  #148647 opened Mar 6, 2025

• Bump Clang-tidy to 19.1.4

  #148648 opened Mar 6, 2025

• [Intel GPU] Fix SDPA dummy LSE output to match meta function

  #148652 opened Mar 6, 2025

• Enable qint8 and quint8 add for AArch64 using ACL directly

  #148653 opened Mar 6, 2025

• Enable ruff check for `torch/utils/data/typing.ipynb`

  #148654 opened Mar 6, 2025

• Allow to run flex_attention on HPU

  #148656 opened Mar 6, 2025

• [associative_scan] Refactoring of input checking and dynamo invocation

  #148657 opened Mar 6, 2025

• [docs] fix autograd description on convex function case

  #148658 opened Mar 6, 2025

• [HPU] Add HPU as a supported device for NestedTensor

  #148659 opened Mar 6, 2025

• Remove deprecated std::aligned_storage_t

  #148660 opened Mar 6, 2025

• removed tocm triton template cond

  #148662 opened Mar 6, 2025

• [Profiler][HPU] Fix incorrect availabilities for HPU

  #148663 opened Mar 6, 2025

• [AOTInductor] Codegen fix

  #148664 opened Mar 6, 2025

149 Issues closed by 52 people

• [triton 3.3] inductor/test_torchinductor_opinfo.py::TestInductorOpInfoCUDA::test_comprehensive_floor_divide_cuda_float16

  #148541 closed Mar 6, 2025

• [triton 3.3] inductor/test_aot_inductor.py::AOTInductorTestABICompatibleGpu::test_none_args_aot_codegen_cuda

  #148540 closed Mar 6, 2025

• DISABLED test_slice_scatter_reinplace_cuda (__main__.GPUTests)

  #145189 closed Mar 6, 2025

• torch.conj behaves differently on cpu and mps

  #148599 closed Mar 6, 2025

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

  #148547 closed Mar 6, 2025

• RuntimeError on MPS: [srcBuf length] > 0 INTERNAL ASSERT FAILED – Placeholder tensor is empty using huggingface model ibm-granite/granite-timeseries-ttm-r2

  #148589 closed Mar 6, 2025

• 'CUDA error: an illegal memory access was encountered' when using forced_align on cuda device > 0

  #148438 closed Mar 6, 2025

• Inductor layout constraints for custom operators changed from 2.5->2.6, breaking BC

  #148356 closed Mar 6, 2025

• [ROCm] sorting torch.bool tensor viewed from torch.uint8 type produces incorrect results

  #139972 closed Mar 6, 2025

• test_reference_numerics_normal fails with certain versions of numpy/scipy

  #148143 closed Mar 5, 2025

• [ONNX] stft export fails with dynamo_export

  #113067 closed Mar 5, 2025

• [ONNX] Record the capture strategy in onnx program

  #147674 closed Mar 5, 2025

• [aot autograd][inline inbuilt nn modules] AOT Autograd changes _version different from eager

  #128198 closed Mar 5, 2025

• [Inductor-CPU] ATen SDPA kernel runtime is not captured in profiling results

  #148225 closed Mar 5, 2025

• [ONNX] aten_unfold needs to support symint

  #148337 closed Mar 5, 2025

• Decorators like `torch.compiler.allow_in_graph` doesn't account for id reuse

  #147777 closed Mar 5, 2025

• The `sympy` dependency spec for pytorch on PyPi wheel is still unchanged.

  #145225 closed Mar 5, 2025

• The latest PyTorch XPU wheel 2.7.0.dev20250117+xpu does not work on Windows

  #145155 closed Mar 5, 2025

• AOTI doesn't account for constant tensors

  #148370 closed Mar 5, 2025

• Redundant Try Block in backward()

  #148115 closed Mar 5, 2025

• dataclasses.replace not supported by dynamo

  #136481 closed Mar 5, 2025

• `torch.compile` fails with customized Triton Operator on Triton 2.2

  #128039 closed Mar 5, 2025

• Compile with non-default mode + triton kernel fails

  #126864 closed Mar 5, 2025

• torch.compile generates wrong code on CPU and compiled code replaces original function

  #126848 closed Mar 5, 2025

• SIGSEGV error when passing a 0-sized tensor to `_local_scalar_dense`

  #145066 closed Mar 5, 2025

• DISABLED test_dropout_deterministic_dynamic_shapes_cuda (__main__.DynamicShapesGPUTests)

  #132957 closed Mar 5, 2025

• DISABLED test_remote_cache_load_function_device_cuda_bfloat16_dynamic_False_bundle_triton_False (__main__.TestFxGraphCache)

  #145182 closed Mar 5, 2025

• DISABLED test_tensor_subclass_basic (__main__.TestCompiledAutograd)

  #145457 closed Mar 5, 2025

• DISABLED test_precompilations (__main__.TestMaxAutotune)

  #124265 closed Mar 5, 2025

• nn.Transformer out[0:-1] not precisely equal to last_out when predicting in tgt mask

  #100052 closed Mar 5, 2025

• [Flex Attention] Errors with Dynamic Shapes (Cannot determine truth value of Relational)

  #146745 closed Mar 5, 2025

• DISABLED test_recompile_on_global_state_change_dynamic_shapes (__main__.DynamicShapesMiscTests)

  #144896 closed Mar 5, 2025

• [xpu] Compilation of pytorch failed, unable to generate RegisterSparseXPU.cpp

  #144718 closed Mar 5, 2025

• torch.accelerator.is_available() raise RuntimeError if no available CUDA/XPU devices

  #144567 closed Mar 5, 2025

• [Profiler] Add profiler activity for HPU devices

  #148181 closed Mar 5, 2025

• [dynamic-shapes][dynamo] Iist index out of range for constraint with out=

  #130066 closed Mar 5, 2025

• PyTorch defaults to using libuv but is built without support for it on Windows

  #139990 closed Mar 5, 2025

• Switch to using Docker Images from ECR instead of Docker Hub

  #147748 closed Mar 4, 2025

• Lintrunner results inconsistent

  #128588 closed Mar 4, 2025

• [ONNX] torch.matmul() breaks dynamic shapes during export

  #148192 closed Mar 4, 2025

• Release 2.6.0 validations checklist and cherry-picks

  #144503 closed Mar 4, 2025

• [RFC] Raising minimal glibc support to: glibc2_28 . Deprecation support for Amazon Linux 2 support

  #126551 closed Mar 4, 2025

• [DTensor] `clip_grad_norm_` follow-ups

  #121020 closed Mar 4, 2025

• RuntimeError when using Adam(fused=True) with torch.compile

  #126585 closed Mar 4, 2025

• [DSD] Test could fail in test_fsdp_dsd.py

  #134212 closed Mar 4, 2025

• Dtensor shard uses more gpu memory than raw tensor

  #133549 closed Mar 4, 2025

• [dynamo] Issue with construction nn.Parameter

  #127697 closed Mar 4, 2025

• Autotuning failure: `Triton Error [CUDA]: invalid argument`

  #145984 closed Mar 4, 2025

• Adding Small Epsilon in linalg_eig_backward to Improve Numerical Stability on GPU

  #147544 closed Mar 4, 2025

• I don't use FSDP，it can train.

  #148409 closed Mar 4, 2025

• built from source windows static library with multiple "unresolved external symbol"

  #87499 closed Mar 4, 2025

• When using torch.compile to compile the function _kernel_make_viewless_tensor, an error occurs：AssertionError: wrong number of dimensions

  #143895 closed Mar 4, 2025

• [Triton upstream] [Inductor] [ROCm] OpInfo quantile UT accuracy issues

  #147736 closed Mar 4, 2025

• [Triton upstream] [Inductor] [ROCm] cpp_wrapper segfaults

  #147734 closed Mar 4, 2025

• DISABLED test_cache_load_function_device_cuda_bfloat16_dynamic_False_bundle_triton_True_grad_True (__main__.TestFxGraphCache)

  #145181 closed Mar 4, 2025

• There is a problem with the wording here.

  #147696 closed Mar 4, 2025

• Triton aarch64 and triton sbsa

  #147857 closed Mar 4, 2025

• torch.compile reorder_for_compute_comm_overlap sink_waits pass does not work

  #127324 closed Mar 4, 2025

• Invalid ONNX graph if using float16 dtype for `torch.arange`

  #148041 closed Mar 3, 2025

• export + aot_export_module on models with non-parameter/buffer tensor state show up as getattrs in the graph without meta['val'] fields

  #146074 closed Mar 3, 2025

• How to get last layer hidden state of transformer model while convert model to onnx format?

  #146682 closed Mar 3, 2025

• Unable to export to ONNX | The serialized model is larger than the 2GiB limit imposed by the protobuf library.

  #147259 closed Mar 3, 2025

• when convert to onnx with dynamix_axis, the Reshape op value is always the same as static, dynamic_axis is useless, it cant't inference right shape dynamically

  #99701 closed Mar 3, 2025

• Exporting onnx model to a buffer causes "TypeError: expected str, bytes or os.PathLike object, not BytesIO"

  #147909 closed Mar 3, 2025

• DISABLED test_custom_hook_custom_stream (__main__.TestHSDPWithCustomHook)

  #147767 closed Mar 3, 2025

• [inductor] [silence] `torch.cdist` outputs inconsistent results with eager

  #148064 closed Mar 3, 2025

• IndexError: tuple index out of range when running vLLM script

  #147839 closed Mar 3, 2025

• `torch.multinomial` outputs inconsistency on ARM and x86

  #148247 closed Mar 3, 2025

• Huge numerical precision error when `torch.tensor(3811, dtype=torch.float16)`

  #148321 closed Mar 3, 2025

• The issue where opt_output in fx_graph_runnable.py is inconsistent with the actual output when testing run_repro(acc=True)

  #147850 closed Mar 3, 2025

• HIP error: invalid device function on ROCm RX 7600XT

  #147626 closed Mar 3, 2025

• `torch.Tensor.pinverse` can cause an `INTERNAL ASSERT FAILED`

  #148300 closed Mar 3, 2025

• `torch.linalg.cond` can cause an `INTERNAL ASSERT FAILED`

  #148299 closed Mar 3, 2025

• `torch.linalg.pinv` can cause an `INTERNAL ASSERT FAILED`

  #148298 closed Mar 3, 2025

• `torch.nansum` can cause a `Segmentation fault (core dumped)`

  #148297 closed Mar 3, 2025

• INTERNAL ASSERT FAILED at "/pytorch/aten/src/ATen/NamedTensorUtils.cpp":163, please report a bug to PyTorch

  #148278 closed Mar 3, 2025

• `torch.sparse.sum` can cause a `Segmentation fault (core dumped)`

  #148276 closed Mar 3, 2025

• `torch.nn.LazyConvTranspose1d` can cause a `Floating point exception (core dumped)`

  #148275 closed Mar 3, 2025

• `torch.nn.functional.conv1d` can cause a `Floating point exception (core dumped)`

  #147458 closed Mar 3, 2025

• `torch.svd` can cause an `INTERNAL ASSERT FAILED`

  #147457 closed Mar 3, 2025

• [inductor] [cpu] `nn.Tanhshrink-atan2` output inconsistent results with eager

  #148241 closed Mar 3, 2025

• ncclUnhandledCudaError

  #147575 closed Mar 3, 2025

• [Break XPU] Newly added test case with CUDA hard code failed on XPU.

  #143479 closed Mar 3, 2025

• xpu: support triton against clang with nightly wheels

  #137518 closed Mar 3, 2025

• Error loading ".venv\Lib\site-packages\torch\lib\c10_xpu.dll" or one of its dependencies

  #138986 closed Mar 3, 2025

• xpu: clarify which Intel GPUs are supported by PyTorch 2.5

  #138347 closed Mar 3, 2025

• xpu: intel conda channel is not available

  #131802 closed Mar 3, 2025

• xpu: add check for supported devices to xpu initialization and torch.xpu.is_available()

  #131799 closed Mar 3, 2025

• xpu: can't build XPU backend without sourcing oneAPI environment variables (/opt/intel/oneapi/setvars.sh)

  #127008 closed Mar 3, 2025

• xpu: python hangs on exit after check for xpu on multi-dev system

  #126259 closed Mar 3, 2025

• DISABLED test_cond_autograd_zeros_unused_branch_complex_compile_mode_compile (__main__.TestControlFlow)

  #148309 closed Mar 3, 2025

• DISABLED test_cond_autograd_zeros_unused_branch_complex_compile_mode_compile (__main__.TestControlFlow)

  #148308 closed Mar 3, 2025

• [RFC] Add CPP INT8 SDPA Template for Inductor CPU

  #144941 closed Mar 3, 2025

• DISABLED test_add_tuple_non_optional (__main__.TestScript)

  #146136 closed Mar 3, 2025

• DISABLED test_non_final_return (__main__.TestScript)

  #145975 closed Mar 3, 2025

• DISABLED test_scriptable_fn_as_attr (__main__.TestScript)

  #145972 closed Mar 3, 2025

• DISABLED test_pt2_traceable_aot_eager_cpu_float8_e5m2 (__main__.TestFloat8DtypeCPUOnlyCPU)

  #144934 closed Mar 3, 2025

• DISABLED test_reassign_module_lhs (__main__.TestScript)

  #145973 closed Mar 3, 2025

• DISABLED test_memory_stats_multigpu (__main__.TestCudaMultiGPU)

  #129860 closed Mar 3, 2025

• DISABLED test_ternary_right_associative (__main__.TestScript)

  #146137 closed Mar 3, 2025

• DISABLED test_tensor_subclasses (__main__.TestScript)

  #119949 closed Mar 3, 2025

• DISABLED test_pybind_type_comparisons (__main__.TestScript)

  #145971 closed Mar 3, 2025

• DISABLED test_script_outputs (__main__.TestScript)

  #145976 closed Mar 3, 2025

• DISABLED test_script_annotation (__main__.TestScript)

  #145974 closed Mar 3, 2025

• DISABLED test_mm_batching (__main__.TestScript)

  #119747 closed Mar 3, 2025

• [distributed] Register sharding strategy for aten.amax.default to support float8 rowwise scaling

  #147578 closed Mar 2, 2025

• Excessive memory usage during compilation start up for (atleast some) in place ops

  #148165 closed Mar 2, 2025

• [Inductor][CPU] SIGSEGV in `torch.slice_copy` with large step value

  #147071 closed Mar 2, 2025

• Cudnn header files should be copied into build package as well

  #47743 closed Mar 2, 2025

• The unevenness of torch.randint() during large range(3e9) sampling.

