21 Pull requests merged by 7 people

• [aarch64] Add sm_80 to CUDA SBSA build

  #158118 merged Jul 11, 2025

• [user triton] AOT inductor support for device-side TMA

  #157241 merged Jul 11, 2025

• Add sm_70 arch for linux cuda 12.8 and 12.9 builds

  #157968 merged Jul 11, 2025

• Revert "Turn on compile with NVSHMEM (#154538)"

  #158040 merged Jul 11, 2025

• Revert "Add NVSHMEM to PYTORCH_EXTRA_INSTALL_REQUIREMENTS (#154568)"

  #158039 merged Jul 11, 2025

• [cherry-pick] revert #156552

  #156767 merged Jul 10, 2025

• cherrypick revert of #152932 for release 2.8

  #158031 merged Jul 10, 2025

• [inductor][user triton] sanitize triple-quoted docstrings in kernel definitions

  #157454 merged Jul 9, 2025

• [release] Triton pin update to 3.4

  #157752 merged Jul 8, 2025

• Bump urllib3 from 2.2.2 to 2.5.0 in /tools/build/bazel

  #156390 merged Jul 8, 2025

• [inductor][static launcher] Skip correctness test for test_floats

  #157200 merged Jul 7, 2025

• [ONNX] Bump onnxscript api for torch 2.8

  #157137 merged Jul 7, 2025

• Fix macOS build with USE_MPS=OFF

  #156932 merged Jul 7, 2025

• [dynamo] do not issue lru_cache warning for functions in the top-level torch namespace

  #157718 merged Jul 7, 2025

• [dynamo] Fix source for lru_cache method

  #157308 merged Jul 7, 2025

• [cherry-pick] Organize BUCK for torch/standalone and Rename torch::standalone to headeronly

  #157418 merged Jul 7, 2025

• [PowerPC] Fixed build issue for vsx vec256 complexfloat and scaled_mm_out_cpu

  #157422 merged Jul 7, 2025

• [ONNX] Fix conversion of attention - 4D

  #157509 merged Jul 7, 2025

• [dynamo] Fix bug in dict(mapping_proxy)

  #157515 merged Jul 7, 2025

• [cherry-pick] [fake tensor] fix issue of no attribute tags (#156689)

  #157519 merged Jul 7, 2025

• Add einops x torch.compile testing in PyTorch CI (#157416)

  #157588 merged Jul 7, 2025

185 Pull requests opened by 106 people

• [pt2 event logging] add configurable prefix

  #157678 opened Jul 6, 2025

• [Inductor][Float8] Add float8_e4m3fn into assertion dtype list.

  #157684 opened Jul 7, 2025

• [BE] add `SHFMT` linter to format shell scripts

  #157685 opened Jul 7, 2025

• [BE][1/4] format shell scripts with `SHFMT`

  #157686 opened Jul 7, 2025

• [BE][2/4] format shell scripts with `SHFMT` in .circleci/ and .github/

  #157687 opened Jul 7, 2025

• [BE][3/4] format shell scripts with `SHFMT` in .ci/

  #157688 opened Jul 7, 2025

• [BE][4/4] format shell scripts with `SHFMT` in scripts/

  #157689 opened Jul 7, 2025

• [inductor] initial triton static config lookup table

  #157699 opened Jul 7, 2025

• wip, lookup table for reduction configs

  #157700 opened Jul 7, 2025

• Enhance APoT quantizer

  #157710 opened Jul 7, 2025

• Add support moviepy 2.x

  #157712 opened Jul 7, 2025

• Avoid writing temporary modules to disk

  #157713 opened Jul 7, 2025

• [c10d] Prototype of `group_split` for dist2 work

  #157716 opened Jul 7, 2025

• Address NaNs if SDPA is called with all values masked from query

  #157727 opened Jul 7, 2025

• Expose opt_einsum in torch.backends

  #157740 opened Jul 7, 2025

• Fuse matmul

  #157743 opened Jul 7, 2025

• Migrate pytorch-linux-jammy-cuda12.8-cudnn9-py3-gcc11 -> pytorch-linux-jammy-cuda12.8-cudnn9-py3-gcc13

  #157748 opened Jul 7, 2025

• [export] Update docs

  #157750 opened Jul 8, 2025

• Fix einops x torch.compile interaction

  #157754 opened Jul 8, 2025

• [SymmMem] Find NVSHMEM from system installation

  #157755 opened Jul 8, 2025

• [SymmMem] find_path does not search /usr/local/lib

  #157765 opened Jul 8, 2025

• Ensure large tensor int32 -> int64 indexing is enabled

  #157767 opened Jul 8, 2025

• [inductor][lookup_table] log entries

  #157768 opened Jul 8, 2025

• Enable _int_mm on Intel GPU

  #157769 opened Jul 8, 2025

• [BE]: Fix NVSHMEM builds, add missing 12.9 dependency and update to latest for 2.8RC

  #157774 opened Jul 8, 2025

• [Test][Do Not Merge] Update Ideep to latest oneDNN commit

  #157782 opened Jul 8, 2025

• S390X: fix detection of magic number placeholder in inductor

  #157784 opened Jul 8, 2025

• unskipped mobilenet_v3 quantization and mobilenet_v2 quantization plus tests from https://github.com/pytorch/pytorch/issues/125438

  #157786 opened Jul 8, 2025

• [BE]: Reduce binary size 40% using aggressive fatbin compression.

  #157791 opened Jul 8, 2025

• Add flag to fx.passes.split_module to normalize input names

  #157793 opened Jul 8, 2025

• [PP] Allow schedules to run under torch.no_grad()

  #157795 opened Jul 8, 2025

• [claude-code] Add top-level module doc for torch/distributed/tensor/_op_schema.py

  #157804 opened Jul 8, 2025

• feat(dynamo): raise UnsupportedError for ndarray.astype(object)

  #157810 opened Jul 8, 2025

• Fix broken docs requirements symlink

  #157811 opened Jul 8, 2025

• Allow docker builds to deal with symlinks

  #157812 opened Jul 8, 2025

• Make functorch notebook symlinks PEP 517 valid

  #157813 opened Jul 8, 2025

• Improve MANIFEST.in for source distribution

  #157814 opened Jul 8, 2025

• Add PEP 517 compliant Python source distribution to release process

  #157815 opened Jul 8, 2025

• [canary] dedupe args + on by default

  #157817 opened Jul 8, 2025

• [Inductor] [Triton] Enabling TMA for flex-attention for supported device types

  #157822 opened Jul 8, 2025

• [WIP][Inductor][Intel GPU] Always use channel last for only freezing mode.

  #157828 opened Jul 8, 2025

• [SymmMem] Avoid library mismatch in CMake search

  #157836 opened Jul 8, 2025

• partial reads

  #157838 opened Jul 8, 2025

• remove allow-untyped-defs from torch/ao/pruning/sparsifier/nearly_diagonal_sparsifier.py

  #157846 opened Jul 8, 2025

• remove allow-untyped-defs from torch/_higher_order_ops/run_const_graph.py

  #157847 opened Jul 8, 2025

• remove allow-untyped-defs from torch/ao/nn/intrinsic/quantized/dynamic/modules/linear_relu.py

  #157848 opened Jul 8, 2025

• remove allow-untyped-defs from torch/fx/passes/backends/cudagraphs.py

  #157849 opened Jul 8, 2025

• remove allow-untyped-defs from torch/utils/tensorboard/_onnx_graph.py

  #157850 opened Jul 8, 2025

• remove allow-untyped-defs from torch/nn/utils/_expanded_weights/linear_expanded_weights.py

  #157851 opened Jul 8, 2025

• remove allow-untyped-defs from torch/ao/quantization/experimental/apot_utils.py

  #157852 opened Jul 8, 2025

• remove allow-untyped-defs from torch/autograd/_functions/utils.py

  #157853 opened Jul 8, 2025

• Make compiled mode broadcasting use 0 strides

  #157854 opened Jul 8, 2025

• Add functions to setup PrivateUse1 as a python backend device.

  #157859 opened Jul 8, 2025

• ScannedModule

  #157864 opened Jul 8, 2025

• [Draft][CUDA][CI] Test B200 Runner with Nightly Inductor Perf Test

  #157870 opened Jul 9, 2025

• [MPS] Improve performance of max_pool3d

  #157875 opened Jul 9, 2025

• [MPS] Move max_pool2d to Metal

  #157876 opened Jul 9, 2025

• [Inductor][Triton] Update TMA Compatibility Requirements

  #157881 opened Jul 9, 2025

• use maybe_mark_dynamic instead of mark_dynamic for -dynamic-batch-only option

  #157885 opened Jul 9, 2025

• Back out "[Inductor] Fix epilogue fusion decision with 1 Triton caller as choice"

  #157887 opened Jul 9, 2025

• [profiler] update CUDA runtime kernel identification logic

  #157890 opened Jul 9, 2025

• Partitioner: Fix to align partition node order with original graph

  #157892 opened Jul 9, 2025

• [CPU] Support GQA for flash attention

  #157893 opened Jul 9, 2025

• [Inductor] optimize welford reduction

  #157902 opened Jul 9, 2025

• [inductor] [cpu] fix the dype hardcoded to int64 in store_reduction

  #157904 opened Jul 9, 2025

• Fix logdet returning finite values for singular matrices on CUDA

  #157910 opened Jul 9, 2025

• [autograd] Avoid creating and recording event when unnecessary

  #157914 opened Jul 9, 2025

• Normalize placeholder names in AOTAutogradCache

  #157916 opened Jul 9, 2025

• Add type assert for tensor_meta, based on real bug in autoparallel.

  #157927 opened Jul 9, 2025

• [WIP] try compress-mode balance on CUDA12.6

  #157928 opened Jul 9, 2025

• Reduce random reads for offset metadata when calling torch.load under FakeTensorMode

  #157931 opened Jul 9, 2025

• Adding a change to kick off the theme pull

  #157932 opened Jul 9, 2025

• [WIP] WindowsArm64 CI changes

  #157935 opened Jul 9, 2025

• Add cuda 12.9 periodic tests

  #157939 opened Jul 9, 2025

• Fix early return in `_EmptyStateDictLoadPlanner`

  #157940 opened Jul 9, 2025

• Small fix to the release-feature-request.yml

  #157941 opened Jul 9, 2025

• [WIP][dynamic shapes] unbacked-safe slicing

  #157944 opened Jul 9, 2025

• [AOTI][CPP] add flag TORCHINDUCTOR_CPP_FORCE_INLINE_KERNEL

  #157949 opened Jul 9, 2025

• [cherry-pick] temporarily disabling generation of weblinks for torch v2.8 & removing string literals for weblink generation

  #157951 opened Jul 9, 2025

• Allow dynamic shapes for DTensor slice

  #157953 opened Jul 9, 2025

• Refactor AOTInductorModelBase for Device Interface Abstraction

  #157954 opened Jul 9, 2025

• Add method get_fqn_to_index_mapping_from_metadata for HF storage reader

  #157956 opened Jul 9, 2025

• Add cuda 12.4 build in CI

  #157958 opened Jul 9, 2025

• improve typing in torch/test

  #157962 opened Jul 9, 2025

• [DCP][Prototype] Checkpoint replication via PGTransport

  #157963 opened Jul 9, 2025

• Ck recipe

  #157964 opened Jul 9, 2025

• [WIP]What does it take to avoid specializations in this program with out unbacked semantics,

  #157965 opened Jul 9, 2025

• [do not review][benchmark] test all_gather communication time

  #157967 opened Jul 9, 2025

• [dynamo] Support more basic output types for `nonstrict_trace`

  #157969 opened Jul 9, 2025

• [Do not land] Test NVSHMEM build

  #157970 opened Jul 9, 2025

• [dynamo, nested graph breaks] implement new resume frame stack/locals/cell layout convention

  #157971 opened Jul 9, 2025

• [Dynamo][Hierarchical Compile] Allow parameters to be propagated to submodules

  #157979 opened Jul 9, 2025

• Define header type directly instead of using DeviceStreamType

  #157982 opened Jul 10, 2025

• [dynamo, docs] programming model dynamo core concepts

  #157985 opened Jul 10, 2025

• [Dynamo] Use proper sources for constructing dataclass defaults

  #157993 opened Jul 10, 2025

• [test][do not merge] Upgrade oneDNN to v3.9

  #157994 opened Jul 10, 2025

• Slightly improve error message from repeat_interleave kernel

  #157996 opened Jul 10, 2025

• [WIP][export] turn on backed_size_oblivious

  #158004 opened Jul 10, 2025

• Bump requests from 2.32.2 to 2.32.4 in /tools/build/bazel

  #158006 opened Jul 10, 2025

• Enable nightly PT2 benchmark on B200

  #158011 opened Jul 10, 2025

• [inductor] consolidate common GEMM triton param retrieval

  #158015 opened Jul 10, 2025

• remove unnecessary sync point in AveragedModel update

  #158017 opened Jul 10, 2025

• Update upstream opinfo to generate appropriately scaled sample inputs

  #158018 opened Jul 10, 2025

• [TESTING] Update pin to disable AMD triton buffer ops

  #158019 opened Jul 10, 2025

• Add property and setter annotations for Tensor attribute stubs

  #158020 opened Jul 10, 2025

• [build] remove `wheel` from build requirements

  #158027 opened Jul 10, 2025

• [precompile][wip] Increment frame and add compile ids when loading packages

  #158028 opened Jul 10, 2025

• Migrate c10/macros/cmake_macros.h.in to torch/headeronly

  #158035 opened Jul 10, 2025

• Support DeepSeek-style blockwise scaling scaled-mm for fp8 on Hopper+

  #158037 opened Jul 10, 2025

• Tag CPython test files with the commit or tag they were copied from.