  #148175 closed Mar 2, 2025

• nn.Matmul return different ret within Parameter and Tensor

  #148280 closed Mar 2, 2025

• Wrong macro used when building c10/util/bit_cast.h with std::bit_cast

  #148263 closed Mar 1, 2025

• [dtensor] write aten.split_tensor using op strategy

  #130758 closed Mar 1, 2025

• PyTorch nightly MPS SDPA op is unusable

  #148194 closed Mar 1, 2025

• [Inductor UT] RuntimeError: Tried to register an operator with the same name and overload name multiple times.

  #148148 closed Mar 1, 2025

• How to install Torch version that supports RTX 5090 on Windows? - CUDA kernel errors might be asynchronously reported at some other API call

  #146977 closed Mar 1, 2025

• Placeholder tensor is empty!

  #123171 closed Mar 1, 2025

• CPU-specific Inductor Error with `view` on `torch.nn.Embedding` output

  #146390 closed Mar 1, 2025

• [Triton upstream] [Inductor] Widespread failures in UTs: AttributeError: 'dict' object has no attribute 'equal_to_1'

  #147375 closed Mar 1, 2025

• LoweringException: AttributeError: 'Constant' object has no attribute 'get_name'

  #141197 closed Feb 28, 2025

• Could not guard on data-dependent expression u0 - 7 < 0 (unhinted: u0 - 7 < 0). (Size-like symbols: u0)

  #128644 closed Feb 28, 2025

• [inductor][cpu]transformers models static/dynamic quant performance/accuracy crash in 2024-06-17 nightly release

  #128933 closed Feb 28, 2025

• randperm + torch.compile + SAC + CUDA graphs doesn't work

  #130123 closed Feb 28, 2025

• Extending more info on fake tensor when compiling

  #130234 closed Feb 28, 2025

• DISABLED test_triton_kernel_multiple_out (__main__.AutogradFunctionTests)

  #147214 closed Feb 28, 2025

• Rule-based reconfig-and-recompile

  #127999 closed Feb 28, 2025

• "grapharg" key is missing in FXGraph placeholder nodes when calling torch APIs with "out" syntax in torch.cond and torch.while_loop

  #130176 closed Feb 28, 2025

• TORCH_COMPILE_CPROFILE=1 broken (strobelight might always be on internally?)

  #131953 closed Feb 28, 2025

• torch._dynamo.exc.Unsupported: call_function args: TensorVariable() UserDefinedObjectVariable(_tuplegetter)

  #131411 closed Feb 28, 2025

• [triton 3.3][cpp_wrapper] TypeError: 'NoneType' object is not subscriptable

  #148111 closed Feb 28, 2025

• F.interpolate returns NAN on MPS if align_corner is True.

  #144245 closed Feb 28, 2025

• `AssertionError` in `torch.compile`

  #147840 closed Feb 28, 2025

• Issue with FBGEMM Operators in Exported PyTorch AOT Model When Running in C++: Cound not find schema for fbgemm:xxx

  #147065 closed Feb 28, 2025

• DISABLED test_pt2_traceable_aot_eager_cpu_float8_e4m3fn (__main__.TestFloat8DtypeCPUOnlyCPU)

  #144903 closed Feb 28, 2025

• [TensorDict - compile] Dynamo doens't like a simple class decorator - but a function is fine

  #130533 closed Feb 28, 2025

• [Inductor-CPU] LLaMA doesn't use templated GEMMs for da8w8 quantization for next-token generation

  #147954 closed Feb 28, 2025

• DISABLED test_remote_cache_load_function_device_cuda_float32_dynamic_False_bundle_triton_False (__main__.TestFxGraphCache)

  #145212 closed Feb 28, 2025

• DISABLED test_profile_all_threads (__main__.TestProfiler)

  #145951 closed Feb 28, 2025

• Very large memory increase when combining bfloat16 autocast with torch.compile

  #133637 closed Feb 28, 2025

• Importing torch_tensorrt causes warning for implicitly cleaned up file

  #147744 closed Feb 28, 2025

• DISABLED test_arange_dynamic_cuda (__main__.TestInductorDynamicCUDA)

  #127067 closed Feb 28, 2025

• Long queue for macOS runners

  #148127 closed Feb 28, 2025

• Torch 2.7.0 nightly cuda 12.6 and cuda 12.8 builds are broken on Amazon linux 2023

  #148120 closed Feb 28, 2025

• torch.clear_autocast_cache is not traceable

  #140759 closed Feb 27, 2025

• SubgraphLoweringException in flex_attention when using custom score_mod with torch.dot (MLA)

  #148107 closed Feb 27, 2025

• DISABLED test_insignificant_strides (__main__.SDPAPatternRewriterCudaDynamicTests)

  #146959 closed Feb 27, 2025

• FUNC_INLINELIST doesn't exist

  #144868 closed Feb 27, 2025

• td does not detect required test for mkl-dnn OneDNN update

  #148085 closed Feb 27, 2025

• DISABLED test_mismatched_global_state (__main__.GraphRegionTrackerTests)

  #144895 closed Feb 27, 2025

170 Issues opened by 96 people

• [Profiler][HPU] Incorrect availabilities for the HPU device

  #148661 opened Mar 6, 2025

• Extra onnx::Neg_2 input after torch.onnx.export

  #148655 opened Mar 6, 2025

• Avoid fork for TORCHINDUCTOR_COMPILE_THREADS > 1

  #148651 opened Mar 6, 2025

• [Inductor-CPU] With cpp-wrapper, some ATen ops don't get profiled with PyTorch profiler

  #148650 opened Mar 6, 2025

• [inductor][torchbench][CI] timm models got obvious performance drop with --ci flag

  #148645 opened Mar 6, 2025

• DISABLED test_set_stance_aot_eager_then_compile (__main__.DecoratorTests)

  #148644 opened Mar 6, 2025

• DISABLED test_symint_in_slice_dynamic_shapes (__main__.DynamicShapesHigherOrderOpTests)

  #148643 opened Mar 6, 2025

• Feature request: throw `torch.cuda.OutOfMemoryError` for TorchScript OOM

  #148642 opened Mar 6, 2025

• [inductor][cpu] poolformer_m36 AMP static shape multiple thread performance regression in 2025-03-03 nightly release

  #148639 opened Mar 6, 2025

• [inductor][cpu] basic_gnn_gin and basic_gnn_sage AMP single thread performance regression in 2025-03-03 nightly release

  #148636 opened Mar 6, 2025

• README doesn't explain how to run tests in the "Test PyTorch" section

  #148634 opened Mar 6, 2025

• DISABLED test_sdpa_rewriter_11_cuda (__main__.SDPAPatternRewriterCudaDynamicTests)

  #148631 opened Mar 6, 2025

• Dynamo export: dynamic dims are not exported with the specified names

  #148629 opened Mar 6, 2025

• DISABLED test_return_captured_var_used_multiple_times_dynamic_shapes (__main__.DynamicShapesHigherOrderOpTests)

  #148624 opened Mar 6, 2025

• DISABLED test_host_memory_stats (__main__.TestCuda)

  #148607 opened Mar 5, 2025

• DISABLED test_nested_tuple_output_dynamic_shapes (__main__.DynamicShapesHigherOrderOpTests)

  #148606 opened Mar 5, 2025

• AOTI takes very long time to compile (1:40 hours)

  #148572 opened Mar 5, 2025

• Suggested fixes sometimes not enough in export

  #148568 opened Mar 5, 2025

• [Feature Request] Dynamic shapes API requires spec for all arguments.

  #148564 opened Mar 5, 2025

• DISABLED test_internal_nonlocal_dynamic_shapes (__main__.DynamicShapesHigherOrderOpTests)

  #148558 opened Mar 5, 2025

• AOTI is OOM-ing when eager doesn't

  #148557 opened Mar 5, 2025

• Return type annotation of `Tensor.long()` etc is not narrowed down to dtype-specific names `LongTensor` etc

  #148552 opened Mar 5, 2025

• Issue with torch.compile

  #148551 opened Mar 5, 2025

• Name 'equal_valued' cannot be imported in pytorch 2.5.0

  #148550 opened Mar 5, 2025

• Name 'equal_valued' cannot be imported in pytorch 2.5.0

  #148549 opened Mar 5, 2025

• Issue with Sparse Tensor Matrix Multiplication and Broadcasting

  #148548 opened Mar 5, 2025

• Pytorch nightly broken Flash Attention 3 compile with sycl commit - TypeError: _write_ninja_file() got an unexpected keyword argument 'sycl_cflags'

  #148546 opened Mar 5, 2025

• F.scaled_dot_product_attention calculation output is nan when in dynamic dim under torch.compile mode

  #148545 opened Mar 5, 2025

• DISABLED test_inlined_functions_dynamic_shapes (__main__.DynamicShapesHigherOrderOpTests)

  #148544 opened Mar 5, 2025

• When Ues torch::jit::Module in UE env, there are some structured exception caused by calling module.forward()

  #148543 opened Mar 5, 2025

• exported modules using custom autograd functions will ignore custom backward function

  #148539 opened Mar 5, 2025

• Computing only the n first rows of a distance matrix with pdist

  #148536 opened Mar 5, 2025

• [inductor][cpu]speech_transformer failure in 2025-03-02 nightly release

  #148535 opened Mar 5, 2025

• [FSDP2] CPU Offload Doest Not Work with `torch.nn.utils.clip_grad_norm`

  #148532 opened Mar 5, 2025

• macos15 M4 can not install torch-2.6.0-cp310-none-macosx_11_0_arm64.whl

  #148528 opened Mar 5, 2025

• [FlexAttention] Error using create_block_mask with mask head number greater than 1

  #148527 opened Mar 5, 2025

• DISABLED test_sdpa_rewriter_11_cuda (__main__.SDPAPatternRewriterCudaTests)

  #148525 opened Mar 5, 2025

• DISABLED test_graph_break_before___enter__ (__main__.ContextlibContextManagerTests)

  #148524 opened Mar 5, 2025

• DISABLED test_globals_change_in_other_file (__main__.ContextlibContextManagerTests)

  #148523 opened Mar 5, 2025

• reshape is decomposed to view setting allow_copy=False making it fail in some case!

  #148519 opened Mar 5, 2025

• Preview (Nightly) version cuda12.8 cannot find torchaudio file

  #148518 opened Mar 5, 2025

• DISABLED test_set_stance_eager_then_compile (__main__.DecoratorTests)

  #148515 opened Mar 5, 2025

• DISABLED test_freevars_as_inputs_to_wrap_dynamic_shapes (__main__.DynamicShapesHigherOrderOpTests)

  #148514 opened Mar 5, 2025

• [MAIA][Autocast] torch.autocast doesn't work on MAIA device

  #148510 opened Mar 5, 2025

• [FSDP2] improve error msg for duplicate wraps

  #148504 opened Mar 4, 2025

• `distributed.checkpoint.async_save` leading to `TypedStorage is deprecated.`

  #148498 opened Mar 4, 2025

• UNSTABLE trunk / libtorch-linux-focal-cuda12.4-py3.10-gcc9-debug / build

  #148495 opened Mar 4, 2025

• [Inductor-CPU] Templated int8 WoQ GEMMs (with BF16 activation) may cause regressions for next-token generation of LLMs

  #148494 opened Mar 4, 2025

• export lift_constants_pass creates ugly warning

  #148487 opened Mar 4, 2025

• Export shouldn't warn when registering constant tensor attribute on graph module.

  #148482 opened Mar 4, 2025

• export is emitting too many not actionable warnings.

  #148479 opened Mar 4, 2025

• export dynamic shapes API throws weird error on upper bound.

  #148478 opened Mar 4, 2025

• XPU not available until I sign into server locally

  #148477 opened Mar 4, 2025

• Illegal memory access in scaled_dot_product_attention if only attn_mask requires grad

  #148476 opened Mar 4, 2025

• Dynamo replaces exception by hard error in `run_node`

  #148475 opened Mar 4, 2025

• DISABLED test_capture_untracked_nonlocal_dynamic_shapes (__main__.DynamicShapesHigherOrderOpTests)

  #148464 opened Mar 4, 2025

• DISABLED test_set_stance_eager_then_compile_with_graph_break (__main__.DecoratorTests)

  #148463 opened Mar 4, 2025

• [dynamo] Memory leak

  #148460 opened Mar 4, 2025

• backport torch.library.custom_op (and improvements) to older versions of PyTorch

  #148457 opened Mar 4, 2025

• BC-linter should ignore testing/linter/adapters/

  #148451 opened Mar 4, 2025

• Union type raise error when running python with argument "-O" for torch script.

  #148447 opened Mar 4, 2025

• DISABLED test_user_defined_binop (__main__.MiscTests)

  #148443 opened Mar 4, 2025

• DISABLED test_capture_untracked_global_dynamic_shapes (__main__.DynamicShapesHigherOrderOpTests)

  #148442 opened Mar 4, 2025

• torch.distributed hangs between 2 Mac Devices

  #148440 opened Mar 4, 2025

• [cudagraph_trees]RuntimeError: Error: accessing tensor output of CUDAGraphs that has been overwritten by a subsequent run

  #148439 opened Mar 4, 2025

• ERROR: I got an error about FSDP, when I trained flux model of sparsity with NVIDIA TensorRT Model Optimizer

  #148434 opened Mar 4, 2025

• DISABLED test_sys_modules (__main__.MiscTests)

  #148428 opened Mar 4, 2025

• DISABLED test_capture_tracked_dynamic_shapes (__main__.DynamicShapesHigherOrderOpTests)

  #148427 opened Mar 4, 2025

• DISABLED test_empty_graph_nested_calls_fullgraph_False_dynamic_shapes (__main__.DynamicShapesReproTests)

  #148426 opened Mar 4, 2025

• [ROCM] `linalg.eigh` crash with `float64` dtype and shape `[8192,8192]`

  #148425 opened Mar 4, 2025

• Float8_e4m3fn

  #148420 opened Mar 4, 2025

• Torch 2.6 doesn't have TCPStore::TCPStore symbol in cu126 binary, but it's available in headers

  #148411 opened Mar 4, 2025

• The apis in torch._C._nn.pyi is nonexhaustive

  #148404 opened Mar 4, 2025

• Generate two reduction loops for vectorization

  #148402 opened Mar 4, 2025

• [inductor][fuzzer] `IndexError` error at `torch.dstack`

  #148397 opened Mar 4, 2025

• DISABLED test_shape_int_inplace_binops (__main__.MiscTests)

  #148392 opened Mar 4, 2025

• DISABLED test_untracked_inputs_in_constraints_dynamic_shapes (__main__.DynamicShapesExportTests)

  #148390 opened Mar 4, 2025

• DISABLED test_sdpa_rewriter_14_cuda (__main__.SDPAPatternRewriterCudaTests)

  #148391 opened Mar 4, 2025

• [export] Unable to trace ops like min/pow

  #148389 opened Mar 4, 2025

• [Inductor-CPU] Debug util request: fine-grained mechanism to disable out-of-template epilogues

  #148382 opened Mar 3, 2025

• Improve nested jagged tensor select performance on batch dim

  #148379 opened Mar 3, 2025

• DISABLED test_param_shape_binops (__main__.MiscTests)

  #148369 opened Mar 3, 2025

• DISABLED test_export_with_cond_dynamic_shape_pred_dynamic_shapes (__main__.DynamicShapesExportTests)

  #148368 opened Mar 3, 2025

• `torch.nn.functional` inconsistent documentation

  #148353 opened Mar 3, 2025

• torch._check(x > 0) should do something sane when x is a Tensor

  #148349 opened Mar 3, 2025

• Installation of `pytorch==2.6.0+cu124` doesn't install `triton` and `nvidia` libraries

  #148345 opened Mar 3, 2025

• [RFC][PGNCCL] Add Float8 support

  #148344 opened Mar 3, 2025

• [CI] [anaconda] CI Perf Tests

  #148342 opened Mar 3, 2025

• [CI] [anaconda] Review Devcontainer anaconda usage

  #148341 opened Mar 3, 2025

• [CI] [anaconda] CI Build and Test scripts MacOS

  #148340 opened Mar 3, 2025

• [Docs] [anaconda] Review and update

  #148339 opened Mar 3, 2025

• [CI] [anaconda] CI Build and Test scripts Windows

  #148338 opened Mar 3, 2025

• [CI] [anaconda] CI Build and Test scripts Linux

  #148336 opened Mar 3, 2025

• [CI] [anaconda] Docker files have conda environment installed

  #148335 opened Mar 3, 2025

• [FSDP2] Issues with model not running on all ranks - Grads not matching fairscale implementation

  #148334 opened Mar 3, 2025

• [export][torchbench] moco fails

  #148333 opened Mar 3, 2025

• DISABLED test_mark_unbacked_strict (__main__.MiscTests)

  #148332 opened Mar 3, 2025

• DISABLED test_export_defaults_ok_dynamic_shapes (__main__.DynamicShapesExportTests)

  #148331 opened Mar 3, 2025

• DISABLED test_sys_modules_dynamic_shapes (__main__.DynamicShapesMiscTests)

  #148330 opened Mar 3, 2025

• [AOTI][torchbench] microbench_unbacked_tolist_sum fails

  #148329 opened Mar 3, 2025

• fx graph fails to recognize tensor.T as a 'call_method' node

  #148326 opened Mar 3, 2025

• `torch.linalg` routines break for inputs of more than 2**32 elements

  #148324 opened Mar 3, 2025

• compile SageAttention faing error C2872: “std” for latest torch nightly

  #148317 opened Mar 3, 2025

• [Doc] [Win] libuv installation doc is not correct.

  #148315 opened Mar 3, 2025

• The recorded step number in profiler is wrong

  #148314 opened Mar 3, 2025

• DISABLED test_int_shape_inplace_binops (__main__.MiscTests)

  #148312 opened Mar 3, 2025

• DISABLED test_empty_graph_nested_calls_fullgraph_True_dynamic_shapes (__main__.DynamicShapesReproTests)

  #148311 opened Mar 3, 2025

• batching rule for `aten::scatter_add_`

  #148307 opened Mar 3, 2025

• torch.vmap incompatibility with DLPack functions

  #148306 opened Mar 3, 2025

• broadcast_object_list not release GPU

  #148302 opened Mar 3, 2025

• DISABLED test_int_shape_binops (__main__.MiscTests)

  #148296 opened Mar 3, 2025

• DISABLED test_dont_aggressively_write_assert_dynamic_shapes (__main__.DynamicShapesReproTests)

  #148295 opened Mar 3, 2025

• Triton Kernel Rejects NamedTupleVariable Arguments

  #148289 opened Mar 2, 2025

• RuntimeError: use_libuv was requested but PyTorch was build without libuv support

  #148283 opened Mar 2, 2025

• SIGSEGV due to insufficient return value checking for PyFrame_GetLocals

  #148273 opened Mar 1, 2025

• Should DTensor support `Shard()` placement without dim requirement?

  #148269 opened Mar 1, 2025

• ONNX Export Produces main_graph Instead of torch_jit and Fails on aten::format in PyTorch 2.x

  #148268 opened Mar 1, 2025

• Raise a warning when `torch.nn.utils.clip_grad_norm_` receives an exhausted generator

  #148259 opened Mar 1, 2025

• [FSDP2] HSDP with globally sharded fp32 weights and optimizer states

  #148257 opened Mar 1, 2025

• Simplify package_data handling in setup.py

  #148256 opened Mar 1, 2025

• Improve Notation for Score Function in Documentation

  #148253 opened Mar 1, 2025

• Can't pass `strict=False` when loading a distributed checkpoint. Succeeds without warnings for "unexpected" keys, fails for "missing" keys.

  #148252 opened Mar 1, 2025

• Errors: train a model of sparsity with tensorrt-model-optimization and FSDP.

  #148251 opened Mar 1, 2025

• BrokenPipeError: [Errno 32] Broken pipe when lacking Numpy package

  #148250 opened Mar 1, 2025

• [inductor] `nn.Upsample-torch.linalg.lu_factor` outputs inconsistent results with eager

  #148244 opened Mar 1, 2025

• [FSDP2] Unclear behavior of `ignored_params` in `fully_shard`

  #148242 opened Mar 1, 2025

• Add Structured Knowledge Accumulation (SKA) Layer to PyTorch

  #148232 opened Mar 1, 2025

• make saved_tensor_hooks work better in compile for doing activation compression

  #148222 opened Feb 28, 2025

• MPS vs Metal vs CPU performance comparison

  #148219 opened Feb 28, 2025

• DISABLED test_dynamic_sources_dynamic_override (__main__.MiscTests)

  #148218 opened Feb 28, 2025

• DISABLED test_guard_failure_fn2 (__main__.MiscTests)

  #148217 opened Feb 28, 2025

• DISABLED test_guard_failure_fn_shape_control_dynamic_shapes (__main__.DynamicShapesMiscTests)

  #148216 opened Feb 28, 2025

• DISABLED test_mark_unbacked_strict_dynamic_shapes (__main__.DynamicShapesMiscTests)

  #148215 opened Feb 28, 2025

• DISABLED test_dynamic_sources_dynamic_override_dynamic_shapes (__main__.DynamicShapesMiscTests)

  #148214 opened Feb 28, 2025

• Regression: Missing Symbols in PyTorch DLL (torch_python)

  #148208 opened Feb 28, 2025

• Add option to shut down idle async_compile workers after timeout

  #148207 opened Feb 28, 2025

• Compile breaks flex-attention with jagged tensors

  #148201 opened Feb 28, 2025

• [Inductor-CPU] qlinear_binary output may have undefined strides with dynamic shape support

  #148199 opened Feb 28, 2025

• [inductor][triton] Decide how to deprecate "old triton versions"

  #148196 opened Feb 28, 2025

• Implement batching rule for masked_fill_

  #148183 opened Feb 28, 2025

• Dynamo failure on handling list comparisons

  #148179 opened Feb 28, 2025

• [Inductor] Layout created with non-sympy.Expr sizes

  #148172 opened Feb 28, 2025

• Inference llama after Export PTQ

  #148171 opened Feb 28, 2025

• PyTorch's nightly version no longer includes the CU118, CU124, and CU121 versions

  #148169 opened Feb 28, 2025

• Build pytorch for rocm failed

  #148167 opened Feb 28, 2025

• [Request Help] “torch._dynamo.exc.UserError: Dynamic control flow is not supported at the moment.” “torch._dynamo.exc.UncapturedHigherOrderOpError: Cond doesn't work unless it is captured completely with torch.compile.”