  #158038 opened Jul 10, 2025

• [1/N] support of replication fallback strategy

  #158046 opened Jul 10, 2025

• Move AOTI static linkage header file generation to cpp_wrapper_cpu

  #158047 opened Jul 10, 2025

• Still run TritonBundler with BundledAOTAutogradCache, save autotune results

  #158048 opened Jul 10, 2025

• Enable tracing `LOAD_BUILD_CLASS` on CPython tests

  #158049 opened Jul 10, 2025

• Documentation Fix: torch.empty_like memory preservation

  #158050 opened Jul 10, 2025

• [DTensor] have split_strategy return OpStrategy instead of TupleStrategy

  #158051 opened Jul 10, 2025

• [cutlass backend][BE] remove force disable cache in tests

  #158053 opened Jul 10, 2025

• [docs, dynamo] add fullgraph=True, common graph breaks docs

  #158055 opened Jul 10, 2025

• [dict] Support `dict.update()` with no args

  #158061 opened Jul 10, 2025

• [simple_fsdp][inductor_collectives] rewrite reorder_collectives, sink_waits_iterative

  #158062 opened Jul 10, 2025

• [WIP] Rewrite reorder collectives/sink_wait to preserve peak memory

  #158063 opened Jul 10, 2025

• adding types to nn module init

  #158065 opened Jul 10, 2025

• [hop] call mode.__dispatch__ when no mode reigstered for the hop

  #158067 opened Jul 10, 2025

• c10d inductor tests: do not re initialize dist process group

  #158068 opened Jul 10, 2025

• Add torch compile force disable caches alias

  #158072 opened Jul 10, 2025

• [benchmarks] Add scalar loss as model output when training

  #158074 opened Jul 10, 2025

• Grab bag of (mostly) typing improvements

  #158075 opened Jul 10, 2025

• [hop] add supports_higher_order_operators flag to TorchDispatchMode

  #158077 opened Jul 10, 2025

• Remove references to TorchScript in PyTorch docs.

  #158079 opened Jul 10, 2025

• updated test cases to use MultithreadTestCase

  #158082 opened Jul 11, 2025

• [pickle] Add polyfills for pickle

  #158084 opened Jul 11, 2025

• Refactor and Improve the OpenReg Module

  #158090 opened Jul 11, 2025

• [inductor] add template hashing for template lookup table

  #158091 opened Jul 11, 2025

• [simplefsdp auto-bucketing] add ir node bucket helper function

  #158097 opened Jul 11, 2025

• [simplefsdp auto-bucketing] add ir node reorder helper function

  #158098 opened Jul 11, 2025

• cache sympy.expand in sizevars simplify

  #158099 opened Jul 11, 2025

• try cache_on_self on ir.py Layout class __str__ method

  #158100 opened Jul 11, 2025

• [DO NOT MERGE] Stress test #1 new capacity.

  #158102 opened Jul 11, 2025

• [DO NOT MERGE] Stress test #2 new capacity.

  #158103 opened Jul 11, 2025

• [build] pin `setuptools>=77` to enable PEP 639

  #158104 opened Jul 11, 2025

• [BE][Easy] split build system `requirements.txt` to a separate file

  #158111 opened Jul 11, 2025

• [DTensor][BE] imporve DTensor ops correctness check utils

  #158112 opened Jul 11, 2025

• Fix AArch64 segfaults by disabling strict-aliasing in GridSamplerKernel for GCC 12 and above

  #158117 opened Jul 11, 2025

• [inductor_collectives] brute force preserve peak memory for sink_waits

  #158119 opened Jul 11, 2025

• Clarify that Store.check() was added in PyTorch v2.8

  #158124 opened Jul 11, 2025

• Add sm_70 to windows 12.9 build

  #158126 opened Jul 11, 2025

• [Optimus][fp8_activation_quantization] Only log when there's some node to be quantized

  #158129 opened Jul 11, 2025

• [DTensor] Assert DTensorSpec has valid placements

  #158133 opened Jul 11, 2025

• [Inductor] addmm + activation function fusion

  #158137 opened Jul 11, 2025

• Standalone compile API in _Exporter

  #158139 opened Jul 11, 2025

• [testing only] Layout optimization on

  #158141 opened Jul 11, 2025

• [testing only] Layout optimization off

  #158142 opened Jul 11, 2025

• Fix grouped MM output strides when compiled but not max-autotuned

  #158143 opened Jul 11, 2025

• Use a fixed size of a buffer in ShufflerIterDataPipe to not use append() and len()

  #158144 opened Jul 11, 2025

• [DO NOT MERGE] Add volta tests to periodic and pull

  #158145 opened Jul 11, 2025

• dist2: add support for passing custom configs directly to PG

  #158147 opened Jul 11, 2025

• [DRAFT][DELETE] Y do we have 7 build systems

  #158148 opened Jul 11, 2025

• Avoid AOTAutogradCache.load in stack trace on cache miss path

  #158149 opened Jul 11, 2025

• Inline dispatch_and_compile into its call site.

  #158150 opened Jul 11, 2025

• Modify c10::complex for CUDA 12.9 Win OOM

  #158151 opened Jul 11, 2025

• [CI] Update mobile build docker image

  #158153 opened Jul 11, 2025

• For discussion

  #158156 opened Jul 11, 2025

• [WIP] _convert_element_type_meta test what fails

  #158157 opened Jul 11, 2025

• [cutlass backend] cache a few things for codegen and properties

  #158158 opened Jul 11, 2025

• Fix torchrec multiprocess tests

  #158159 opened Jul 11, 2025

• Add transpose to torch/csrc/stable

  #158160 opened Jul 11, 2025

• [CI] Move main branch rocm binary builds to its own workflow

  #158161 opened Jul 11, 2025

• [CI] Do not run inductor rocm on ciflow/inductor

  #158162 opened Jul 11, 2025

• [CI][TD] Enable TD on all test configs

  #158163 opened Jul 11, 2025

• [ROCm] Fix tensor.item() for ROCm

  #158165 opened Jul 11, 2025

• [SymmMem] Fix NCCL Hang in NVSHMEM Triton Wait Until Test

  #158167 opened Jul 11, 2025

• [scan][cse] avoid cse zeros like gradient buffers

  #158168 opened Jul 12, 2025

• add eq function to NodeSource

  #158170 opened Jul 12, 2025

• [dynamo] turn off sys.monitoring if eval_frame is set

  #158171 opened Jul 12, 2025

• Pipeline _create_aot_dispatcher_function

  #158173 opened Jul 12, 2025

• Add inputs and outputs in Triton Kernel FX Graph segment

  #158174 opened Jul 12, 2025

• Hoist choose_dispatcher to top level, remove unnecessary returns

  #158176 opened Jul 12, 2025

• Update pr_time_benchmarks/expected_results.csv

  #158177 opened Jul 12, 2025

• [BE] Move repeated code into helper functions

  #158178 opened Jul 12, 2025

• [MPS] Extend atomic operations to more int types

  #158179 opened Jul 12, 2025

• don't error out in empty_cache under mempool context

  #158180 opened Jul 12, 2025

• Reproduce issue from #156097

  #158181 opened Jul 12, 2025

• [WIP][Inductor] Add cpu_max_other_dimension_decomposition for decompose_mm_pass

  #158183 opened Jul 12, 2025

• Fix compilation and "import torch" issues for cpython 3.14

  #158184 opened Jul 12, 2025

• Remove unnecessary CMake checks

  #158185 opened Jul 12, 2025

163 Issues closed by 39 people

• Inconsistent behavior between eager and compiled mode for `F.conv_transpose2d`

  #157909 closed Jul 12, 2025

• Question: Is it expected that `QuantStub` and `DeQuantStub` are skipped in `torch.compile`?

  #157998 closed Jul 12, 2025

• Discrepancy between Numpy and PyTorch advanced indexing

  #158134 closed Jul 12, 2025

• torch.compile with numpy code differs from numpy's behavior

  #157569 closed Jul 12, 2025

• dynamo+numpy 2.0 issue: TypeError: 'numpy.bool' object cannot be interpreted as an integer

  #157973 closed Jul 12, 2025

• Slicing of large tensors is wrong on MPS

  #153560 closed Jul 11, 2025

• torch.compile produces incorrect output

  #155690 closed Jul 11, 2025

• RuntimeError: Placeholder storage has not been allocated on MPS device! - when calling torch.export.export on mps

  #158121 closed Jul 11, 2025

• Bug in cmake/public/cuda.cmake: Incorrect use of set(${...}) leads to missing CUDA version in error message

  #157354 closed Jul 11, 2025

• [inductor] TorchInductor does not correctly recognize the grad status of model code

  #125474 closed Jul 11, 2025

• DISABLED test_op_has_batch_rule_linalg_vecdot_cuda_float32 (__main__.TestVmapOperatorsOpInfoCUDA)

  #142924 closed Jul 11, 2025

• DISABLED test_forward_backward (__main__.CudaGraphTreeTests)

  #156957 closed Jul 11, 2025

• DISABLED test_sdpa_with_packed_in_proj_cuda_float32 (__main__.TestNestedTensorSubclassCUDA)

  #120029 closed Jul 11, 2025

• DISABLED test_invalid_status_for_legacy_api (__main__.TestCuda)

  #157110 closed Jul 11, 2025

• DISABLED test_remove_noop_view_default_cpu (__main__.CpuTests)

  #151512 closed Jul 11, 2025

• DISABLED test_foreach_reduce_large_input__foreach_max_w_empty_False_cuda_bfloat16 (__main__.TestForeachCUDA)

  #150932 closed Jul 11, 2025

• Disable bot spamming

  #117132 closed Jul 11, 2025

• Optimizers should use learning rates passed as tensors directly

  #106802 closed Jul 11, 2025

• Support TorchBind x PT2

  #121266 closed Jul 11, 2025

• 38 Dynamo test are failing with "BuiltinVariable.tensor_args() got multiple values for argument 'self'".

  #120643 closed Jul 11, 2025

• Track the accurate check regress for DebertaForQuestionAnswering and nanogpt

  #122987 closed Jul 11, 2025

• Tracking issue for PT2 dashboard training passrate

  #120129 closed Jul 11, 2025

• [Dynamo] Prints, logging, and warnings

  #93739 closed Jul 11, 2025

• torch.compile fails with torch._dynamo.exc.TorchRuntimeError on a function that contains a torch script module

  #97784 closed Jul 11, 2025

• torch.compile slower than eager on simple MLP

  #119611 closed Jul 11, 2025

• Compilation Failure of torch.special.exp2 in torch.compile Optimized Mode

  #112495 closed Jul 11, 2025

• Compilation Failure of torch.cumsum in torch.compile Optimized Mode

  #112492 closed Jul 11, 2025

• Compile targts cuda:0 rather than the device the model is on

  #97693 closed Jul 11, 2025

• explain() has confusing explanation of graph breaks

  #93656 closed Jul 11, 2025

• [inductor] `aten.index_put_` runtime shape mismatch on H100 but not on A100

  #126614 closed Jul 11, 2025

• AOT inductor should generate source code instead of a library

  #115965 closed Jul 11, 2025

• DISABLED test_op_has_batch_rule_inner_cuda_float32 (__main__.TestVmapOperatorsOpInfoCUDA)

  #157090 closed Jul 11, 2025

• DISABLED test_fallback_to_eager_if_recompiling_too_many_times_warn_only_once (__main__.CudaGraphTreeTests)

  #156954 closed Jul 11, 2025

• Massive initial memory overhead GPU

  #12873 closed Jul 11, 2025

• Error in Qwen inference: NVML_SUCCESS == r INTERNAL ASSERT FAILED at "/pytorch/c10/cuda/CUDACachingAllocator.cpp

  #157535 closed Jul 11, 2025

• DISABLED test_sdpfa_xpu.(__main__.TestUnbackedSymintsXPU)

  #158095 closed Jul 11, 2025

• DISABLED test_foreach_check_stride_ignore_dims_of_one_cuda_float32 (__main__.TestForeachCUDA)

  #150026 closed Jul 11, 2025

• DISABLED test_dynamic_scalar_cuda (__main__.AOTInductorTestABICompatibleGpu)

  #156982 closed Jul 11, 2025

• DISABLED test_hessian_vectorize_correctness_multi_input_cuda (__main__.TestHessianCUDA)

  #157059 closed Jul 11, 2025

• DISABLED test_op_has_batch_rule_dot_cuda_float32 (__main__.TestVmapOperatorsOpInfoCUDA)