  #148193 opened Feb 28, 2025

• DISABLED test_nonstrict_trace_pre_existing_custom_class (__main__.DecoratorTests)

  #148166 opened Feb 28, 2025

• Specifying device_id in init_process_group causes tensor parallel + pipeline parallel to fail

  #148162 opened Feb 28, 2025

• [MPS][Complex] Conjugations are broken

  #148156 opened Feb 28, 2025

• [inductor] [cuda] [MultiheadAttention] `nn.MultiheadAttention-torch.reciprocal` outputs a big difference with eager

  #148153 opened Feb 28, 2025

• [inductor] `F.fractional_max_pool2d` throws `LoweringException: ZeroDivisionError: division by zero` while eager passes the check

  #148152 opened Feb 28, 2025

• [pytorch elastic] [RHEL] multiple processes call to dist.destroy_process_group() cause an RuntimeError: CUDA error: CUDA-capable device(s) is/are busy or unavailable

  #148150 opened Feb 28, 2025

• Unify OpOverload._get_dispatch and HigherOrderOperator.dispatch

  #148146 opened Feb 28, 2025

• Torch export does not preserve original edges between nodes

  #148144 opened Feb 28, 2025

• Applying online softmax patterns on joint_graph cause 1.2x peak memory regression for TB hf_T5_base model

  #148141 opened Feb 28, 2025

• Conv/pool doc on ceilmode wrong

  #148123 opened Feb 27, 2025

• [cutlass backend] C++ compile error for CUTLASS config only get resolved in autotuning stage

  #148122 opened Feb 27, 2025

• Add int32 support to torch.gather

  #148119 opened Feb 27, 2025

• [inductor][triton] Explicit kernel-arg mismatch checks

  #148116 opened Feb 27, 2025

• We should never throw vanilla C++ exceptions

  #148114 opened Feb 27, 2025

• [inductor][triton] introduce better "APIs" in triton that can clean up our triton/inductor integration

  #148113 opened Feb 27, 2025

• MLA with Learnable RoPE Tensors is Broken with Flex Attention

  #148112 opened Feb 27, 2025

• [CI] Remove conda usage from lint related jobs

  #148110 opened Feb 27, 2025

• NotImplementedError: FlexAttentionHigherOrderVariable() has no type

  #148106 opened Feb 27, 2025

• Initial investigation for removing MOD_SKIPLIST

  #148103 opened Feb 27, 2025

• onnx dynamo export does not support aten bucketize

  #148098 opened Feb 27, 2025

• DISABLED test_njt_causal_bfloat16 (__main__.TestFlexAttention)

  #148095 opened Feb 27, 2025

• DISABLED test_split_dynamic (__main__.AutoFunctionalizeTests)

  #148094 opened Feb 27, 2025

• DISABLED test_dynamo_timed (__main__.TestDynamoTimed)

  #148093 opened Feb 27, 2025

• FSDP2 without sharding works slower than DDP

  #148086 opened Feb 27, 2025

490 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 Windows Arm64 Nightly Builds

  #139760 commented on Mar 6, 2025 • 29 new comments

• [HOP] Mutation and alias rework

  #146658 commented on Mar 6, 2025 • 20 new comments

• Add `__context/cause/suppress_context/traceback__` to Exception

  #146499 commented on Mar 5, 2025 • 17 new comments

• [export][dynamic shapes] add Dim._OBLIVIOUS, _mark_oblivious()

  #147881 commented on Mar 3, 2025 • 15 new comments

• [inductor] Fix issue with set_linter, improve linter framework

  #144620 commented on Mar 4, 2025 • 13 new comments

• [pytree] Add public pytree module `torch.utils.pytree`

  #137400 commented on Mar 5, 2025 • 12 new comments

• fix indirect broadcast

  #145992 commented on Mar 6, 2025 • 10 new comments

• Fix `torch.nn.functional.hardswish` gradients corner case

  #148049 commented on Mar 6, 2025 • 10 new comments

• Support subclass constructor capturing in export

  #147014 commented on Mar 6, 2025 • 10 new comments

• [pytree] add APIs to determine a class is a namedtuple or PyStructSequence

  #113257 commented on Mar 5, 2025 • 9 new comments

• Enable Accelerator to perform streaming backward

  #142097 commented on Mar 6, 2025 • 9 new comments

• Enable fast qlinear_dynamic path for AArch64 through ACL directly

  #145942 commented on Mar 5, 2025 • 9 new comments

• Custom ops support arbitrary input types by migrating to python dispatcher

  #147927 commented on Mar 6, 2025 • 8 new comments

• [Inductor][CPP] Add float16 support for CppMicroGemmAMX

  #147368 commented on Mar 6, 2025 • 7 new comments

• [Intel GPU] int4 WOQ gemm XPU Support

  #137566 commented on Mar 4, 2025 • 7 new comments

• torch.utils.checkpoint preserves torch function mode stack during recompute

  #148023 commented on Mar 4, 2025 • 4 new comments

• Replace `unimplemented` with `unimplemented_v2' in `codegen.py`

  #148069 commented on Mar 5, 2025 • 4 new comments

• Parallelize bf16->f32 conversion for gemm(bf16:bf16->bf16)

  #147864 commented on Mar 3, 2025 • 4 new comments

• Change arg_kwarg_vals propagation strategy

  #148046 commented on Mar 6, 2025 • 4 new comments

• [export] Add support for invoke_subgraph

  #147863 commented on Mar 4, 2025 • 4 new comments

• [export] Add export_cache

  #147992 commented on Mar 4, 2025 • 3 new comments

• [ci][anaconda] Remove conda from linter docker images

  #147789 commented on Mar 5, 2025 • 3 new comments

• Increase reference count of state tensor in `THPGenerator_reduce` to avoid premature garbage collection in `multiprocessing` spawn method `"forkserver"` and `"spawn"`

  #147907 commented on Mar 5, 2025 • 3 new comments

• cpu: aarch64: enable gemm-bf16f32

  #140159 commented on Mar 6, 2025 • 2 new comments

• Enable onednn in pytorch for ppc64le architecture

  #143743 commented on Mar 6, 2025 • 2 new comments

• Remove unused rand call if not fallback to eager for rand

  #147790 commented on Mar 6, 2025 • 2 new comments

• Correctly propagate exception to parent tx

  #146502 commented on Mar 5, 2025 • 2 new comments

• add grad_output shape check for adaptive_avg_pool2d_backward

  #145241 commented on Mar 5, 2025 • 2 new comments

• [import][inductor] Simplify grid handling

  #147583 commented on Mar 4, 2025 • 2 new comments

• add `torch.float4_e2m1fn_x2` to PyTorch

  #146578 commented on Feb 28, 2025 • 2 new comments

• [CD] Enable triton xpu windows build

  #147637 commented on Mar 6, 2025 • 2 new comments

• Add torch._library.utils.normalize_args_kwargs

  #148062 commented on Feb 27, 2025 • 2 new comments

• Deprecate DataLoader pin_memory_device param

  #146821 commented on Mar 4, 2025 • 2 new comments

• [export] check non-negative modulus, avoid unnecessary congruences, in export solver

  #144925 commented on Mar 4, 2025 • 2 new comments

• cpp_wrapper: reduce memory usage by removing unneeded temporaries

  #147403 commented on Mar 6, 2025 • 2 new comments

• experimental proposal DCP v2

  #146999 commented on Mar 3, 2025 • 2 new comments

• Enable a fast path for (static) qlinear for AArch64 through ACL directly.

  #147337 commented on Mar 5, 2025 • 2 new comments

• [torch/elastic][upstream] Fix the wrong order when start_index is not 0

  #147967 commented on Mar 5, 2025 • 2 new comments

• [ROCm] opportunistic fastatomics for ReduceAdd operations for MI300 GPUs

  #146264 commented on Feb 28, 2025 • 2 new comments

• Add ppc64le wheel build support

  #147194 commented on Mar 4, 2025 • 1 new comment

• [Dtensor] Pass device information in OffsetBasedRNGTracker

  #147594 commented on Mar 5, 2025 • 1 new comment

• Fix the Problems About Defining Static Variable in Inline Function

  #147095 commented on Mar 4, 2025 • 1 new comment

• [ONNX] Add draft_export as a strategy

  #147529 commented on Mar 5, 2025 • 1 new comment

• Sticky cache API for torch.compile

  #147528 commented on Feb 28, 2025 • 1 new comment

• Small scheduler refactors

  #147410 commented on Feb 28, 2025 • 1 new comment

• Make codegen dynamic shapes more device agnostic

  #146830 commented on Mar 5, 2025 • 1 new comment

• Enable pt2e quantization path for arm

  #146690 commented on Mar 5, 2025 • 1 new comment

• [CD] Annotate linux/arm64 cuda wheels with consistent nvidia dependencies

  #145021 commented on Mar 6, 2025 • 1 new comment

• Use the device interface for detecting Triton availability

  #139171 commented on Feb 28, 2025 • 1 new comment

• [WIP][ptd][nccl] use current-stream as nccl-stream under async=False mode

  #147820 commented on Mar 3, 2025 • 1 new comment

• test 0-dim squeeze in basic.TestSqueeze

  #147928 commented on Mar 3, 2025 • 1 new comment

• Introduce `UserDefinedExceptionClassVariable`

  #146504 commented on Mar 5, 2025 • 1 new comment

• Facilitate at::_weight_int4pack_mm_with_scale_and_zeros related registration

  #147962 commented on Mar 4, 2025 • 1 new comment

• fix simple-spec crash

  #147723 commented on Feb 28, 2025 • 1 new comment

• Add needs_exact_strides operator tag for Inductor to force exact strides

  #148063 commented on Mar 6, 2025 • 1 new comment

• [Intel CPU] Fix issue #143483.

  #144854 commented on Mar 3, 2025 • 0 new comments

• [Intel CPU] Fix issue #143482.

  #144760 commented on Mar 3, 2025 • 0 new comments

• optimize the decomposition of aten.native_group_norm

  #144733 commented on Mar 2, 2025 • 0 new comments

• Generalize poison fork logic for each device backend

  #144664 commented on Mar 5, 2025 • 0 new comments

• [will-not-merge] tuning

  #145798 commented on Mar 6, 2025 • 0 new comments

• [AsyncMM] re-enable and adapt to cutlass 3.6.0 (#144011)

  #145811 commented on Mar 6, 2025 • 0 new comments

• [BE][PYFMT] migrate PYFMT for `torch/[a-c]*/` to `ruff format`

  #144554 commented on Mar 2, 2025 • 0 new comments

• [inductor] Enable docstring_linter on _inductor

  #144622 commented on Mar 4, 2025 • 0 new comments

• [BE][PYFMT] migrate PYFMT for `test/[a-h]*/` to `ruff format`

  #144555 commented on Mar 3, 2025 • 0 new comments

• [BE][PYFMT] migrate PYFMT for `test/[i-z]*/` to `ruff format`

  #144556 commented on Mar 5, 2025 • 0 new comments

• [inductor] Add tests for new docstring_linter features (fix #142496)

  #144621 commented on Mar 4, 2025 • 0 new comments

• [Test][Linalg][CUDA] Increase niter in test_svd_lowrank_cuda_float64

  #145930 commented on Mar 3, 2025 • 0 new comments

• [BE][PYFMT] remove `black`: finish `black -> ruff format` migration

  #144557 commented on Feb 28, 2025 • 0 new comments

• Remove unneeded CUDA logic from _create_build_env

  #145822 commented on Mar 4, 2025 • 0 new comments

• [inductor] Add features to docstring_linter (see #142496)

  #145834 commented on Mar 4, 2025 • 0 new comments

• Guard the CPU cpp wrapper tests on having a cpp wrapper

  #145847 commented on Mar 4, 2025 • 0 new comments

• [micro_pipeline_tp] add logging for all-gather-matmul fusion

  #145594 commented on Mar 6, 2025 • 0 new comments

• Open up PT UTs to cover additional devices

  #145589 commented on Mar 6, 2025 • 0 new comments

• [micro_pipeline_tp] support pattern matching row-wise scaled_mm with sharded scale

  #145595 commented on Mar 6, 2025 • 0 new comments

• Add device support for chunk_cat, all_gather_copy_in, and split_with_…

  #145600 commented on Mar 3, 2025 • 0 new comments

• General Changes for multi accelerators

  #145521 commented on Mar 6, 2025 • 0 new comments

• Simplify functional composition in _aot_autograd/dispatch_and_compile_graph.py

  #145636 commented on Mar 6, 2025 • 0 new comments

• Remove unnecessary "special linking" for `BLAS_LIBRARIES`

  #145487 commented on Mar 5, 2025 • 0 new comments

• removed check for ConvTranspose3D on MPS

  #145366 commented on Feb 28, 2025 • 0 new comments

• [c10d] implement ReduceOp.unbox()

  #145652 commented on Mar 6, 2025 • 0 new comments

• Fix incorrect citation of authors in documentation

  #145209 commented on Mar 6, 2025 • 0 new comments

• [Inductor] optimize welford reduction

  #145061 commented on Mar 5, 2025 • 0 new comments

• [Easy] update pip sources for ROCm in nightly pull tool

  #145685 commented on Feb 28, 2025 • 0 new comments

• Fix support for nccl < 2.17

  #145719 commented on Mar 1, 2025 • 0 new comments

• [Async-TP] Port _fused_all_gather_matmul_native to cpp to reduce launching overhead

  #145794 commented on Mar 6, 2025 • 0 new comments

• Made partitioning more(?) deterministic

  #145024 commented on Feb 28, 2025 • 0 new comments

• [AsyncMM] preliminary tuning

  #145795 commented on Mar 6, 2025 • 0 new comments

• [test] fix unit test

  #144977 commented on Mar 3, 2025 • 0 new comments

• [Async-TP] _pipelined_multi_all_gather_and_consume reduce overhead

  #145796 commented on Mar 6, 2025 • 0 new comments

• [Async-TP] improve algo selection

  #145797 commented on Mar 6, 2025 • 0 new comments

• Add the max_autotune tests in the periodic jobs.