  #157069 closed Jul 11, 2025

• DISABLED test_corrcoef_cuda_float32 (__main__.TestTorchDeviceTypeCUDA)

  #156987 closed Jul 11, 2025

• DISABLED test_inplace_on_view_non_contig_cpu (__main__.TestAutogradDeviceTypeCPU)

  #156265 closed Jul 11, 2025

• DISABLED test_inplace_on_view_backprop_base_cpu (__main__.TestAutogradDeviceTypeCPU)

  #156143 closed Jul 11, 2025

• DISABLED test_inplace_on_view_backprop_view_cpu (__main__.TestAutogradDeviceTypeCPU)

  #156163 closed Jul 11, 2025

• DISABLED test_op_has_batch_rule_addmv_cuda_float32 (__main__.TestVmapOperatorsOpInfoCUDA)

  #157037 closed Jul 11, 2025

• DISABLED test_inplace_on_view_undefined_grad_output_cpu (__main__.TestAutogradDeviceTypeCPU)

  #156363 closed Jul 11, 2025

• DISABLED test_against_reference_multi_input_multi_output_jacfwd_cuda (__main__.TestJacCUDA)

  #157036 closed Jul 11, 2025

• DISABLED test_fallback_to_eager_if_recompiling_too_many_times_due_to_cudagraph_managed_tensor (__main__.CudaGraphTreeTests)

  #156922 closed Jul 11, 2025

• DISABLED test_inplace_on_view_then_no_grad_cpu (__main__.TestAutogradDeviceTypeCPU)

  #156306 closed Jul 11, 2025

• DISABLED test_inplace_on_view_of_view_cpu (__main__.TestAutogradDeviceTypeCPU)

  #156289 closed Jul 11, 2025

• DISABLED test_inplace_on_view_backprop_view_of_view_cpu (__main__.TestAutogradDeviceTypeCPU)

  #156180 closed Jul 11, 2025

• DISABLED test_foreach_reduce_large_input__foreach_max_w_empty_False_cuda_int8 (__main__.TestForeachCUDA)

  #156960 closed Jul 11, 2025

• miss doc for torch.segment_reduce

  #153138 closed Jul 11, 2025

• torch.compile with flex_attention: 'ShapeAsConstantBuffer' object has no attribute 'dtype'

  #157833 closed Jul 11, 2025

• test_generate_tensor_from_list_of_numpy_primitive_type fails if run under pytest

  #103439 closed Jul 11, 2025

• tensor.to_sparse() handling indices incorrectly under dynamo/fake tensor

  #93493 closed Jul 11, 2025

• [inductor] [dynamic shape] 5 HF models fails with `Constraints violated` using transformers v4.31.0

  #107200 closed Jul 11, 2025

• ShapeEnv.produce_guards uses a lot of (CPU) memory

  #118222 closed Jul 11, 2025

• export does not support boolean tensor indexing

  #91990 closed Jul 11, 2025

• Dynamo can not trace 'int(a_scalar_tensor.item())'

  #93515 closed Jul 11, 2025

• [PT2.0] [.Compile] [Dynamic] Pytorch FX/JIT graph's inputs/nodes ordering is changed when FX recompiles even though the graph operations are same

  #108745 closed Jul 11, 2025

• _assert_bound_is_rational can fail

  #109514 closed Jul 11, 2025

• dlrm and hf_T5_generate fails aot_eager with bfloat16+dynamic_shapes

  #103760 closed Jul 11, 2025

• Attempt to use minifier on sam model fails

  #104301 closed Jul 11, 2025

• detectron2_fcos_r_50_fpn and other models have enough graph breaks that we end up with multiple cache entries on module blocks

  #103672 closed Jul 11, 2025

• mark_dynamic may error too aggressively

  #102814 closed Jul 11, 2025

• [inductor] test_fft_real_inputs fails with dynamic shapes

  #103194 closed Jul 11, 2025

• tensor with dims marked with torch._dynamo.mark_dynamic loses dynamic dim marks after being moved to a different device

  #100220 closed Jul 11, 2025

• Dynamic shapes exhaustive tests should fail (not xfail) if data mismatch

  #87576 closed Jul 11, 2025

• SymIntType gets translated to int when going through pybind

  #91753 closed Jul 11, 2025

• Functionalization does something wrong with pad backward when it uses as_strided

  #87575 closed Jul 11, 2025

• Symbolic tensors are not printable

  #82517 closed Jul 11, 2025

• test_make_fx_symbolic_exhaustive should pass dynamic ints for shape arguments

  #82318 closed Jul 11, 2025

• [Inductor] AOTInductorTestABICompatibleGpu.test_on_gpu_device1_cuda fails

  #157737 closed Jul 11, 2025

• torch.logdet produces incorrect results for singular matrices on CUDA vs CPU

  #154312 closed Jul 11, 2025

• [Inductor] StableDiffusion unet with `cudagraphs` backend raises fake tensor mismatch error

  #114525 closed Jul 10, 2025

• Custom backend not called for compiling backward graph

  #114189 closed Jul 10, 2025

• [inductor] Assert that Inductor preserves output strides if `TORCHINDUCTOR_LAYOUT_OPTIMIZATION=0`

  #114070 closed Jul 10, 2025

• [inductor] [silent incorrectness] Multiple internal `torch.rand` can lead to inconsistent results with eager

  #151524 closed Jul 10, 2025

• In the torch.Tensor.scatter_ documentation, self, index, and src (if it is a tensor) should have the same number of dimensions, but in practice, the CPU、Gpu does not add a check. Validation needs to be added.

  #157419 closed Jul 10, 2025

• The torch.gather documentation states that input and index must have the same number of dimensions. However, no corresponding validation is added.

  #157425 closed Jul 10, 2025

• Conjugate bit not handled properly in wrapped subclasses

  #130646 closed Jul 10, 2025

• Enable an MPS benchmark

  #115201 closed Jul 10, 2025

• Improve cuda OOM message: include by default more useful aggregate memory usage statistics, measures of memory fragmentation, semantic categories (like activations / weights / scratch workspaces), allocator states

  #32101 closed Jul 10, 2025

• DISABLED test_circular_dependencies (__main__.TestImports)

  #110040 closed Jul 10, 2025

• TransformerEncoderLayer precision loss when fast path is enabled

  #158012 closed Jul 10, 2025

• DISABLED test_mps_event_module (__main__.TestMPS)

  #145052 closed Jul 10, 2025

• [export] run_decompositions generates inefficient operations

  #157289 closed Jul 10, 2025

• `FractionalMaxPool3d` INTERNAL ASSERT FAILED when computing `jacrev`

  #96316 closed Jul 10, 2025

• CONTRIBUTING.md install command incorrect

  #157680 closed Jul 10, 2025

• FFT regression caused by MKL upgrading: MKL FFT error: Intel oneMKL DFTI ERROR: Inconsistent configuration parameters

  #154477 closed Jul 10, 2025

• Dynamo guard source not implemented due to int specialization

  #157992 closed Jul 10, 2025

• DISABLED test_foreach_l2_large_value_input__foreach_norm_cuda_float16 (__main__.TestForeachCUDA)

  #150509 closed Jul 10, 2025

• Wrong error message for wrong dtypes in `torch.binomial`

  #157195 closed Jul 10, 2025

• Calling unbind on 2D NestedTensor throws RuntimeError

  #157404 closed Jul 10, 2025

• XPU build failure with DLE 2025.1.0

  #150047 closed Jul 9, 2025

• UNSTABLE rocm-mi300 / linux-jammy-rocm-py3.10-mi300 / test (default)

  #156360 closed Jul 9, 2025

• ResNet Onnx export dynamic batch size exported as fixed batch size

  #157621 closed Jul 9, 2025

• `FSDPModule.set_reduce_scatter_divide_factor` on subset of parameters is broken?

  #157485 closed Jul 9, 2025

• MPSInductor (aka torch.compile for Apple GPUs)

  #157957 closed Jul 9, 2025

• Triton pin update for PyTorch 2.8 / Triton 3.4

  #154206 closed Jul 9, 2025

• High-performance LLM quantization on X86 CPU with native PyTorch

  #155435 closed Jul 9, 2025

• [RFC][API-Unstable]A16W4 on XPU Device

  #153019 closed Jul 9, 2025

• [RFC][API-Unstable] Support 3rd party SYCL kernels with CPP Extension API

  #153265 closed Jul 9, 2025

• test_gradient_all Device Type test regression with Numpy >= 2.0.0

  #132450 closed Jul 9, 2025

• `setup.py develop` command is disappearing soon from `setuptools`

  #152276 closed Jul 9, 2025

• DISABLED test_foreach_l2_large_value_input__foreach_norm_cuda_bfloat16 (__main__.TestForeachCUDA)

  #150467 closed Jul 9, 2025

• DISABLED test_graph_partition_reorder_custom_op_with_no_dependency1 (__main__.CudaGraphTreeTests)

  #157900 closed Jul 9, 2025

• Add RMS Norm layer

  #128713 closed Jul 9, 2025

• [Intel GPU] The failures do not block new pull request merge if the failures are also in the main branch for Intel GPU

  #137701 closed Jul 9, 2025

• [inductor] [cpu] cpu inductor incorrectly processes `.to(torch.uint8)`, resulting in numerical inconsistency

  #156788 closed Jul 9, 2025

• Inductor throws UnicodeDecodeError when compiling a simple model on Windows with MSVC

  #157673 closed Jul 9, 2025

• [autograd] Slowdown in backward after #151079

  #157407 closed Jul 9, 2025

• return type of torch.nn.functional.interpolate not working

  #129053 closed Jul 9, 2025

• Build error: unrecognizable insn with using gcc-14 on aarch64

  #157842 closed Jul 9, 2025

• DISABLED test_comprehensive_logcumsumexp_xpu_float16 (__main__.TestInductorOpInfoXPU)

  #157697 closed Jul 9, 2025

• aten._cdist_backward

  #105561 closed Jul 9, 2025

• aten.multilabel_margin_loss_backward

  #105562 closed Jul 9, 2025

• Reflection_pad1d

  #105566 closed Jul 9, 2025

• Move Inductor-specific decompositions to general decomposition registrations.

  #105568 closed Jul 9, 2025

• Lowering topk to reductions and pointwise when k is small

  #105569 closed Jul 9, 2025

• Using scans

  #105570 closed Jul 9, 2025

• Add color-coding to fx graph readable printouts :)

  #105572 closed Jul 9, 2025

• TorchInductor Hack-a-Day on July 19th

  #105328 closed Jul 9, 2025

• How to compose HSDP with CP?

  #157393 closed Jul 9, 2025

• [aarch64] Inductor benchmark - fail with IDEEP update

  #157785 closed Jul 8, 2025

• torch.ops._c10d_functional_autograd.all_to_all_single missing dynamic shapes support

  #157479 closed Jul 8, 2025

• PyTorch fails to detect AVX through it's detected

  #157538 closed Jul 8, 2025

• extern declaration of the entity XXX is treated as a static definition

  #157674 closed Jul 8, 2025

• PyTorch wheel binary size increase ~80mb

  #150647 closed Jul 8, 2025

• DISABLED test_progressive (__main__.TestSubprocess)

  #157787 closed Jul 8, 2025

• AOTI: Failure in compile_fx.py with FakeScriptObject (with possible fix)

  #157401 closed Jul 8, 2025

• MPS internal assertion with jacfwd and concatenation

  #152701 closed Jul 8, 2025

• The opp is not compatible with compile mode="reduce-overhead" and linear layers for large inputs.

  #157363 closed Jul 8, 2025

• UNSTABLE trunk / macos-py3-arm64 / test (mps)

  #156833 closed Jul 8, 2025

• DISABLED test_graph_partition_reorder_cpu_and_gpu (__main__.CudaGraphTreeTests)

  #157760 closed Jul 8, 2025

• Win32 Build crashes on startup (C++).

  #146240 closed Jul 8, 2025

• DISABLED test_non_contiguous_input_mm_plus_mm (__main__.TestMaxAutotune)

  #126867 closed Jul 8, 2025

• DISABLED test_add_complex_conj (__main__.ReproTests)

  #156579 closed Jul 8, 2025

• [Segfault Bug] Out-of-bounds write of at::native::cpubla::gemm (bfloat16) in at::native::cpu_flash_attention_backward

  #156022 closed Jul 8, 2025

• Vmap error raised by mask_mod of FlexAttention

  #157543 closed Jul 8, 2025

• DISABLED test_graph_partition_forward_with_skipped_cudagraphed_backward (__main__.CudaGraphTreeTests)

  #157722 closed Jul 8, 2025

• Cannot compile a block that contains Flex attention without graph breaks

  #143163 closed Jul 8, 2025

• [FlexAttention] Support the number of shared query heads in GQA to not be the power of 2

  #143117 closed Jul 8, 2025

• FlexAttention gives me an INTERNAL_ASSERT_FAILED during mask_mod

  #140363 closed Jul 8, 2025

• Dynamo's einops version check is bogus

  #157451 closed Jul 7, 2025

• Compiling the pytorch project produces unnecessary integer comparison warnings, which can lead to a poor user experience.

  #157701 closed Jul 7, 2025

• test_tensor_with_grad_to_scalar_warning failure

  #157252 closed Jul 7, 2025

• Cannot create a mask for each sequence in a batch with Flex Attention

  #157675 closed Jul 7, 2025

• DISABLED test_foreach_reduce_large_input__foreach_max_w_empty_False_cuda_int64 (__main__.TestForeachCUDA)

  #156612 closed Jul 7, 2025

• [RFC] Integrate NCCL scalable init API

  #136539 closed Jul 7, 2025

• [FSDP2] set_reduce_scatter_divide_factor errors with non-trivial MixedPrecisionPolicy

  #155223 closed Jul 7, 2025

• [CI] s390x-periodic tests broken with "No matching distribution found for cuda-bindings<13.0,>=12.0"

  #157409 closed Jul 7, 2025

• pytorch

  #157531 closed Jul 7, 2025

• DISABLED test_tracker_with_activation_checkpointing (__main__.TestTrackerFullyShard1DTrainingCompose)

  #139814 closed Jul 7, 2025

• DISABLED test_tracker_non_root_forward_backward (__main__.TestTrackerFullyShard1DTrainingCore)

  #129692 closed Jul 7, 2025

• DISABLED test_aoti (__main__.TestMemoryPlanning)

  #145211 closed Jul 7, 2025

• DISABLED test_graph_partition_forward_backward_not_called (__main__.CudaGraphTreeTests)

  #157642 closed Jul 7, 2025

• Will the Metal4 update bring significant optimizations for future pytorch mps performance and compatibility?

  #157660 closed Jul 6, 2025

• `torch.compile` fails with `UnicodeDecodeError` when model contains extreme value injection

  #156451 closed Jul 6, 2025

• torch.utils.cpp_extension fails to parse clang version 20.1.7+libcxx

  #157665 closed Jul 6, 2025

• Mispelled "paramter" in test_fully_shard_training.py

  #157564 closed Jul 5, 2025

• torch.nonzero(t, as_tuple=...) does not work with the JIT because the as_tuple signatures are not exposed properly

  #45499 closed Jul 5, 2025

119 Issues opened by 66 people

• Flash Attention 2 + Dynamo + FSDP accelerate plugin + torch.compile error

  #158186 opened Jul 12, 2025

• UNSTABLE rocm-mi300 / linux-noble-rocm-py3.12-mi300 / test (default)

  #158182 opened Jul 12, 2025

• `torch.fmin` has inconsistent overflow behavior on CPU and GPU

  #158172 opened Jul 12, 2025

• Invalid `torch.linalg.lstsq` check during vmapped implementation

  #158169 opened Jul 12, 2025

• `torch.nn.LPPool2d` throws inconsistent error on CPU and GPU

  #158166 opened Jul 11, 2025

• [dynamo] Incompatibility with sys.monitoring

  #158164 opened Jul 11, 2025

• `torch.native_channel_shuffle` throws FPE when `groups` > size of the second dimension of the input tensor

  #158154 opened Jul 11, 2025

• DISABLED test_one_shot_all_reduce (__main__.SymmMemCollectiveTest)

  #158138 opened Jul 11, 2025

• DISABLED test_cuda_nvlink_connectivity_detection (__main__.SymmetricMemoryTest)

  #158136 opened Jul 11, 2025

• DISABLED test_allow_overlapping_devices (__main__.SymmetricMemoryTest)

  #158135 opened Jul 11, 2025

• Eager success but inductor failed on torch.ops.aten._cudnn_rnn.default

  #158131 opened Jul 11, 2025

• A "Merge-when-ready" label when ci finishes?

  #158127 opened Jul 11, 2025

• index.Tensor doesn't properly account for boolean masks acting on multiple dimensions.

  #158125 opened Jul 11, 2025

• [RFC] A Distributed CUDA Unified Memory Backend for PyTorch

  #158122 opened Jul 11, 2025

• AttributeError: partially initialized module 'torch._dynamo' has no attribute 'disable' (most likely due to a circular import)

  #158120 opened Jul 11, 2025

• DISABLED test_gru (__main__.TestXNNPACKQuantizer)

  #158116 opened Jul 11, 2025

• DISABLED test_conv_transpose_unary_fusion_ops (__main__.TestMkldnnFusion)

  #158115 opened Jul 11, 2025

• DISABLED test_triton_signal_wait_until (__main__.NVSHMEMTritonTest)

  #158114 opened Jul 11, 2025

• Combination of USE_MEM_EFF_ATTENTION and AOTRITON_INSTALLED_PREFIX misbehaves

  #158109 opened Jul 11, 2025

• DISABLED test_vmap_exhaustive_nn_functional_conv_transpose3d_cuda_float32 (__main__.TestVmapOperatorsOpInfoCUDA)

  #158108 opened Jul 11, 2025

• DISABLED test_pool_method_subprocess (__main__.TestAsyncCompile)

  #158107 opened Jul 11, 2025

• DISABLED test_triton_quiet (__main__.NVSHMEMTritonTest)

  #158106 opened Jul 11, 2025

• Exporting ONNX Model with Operator Name as Output Name Produces Invalid Model When dynamo=True

  #158094 opened Jul 11, 2025

• kvstore Documentation Inconsistency: check() Method Claimed in v2.7 Docs But Missing in Implementation

  #158093 opened Jul 11, 2025

• `torch.compile` doesn't properly raise eager fake tensor exception

  #158088 opened Jul 11, 2025

• `torch.compile` ignores nan check in `int(torch.tensor(torch.nan))`

  #158087 opened Jul 11, 2025

• `torch.compile` on `.sum() `and `.item()` calls errors from `tensorify_python_scalars`

  #158083 opened Jul 11, 2025

• `torch.compile` errors with inductor raised `GuardOnDataDependentSymNode` exception

  #158081 opened Jul 11, 2025

• loaded storage won't be cached during deserialization if location is meta

  #158080 opened Jul 10, 2025

• torch.compile on BFloat16 Segment Anything segfaults in cpp_CppMicroGemmRef_micro_gemm<false, false> on Mac

  #158076 opened Jul 10, 2025

• Mutating a tensor while serializing with safetensors crashes free-threaded PyTorch

  #158071 opened Jul 10, 2025

• InductorError: CppCompileError: C++ compile error on a function with a single `item` call

  #158060 opened Jul 10, 2025

• `torch.compile` on `torch.arange` hard errors with `PendingUnbackedSymbolNotFound`

  #158058 opened Jul 10, 2025

• DISABLED test_generalized_sq_cases (__main__.TestSolve)

  #158054 opened Jul 10, 2025

• Add GoLU Activation Function 🚀

  #158043 opened Jul 10, 2025

• [inductor] grouped_mm is autotuning under torch.compile default mode

  #158042 opened Jul 10, 2025

• pow_test fails on Aarch64

  #158041 opened Jul 10, 2025

• extension-cpp claims ROCM support with no additional changes; we should test this.

  #158032 opened Jul 10, 2025

• NCCL + cudagraphs + expandable segments result in IMA

  #158029 opened Jul 10, 2025

• DISABLED test_vmap_exhaustive_nn_functional_conv_transpose2d_cuda_float32 (__main__.TestVmapOperatorsOpInfoCUDA)

  #158025 opened Jul 10, 2025

• DISABLED test_pool_method_spawn (__main__.TestAsyncCompile)

  #158024 opened Jul 10, 2025

• DISABLED test_triton_put_signal_set (__main__.NVSHMEMTritonTest)

  #158023 opened Jul 10, 2025

• empty_like(tensor, memory_format=torch.preserve_format) does not preserve strides for views

  #158022 opened Jul 10, 2025

• [2.8 regression] CUDAAllocator has BC-breaking changes

  #158021 opened Jul 10, 2025

• dist.all_to_all_single, when input tensors of different shapes result in undefined output behavior

  #158016 opened Jul 10, 2025

• DISABLED test_vmap_exhaustive_nn_functional_conv_transpose1d_cuda_float32 (__main__.TestVmapOperatorsOpInfoCUDA)

  #158010 opened Jul 10, 2025

• DISABLED test_pool_method_fork (__main__.TestAsyncCompile)

  #158009 opened Jul 10, 2025

• DISABLED test_triton_put_signal_add (__main__.NVSHMEMTritonTest)

  #158008 opened Jul 10, 2025

• `torch.compile` cannot compile a model with a basic `LSTM`, even on latest main

  #158007 opened Jul 10, 2025

• BUG: numpy very slow after import torch

  #158005 opened Jul 10, 2025

• [inuctor] [triton] `torch.cumsum` outputs inconsistent results when meeting large tensors

  #158003 opened Jul 10, 2025

• [inductor] performance of torchbench alexnet has regression compare to 2.7 due to inductor changes a triton kernel

  #158000 opened Jul 10, 2025

• [XPU] torch.xpu.mem_get_info() query failed on BMG

  #157989 opened Jul 10, 2025

• DISABLED test_vmap_exhaustive_nn_functional_conv2d_strided_padding_dilation_with_bias_cuda_float32 (__main__.TestVmapOperatorsOpInfoCUDA)

  #157987 opened Jul 10, 2025

• DISABLED test_triton_put (__main__.NVSHMEMTritonTest)

  #157986 opened Jul 10, 2025

• support DTensor's local_map in compile

  #157976 opened Jul 9, 2025

• [Inductor] Online softmax disabled due to reduction split – Unexpected performance warning

  #157975 opened Jul 9, 2025

• Torch compile don't work correctly with divide by scalar

  #157959 opened Jul 9, 2025

• [RFC]: PyTorch Low-Precision GEMMs Public API

  #157950 opened Jul 9, 2025

• Dim argument of torch.max can currently be only int or name, but documentation says it can be int or tuple

  #157948 opened Jul 9, 2025

• torch.compile fails with InternalTorchDynamoError when slicing torch.linalg.svd results

  #157945 opened Jul 9, 2025

• torch.compile() disables torch.distributions parameter validation globally

  #157926 opened Jul 9, 2025

• TypeError: message must be a callable when calling grouped_mm with incompatible batch size for offsets

  #157922 opened Jul 9, 2025

• BC-breaking change to symint range constraints from 2.7 -> 2.8

  #157921 opened Jul 9, 2025

• tighten pt2 FX graph invariants

  #157919 opened Jul 9, 2025

• [Precompile] Umbrella task

  #157918 opened Jul 9, 2025

• tlparse symint guard user provenance slightly off

  #157915 opened Jul 9, 2025

• DISABLED test_qlinear_add_int8_mixed_bf16_use_relu_False_is_qat_False_is_dynamic_True (__main__.TestPatternMatcher)

  #157911 opened Jul 9, 2025

• DISABLED test_graph_partition_reorder_custom_op_with_no_dependency1 (__main__.CudaGraphTreeTests)

  #157901 opened Jul 9, 2025

• DISABLED test_vmap_exhaustive_nn_functional_conv2d_strided_padding_dilation_no_bias_cuda_float32 (__main__.TestVmapOperatorsOpInfoCUDA)

  #157899 opened Jul 9, 2025

• DISABLED test_triton_get_ring (__main__.NVSHMEMTritonTest)

  #157898 opened Jul 9, 2025

• DISABLED test_reduce_scatter_float8 (__main__.ProcessGroupNCCLOpTest)

  #157897 opened Jul 9, 2025

• DISABLED test_nccl_watchdog_cudagraph (__main__.ProcessGroupNCCLOpTest)

  #157896 opened Jul 9, 2025

• DISABLED test_2d_reductions_mixed_indexing_reduction_op0_cpu (__main__.TritonBlockPointerTestCPU)

  #157895 opened Jul 9, 2025

• Partitioner: Fix to align partition node order with original graph

  #157891 opened Jul 9, 2025

• Way to customize HOPs overrided arguments selection

  #157888 opened Jul 9, 2025

• DISABLED test_vmap_exhaustive_nn_functional_conv2d_cuda_float32 (__main__.TestVmapOperatorsOpInfoCUDA)

  #157880 opened Jul 9, 2025

• DISABLED test_allreduce_float8 (__main__.ProcessGroupNCCLOpTest)

  #157879 opened Jul 9, 2025

• DISABLED test_triton_template_generated_code_caching_mm_plus_mm (__main__.TestMaxAutotune)

  #157878 opened Jul 9, 2025

• DISABLED test_triton_get (__main__.NVSHMEMTritonTest)

  #157877 opened Jul 9, 2025

• DISABLED test_int64_index_intermediate (__main__.CudaReproTests)

  #157872 opened Jul 9, 2025

• DISABLED test_graph_partition_reorder_custom_op_with_no_dependency (__main__.CudaGraphTreeTests)

  #157871 opened Jul 9, 2025

• Exporting tensor.to("cuda") under FakeTensorMode doesn't work on a CPU-only machine

  #157869 opened Jul 8, 2025

• Spurious "Grad strides do not match bucket view strides" warning for 1x1 convolution

  #157862 opened Jul 8, 2025

• CUDA ec2 runner with no cuda runtime

  #157844 opened Jul 8, 2025

• [FSDP2] should pass args as is instead of creating new ones

  #157832 opened Jul 8, 2025

• sparse_csr bfloat16 matrix multiplication backward is 10x slower than float16

  #157808 opened Jul 8, 2025

• [RFC] Replace setuptools build backend with scikit-build-core

  #157807 opened Jul 8, 2025

• Consider changing AOTAutograd cache to hit on graphs with different input and node names

  #157792 opened Jul 8, 2025

• DISABLED test_vmap_exhaustive_matmul_cuda_float32 (__main__.TestVmapOperatorsOpInfoCUDA)

  #157790 opened Jul 8, 2025

• DISABLED test_triton_fence (__main__.NVSHMEMTritonTest)

  #157789 opened Jul 8, 2025

• DISABLED test_progressive (__main__.TestSubprocess)

  #157788 opened Jul 8, 2025

• torch.distributed.checkpoint.state_dict.set_state_dict stucks with StateDictOptions(full_state_dict=True)

  #157781 opened Jul 8, 2025

• test_optimizer_non_static_param got failed on Intel GPU

  #157778 opened Jul 8, 2025

• Does FSDP works with DDP ?