  #143560 commented on Mar 6, 2025 • 0 new comments

• Fix space typo in warning message

  #143473 commented on Mar 4, 2025 • 0 new comments

• [while_loop][jit inductor] auto-unspecialize int input and output to unbacked symints

  #143457 commented on Mar 1, 2025 • 0 new comments

• Remove all dead type ignores (round 2)

  #143256 commented on Mar 3, 2025 • 0 new comments

• Enable AArch64 CI scripts to be used for local dev

  #143190 commented on Feb 28, 2025 • 0 new comments

• Support unique id for Tensor Storage Object

  #143093 commented on Mar 4, 2025 • 0 new comments

• Update low prec codegen for div/mod

  #142350 commented on Mar 3, 2025 • 0 new comments

• Fix more undefined errors in TypeCast.h

  #142346 commented on Mar 6, 2025 • 0 new comments

• [DO NOT MERGE][WIP] CI: Dispatch PR events to the out-of-tree test infra

  #142114 commented on Mar 4, 2025 • 0 new comments

• Fix platform detection in MKLDNN CMake file

  #142067 commented on Mar 3, 2025 • 0 new comments

• Add AOT inductor support for _scaled_mm for CPU

  #141961 commented on Mar 6, 2025 • 0 new comments

• Banned ever saving unclaimed nodes

  #141940 commented on Feb 28, 2025 • 0 new comments

• Enable CUDA 12.6 OSS CI

  #140793 commented on Mar 5, 2025 • 0 new comments

• ILP for auto FSDP wrapping

  #140298 commented on Feb 28, 2025 • 0 new comments

• [Environment Variable][6/N] Use thread-safe getenv functions

  #140200 commented on Mar 5, 2025 • 0 new comments

• Add torch._scaled_mm for CPU

  #139975 commented on Mar 6, 2025 • 0 new comments

• [cuDNN] Add an option to force cuDNN usage (incl. SDPA)

  #139699 commented on Mar 6, 2025 • 0 new comments

• Refactor CMake to install header by build option

  #139469 commented on Mar 3, 2025 • 0 new comments

• Fix `USE_STATIC_MKL` lost functionality

  #138996 commented on Feb 27, 2025 • 0 new comments

• [Docker] Create an independent dependecies layer

  #138612 commented on Mar 3, 2025 • 0 new comments

• [Inductor] introduce comm buffer planning

  #138519 commented on Mar 6, 2025 • 0 new comments

• Extending SVE VEC Backend Support in PyTorch to SVE128 and SVE512.

  #138388 commented on Feb 28, 2025 • 0 new comments

• [WIP] Add CachingDeviceAllocatorInterface as the base device allocator

  #138222 commented on Mar 4, 2025 • 0 new comments

• [pytree] add `treespec_{leaf,tuple,dict}` functions for args_spec modification

  #138214 commented on Mar 3, 2025 • 0 new comments

• [POC][FX][pytree] cleanup fx pytree implementation

  #138202 commented on Mar 3, 2025 • 0 new comments

• Add TORCH_CHECK_INDEX in convert_indices_from_coo_to_csr_cpu

  #138068 commented on Feb 28, 2025 • 0 new comments

• DISABLED test_repeated_calling_cuda (__main__.AOTInductorTestABICompatibleGpu)

  #146185 commented on Mar 3, 2025 • 0 new comments

• [BE][PYFMT] migrate PYFMT for `torch/[e-n]*/` to `ruff format`

  #144553 commented on Mar 2, 2025 • 0 new comments

• [BE][PYFMT] migrate PYFMT for `torch/[p-z]*/` to `ruff format`

  #144552 commented on Mar 2, 2025 • 0 new comments

• [BE][PYFMT] migrate PYFMT for `torch/_[a-h]*/` to `ruff format`

  #144551 commented on Mar 2, 2025 • 0 new comments

• [BE][PYFMT] migrate PYFMT for `{torch,test}/{nn,optim}/**` to `ruff format`

  #144548 commented on Mar 3, 2025 • 0 new comments

• [20/N] Fix extra warnings brought by clang-tidy-17

  #144473 commented on Mar 4, 2025 • 0 new comments

• Support Swiglu for Module and functional

  #144465 commented on Mar 5, 2025 • 0 new comments

• improve WOQ first token performance on CPU

  #144463 commented on Mar 5, 2025 • 0 new comments

• [BE][pytree][Easy] change imports `torch.utils._pytree` -> `torch.utils.pytree.python`

  #144405 commented on Mar 3, 2025 • 0 new comments

• codecache.py: Utilize precompiled headers for CPP python bindings

  #144349 commented on Mar 1, 2025 • 0 new comments

• [pytree][1/N] change pytree usages to implementation agnostic: `torch.distributed`

  #144332 commented on Mar 3, 2025 • 0 new comments

• codecache: Remove cpp_prefix.h duplication per build, then precompile it

  #144293 commented on Mar 1, 2025 • 0 new comments

• Add CUDA aarch64 triton wheel build

  #144049 commented on Mar 4, 2025 • 0 new comments

• Fix to torch.hub documentation grammar mistakes.

  #144016 commented on Mar 3, 2025 • 0 new comments

• Brister/always tiled reduction

  #144008 commented on Mar 2, 2025 • 0 new comments

• Fix typo: change 'recieve' into 'receive'

  #143981 commented on Feb 28, 2025 • 0 new comments

• Fixed bug in FindMKL.cmake

  #143980 commented on Mar 1, 2025 • 0 new comments

• [ci] Add riscv opt-int build

  #143979 commented on Mar 6, 2025 • 0 new comments

• [Submodule] Bump flatbuffers to v24.12.23

  #143964 commented on Mar 5, 2025 • 0 new comments

• using more descriptive alt text for accessibility

  #143958 commented on Feb 28, 2025 • 0 new comments

• Remove aten/src/ATen/core/Array.h

  #143950 commented on Mar 1, 2025 • 0 new comments

• Add `_benchmark_func` convenience method

  #143911 commented on Feb 28, 2025 • 0 new comments

• Remove remove_non_owning_ref_types

  #143805 commented on Mar 3, 2025 • 0 new comments

• [CI] enable operator benchmark on CPU

  #143733 commented on Feb 27, 2025 • 0 new comments

• Apply clang-format for ATen/core/op_registration headers

  #143730 commented on Mar 3, 2025 • 0 new comments

• ci: Add scaffolding for buidling wheels sequentially

  #143672 commented on Mar 4, 2025 • 0 new comments

• Extend vec backend with BF16 SVE intrinsics

  #143666 commented on Mar 5, 2025 • 0 new comments

• Attempt to speed up MPS getTensorStringKey

  #143630 commented on Mar 4, 2025 • 0 new comments

• Fix test_tensorboard when started w/o tensorboard package

  #148079 commented on Feb 27, 2025 • 0 new comments

• Onednn pri cache

  #147693 commented on Mar 3, 2025 • 0 new comments

• [DCP] fix dcp gather_object/scatter_object_list

  #147675 commented on Mar 4, 2025 • 0 new comments

• Attempt a mixed precision fused adam

  #147653 commented on Feb 27, 2025 • 0 new comments

• [CUDAGraph] Graph Partition

  #147648 commented on Mar 6, 2025 • 0 new comments

• Implement metal kernel for MPS binary ops using TensorIterator

  #147644 commented on Feb 28, 2025 • 0 new comments

• [ROCm] Improve backwards indexing when stride is not one

  #147630 commented on Feb 27, 2025 • 0 new comments

• Fixed abnormal behavior of LazyLinear when using LayzLinear and load_state together

  #147599 commented on Feb 27, 2025 • 0 new comments

• Define USE_C10D_XCCL and USE_XCCL in pytorch

  #147593 commented on Mar 5, 2025 • 0 new comments

• Fix log2, PowByNatural printing

  #147592 commented on Mar 4, 2025 • 0 new comments

• [ONNX][demo] Rotary embedding

  #147576 commented on Mar 5, 2025 • 0 new comments

• [partitioner] always ban compiler-driven recompute of collectives by default

  #147561 commented on Feb 27, 2025 • 0 new comments

• ROCm MX-FP8 Gemm

  #147553 commented on Mar 6, 2025 • 0 new comments

• [ROCm] Input vectorization in elementwise kernels for tensors with heterogeneous types

  #147527 commented on Feb 27, 2025 • 0 new comments

• Update pybind11 submodule to 3.0.0-dev test

  #147524 commented on Mar 1, 2025 • 0 new comments

• Enable UBSAN test

  #147511 commented on Mar 4, 2025 • 0 new comments

• Document poison fork note for accelerator APIs

  #147507 commented on Mar 4, 2025 • 0 new comments

• modified device check logic and added tests

  #147501 commented on Mar 2, 2025 • 0 new comments

• Optimize `dynamo` typing

  #147499 commented on Mar 4, 2025 • 0 new comments

• Upgrade submodule oneDNN to v3.7

  #147498 commented on Feb 28, 2025 • 0 new comments

• [CUDA] Replace deprecated usages of cub iterators and thread operators

  #147493 commented on Feb 28, 2025 • 0 new comments

• [ONNX] Migrate onnx ops decomp functions

  #147469 commented on Mar 4, 2025 • 0 new comments

• [Inductor] Fix `torch.polygamma()` when n == 1

  #147453 commented on Feb 28, 2025 • 0 new comments

• Reland "Introduce new template heuristic for triton autotune configs"

  #147452 commented on Mar 5, 2025 • 0 new comments

• [5/N] Remove unnecessary once flag usage

  #147445 commented on Feb 28, 2025 • 0 new comments

• [executorch hash update] update the pinned executorch hash

  #147422 commented on Mar 6, 2025 • 0 new comments

• Fix atomic operation compatibility for ARMv8-A (Raspberry Pi 4) by adjusting compilation flags

  #148070 commented on Feb 27, 2025 • 0 new comments

• use identity op for alpha=inf in torch.celu and quantized_celu

  #148066 commented on Feb 28, 2025 • 0 new comments

• [ROCm] Skip Navi4 Row-Wise F8 Tests

  #148037 commented on Feb 28, 2025 • 0 new comments

• [CUDA][complex] skip `test_reference_numerics_large_jiterator_unary_cuda_complex64` on CUDA

  #148024 commented on Mar 5, 2025 • 0 new comments

• [while_loop] require stride to be the same as input for body_fn

  #148002 commented on Mar 6, 2025 • 0 new comments

• [ROCm] Use generated CK config.h rather than system

  #147993 commented on Mar 5, 2025 • 0 new comments

• Support `contextlib.suppress`

  #147990 commented on Mar 5, 2025 • 0 new comments

• Bump Protobuf to 5.29

  #147963 commented on Feb 28, 2025 • 0 new comments

• [Don't merge]Upgrade submodule oneDNN to v3.7 (#147498)(Zi)

  #147917 commented on Mar 3, 2025 • 0 new comments

• [Draft] Enable cpu_offload for _distribute_state_dict

  #147916 commented on Mar 3, 2025 • 0 new comments

• Change persistent reduction threshold to 32

  #147899 commented on Feb 27, 2025 • 0 new comments

• [PT2] Allow tensor type in allowed_getattr_types_for_subgm when verifiying ep

  #147898 commented on Mar 4, 2025 • 0 new comments

• [targets2buck] Remove tombstone messages proactively

  #147897 commented on Feb 28, 2025 • 0 new comments

• [Not4Land] test `optree` with HEAD version

  #147870 commented on Mar 6, 2025 • 0 new comments

• Set disable_clone=True when running opt_gm

  #147845 commented on Mar 3, 2025 • 0 new comments

• Use /permissive- for MSVC build of torch libraries

  #147825 commented on Mar 3, 2025 • 0 new comments

• Use torch_compile_options for c10 libraries

  #147821 commented on Mar 5, 2025 • 0 new comments

• test

  #147800 commented on Feb 27, 2025 • 0 new comments

• [ROCm] CK Memory-Efficient Attention (attention bias support)

  #147778 commented on Mar 1, 2025 • 0 new comments

• [cuda] Added a correctness test for layernorm backwards

  #147763 commented on Mar 6, 2025 • 0 new comments

• [DCP] Work in progress: Demonstrate rank local checkpointing in DCP

  #147758 commented on Mar 3, 2025 • 0 new comments

• Update CPU tolerance for f16 triplet margin loss

  #147742 commented on Mar 4, 2025 • 0 new comments

• [WIP][XPU][Inductor] Update Intel triton for release 2.7.

  #147727 commented on Mar 5, 2025 • 0 new comments

• Avoid linking multiple OMP runtimes in libtorch_cpu.so if BLAS used is OpenBLAS.

  #147725 commented on Mar 5, 2025 • 0 new comments

• Skip test_dtypes xpu test on bmm and addbmm

  #147721 commented on Mar 4, 2025 • 0 new comments

• [ROCm][Windows] Enable torchvision build with ROCm on Windows

  #147382 commented on Mar 3, 2025 • 0 new comments

• Enable qint8 and quint8 add for AArch64 using ACL directly

  #146620 commented on Mar 5, 2025 • 0 new comments

• [WIP] BaseSubclass

  #146612 commented on Mar 6, 2025 • 0 new comments

• [NOT FOR LANDING] experimental NVSHMEM integration

  #146593 commented on Mar 6, 2025 • 0 new comments

• clang-format CUDASymmetricMemory.cu

  #146592 commented on Mar 6, 2025 • 0 new comments

• [Partitioner] Reduce time consuming of partitions merger

  #146582 commented on Mar 1, 2025 • 0 new comments

• [Partitioner] Remove unnecessary upstream nodes in dependency viewer

  #146580 commented on Mar 1, 2025 • 0 new comments

• add python root bin to windows load path.

  #146573 commented on Mar 2, 2025 • 0 new comments

• [pt2d] Add reorder_comms_preserving_peak_memory pass

  #146562 commented on Mar 6, 2025 • 0 new comments

• Improve comms debug visualization

  #146561 commented on Mar 6, 2025 • 0 new comments

• [not for land] temp changes to enable 'simple_fsdp'

  #146558 commented on Mar 3, 2025 • 0 new comments

• Support contextlib.ExitStack

  #146506 commented on Mar 5, 2025 • 0 new comments

• Allow setting attribute to NestedUserFunctionVariable

  #146505 commented on Mar 4, 2025 • 0 new comments

• Update CPython tests for ctx manager to use unittest

  #146501 commented on Mar 5, 2025 • 0 new comments

• Allow trace through unittest

  #146500 commented on Mar 5, 2025 • 0 new comments

• Update code_template.py re.compile() is directly applied to the regex…

  #146489 commented on Feb 28, 2025 • 0 new comments

• Update quantile doc

  #146485 commented on Mar 6, 2025 • 0 new comments

• [1/N] Use std::string_view in torchgen

  #146403 commented on Mar 3, 2025 • 0 new comments

• [WIP][dynamic shapes] mark backed size symbols as size-like

  #146335 commented on Mar 6, 2025 • 0 new comments

• Use device agnostic APIs for device_count and backend in common_fsdp

  #146289 commented on Mar 6, 2025 • 0 new comments

• [dcp] Minor improvements to filesystem writer

  #146273 commented on Mar 3, 2025 • 0 new comments

• Format tests by PYFMT

  #146267 commented on Mar 3, 2025 • 0 new comments

• [2/N] Fix cppcoreguidelines-init-variables suppression

  #146237 commented on Mar 6, 2025 • 0 new comments

• Subprocess compile

  #146134 commented on Mar 6, 2025 • 0 new comments

• Move get accelerator to use build time flags when possible

  #146098 commented on Mar 4, 2025 • 0 new comments

• [ARM] Fix TestDataLoader.test_segfault unexpected success on Aarch6[4

  #146090 commented on Feb 28, 2025 • 0 new comments

• Force build to conform C++ standard on windows by adding `/permissive-` flag

  #147367 commented on Mar 4, 2025 • 0 new comments

• Add meta function for out variants of ones,zeros,empty

  #147350 commented on Mar 6, 2025 • 0 new comments

• Refine XPU oneDNN context manager API

  #147349 commented on Mar 6, 2025 • 0 new comments

• [ROCm][Windows] Disable Composable Kernels and Triton for Windows builds

  #147334 commented on Mar 4, 2025 • 0 new comments

• [TESTING] [NO MERGE] Testing new triton commit for release/2.7

  #147320 commented on Mar 6, 2025 • 0 new comments

• Add the memory and dispatch to the logging module.

  #147262 commented on Mar 4, 2025 • 0 new comments

• add PrivateUse1 backend in fsdp collecitves

  #147260 commented on Feb 28, 2025 • 0 new comments

• logging: close handler after removing it

  #147235 commented on Mar 4, 2025 • 0 new comments

• cpp_wrapper: Fix even more tests

  #147225 commented on Mar 6, 2025 • 0 new comments

• Fix rms_norm in fp16/bf16

  #147203 commented on Mar 3, 2025 • 0 new comments

• dynamo: Count number of opcodes processes

  #147149 commented on Mar 6, 2025 • 0 new comments

• [fsdp] add an experimental allocator hook for buffers that participate in collective communication

  #147146 commented on Mar 6, 2025 • 0 new comments

• ROCm F8 Datatype Selector

  #147142 commented on Feb 28, 2025 • 0 new comments

• fake_tensor: Handle op errors more gracefully

  #147049 commented on Mar 5, 2025 • 0 new comments

• [ROCm] [TunableOp] Enable logging of BLAS parameters

  #147034 commented on Mar 6, 2025 • 0 new comments

• [BE]: Try to remove unused type ignores - attempt 1

  #146989 commented on Mar 3, 2025 • 0 new comments

• [Inductor] Unify the data type propagation between Triton and CPP Backend

  #146970 commented on Mar 6, 2025 • 0 new comments

• cpp_wrapper: Precompile device-specific header files

  #146928 commented on Mar 3, 2025 • 0 new comments

• Adjust TestInductorOpInfo to depend on backend, not device

  #146911 commented on Mar 5, 2025 • 0 new comments

• Optimize transformer encoder/decoder init suggestion

  #146882 commented on Mar 3, 2025 • 0 new comments

• update types on dynamo configs

  #146873 commented on Feb 28, 2025 • 0 new comments

• Don't look at TESTING_ONLY in fuzzer

  #146870 commented on Feb 28, 2025 • 0 new comments

• Enable explicitly vectorized `_weight_int8pack_mm` op for FP16 dtype on x86_64 CPU

  #146777 commented on Feb 28, 2025 • 0 new comments

• cpp_wrapper: persist autotune example tensors until last use

  #146706 commented on Mar 6, 2025 • 0 new comments

• Enable Windows tests

  #146695 commented on Mar 5, 2025 • 0 new comments

• AOTI packaged model fails with generic error when run in for loop but succeeds on individual sample

  #146524 commented on Mar 3, 2025 • 0 new comments

• Cannot import PyTorch in Alpine Docker Container

  #71381 commented on Mar 3, 2025 • 0 new comments

• expandable_segments does not work for CUDAPluggableAllocator + MemPool

  #147851 commented on Mar 3, 2025 • 0 new comments

• [ONNX] Migrate torchlib from onnxscript

  #139301 commented on Mar 3, 2025 • 0 new comments

• No gradient for `residuals` in the return value of `torch.linalg.lstsq`

  #147543 commented on Mar 3, 2025 • 0 new comments

• Does CUDACachingAllocator.cpp still require deferred event creation?

  #147874 commented on Mar 3, 2025 • 0 new comments

• torch._scaled_mm reproductibility

  #147972 commented on Mar 3, 2025 • 0 new comments

• Fp8 scaled-mm row-wise is substantially slower than tensor-wise

  #147971 commented on Mar 3, 2025 • 0 new comments

• Adam doesn't work with nonzero-dim Tensor betas

  #147921 commented on Mar 3, 2025 • 0 new comments

• `FxGraphDrawer` fails on `einsum` nodes

  #147884 commented on Mar 3, 2025 • 0 new comments

• [CI/CD] Deprecating PyTorch’s official Anaconda channel

  #138696 commented on Mar 3, 2025 • 0 new comments

• MPS support `torch.linalg.norm` on complex numbers

  #146691 commented on Mar 3, 2025 • 0 new comments

• MPS operator coverage tracking issue (2.6+ version)

  #141287 commented on Mar 3, 2025 • 0 new comments

• "unhashable type: non-nested SymInt" error when using DTensor and Compiled Autograd together

  #127797 commented on Mar 3, 2025 • 0 new comments

• Torch Compile mode giving Graph Break for Higher Order Op for ProcessGroup param

  #129942 commented on Mar 3, 2025 • 0 new comments

• PyTorch C++ API binary compiled with xmake crashes

  #129305 commented on Mar 3, 2025 • 0 new comments

• Model parameter and gradient memory formats are inconsistent with compiled autograd

  #127922 commented on Mar 3, 2025 • 0 new comments

• WSL2 RTX A6000 , CUDA out of memory.

  #117197 commented on Mar 3, 2025 • 0 new comments

• Error loading "torch\lib\aoti_custom_ops.dll" or one of its dependencies, when importing Torch, when building from Source on Windows 11 with cuDNN.

  #144931 commented on Mar 3, 2025 • 0 new comments

• CMake Error: When installing PyTorch from source, CUDA not being detected.

  #134331 commented on Mar 3, 2025 • 0 new comments

• User-defined way to hook into Inductor autotuning

  #142388 commented on Mar 3, 2025 • 0 new comments

• [JIT] support list of nn.Module in torchscript

  #36061 commented on Mar 3, 2025 • 0 new comments

• [XPU] torch.nn.functional.pad brings wrong results with torch.compile on Intel GPU

  #145372 commented on Mar 3, 2025 • 0 new comments

• DISABLED test_input_mutation2_dynamic_shapes_cpu (__main__.DynamicShapesCpuTests)

  #135295 commented on Mar 3, 2025 • 0 new comments

• [DTensor] [distributed]: Operator aten.select.int does not have a sharding strategy registered

  #147724 commented on Mar 3, 2025 • 0 new comments

• DISABLED test_vdd_clamp_dynamic_shapes_cpu (__main__.DynamicShapesCpuTests)

  #134445 commented on Mar 3, 2025 • 0 new comments

• DISABLED test_embedding_dynamic_shapes_cpu (__main__.DynamicShapesCpuTests)

  #135250 commented on Mar 3, 2025 • 0 new comments

• Regression: Multiple OpenMP runtimes linked to libtorch_cpu.so

  #146603 commented on Mar 3, 2025 • 0 new comments

• Get `aot_autograd`'ed graph without `torch.compile` and freeze constants without Inductor context

  #140205 commented on Mar 3, 2025 • 0 new comments

• [export] Fail fast on pytorch with `aoti_load_package`

  #145730 commented on Feb 28, 2025 • 0 new comments

• torch.device context manager change doesn't show in torch.get_default_device

  #131328 commented on Mar 6, 2025 • 0 new comments

• [RFC] Test Cases Enabling for Accelerators

  #146898 commented on Mar 4, 2025 • 0 new comments

• DISABLED test_reentrant_parent_error_on_cpu_cuda (__main__.TestAutogradDeviceTypeCUDA)

  #86735 commented on Mar 4, 2025 • 0 new comments

• partitioner doesn't appear to respect SAC region

  #128730 commented on Mar 3, 2025 • 0 new comments

• Add a hash function for tensor data

  #2569 commented on Mar 3, 2025 • 0 new comments

• SerializeError for ScriptObject in AOTInductor

  #147283 commented on Mar 3, 2025 • 0 new comments

• [ONNX] slice complex tensor needs implementation

  #147896 commented on Mar 3, 2025 • 0 new comments

• Inconsistent results from `is_compile_supported ` with equivalent device identifiers

  #147826 commented on Mar 3, 2025 • 0 new comments

• [BUG][PyTorch 2.0 Export][quant]:get_source_partitions() may return different matches with same input graph

  #147170 commented on Mar 3, 2025 • 0 new comments

• [dynamo] Activation checkpointing tests erroring at runtime

  #127115 commented on Mar 3, 2025 • 0 new comments

• Support AC with graph break

  #139989 commented on Mar 3, 2025 • 0 new comments

• Compiled Autograd + Activation Checkpointing/Offloading

  #143176 commented on Mar 3, 2025 • 0 new comments

• CUDA error when compiling loss function

  #143774 commented on Mar 3, 2025 • 0 new comments

• Partitioner's auto-AC misbehaves with mixed dtypes

  #144470 commented on Mar 3, 2025 • 0 new comments

• CheckpointError with `torch.distributed.algorithms._checkpoint.checkpoint_wrapper` and `torch.compile`

  #144637 commented on Mar 3, 2025 • 0 new comments

• Activation Checkpointing composability with split backward computation

  #145511 commented on Mar 3, 2025 • 0 new comments

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

  #24422 commented on Mar 3, 2025 • 0 new comments

• Missing grad_fn information while torch.compile with customized gradient function

  #131974 commented on Mar 3, 2025 • 0 new comments

• `RuntimeError` not raised for `out=` argument in `torch.tensordot` with `requires_grad` tensors

  #147846 commented on Mar 3, 2025 • 0 new comments

• Unable to checkpoint model and optimizer state when using Hybrid Sharding Strategy

  #102904 commented on Mar 3, 2025 • 0 new comments

• [DCP] Native S3/object storage interface

  #116198 commented on Mar 3, 2025 • 0 new comments

• [DCP] Review and update examples and docstring to reflect DCP save/load API updates

  #119070 commented on Mar 3, 2025 • 0 new comments

• Distributed checkpoint state_dict load may report nccl error and hide the real root-cause exception

  #122529 commented on Mar 3, 2025 • 0 new comments

• `torch_save_to_dcp`'s produced dcp checkpoint doesn't fit the load interface for FSDP

  #126047 commented on Mar 3, 2025 • 0 new comments

• [DCP] DCP does not support objects which are lazy initialized.

  #126881 commented on Mar 3, 2025 • 0 new comments

• PyTorch's Distributed Checkpoint Cannot Save a Parameter of Size 1

  #132366 commented on Mar 3, 2025 • 0 new comments

• Incomplete Parameter Gathering on Rank 0 with FSDP Model Saving

  #136950 commented on Mar 3, 2025 • 0 new comments

• [torch2.4][Distributed Checkpoint] new flatten logic is error-prone

  #137327 commented on Mar 3, 2025 • 0 new comments

• FSDP learning hangs when the program tries to save the model

  #143536 commented on Mar 3, 2025 • 0 new comments

• What is the recommended way to use Distributed Checkpointing Save/Load with HSDP?