  #157917 opened Jul 8, 2025

• Support for _Float16/C++23 std::float16_t

  #157776 opened Jul 8, 2025

• Fatal on torch.xpu

  #157775 opened Jul 8, 2025

• Drop SSE4 support in oneDNN

  #157764 opened Jul 8, 2025

• Nightly C++ docs build timeout in CI after 4 hours

  #157763 opened Jul 8, 2025

• DISABLED test_vmap_exhaustive_linalg_vecdot_cuda_float32 (__main__.TestVmapOperatorsOpInfoCUDA)

  #157762 opened Jul 8, 2025

• DISABLED test_graph_partition_reorder_cpu_and_gpu (__main__.CudaGraphTreeTests)

  #157761 opened Jul 8, 2025

• [dynamo] Missing meta kernel for `aten::quantize_per_tensor.tensor_qparams`

  #157729 opened Jul 7, 2025

• fill_ overflows on uint64 in range [2**63, 2**64) when profiler is engaged

  #157728 opened Jul 7, 2025

• DISABLED test_vmap_exhaustive_inner_cuda_float32 (__main__.TestVmapOperatorsOpInfoCUDA)

  #157726 opened Jul 7, 2025

• DISABLED test_resnet (__main__.TestBlockStateAbsorption)

  #157725 opened Jul 7, 2025

• DISABLED test_progressive (__main__.GPUTests)

  #157724 opened Jul 7, 2025

• DISABLED test_graph_partition_forward_with_skipped_cudagraphed_backward (__main__.CudaGraphTreeTests)

  #157723 opened Jul 7, 2025

• [Precompile] Caching precompile and presence of packages in general should be safe re: guard serialization

  #157721 opened Jul 7, 2025

• [dynamo] `torch.compile` errors on numpy `astype("O")`

  #157720 opened Jul 7, 2025

• AssertHandler::printMessage On Intel GPU

  #157714 opened Jul 7, 2025

• /lib/x86_64-linux-gnu/libstdc++.so.6: version `GLIBCXX_3.4.30' not found when IMPORT TORCH

  #157709 opened Jul 7, 2025

• `torch.randint` raises a `RuntimeError` if `dtype=torch.uint64` and `high >= 2**63`

  #157707 opened Jul 7, 2025

• Seg faults on macos / OSX

  #157704 opened Jul 7, 2025

• [inductor][fuzzer] Compilation Error in complex64+toint

  #157683 opened Jul 7, 2025

• Flex Attention breaks in certain cases when used with a learned bias

  #157677 opened Jul 6, 2025

• Feedback about Getting Started on Intel GPU

  #157672 opened Jul 6, 2025

• NCCL error caused due to use of NVLS in torch 2.7.1-cu128 on aarch64 gb200 cluster

  #157668 opened Jul 6, 2025

• ConvNd ops in channel last layout (N,L,C) / (N,H,W,C) / (N,D,H,W,C)

  #157663 opened Jul 5, 2025

• OffsetBasedRNGTracker's run_state_sync causes deadlock due to inconsistent broadcast order across ranks

  #157662 opened Jul 5, 2025

565 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.

• DDE-Free select with unbacked index.

  #157605 commented on Jul 12, 2025 • 16 new comments

• [Device] Add support for PrivateUse1 device type in parse_type function

  #157609 commented on Jul 12, 2025 • 13 new comments

• [HOP, map] Rework of map autograd to the new interface

  #153343 commented on Jul 11, 2025 • 11 new comments

• Enable `_lazy_clone` between CPU and MPS

  #148408 commented on Jul 11, 2025 • 9 new comments

• [cutlass backend] cache maybe_append_choices

  #156781 commented on Jul 12, 2025 • 7 new comments

• Update the signature and test of torch.hamming_window()

  #152682 commented on Jul 10, 2025 • 7 new comments

• NUMA Binding Integration with torchrun

  #149334 commented on Jul 12, 2025 • 6 new comments

• [nativert] Move ModelRunnerBase to oss.

  #157633 commented on Jul 7, 2025 • 5 new comments

• Add test for user-managed weights with load_state_dict

  #157496 commented on Jul 11, 2025 • 5 new comments

• [dynamo] [guard] Add caching for inside torch.compile.disable function to avoid unnecessary recompilation.

  #157566 commented on Jul 12, 2025 • 5 new comments

• Deprecate DataLoader pin_memory_device param

  #146821 commented on Jul 11, 2025 • 4 new comments

• CI for Windows Arm64

  #148753 commented on Jul 8, 2025 • 4 new comments

• [PT2][fusion] ban fusions with large accumulated reads

  #157563 commented on Jul 12, 2025 • 4 new comments

• [simplefsdp auto-bucketing] ir node runtime estimation

  #157572 commented on Jul 12, 2025 • 4 new comments

• allow user to pass in custom partitioner function

  #157580 commented on Jul 8, 2025 • 3 new comments

• allow _size_of to return individual element's size

  #157582 commented on Jul 8, 2025 • 3 new comments

• [Draft][CUDA] Upgrade torch._scaled_grouped_mm to SM100+

  #156806 commented on Jul 9, 2025 • 3 new comments

• [cpp_wrapper] Build main and kernel code in separate threads

  #154551 commented on Jul 11, 2025 • 3 new comments

• [struct] Add `struct.pack` and `struct.unpack` polyfills

  #156977 commented on Jul 10, 2025 • 3 new comments

• Always disable ShardingPropagation cache if compiling

  #156868 commented on Jul 11, 2025 • 3 new comments

• Parameterized CUDA Graph Launch

  #152622 commented on Jul 11, 2025 • 3 new comments

• [NOT FOR MERGE] Exploratory work on AOTInductor training

  #155877 commented on Jul 11, 2025 • 3 new comments

• [Misc] skip the case test_foreach_add_different_mesh if world size is…

  #155563 commented on Jul 8, 2025 • 2 new comments

• Feature: Implement support for `cudnn_batch_norm_out` kernel to replace the autogen approach.

  #123020 commented on Jul 7, 2025 • 2 new comments

• [iter] support `iter(callable, sentinel)`

  #156416 commented on Jul 10, 2025 • 2 new comments

• [DTensor][FSDP2] necessary changes to FSDP and TP to unblock EP

  #157216 commented on Jul 11, 2025 • 2 new comments

• [BE] always use `uv pip` if possible in `pip_init.py` for `lintrunner init`

  #157199 commented on Jul 10, 2025 • 2 new comments

• Add DeviceAllocator as the base device allocator

  #138222 commented on Jul 11, 2025 • 2 new comments

• Do not checkout nccl for `USE_SYSTEM_LIBS`

  #153807 commented on Jul 9, 2025 • 2 new comments

• [Inductor] Set the default value of min_chunk_size to 512

  #150762 commented on Jul 12, 2025 • 2 new comments

• [generator] Raise `StopIteration(value)` with value from the return stmt

  #157152 commented on Jul 9, 2025 • 2 new comments

• handling special case for pow(3) for GPU

  #157537 commented on Jul 11, 2025 • 2 new comments

• Add cascade sum support for Inductor CPP backend

  #156296 commented on Jul 11, 2025 • 2 new comments

• Deprecate c10::string_view

  #156798 commented on Jul 9, 2025 • 2 new comments

• Optimize scatter/gather kernel for ARM.

  #156161 commented on Jul 11, 2025 • 2 new comments

• [doc] Updates to distributed.md for XCCL backend

  #155834 commented on Jul 8, 2025 • 2 new comments

• Add cuda 12.9 periodic tests

  #156900 commented on Jul 10, 2025 • 1 new comment

• Add basic torch.hash_tensor op

  #154149 commented on Jul 12, 2025 • 1 new comment

• [Don't Review] Test CI

  #139971 commented on Jul 11, 2025 • 1 new comment

• Make open device registration tests standalone

  #153855 commented on Jul 9, 2025 • 1 new comment

• [PrivateUse1] Optimize 3rd party backend experiences

  #155215 commented on Jul 9, 2025 • 1 new comment

• [Easy] Show some clear error when torch.ops.load_library fails.

  #157524 commented on Jul 6, 2025 • 1 new comment

• [dynamo, docs] add dynamo programming model docs

  #157527 commented on Jul 10, 2025 • 1 new comment

• Enable TF32 as fp32 internal precision for matmul/linear/conv

  #157520 commented on Jul 9, 2025 • 1 new comment

• [BE][1/5] fix typos in aten/

  #157550 commented on Jul 12, 2025 • 1 new comment

• [Refactor][XPU] Refactor XPU quantization op and add header files.

  #157430 commented on Jul 8, 2025 • 1 new comment

• [BE][2/5] fix typos in aten/ (aten/src/ATen/native/)

  #157551 commented on Jul 12, 2025 • 1 new comment

• [BE][3/5] fix typos in aten/ (aten/src/ATen/native/)

  #157552 commented on Jul 12, 2025 • 1 new comment

• Preserve current stream in TestCuda::test_stream_compatibility

  #157421 commented on Jul 7, 2025 • 1 new comment

• [BE][4/5] fix typos in aten/ (aten/src/ATen/native/)

  #157553 commented on Jul 12, 2025 • 1 new comment

• Fix doc issue 153531 by adding further explanation of STFT equation

  #157595 commented on Jul 8, 2025 • 1 new comment

• Add inductor lowerings for adaptive_avg_pool3d/adaptive_max_pool3d

  #157331 commented on Jul 9, 2025 • 1 new comment

• [BE] Replace `std::runtime_error` with `TORCH_CHECK` [2/N]

  #152080 commented on Jul 11, 2025 • 1 new comment

• enable windows inductor UT in CI

  #151777 commented on Jul 11, 2025 • 0 new comments

• Add adaptive_avg_pool2d input and output_size check

  #151769 commented on Jul 10, 2025 • 0 new comments

• [bazel] Fix unusual reference to cpuinfo workspace

  #151578 commented on Jul 9, 2025 • 0 new comments

• Allow to byteswap data when reading saved torch jit data

  #151447 commented on Jul 9, 2025 • 0 new comments

• [Environment Variable] Use thread-safe getenv functions

  #152609 commented on Jul 12, 2025 • 0 new comments

• [2/N] Use std::filesystem

  #152586 commented on Jul 10, 2025 • 0 new comments

• [BE] Update numba versions

  #152557 commented on Jul 6, 2025 • 0 new comments

• Remove Conda Instructions

  #152546 commented on Jul 11, 2025 • 0 new comments

• [compile async] [cache] testing

  #152523 commented on Jul 6, 2025 • 0 new comments

• Horizontal

  #151780 commented on Jul 9, 2025 • 0 new comments

• fix: outdated contents in dynamo overview

  #152382 commented on Jul 6, 2025 • 0 new comments

• [reland][ROCm] remove caffe2 from hipify

  #151845 commented on Jul 8, 2025 • 0 new comments

• [FP8][CUTLASS] xFail `honor_sm_carveout` on `sm100`

  #152378 commented on Jul 12, 2025 • 0 new comments

• Enable type promotions in slice_scatter (pytorch#147842)

  #151911 commented on Jul 11, 2025 • 0 new comments

• Avoid rounding errors in _get_total_norm for a dTensor by using torch Dynamo

  #152234 commented on Jul 11, 2025 • 0 new comments

• Add dynamo config to HOP-ify context managers

  #152159 commented on Jul 8, 2025 • 0 new comments

• docs: add torch.e and torch.pi to constants table (#134964)