  #145978 commented on Mar 3, 2025 • 0 new comments

• [torch.export] RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cpu!

  #126674 commented on Mar 3, 2025 • 0 new comments

• DISABLED test_linear_backward_memory_usage_cuda_float32 (__main__.TestNestedTensorSubclassCUDA)

  #141292 commented on Feb 28, 2025 • 0 new comments

• Hooks not working in version 2.0.1+cu118

  #102374 commented on Feb 28, 2025 • 0 new comments

• DISABLED test_rng (__main__.TestCompilerBisector)

  #139590 commented on Feb 28, 2025 • 0 new comments

• DISABLED test_while_loop_schema_gen (__main__.TestHopSchema)

  #141202 commented on Feb 28, 2025 • 0 new comments

• CTCLoss gradient is incorrect

  #52241 commented on Feb 28, 2025 • 0 new comments

• DISABLED test_reorder_peak_memory_dfs (__main__.TestOperatorReorderForPeakMemory)

  #145183 commented on Feb 28, 2025 • 0 new comments

• AdamW is CPU-bottlenecked with FSDP2, with slow foreach kernels

  #134191 commented on Feb 28, 2025 • 0 new comments

• DISABLED test_split (__main__.AutoFunctionalizeTests)

  #148080 commented on Feb 28, 2025 • 0 new comments

• [discussion] Support other index dtypes for scatter, scatter_reduce and other indexing functions in addition to int64: uint8, int16, int32, uint16 etc (without casting copies/reallocations)

  #61819 commented on Feb 28, 2025 • 0 new comments

• Potential rooms for fewer recompilations by introducing higher-level guards

  #143670 commented on Feb 27, 2025 • 0 new comments

• DISABLED test_item_to_inputs_kernel_nobreak_cuda (__main__.TestInductorDynamicCUDA)

  #119538 commented on Feb 27, 2025 • 0 new comments

• compilation error on SequenceParallel'ed Dropout

  #147757 commented on Feb 27, 2025 • 0 new comments

• More informative variable names in AOTAutograd

  #110700 commented on Feb 27, 2025 • 0 new comments

• `torch.mul` uses `OpMathType` for computation.

  #147134 commented on Feb 27, 2025 • 0 new comments

• DISABLED test_int64_upsample3d_cuda_bfloat16 (__main__.TestTorchDeviceTypeCUDA)

  #146007 commented on Feb 27, 2025 • 0 new comments

• make_graphed_callables don't work with FSDP at all even on a simple network

  #127225 commented on Feb 27, 2025 • 0 new comments

• Let torch.compiler.allow_in_graph work in more situations

  #140437 commented on Feb 27, 2025 • 0 new comments

• inspect.Signature with functools.partial partially applying tensors doesn't work

  #139374 commented on Feb 27, 2025 • 0 new comments

• require_exact_stride better handling of expanded dims

  #147156 commented on Feb 27, 2025 • 0 new comments

• Failed to Open libnvrtc-builtins.so.11.7

  #93134 commented on Feb 27, 2025 • 0 new comments

• Partitioner stores fp8 copy of all weights between fwd and bwd, causing OOM

  #141881 commented on Feb 27, 2025 • 0 new comments

• Checkpoint doesn't work with torch_function if torch_function change tensor metadata

  #147995 commented on Feb 27, 2025 • 0 new comments

• Run Performance Regression Tests on new Version of Triton

  #148045 commented on Feb 27, 2025 • 0 new comments

• [RFC] Customization of ElasticAgent for fault-tolerance and node-replacement in big ddp job

  #148048 commented on Feb 27, 2025 • 0 new comments

• Support alpha=inf consistently for torch.celu

  #148065 commented on Feb 27, 2025 • 0 new comments

• DISABLED test_recompile (__main__.AutoFunctionalizeTests)

  #148013 commented on Feb 27, 2025 • 0 new comments

• Feature Request: CUDA-Optimized Queue Buffer for PyTorch

  #148077 commented on Feb 27, 2025 • 0 new comments

• Track follow ups to #147354

  #147867 commented on Feb 27, 2025 • 0 new comments

• torch/_inductor/cpp_builder.py : _is_gcc Function Incorrectly Classifies clang++ as g++

  #146712 commented on Feb 27, 2025 • 0 new comments

• DISABLED test_slice (__main__.AutoFunctionalizeTests)

  #148035 commented on Feb 27, 2025 • 0 new comments

• [RFC] Cuda support matrix for Release 2.7

  #145544 commented on Mar 2, 2025 • 0 new comments

• MPS Error on sequoia 15.3: NDArray dimension length > INT_MAX'

  #146769 commented on Mar 2, 2025 • 0 new comments

• Inductor-CPU might load (and store) fewer elements than the vector-width

  #146824 commented on Mar 1, 2025 • 0 new comments

• [typing] Add static type hints to `torch.distributions`.

  #144196 commented on Mar 1, 2025 • 0 new comments

• torch.export does not support torchaudio.transforms.Spectrogram

  #112844 commented on Mar 1, 2025 • 0 new comments

• FlexAttention uses much more GPU memory than FlashAttention-2

  #144537 commented on Mar 1, 2025 • 0 new comments

• Unable to print in a branch run by torch.cond

  #147115 commented on Mar 1, 2025 • 0 new comments

• The value of requires_grad is not set when creating the tensor using TensorMaker

  #146419 commented on Mar 1, 2025 • 0 new comments

• General MPS op coverage tracking issue

  #77764 commented on Mar 1, 2025 • 0 new comments

• Error using the "index_put_" function

  #124288 commented on Mar 1, 2025 • 0 new comments

• PyPy support

  #17835 commented on Mar 1, 2025 • 0 new comments

• DISABLED test_weight_norm_bwd_dynamic_shapes_cpu (__main__.DynamicShapesCpuTests)

  #141484 commented on Mar 1, 2025 • 0 new comments

• DISABLED test_embedding_bag_dynamic_shapes_cpu (__main__.DynamicShapesCpuTests)

  #135215 commented on Mar 1, 2025 • 0 new comments

• DISABLED test_embedding_dynamic_shapes_cpu (__main__.DynamicShapesCodegenCpuTests)

  #135456 commented on Mar 1, 2025 • 0 new comments

• DISABLED test_wait_tensor (__main__.CompileTest)

  #148014 commented on Feb 28, 2025 • 0 new comments

• DISABLED test_partitioning_unremat_bw (__main__.MinCutPartitioningTests)

  #145343 commented on Feb 28, 2025 • 0 new comments

• DistributedDataParallel with compile(..., mode="max-autotune") hangs in 2.5+

  #140395 commented on Feb 28, 2025 • 0 new comments

• Improving expand w/ unbacked symints

  #128645 commented on Feb 28, 2025 • 0 new comments

• compile_fx inplace modifly the input graph

  #138980 commented on Feb 28, 2025 • 0 new comments

• DTensor support for fused qkv matmul

  #140069 commented on Feb 28, 2025 • 0 new comments

• torch.nn.AvgPool2d fails with stride >= 2^31 on CUDA

  #147613 commented on Feb 28, 2025 • 0 new comments

• [feature][cudagraph] API to clear a bad recording

  #127147 commented on Feb 28, 2025 • 0 new comments

• [compiled autograd][aot autograd] accumulate grad (on param with non empty grad) mutates inputs and prevents cudagraph

  #126938 commented on Feb 28, 2025 • 0 new comments

• autograd.function with `setup_context` has a number of issues with `torch.compile`

  #130051 commented on Feb 28, 2025 • 0 new comments

• inductor error when torch.compile on distrifuser

  #128073 commented on Feb 28, 2025 • 0 new comments

• Support AOT Autograd level Caching

  #125958 commented on Feb 28, 2025 • 0 new comments

• torch._dynamo.exc.UserError: Could not guard on data-dependent expression Eq(256*u0, 256) (unhinted: Eq(256*u0, 256)).

  #147672 commented on Feb 28, 2025 • 0 new comments

• AOTAutogradCache implementation

  #128234 commented on Feb 28, 2025 • 0 new comments

• Triton pin update for PyTorch 2.7 / Triton 3.3: Upgrading PyTorch-Triton to a version that Supports Blackwell

  #146518 commented on Feb 28, 2025 • 0 new comments

• torch.export.export creates guards that denies exporting.

  #147623 commented on Feb 28, 2025 • 0 new comments

• [inductor] [cpu] `torch.scatter` throws `AttributeError: 'int' object has no attribute 'find'` on CPP backend

  #148058 commented on Feb 28, 2025 • 0 new comments

• Sweep through Potentially BC Breaking Commits in Triton

  #148044 commented on Mar 5, 2025 • 0 new comments

• Add docs for __tensor_flatten__ / __tensor_unflatten__

  #113089 commented on Mar 5, 2025 • 0 new comments

• [dynamo] Replace `unimplemented` with `unimplemented_v2`

  #147913 commented on Mar 5, 2025 • 0 new comments

• [compile] DDPOptimizer + activation checkpointing not supported

  #104674 commented on Mar 5, 2025 • 0 new comments

• Preload CUDA fails if CUDA libs in different PYTHONPATH

  #147001 commented on Mar 5, 2025 • 0 new comments

• Looking for valid compiling option for extension based on torch-2.1.0+cpu.cxx11.abi

  #143780 commented on Mar 5, 2025 • 0 new comments

• Torch Onnx Export (with Dynamo) does not recognize `Remainder` function

  #147973 commented on Mar 5, 2025 • 0 new comments

• Scaled Dot-Product Attention Invalid Configuration Error on Large batch size

  #142228 commented on Mar 5, 2025 • 0 new comments

• [ONNX] Run report_exportability when report=True

  #139904 commented on Mar 5, 2025 • 0 new comments

• [torchbench] torch._dynamo.exc.Unsupported: Graph break due to unsupported builtin None.morphologyEx

  #145088 commented on Mar 5, 2025 • 0 new comments

• Immediate Global State Mutation After Using `_force_original_view_tracking` Decorator

  #147849 commented on Mar 5, 2025 • 0 new comments

• Torch 2.6.0 cu126 is missing several dependencies in the METADATA-file

  #146679 commented on Mar 5, 2025 • 0 new comments

• Assertion error in Flex Attention backward pass when indexing a parameter

  #146896 commented on Mar 5, 2025 • 0 new comments

• Symbolic shape fails symbol_to_source when caching is enabled

  #127970 commented on Mar 5, 2025 • 0 new comments

• `torch.compile` and complex numbers

  #125718 commented on Mar 5, 2025 • 0 new comments

• libtorch_cuda_linalg.so: undefined symbol: mkl_lapack_dsbrdbn on a source built PyTorch 2.6.0 with USE_STATIC_MKL=1 on CUDA platform

  #146551 commented on Mar 5, 2025 • 0 new comments

• On Linux, passing torch.Generator to multiprocessing.Process crashes for forkserver and spawn start method

  #146828 commented on Mar 5, 2025 • 0 new comments

• [CUDA][Blackwell] Blackwell Tracking Issue

  #145949 commented on Mar 5, 2025 • 0 new comments

• Enable more flake8-bugbear lints

  #106571 commented on Mar 5, 2025 • 0 new comments

• Custom operators registered via decorator slower than ops registered via `torch.Library.{define, impl}`

  #139500 commented on Mar 5, 2025 • 0 new comments

• [inductor] [silence] `nn.ConvTranspose2d-F.dropout` outputs inconsistent results with eager

  #148061 commented on Mar 5, 2025 • 0 new comments

• DISABLED test_profiler_mark_wrapper_call_dynamic_shapes_cpu (__main__.DynamicShapesCpuTests)

  #135294 commented on Mar 5, 2025 • 0 new comments

• DISABLED test_pattern_matcher_multi_user_dynamic_shapes_cpu (__main__.DynamicShapesCpuTests)

  #134433 commented on Mar 5, 2025 • 0 new comments

• DISABLED test_max_autotune (__main__.TestFlexAttention)

  #147200 commented on Mar 5, 2025 • 0 new comments

• Video-Llama (version 1) runs much slower using Float16 than Float32 on Kunpeng CPU

  #148078 commented on Mar 5, 2025 • 0 new comments

• Tensor parallel for convolutions and groupnorm

  #133221 commented on Mar 5, 2025 • 0 new comments

• RuntimeError: expect_autograd_hooks_ INTERNAL ASSERT FAILED at "../torch/csrc/distributed/c10d/reducer.cpp"

  #143580 commented on Mar 5, 2025 • 0 new comments

• Pytorch build fail with GCC 14.1.0 due to third_party/fbgemm/src/UtilsAvx512.cc:970:35: error: ‘r’ may be used uninitialized [-Werror=maybe-uninitialized]

  #129358 commented on Mar 5, 2025 • 0 new comments

• [inductor][cpu]DebertaV2ForMaskedLM, DebertaV2ForQuestionAnswering and eca_halonext26ts max_autotune accuracy failure in 2025-02-24 nightly release

  #148074 commented on Mar 5, 2025 • 0 new comments

• torch._check doesn't work for .item() then select

  #147772 commented on Mar 4, 2025 • 0 new comments

• Load cuda deps more aggressively

  #137059 commented on Mar 5, 2025 • 0 new comments

• Enabling ATen Distribution kernels for AARCH64 using OpenRNG

  #134942 commented on Mar 5, 2025 • 0 new comments

• Avoid args being parsed when common_utils imported

  #134592 commented on Mar 4, 2025 • 0 new comments

• Update Doc for Intel XPU Profiling

  #134515 commented on Mar 4, 2025 • 0 new comments

• [Don't Merge] Try to build custom ops with MKL XPU

  #133658 commented on Feb 28, 2025 • 0 new comments

• Make IPC features extendable on third-party devices

  #133222 commented on Mar 4, 2025 • 0 new comments

• autograd codegen: bump VC properly for mutable ops with no returns

  #133044 commented on Feb 28, 2025 • 0 new comments

• [torch.special] Adding betainc, betaincc, betaincinv, betainccinv, betaln and beta with backward operation

  #132135 commented on Mar 6, 2025 • 0 new comments

• [pytree] implement key path APIs for CXX pytree

  #130141 commented on Feb 28, 2025 • 0 new comments

• [CI] Run `lintrunner` on generated `.pyi` stub files in CI

  #129887 commented on Feb 28, 2025 • 0 new comments

• Add `__all__` to `torch/nn/functional.pyi` and `torch/return_types.pyi`

  #129872 commented on Feb 28, 2025 • 0 new comments

• [torchgen] Refactor `torchgen.utils.FileManager` to accept `pathlib.Path`

  #129871 commented on Feb 28, 2025 • 0 new comments

• Use Generic TypeAlias (PEP 585) and Union Type (PEP 604) in generated `.pyi` stub files

  #129420 commented on Feb 28, 2025 • 0 new comments

• Enable Leak Sanitizer

  #127171 commented on Mar 3, 2025 • 0 new comments

• [inductor] online softmax

  #127011 commented on Mar 4, 2025 • 0 new comments

• [AOTAutograd] tweak min-cut partitioner to avoid saving softmax output

  #126348 commented on Feb 28, 2025 • 0 new comments

• [vision hash update] update the pinned vision hash

  #125806 commented on Mar 6, 2025 • 0 new comments

• Automated submodule update: FBGEMM

  #115316 commented on Mar 6, 2025 • 0 new comments

• [pytree] support PyStructSequence types for Python pytree

  #113258 commented on Mar 3, 2025 • 0 new comments

• Automated submodule update: kineto

  #106149 commented on Mar 5, 2025 • 0 new comments

• NotImplementedError: Output channels > 65536 not supported at the MPS device.

  #144445 commented on Mar 6, 2025 • 0 new comments

• [Feature Request] Release original parameters by layer when turning on `freezing_discard_parameters`

  #147062 commented on Mar 6, 2025 • 0 new comments

• Pytorch DDP across nodes: self._store = TCPStore( # type: ignore[call-arg] RuntimeError: Stop_waiting response is expected

  #114357 commented on Mar 6, 2025 • 0 new comments

• Deprecation of NVTX 2 (`nvToolsExt`): Recommended to move to NVTX 3

  #147011 commented on Mar 6, 2025 • 0 new comments

• DISABLED test_remote_cache_load_function_device_cuda_float32_dynamic_False_bundle_triton_True (__main__.TestFxGraphCache)

  #145191 commented on Mar 6, 2025 • 0 new comments

• [RFC] Request for Feedback and Review on PRs Adding RISC-V and RVV Support

  #147513 commented on Mar 6, 2025 • 0 new comments

• DISABLED test_tensor_dtype_complex (__main__.CommTest)

  #112460 commented on Mar 6, 2025 • 0 new comments

• [ONNX] BitwiseOr was generated for bool inputs (invalid)

  #147854 commented on Mar 6, 2025 • 0 new comments

• Enable CUDA 12.8.0, Disable CUDA 12.4

  #145570 commented on Mar 6, 2025 • 0 new comments

• AttributeError: Can't pickle local object 'make_opaque_bitwise_fn.<locals>.BitwiseFn'

  #147841 commented on Mar 6, 2025 • 0 new comments

• python3 -m torch.utils.collect_env not providing expected output.

  #147669 commented on Mar 6, 2025 • 0 new comments

• DISABLED test_nonstrict_trace_nested_custom_class_error (__main__.DecoratorTests)

  #148031 commented on Mar 4, 2025 • 0 new comments

• DISABLED test_nonstrict_trace_newly_constructed_custom_class_with_side_effects (__main__.DecoratorTests)

  #148032 commented on Mar 4, 2025 • 0 new comments

• DISABLED test_nonstrict_trace_nested_custom_class (__main__.DecoratorTests)

  #148033 commented on Mar 4, 2025 • 0 new comments

• DISABLED test_nonstrict_trace_no_action_at_a_distance (__main__.DecoratorTests)

  #148034 commented on Mar 4, 2025 • 0 new comments

• DISABLED test_nonstrict_trace_on_method (__main__.DecoratorTests)

  #148054 commented on Mar 4, 2025 • 0 new comments

• DISABLED test_nonstrict_trace_inside_compiled_function_kwarg (__main__.DecoratorTests)

  #148055 commented on Mar 4, 2025 • 0 new comments

• DISABLED test_nonstrict_trace_pre_existing_custom_class_with_side_effects (__main__.DecoratorTests)

  #148056 commented on Mar 4, 2025 • 0 new comments

• PyTorch VS2022 official build Windows binary illegal instruction on AVX2(max ISA level) CPU

  #145702 commented on Mar 4, 2025 • 0 new comments

• [inductor] Add Python type annotations to `torch/_inductor`

  #146167 commented on Mar 4, 2025 • 0 new comments

• DISABLED test_per_sample_api_compute_batch_size_not_pytreeable_cpu (__main__.TestExpandedWeightModuleCPU)

  #146972 commented on Mar 4, 2025 • 0 new comments

• torch.compile() on quantized model: No attribute "meta"

  #148072 commented on Mar 4, 2025 • 0 new comments

• Triton Error [CUDA]: invalid device context when autograd.backward a triton kernel

  #124565 commented on Mar 4, 2025 • 0 new comments

• Loading weights using `torch.distributed.checkpoint` leads to large loss values

  #145378 commented on Mar 4, 2025 • 0 new comments

• incorrect _unsafe_index meta

  #139312 commented on Mar 4, 2025 • 0 new comments

• DISABLED test_cache_load_function_device_cuda_float32_dynamic_False_bundle_triton_True_grad_True (__main__.TestFxGraphCache)

  #145336 commented on Mar 4, 2025 • 0 new comments

• TorchInductor CPU Performance Dashboard

  #93531 commented on Mar 4, 2025 • 0 new comments

• Support FP16 accumulation for faster LLM inference on 4090 like GPUs

  #123558 commented on Mar 4, 2025 • 0 new comments

• gradient checkpointing with use_reentrant=False cannot reduce peak memory

  #147449 commented on Mar 4, 2025 • 0 new comments

• [inductor][user triton] support on-device TMA / tensor descriptor API

  #148052 commented on Mar 4, 2025 • 0 new comments

• Adjust test_mm_triton_kernel_benchmark for unpadded tensors

  #147999 commented on Mar 4, 2025 • 0 new comments

• torch.compile supported with GIL disabled

  #147946 commented on Mar 4, 2025 • 0 new comments

• redundant recompilation caused by duplicated Sym()

  #144068 commented on Mar 4, 2025 • 0 new comments

• Strange recompilations on torch 2.5 + FSDP + UNet

  #138813 commented on Mar 4, 2025 • 0 new comments

• automatic_dynamic_shapes for mark_unbacked

  #136605 commented on Mar 4, 2025 • 0 new comments

• dynamo (re)compilation issues: shape (1,1), nn.Parameter, mark_dynamic

  #135011 commented on Mar 4, 2025 • 0 new comments

• Avoid recompilation for inputs integer number

  #132849 commented on Mar 4, 2025 • 0 new comments

• StableDiffusion with dynamic=True still recompiles

  #104913 commented on Mar 4, 2025 • 0 new comments

• Recompilation triggered at each step of the loop involving array indexing

  #114293 commented on Mar 4, 2025 • 0 new comments

• torch.compile support for SeamlessExpressivity/SeamlessM4T in fairseq2

  #114373 commented on Mar 4, 2025 • 0 new comments

• DynamicInt helper structure that is equivalent to mark_dynamic on an int

  #129623 commented on Mar 4, 2025 • 0 new comments

• Verifier (in torch.export.export) does not make use of if-condition inside branches

  #147991 commented on Mar 4, 2025 • 0 new comments

• Build Triton for aarch64

  #130558 commented on Mar 4, 2025 • 0 new comments

• `torch.device(0)` makes CUDA init fail in subprocess since `2.5.0`

  #144152 commented on Mar 4, 2025 • 0 new comments

• Comm reordering can make Inductor use variable before its definition

  #147328 commented on Mar 4, 2025 • 0 new comments

• Web Page do not match the original documentation

  #146683 commented on Mar 4, 2025 • 0 new comments

• [ROCm] scaled_dot_product_attention using mem-efficient backend (aotriton) produces wrong outputs with custom attn_mask on torch 2.6.0+rocm6.2.4

  #147460 commented on Mar 4, 2025 • 0 new comments

• `Illegal instruction (core dumped)` on Raspberry Pi 4 when exporting ONNX with `torch 2.6.0`

  #146792 commented on Mar 4, 2025 • 0 new comments

• XPU - UserWarning: Failed to initialize XPU devices. when run on the host without Intel GPU Driver

  #145433 commented on Mar 4, 2025 • 0 new comments

• CI/CD: Figure out what to do with split build

  #138750 commented on Mar 4, 2025 • 0 new comments

• torch_cuda.dll was built failed to link _cudnn_attention_forward

  #147671 commented on Mar 4, 2025 • 0 new comments

• DISABLED test_inductor_all_gather_into_tensor_coalesced (__main__.CompileTest)

  #146806 commented on Mar 4, 2025 • 0 new comments

• Publish pytorch RC docker images before release

  #145925 commented on Mar 4, 2025 • 0 new comments

• unbind_copy opinformation cause exception while running test_dtensor_ops.py

  #147814 commented on Mar 4, 2025 • 0 new comments

• [inductor] [silence] inconsistent swap wih eager when compiling `torch.rot90-torch.randn_like`

  #147847 commented on Mar 4, 2025 • 0 new comments

• Flex Attention is incompatible with selective AC

  #147879 commented on Mar 4, 2025 • 0 new comments

• torch.compile with the inductor backend slows down (exponentially?) for certain graphs

  #148073 commented on Mar 4, 2025 • 0 new comments

• DISABLED test_slice_dynamic (__main__.AutoFunctionalizeTests)

  #148067 commented on Mar 4, 2025 • 0 new comments

• torch.compile for division gives different numeric output vs eager mode division: torch.tensor/torch.tensor

  #141753 commented on Mar 4, 2025 • 0 new comments

• torch.compile mode="max-autotune" precision appears to be lower

  #96693 commented on Mar 4, 2025 • 0 new comments

• [RFE][Distributed][NCCL] A feature request for stream management API in PG NCCL

  #147729 commented on Mar 4, 2025 • 0 new comments

• context_parallel fails with plain sdpa kernel SDPBackend.MATH

  #147793 commented on Mar 4, 2025 • 0 new comments

• Using DTensor with device meshes that use different devices for input and output

  #126795 commented on Mar 4, 2025 • 0 new comments

• Memory leak when using SequentialLR and ChainedLR schedulers

  #126131 commented on Mar 4, 2025 • 0 new comments

• MX basic dtypes in pytorch/pytorch

  #146414 commented on Mar 4, 2025 • 0 new comments

• grad_fn function disobeys broadcast rules

  #144228 commented on Mar 4, 2025 • 0 new comments

• AOT subclass desugaring adds static input arguments to inductor

  #130502 commented on Mar 4, 2025 • 0 new comments

• [ONNX] GNN model inaccuracy: scatter_reduce need to be fixed

  #147617 commented on Mar 4, 2025 • 0 new comments

• roundtrip cast between float32|bfloat16 and e8m0 should work in torchinductor

  #147875 commented on Mar 4, 2025 • 0 new comments

• Error : torch/utils/_sympy/interp.py:176] [0/2] failed while executing pow_by_natural([VR1, int_oo], VR[-1, -1]])

  #148003 commented on Mar 4, 2025 • 0 new comments

• returning tensors of dtype torch.float8_e8m0fnu should work with torchinductor

  #147873 commented on Mar 4, 2025 • 0 new comments

• [inductor] `torch.slice_scatter` throws `AssertionError` when meeting internal `float32`

  #147842 commented on Mar 4, 2025 • 0 new comments