  #151996 commented on Jul 7, 2025 • 0 new comments

• [UniformValueConstantFolder] deduce value on CPU rather than on device

  #151998 commented on Jul 7, 2025 • 0 new comments

• [BE][3/6] fix typos in test/

  #157637 commented on Jul 12, 2025 • 0 new comments

• try relanding cublaslt autotuning support for TunableOp #

  #153316 commented on Jul 9, 2025 • 0 new comments

• [DEBUG] MTIA Module and Interface

  #153308 commented on Jul 9, 2025 • 0 new comments

• [DEBUG] memory profiler and combined trace

  #153307 commented on Jul 9, 2025 • 0 new comments

• [BE] fix skip_if_lt_x_gpu decorator and add test coverage

  #153295 commented on Jul 9, 2025 • 0 new comments

• [BE][FSDP] fix FSDP to skip tests where #GPUs < world_size before entering into init_pg

  #153291 commented on Jul 9, 2025 • 0 new comments

• test timm_efficientnet pass

  #153290 commented on Jul 12, 2025 • 0 new comments

• [MX] Add more ops to allowed set for e8

  #153271 commented on Jul 8, 2025 • 0 new comments

• [nocommit] bundled autograd cache test

  #153269 commented on Jul 11, 2025 • 0 new comments

• fix dtensor and tensor inconsistent compute mesh

  #153268 commented on Jul 7, 2025 • 0 new comments

• DEBUG PR Issue

  #153267 commented on Jul 9, 2025 • 0 new comments

• MXFP8 Fix broken bias support for mxfp8

  #153254 commented on Jul 8, 2025 • 0 new comments

• Fix integer overflow bug in triu/tril for large diagonal values

  #153240 commented on Jul 12, 2025 • 0 new comments

• Delete .github/workflows/docker-cache-mi300.yml

  #153075 commented on Jul 10, 2025 • 0 new comments

• Add CUDA support for Adagrad(fused=True)

  #153038 commented on Jul 10, 2025 • 0 new comments

• Allow zero sized dimensions in padding operations

  #153037 commented on Jul 11, 2025 • 0 new comments

• [WIP][dynamic shapes] unbacked safer cat, repeat

  #153011 commented on Jul 6, 2025 • 0 new comments

• [Pytorch] Add `torch.cuda.streams.Event` to save torch functions list

  #152978 commented on Jul 6, 2025 • 0 new comments

• [dtensor] Extend Partial partition of replicated tensor for min/max reduce

  #152975 commented on Jul 7, 2025 • 0 new comments

• docs: Improve documentation for NCCL timeout / watchdog variables

  #152959 commented on Jul 6, 2025 • 0 new comments

• [dtensor] add privateuse1 SDPA op support to DTensor

  #152949 commented on Jul 9, 2025 • 0 new comments

• [feature] Channel Wise Parallel API for Conv layers

  #152937 commented on Jul 6, 2025 • 0 new comments

• Allow Inductor backends to attest their own availability

  #152933 commented on Jul 5, 2025 • 0 new comments

• Add unified memory APIs for torch.accelerator

  #152932 commented on Jul 11, 2025 • 0 new comments

• Add overall tensor similarity comparison (#152647)

  #152920 commented on Jul 6, 2025 • 0 new comments

• [DO NOT MERGE] update build tools version

  #152820 commented on Jul 8, 2025 • 0 new comments

• Update CMakeLists.txt

  #152786 commented on Jul 6, 2025 • 0 new comments

• Pattern matcher support for mutable ops with view inputs

  #152776 commented on Jul 12, 2025 • 0 new comments

• Handle less functions than number of segments

  #152753 commented on Jul 6, 2025 • 0 new comments

• [Dynamo] Guard serialization for BUILTIN_MATCH

  #152729 commented on Jul 6, 2025 • 0 new comments

• [pytree] make `tree_*` functions accept both Python and C++ `PyTreeSpec`

  #152624 commented on Jul 9, 2025 • 0 new comments

• [WIP] Generalize device caching allocator

  #151298 commented on Jul 10, 2025 • 0 new comments

• [DO NOT MERGE] Update oneDNN to the latest main branch

  #147073 commented on Jul 8, 2025 • 0 new comments

• fake_tensor: Handle op errors more gracefully

  #147049 commented on Jul 7, 2025 • 0 new comments

• Porting Pytorch to AIX Operating System.

  #146983 commented on Jul 8, 2025 • 0 new comments

• Support contextlib.ExitStack

  #146506 commented on Jul 9, 2025 • 0 new comments

• Fix full_like decomposition to preserve strides

  #144765 commented on Jul 9, 2025 • 0 new comments

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

  #144557 commented on Jul 12, 2025 • 0 new comments

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

  #144556 commented on Jul 12, 2025 • 0 new comments

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

  #144552 commented on Jul 12, 2025 • 0 new comments

• [dynamo, nested graph breaks] add nested graph break tests

  #144516 commented on Jul 10, 2025 • 0 new comments

• [Draft][WIP] Enable XPU path for FlexAttention

  #143553 commented on Jul 7, 2025 • 0 new comments

• Support LOAD_BUILD_CLASS opcode in dynamo

  #139561 commented on Jul 10, 2025 • 0 new comments

• `has_triton`: Use the device interface for detecting Triton availability

  #139171 commented on Jul 7, 2025 • 0 new comments

• Add overflow check for negtive integer div_floor and div_trunc on CPU

  #138684 commented on Jul 12, 2025 • 0 new comments

• Always produce XML

  #138513 commented on Jul 7, 2025 • 0 new comments

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

  #138214 commented on Jul 9, 2025 • 0 new comments

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

  #137400 commented on Jul 9, 2025 • 0 new comments

• Help fix numpy detection in cross compiled layouts

  #137084 commented on Jul 8, 2025 • 0 new comments

• Automated submodule update: FBGEMM

  #115316 commented on Jul 12, 2025 • 0 new comments

• [pytree] support PyStructSequence types for Python pytree

  #113258 commented on Jul 9, 2025 • 0 new comments

• [docs] URL and link format proposal to make function page URLs more concise

  #106664 commented on Jul 12, 2025 • 0 new comments

• Python 3.14 support for PyTorch

  #156856 commented on Jul 12, 2025 • 0 new comments

• Illegal Memory Access when Using Trainable Biases in Flex Attention

  #144511 commented on Jul 12, 2025 • 0 new comments

• addition of muon optimizer to torch.optim

  #148819 commented on Jul 12, 2025 • 0 new comments

• UNSTABLE rocm / linux-jammy-rocm-py3.10 / test (default)

  #156098 commented on Jul 12, 2025 • 0 new comments

• DISABLED test_remove_noop_view_default_cuda (__main__.GPUTests)

  #151511 commented on Jul 12, 2025 • 0 new comments

• [v.2.8.0] Release Tracker

  #156745 commented on Jul 12, 2025 • 0 new comments

• [inductor] [silence] inconsistent swap wih eager when compiling `torch.rot90-torch.randn_like`

  #147847 commented on Jul 12, 2025 • 0 new comments

• Inductor can not fuse cat with a pointwise

  #125075 commented on Jul 12, 2025 • 0 new comments

• [FSDP2] document the contract for modifying DTensor model.parameters()

  #157391 commented on Jul 12, 2025 • 0 new comments

• mps and cpu backends produce different training results with FFT and Adam

  #151740 commented on Jul 8, 2025 • 0 new comments

• [dynamo] Avoid unnecessary `.detach()` call in `_make_subclass` polyfill

  #151265 commented on Jul 5, 2025 • 0 new comments

• NCCL: Fix cmake file when cross compiling.

  #151234 commented on Jul 8, 2025 • 0 new comments

• Implement MKLGenerator

  #151218 commented on Jul 10, 2025 • 0 new comments

• Fix `MaskedTensor` to device ignored mask

  #151205 commented on Jul 10, 2025 • 0 new comments

• [MPS] Get Vmap to work with mps backend

  #151177 commented on Jul 12, 2025 • 0 new comments

• Pin all root requirements to major versions

  #150833 commented on Jul 7, 2025 • 0 new comments

• [draft][distributed] add into 3d composability test at AMD CI test

  #150694 commented on Jul 8, 2025 • 0 new comments

• Make LazyModuleMixin materialize after load_state_dict

  #150593 commented on Jul 10, 2025 • 0 new comments

• Add differentiable ops hint message in Module docs

  #150291 commented on Jul 10, 2025 • 0 new comments

• Add cmake variable USE_ROCM_CK

  #150245 commented on Jul 5, 2025 • 0 new comments

• [DLPack] Add support for missing keyword-arguments.

  #150218 commented on Jul 9, 2025 • 0 new comments

• Add path used by pip's build isolation procedure to DLL search

  #150013 commented on Jul 11, 2025 • 0 new comments

• AOTI freezing: fix test issues and enable by default

  #149961 commented on Jul 11, 2025 • 0 new comments

• DRAFT: Add TMA opt for concat function target hopper and blackwell arch

  #149893 commented on Jul 6, 2025 • 0 new comments

• Configure `cuda.cmake` to ensure consistent behavior downstream

  #149861 commented on Jul 8, 2025 • 0 new comments

• [test] sccache docker build

  #149536 commented on Jul 7, 2025 • 0 new comments

• Fix unexpected keyword argument 'mode' when calling `CompileCounterWithBackend`

  #149271 commented on Jul 6, 2025 • 0 new comments

• Fix AttributeError for `_get_vc_env` with setuptools>=75.9.0

  #148847 commented on Jul 6, 2025 • 0 new comments

• [BE][pytree] cleanup parameterized pytree tests

  #148569 commented on Jul 9, 2025 • 0 new comments

• Implement fast access to individual elements of jagged nested tensors

  #148497 commented on Jul 11, 2025 • 0 new comments

• [triton hash update] update the pinned triton hash

  #148492 commented on Jul 12, 2025 • 0 new comments

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

  #148484 commented on Jul 9, 2025 • 0 new comments

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

  #148474 commented on Jul 9, 2025 • 0 new comments

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

  #148328 commented on Jul 9, 2025 • 0 new comments

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

  #148180 commented on Jul 9, 2025 • 0 new comments

• Support `contextlib.suppress`

  #147990 commented on Jul 9, 2025 • 0 new comments

• [DO NOT MERGE] Migrate from oneDNN Inner Product to oneDNN MatMul for mkldnn_linear and mkldnn_linear_backward

  #147855 commented on Jul 8, 2025 • 0 new comments

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

  #147368 commented on Jul 9, 2025 • 0 new comments

• [DO NOT MERGE][Inductor] Migrate from oneDNN Inner Product to oneDNN MatMul for mkldnn._linear_pointwise and mkldnn._linear_pointwise.binary

  #147360 commented on Jul 8, 2025 • 0 new comments

• switch from deprecated `find_package(CUDA)` to `find_package(CUDAToolkit)`

  #147300 commented on Jul 8, 2025 • 0 new comments

• [nativert] libtorch kernel registry

  #157150 commented on Jul 8, 2025 • 0 new comments

• [generator] Close all open generators in compile_subgraph

  #157149 commented on Jul 9, 2025 • 0 new comments

• [contextlib] Fixes for CPython contextlib tests

  #157148 commented on Jul 10, 2025 • 0 new comments

• [DDP][FSDP2] Add unit test for DDP mixed precision with FSDP2 ignored params

  #157140 commented on Jul 9, 2025 • 0 new comments

• Documenting discrepancy for Numpy dependency versions

  #157132 commented on Jul 11, 2025 • 0 new comments

• updated path to requirements txt in docs

  #157106 commented on Jul 11, 2025 • 0 new comments

• [CI] add decorator for specifying H100-only tests

  #156980 commented on Jul 10, 2025 • 0 new comments

• [math] Trace `float.fromhex`

  #156976 commented on Jul 9, 2025 • 0 new comments

• [math] Raise exception in Dynamo if constant fold call fail

  #156975 commented on Jul 9, 2025 • 0 new comments

• Use `_get_object_coll_device` instead of deprecated API

  #156878 commented on Jul 7, 2025 • 0 new comments

• `fast-autotune`: Model Prediction of Triton Kernel Runtimes

  #156851 commented on Jul 10, 2025 • 0 new comments

• [TESTING] [DO NOT MERGE] Updated triton commit pin - upstream base

  #156841 commented on Jul 7, 2025 • 0 new comments

• Introduce a new API torch.accelerator.get_mem_info

  #156812 commented on Jul 10, 2025 • 0 new comments

• add tests for Thunk utility function

  #156759 commented on Jul 7, 2025 • 0 new comments

• Add back manywheel-py3_9-cuda12_4-build/test

  #156753 commented on Jul 9, 2025 • 0 new comments

• Stop parsing command line arguments every time common_utils is imported.

  #156703 commented on Jul 10, 2025 • 0 new comments

• ReplaceWithCopy graph pass

  #156666 commented on Jul 7, 2025 • 0 new comments

• Adds support for Nested Jagged Tensor in Multihead Attention

  #156660 commented on Jul 8, 2025 • 0 new comments

• Fix torch==2.6 broke nn.Module.dtype typing

  #156631 commented on Jul 9, 2025 • 0 new comments

• multi-kernel matmuls based on varying hint sizes

  #156628 commented on Jul 12, 2025 • 0 new comments

• [BE][15/16] fix typos in torch/ (torch/distributed/tensor/)

  #156605 commented on Jul 12, 2025 • 0 new comments

• [Inductor Dashboard] Enable deterministic algorithms for all models on ROCm

  #156592 commented on Jul 10, 2025 • 0 new comments

• [Doc] remove WSL2 in support matrix for Intel GPU

  #156590 commented on Jul 8, 2025 • 0 new comments

• [2/N] Remove FindPackageHandleStandardArgs.cmake

  #156559 commented on Jul 11, 2025 • 0 new comments

• [docs][typing] Document and type support for dim=None in torch.amin and torch.amax

  #156510 commented on Jul 9, 2025 • 0 new comments

• [Inductor] Fix epilogue fusion decision with 1 Triton caller as choice

  #156500 commented on Jul 10, 2025 • 0 new comments

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

  #156486 commented on Jul 11, 2025 • 0 new comments

• [iter] Wrap iter(..) call in a ObjectIteratorVariable

  #156460 commented on Jul 10, 2025 • 0 new comments

• Change t.is_cuda to t.device.type == 'cuda' in torch/utils/viz

  #156418 commented on Jul 7, 2025 • 0 new comments

• [iter] Add support for sequence protocol in `iter(..)`

  #156371 commented on Jul 9, 2025 • 0 new comments

• [BE][1/6] fix typos in test/

  #157635 commented on Jul 12, 2025 • 0 new comments

• [pruning] Implement Taylor expansion unstructured pruning

  #157620 commented on Jul 12, 2025 • 0 new comments

• [pruning] add more test cases for pruning

  #157613 commented on Jul 7, 2025 • 0 new comments

• [dynamo] Move skipIf decorator to class level in test_fx_graph_runnable

  #157594 commented on Jul 10, 2025 • 0 new comments

• [CUDA][NVTX] use `pytorch` nvtx domain for pytorch ranges

  #157586 commented on Jul 8, 2025 • 0 new comments

• Linux py 3.14 wheel builds

  #157559 commented on Jul 9, 2025 • 0 new comments

• [BE][5/5] fix typos in aten/ (aten/src/ATen/)

  #157554 commented on Jul 12, 2025 • 0 new comments

• [indcutor] pack linear for FP32 dynamic mode

  #157542 commented on Jul 7, 2025 • 0 new comments

• Add a test for checking that the CUDA stubs directory is not in libcaffe2_nvrts.so's RPATH or RUNPATH

  #157437 commented on Jul 7, 2025 • 0 new comments

• [build] bootstrap git repo for build for non-git-clone archive

  #157432 commented on Jul 12, 2025 • 0 new comments

• Add a flag "realized" in IRNode to enable tracking origin_nodes

  #157423 commented on Jul 8, 2025 • 0 new comments

• [CI] Fixes CI for CUDA Version > 12.9

  #157385 commented on Jul 10, 2025 • 0 new comments

• [inductor] Fix memory layout for concatenation of repeated input

  #157380 commented on Jul 7, 2025 • 0 new comments

• [cherry-pick] temporarily disabling generation of weblinks for torch v2.8 …

  #157353 commented on Jul 7, 2025 • 0 new comments

• Making input dynamically adjust.

  #157324 commented on Jul 9, 2025 • 0 new comments

• Using torch.accelerator in comm_mode_features_example.py and visualize_sharding_example.py

  #157317 commented on Jul 10, 2025 • 0 new comments

• Fix inconsistent pybind11 usage across ONNX and Tensorpipe during CMake build

  #157309 commented on Jul 10, 2025 • 0 new comments

• Test re-enabling ET test

  #157298 commented on Jul 9, 2025 • 0 new comments

• adding the ability to record aten arg vals and types

  #157291 commented on Jul 8, 2025 • 0 new comments

• [nativert] add memory overlap debug assertion

  #157290 commented on Jul 7, 2025 • 0 new comments

• Update docs dependencies

  #157287 commented on Jul 12, 2025 • 0 new comments

• [inductor][templates] Finalize all registered hooks

  #157270 commented on Jul 7, 2025 • 0 new comments

• Fix the Problems About Defining Static Variable in Inline Function

  #157269 commented on Jul 9, 2025 • 0 new comments

• Fix init CUDA preload: get correct versions (#147001)

  #157264 commented on Jul 9, 2025 • 0 new comments

• [distributed] build enum for Backend class

  #157263 commented on Jul 7, 2025 • 0 new comments

• [dynamo][fsdp] Consistent behavior of int attributes

  #157262 commented on Jul 11, 2025 • 0 new comments

• Updating default value of eps in RMSNorm documentation

  #157223 commented on Jul 9, 2025 • 0 new comments

• Adding bias argument to NN normalization methods

  #157198 commented on Jul 11, 2025 • 0 new comments

• [DO NOT MERGE] Test new MI300X capacity.

  #157191 commented on Jul 12, 2025 • 0 new comments

• [SymmMem] Enable NVL72

  #157180 commented on Jul 10, 2025 • 0 new comments

• [OrderedDict] Implement `OrderedDict.move_to_end(key, last=False)`

  #155152 commented on Jul 10, 2025 • 0 new comments

• Fix conversion of values in libtorch agnostic tests

  #155115 commented on Jul 9, 2025 • 0 new comments

• [dict] Implement dict.__ior__ and fix return type in dict.__or__

  #155072 commented on Jul 10, 2025 • 0 new comments

• Avoid differing results in `linalg.(tensor_)solve`

  #154983 commented on Jul 10, 2025 • 0 new comments

• [CI] Removing --user flag from all pip install commands

  #154900 commented on Jul 10, 2025 • 0 new comments

• [BE]: Try to enable LTO

  #154819 commented on Jul 5, 2025 • 0 new comments

• [vision hash update] update the pinned vision hash

  #154694 commented on Jul 12, 2025 • 0 new comments

• Add `scale` complex type check in `quantize_per_tensor`

  #154601 commented on Jul 10, 2025 • 0 new comments

• Use official CUDAToolkit module in CMake

  #154595 commented on Jul 7, 2025 • 0 new comments

• Enable Leak Sanitizer

  #154584 commented on Jul 7, 2025 • 0 new comments

• Fixes Issue #154491

  #154561 commented on Jul 11, 2025 • 0 new comments

• implement MKLGenerator

  #154199 commented on Jul 10, 2025 • 0 new comments

• [cuBLASLt][cuBLAS] Support 2D bias and `beta != 1.0` in cuBLASLt

  #154170 commented on Jul 8, 2025 • 0 new comments

• [BE]: Update pybind11 submodule to 3.0.0rc

  #154115 commented on Jul 10, 2025 • 0 new comments

• Fused RMSNorm implementation

  #153666 commented on Jul 12, 2025 • 0 new comments

• [BE]: Update CUTLASS submodule to 4.0.0

  #153541 commented on Jul 9, 2025 • 0 new comments

• [dynamo][compile-time] Cache frame summaries

  #153434 commented on Jul 12, 2025 • 0 new comments

• [PT2][Optimus][fp8 compuation quantizatoin] Add fallback logic

  #153430 commented on Jul 12, 2025 • 0 new comments

• [BE] Move `BUILD_AOT_INDUCTOR_TEST` to build stage

  #153419 commented on Jul 12, 2025 • 0 new comments

• defer to aot eager instead of skip frame

  #153409 commented on Jul 12, 2025 • 0 new comments

• Print correct variable names in cuda.cmake

  #153402 commented on Jul 12, 2025 • 0 new comments

• [ONNX] Cast before calling Softmax when dtype is specified

  #153393 commented on Jul 12, 2025 • 0 new comments

• Remove mut marker for fused_adagrad in native_functions.yaml

  #153376 commented on Jul 12, 2025 • 0 new comments

• [DEBUG] REmove has CUDA

  #153349 commented on Jul 11, 2025 • 0 new comments

• [Dynamo][TVM] Check TVM existence and version

  #153338 commented on Jul 12, 2025 • 0 new comments

• [don't merge] upgrade vs2022 to v17.13.6

  #153322 commented on Jul 9, 2025 • 0 new comments

• [DEBUG] only comment

  #153320 commented on Jul 9, 2025 • 0 new comments

• [DEBUG] only combined_traceback

  #153319 commented on Jul 9, 2025 • 0 new comments

• [DEBUG] dump combined_traceback

  #153318 commented on Jul 9, 2025 • 0 new comments

• [associative_scan] Autograd for additional inputs

  #153317 commented on Jul 9, 2025 • 0 new comments

• [iter] exhaust `ListIterator` when `unpack_var_sequence` is called

  #156370 commented on Jul 9, 2025 • 0 new comments

• [iter] Update some of the tests to not call pickle

  #156369 commented on Jul 9, 2025 • 0 new comments

• [WIP] Add a new API of allocator setting for accelerator

  #156175 commented on Jul 10, 2025 • 0 new comments

• [executorch hash update] update the pinned executorch hash

  #156141 commented on Jul 12, 2025 • 0 new comments

• [CUDA] Use runtime driver API for cuStreamWriteValue32

  #156097 commented on Jul 12, 2025 • 0 new comments

• [build] remove upper version pin for `setuptools<80.0`

  #156049 commented on Jul 12, 2025 • 0 new comments

• Fix atleast_{1,2,3}d() with no arguments description

  #156042 commented on Jul 10, 2025 • 0 new comments

• [DRAFT][cuDNN][SDPA] Introduce `TORCH_CUDNN_SDPA_AVOID_RECOMPILE=1`

  #155958 commented on Jul 12, 2025 • 0 new comments

• [dynamo] Add `-> bool` to functions named `is_*` or `_is_*`

  #155923 commented on Jul 5, 2025 • 0 new comments

• add sfdp pattern

  #155792 commented on Jul 9, 2025 • 0 new comments

• Fix torch.export.export() GPU failure with RNN modules.

  #155734 commented on Jul 9, 2025 • 0 new comments

• docs: clean up docstring for clarity and correctness

  #155712 commented on Jul 9, 2025 • 0 new comments

• [Optimus] add einsum_to_pointwise_pass pattern

  #155666 commented on Jul 11, 2025 • 0 new comments

• Make upsample accept list scale_factor

  #155654 commented on Jul 9, 2025 • 0 new comments

• DOC: update CrossEntropyLoss with note and example of incorrect target specification

  #155649 commented on Jul 8, 2025 • 0 new comments

• [dict] Implement dict subclass `fromkeys` classmethod

  #155608 commented on Jul 11, 2025 • 0 new comments

• [DRAFT] Evaluate feasability of using FunctionalTensor for Example Value

  #155606 commented on Jul 8, 2025 • 0 new comments

• [aoti][mps] Enable test_aot_inductor.py tests

  #155598 commented on Jul 10, 2025 • 0 new comments

• [Misc] fix distributed/_tools/test_sac_ilp.py::TestSACILP::test_sac_i…

  #155548 commented on Jul 8, 2025 • 0 new comments

• [OrderedDict] Add `bool(OrderedDict)`

  #155503 commented on Jul 10, 2025 • 0 new comments

• [OrderedDict] Set the correct dict class in UserDefinedDictVariable

  #155502 commented on Jul 10, 2025 • 0 new comments

• [OrderedDict] Implement `hasattr(..., IteratorVariable)`

  #155501 commented on Jul 10, 2025 • 0 new comments

• [Dynamo] Enable torch function dispatch on HOPs

  #155452 commented on Jul 10, 2025 • 0 new comments

• Use unpack instructions for vec256 (de)interleave2

  #155440 commented on Jul 9, 2025 • 0 new comments

• [scan] Fix issues with scan on CPU and for autograd when implementing an RNN with multiple layers

  #155422 commented on Jul 10, 2025 • 0 new comments

• Convert onnx torchscript rst to md

  #155390 commented on Jul 12, 2025 • 0 new comments

• [einops] Ensure Dynamo can trace through einops

  #155310 commented on Jul 8, 2025 • 0 new comments

• Add UT for torch.accelerator memory-related API

  #155200 commented on Jul 10, 2025 • 0 new comments

• [dict] Implement `__eq__` for dict_items

  #155154 commented on Jul 10, 2025 • 0 new comments

• [OrderedDict] Implement `OrderedDict.popitem(last=...)`

  #155153 commented on Jul 10, 2025 • 0 new comments

• Export always give a value range with max length - 1

  #156882 commented on Jul 9, 2025 • 0 new comments

• DISABLED test_nn_module (__main__.TestGuardSerialization)

  #153120 commented on Jul 9, 2025 • 0 new comments

• DISABLED test_aot_autograd_runtime_wrapper_prologue_profiled (__main__.ReproTests)

  #156678 commented on Jul 9, 2025 • 0 new comments

• DISABLED test_inductor_all_reduce_non_contig_input (__main__.CompileTest)

  #147733 commented on Jul 9, 2025 • 0 new comments

• DISABLED test_rng (__main__.TestCompilerBisector)

  #139590 commented on Jul 9, 2025 • 0 new comments

• DISABLED test_dynamic_warmup (__main__.CudaGraphTreeTests)

  #156693 commented on Jul 9, 2025 • 0 new comments

• DISABLED test_relative_import (__main__.ReproTests)

  #156679 commented on Jul 9, 2025 • 0 new comments

• DISABLED test_vmap_exhaustive_mv_cuda_float32 (__main__.TestVmapOperatorsOpInfoCUDA)

  #142631 commented on Jul 9, 2025 • 0 new comments

• libtorch.so file size is very large

  #34058 commented on Jul 9, 2025 • 0 new comments

• torch.fx.experimental.symbolic_shapes.GuardOnDataDependentSymNode: Could not guard on data-dependent expression Eq(u0, 1) (unhinted: Eq(u0, 1)). (Size-like symbols: none)

  #157355 commented on Jul 9, 2025 • 0 new comments

• Improve Error Message in MultiMarginLoss for Inconsistent Target Size

  #106251 commented on Jul 9, 2025 • 0 new comments

• Training/Fine-tuning fails with PyTorch 2.8 + 4x 5090 GPUs using DDP/FSDP/DeepSpeed

  #150734 commented on Jul 9, 2025 • 0 new comments

• [RFC] Per-Parameter-Sharding FSDP

  #114299 commented on Jul 9, 2025 • 0 new comments

• DISABLED test_simple_multi_arch_embed_kernel_binary_True_cuda (__main__.AOTInductorTestABICompatibleGpu)

  #156930 commented on Jul 9, 2025 • 0 new comments

• DISABLED test_relative_import_no_modulename (__main__.ReproTests)

  #156691 commented on Jul 9, 2025 • 0 new comments

• DISABLED test_bitwise_print_precedence (__main__.ReproTests)

  #156736 commented on Jul 9, 2025 • 0 new comments

• DISABLED test_empty_storage (__main__.CudaGraphTreeTests)

  #156755 commented on Jul 9, 2025 • 0 new comments

• we should graph break on nn.Parameter constructors

  #157452 commented on Jul 8, 2025 • 0 new comments

• torch._dynamo.exc.InternalTorchDynamoError: RuntimeError: Compiler: cl is not found

  #157458 commented on Jul 8, 2025 • 0 new comments

• PyTorch 2.7.1 torch.compile will probably break with einops 0.8.2 or 0.9.0

  #157601 commented on Jul 8, 2025 • 0 new comments

• DISABLED test_ind_worker_queue (__main__.TestIndividualWorkerQueue)

  #68643 commented on Jul 8, 2025 • 0 new comments

• DISABLED test_addr_alpha_beta_out (__main__.ReproTests)

  #156641 commented on Jul 8, 2025 • 0 new comments

• DISABLED test_while_loop_schema_gen (__main__.TestHopSchema)

  #141202 commented on Jul 8, 2025 • 0 new comments

• DISABLED test_graph_break_unsupported_fake (__main__.ReproTests)

  #156629 commented on Jul 8, 2025 • 0 new comments

• DISABLED test_graph_partition_custom_op_dynamoc_shapes (__main__.CudaGraphTreeTests)

  #157428 commented on Jul 8, 2025 • 0 new comments

• DISABLED test_dont_dce_rand (__main__.ReproTests)

  #156580 commented on Jul 8, 2025 • 0 new comments

• DISABLED test_graph_partition_reorder_cpu_and_gpu_interleave (__main__.CudaGraphTreeTests)

  #152561 commented on Jul 8, 2025 • 0 new comments

• canUse32BitIndexMath set to False with efficient net

  #155225 commented on Jul 8, 2025 • 0 new comments

• magma builds should be part of the docker image builds

  #148762 commented on Jul 8, 2025 • 0 new comments

• Cuda-12.9 removed libnvToolsExt.so.* and is now purely header nvtx3

  #152756 commented on Jul 8, 2025 • 0 new comments

• [Release improvements] Have cherry-pick bot always add the current release to the PR

  #152212 commented on Jul 8, 2025 • 0 new comments

• [feature request] torch.mix function to generalize/symmetrize addcmul

  #104849 commented on Jul 8, 2025 • 0 new comments

• DISABLED test_matmul_small_brute_force_1d_Nd_cuda_float32 (__main__.TestLinalgCUDA)

  #125276 commented on Jul 8, 2025 • 0 new comments

• PT2E Quantization Migration Tracker

  #157591 commented on Jul 8, 2025 • 0 new comments

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

  #24422 commented on Jul 8, 2025 • 0 new comments

• DISABLED test_inductor_all_to_all_single (__main__.CompileTest)

  #147795 commented on Jul 9, 2025 • 0 new comments

• DISABLED test_empty_cpu_tensor (__main__.CudaGraphTreeTests)

  #156735 commented on Jul 9, 2025 • 0 new comments

• DISABLED test_dataclass_in_module (__main__.ReproTests)

  #156776 commented on Jul 9, 2025 • 0 new comments

• Add full support for NVIDIA RTX Pro 6000 (Blackwell – SM122 / Compute Capability 12.2)

  #157549 commented on Jul 9, 2025 • 0 new comments

• Allow creation of pseudo devices for testing purposes

  #61654 commented on Jul 9, 2025 • 0 new comments

• Restore CUDA 12.4 manylinux build and test in CI

  #156747 commented on Jul 9, 2025 • 0 new comments

• Importing xgboost before torch + openmp causes seg fault

  #155201 commented on Jul 9, 2025 • 0 new comments

• [CUDA][Compex] `test_reference_numerics_large_jiterator_unary_cuda_complex64` broken after updating to `numpy >= 1.25.0`

  #125198 commented on Jul 9, 2025 • 0 new comments

• FakeTensorUpdater does not trace nodes correctly

  #152548 commented on Jul 9, 2025 • 0 new comments

• Unexpected, batch size and device dependent NaN propagation in Conv1d

  #157237 commented on Jul 9, 2025 • 0 new comments

• Deprecation of NVTX 2 (`nvToolsExt`): Recommended to move to NVTX 3

  #147011 commented on Jul 9, 2025 • 0 new comments

• torch.compile on MPS progress tracker

  #150121 commented on Jul 9, 2025 • 0 new comments

• Export Huggingface models with StaticCache

  #155862 commented on Jul 9, 2025 • 0 new comments

• Add dlpack support for MPS device

  #153789 commented on Jul 9, 2025 • 0 new comments

• torch wheels are unusable if CUDA RPMs are installed on the system (was Import error in nvidia/cuda:12.6.3-cudnn-devel-rockylinux9)

  #150399 commented on Jul 9, 2025 • 0 new comments

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

  #143780 commented on Jul 9, 2025 • 0 new comments

• Tensor.nbytes() returns itemsize * numel for sparse tensors

  #29734 commented on Jul 11, 2025 • 0 new comments

• DISABLED test_TransformerDecoderLayer_gelu_activation_cuda_fp32 (__main__.TestNN)

  #157121 commented on Jul 9, 2025 • 0 new comments

• DISABLED test_graph_partition_custom_op_mutation (__main__.CudaGraphTreeTests)

  #157449 commented on Jul 9, 2025 • 0 new comments

• DISABLED test_Transformer_multilayer_coder_cuda_fp32 (__main__.TestNN)

  #157120 commented on Jul 9, 2025 • 0 new comments

• Pytorch XPU Windows build failed in cmake rerun loop due to the source code deep path

  #134956 commented on Jul 9, 2025 • 0 new comments

• Matmul Triton Template with epilogue fusion can not speed up on XPU.

  #146568 commented on Jul 9, 2025 • 0 new comments

• [XPU][Inductor] Failed to run max-autotune in subprocess.

  #149703 commented on Jul 9, 2025 • 0 new comments

• optree package status in PyTorch

  #152535 commented on Jul 9, 2025 • 0 new comments

• [torch.export] Cannot export TorchVision fasterrcnn_mobilenet_v3_large_fpn

  #146152 commented on Jul 9, 2025 • 0 new comments

• Inductor error with Torch XPU optimizations to StableDiffusion3 Pipeline

  #156303 commented on Jul 9, 2025 • 0 new comments

• `RuntimeError: UR error` with XPU

  #149953 commented on Jul 9, 2025 • 0 new comments

• nn.rmsnorm is super slower than nn.layernorm

  #157345 commented on Jul 9, 2025 • 0 new comments

• Allow clamp to work on absolute values and preserve sign

  #156956 commented on Jul 9, 2025 • 0 new comments

• [inductor][triton] Block ptrs are being removed from Triton

  #154025 commented on Jul 9, 2025 • 0 new comments

• RuntimeError: each element in list of batch should be of equal size

  #42654 commented on Jul 9, 2025 • 0 new comments

• from_blob python api

  #107112 commented on Jul 9, 2025 • 0 new comments

• RuntimeError: Could not find libnvrtc.so. Please make sure CUDA is installed.

  #155378 commented on Jul 9, 2025 • 0 new comments

• get_ema_multi_avg_fn() equation is a little confused

  #155551 commented on Jul 9, 2025 • 0 new comments

• MPS operator coverage tracking issue (2.6+ version)

  #141287 commented on Jul 9, 2025 • 0 new comments

• [feature request] Caching allocator diagnostics and memory allocation tracing/visualization

  #1529 commented on Jul 7, 2025 • 0 new comments

• [Upstream Triton] Handle user-specified triton.set_allocator function

  #155584 commented on Jul 7, 2025 • 0 new comments

• Incremental version of pca_lowrank

  #40770 commented on Jul 7, 2025 • 0 new comments

• Migrating existing backend-MAIA integration toward PrivateUse1 / openReg

  #155864 commented on Jul 7, 2025 • 0 new comments

• `bias: bool` argument for Batch, Instance, Group and RMSNorm

  #157144 commented on Jul 7, 2025 • 0 new comments

• Regression: torch.distributed.gather_object segfaults

  #157627 commented on Jul 7, 2025 • 0 new comments

• Einsum of 2 dtensors fails in inference mode

  #157631 commented on Jul 7, 2025 • 0 new comments

• [DTensor] Better communication cost model for redistribute

  #157585 commented on Jul 7, 2025 • 0 new comments

• `TORCH_DISTRIBUTED_DEBUG=DETAIL` causes DTensors to raise errors

  #157622 commented on Jul 7, 2025 • 0 new comments

• [Export] Non-strict mode can't handle conditionals on tensor subclass types

  #153429 commented on Jul 7, 2025 • 0 new comments

• cmake: add USE_SYSTEM_{KLEIDI,CUDNN_FRONTEND,CUTLASS,FMT} options to USE_SYSTEM_LIBS

  #153863 commented on Jul 7, 2025 • 0 new comments

• DISABLED test_sdpa_mask_fp16_L6_S17_NH23_HS121 (__main__.TestSDPA)

  #138905 commented on Jul 7, 2025 • 0 new comments

• MPS SDPA returns NaN when attention mask blocks all rows

  #156707 commented on Jul 7, 2025 • 0 new comments

• Flex Attention is incompatible with selective AC

  #147879 commented on Jul 7, 2025 • 0 new comments

• [dynamo] Improve trace rules reasoning

  #150435 commented on Jul 7, 2025 • 0 new comments

• `torch.combinations` exhibits excessive memory usage and hangs for moderate `n` and `r` due to `n^r`

  #153337 commented on Jul 7, 2025 • 0 new comments

• [complex] dropout and it's variants should support complex tensors

  #80256 commented on Jul 7, 2025 • 0 new comments

• DISABLED test_graph_partition_cpu_tensor_symints (__main__.CudaGraphTreeTests)

  #157367 commented on Jul 8, 2025 • 0 new comments

• RendezvousConnectionError when use C10d on multi nodes

  #69197 commented on Jul 5, 2025 • 0 new comments

• I want to calculate the matrix multiplication of two Boolean matrices, but torch.mm will report an error. Is there any more efficient alternative?

  #107041 commented on Jul 5, 2025 • 0 new comments

• Make tlparse able to show a summary of distinct graph breaks

  #153669 commented on Jul 5, 2025 • 0 new comments

• ROCm+gcc 15 asserts

  #145608 commented on Jul 5, 2025 • 0 new comments

• Pipeline Parallelism Fails when stage input does not produce gradients in all stages.

  #152827 commented on Jul 6, 2025 • 0 new comments

• ImportError: libcupti.so.11.2: cannot open shared object file: No such file or directory

  #88802 commented on Jul 6, 2025 • 0 new comments

• Add `is_outputs_batched` param to `autograd.grad`

  #156616 commented on Jul 6, 2025 • 0 new comments

• file_name is not correctly read in here

  #157624 commented on Jul 7, 2025 • 0 new comments

• Segmentation fault in torch.repeat_interleave

  #157097 commented on Jul 7, 2025 • 0 new comments

• Incorrect inference of the groups parameter type for channel_stuffle (int misclassified as Tensor)

  #157603 commented on Jul 7, 2025 • 0 new comments

• Deprecation of CUTLASS Python interface

  #157456 commented on Jul 7, 2025 • 0 new comments

• FlexAttention + int64 indexing

  #157446 commented on Jul 7, 2025 • 0 new comments

• nll_loss gives result when both input and target are 1D tensor

  #157420 commented on Jul 7, 2025 • 0 new comments

• Several `torch.*` functions raise uninformative `NotImplementedError`s when called with integer `dtype`

  #157547 commented on Jul 7, 2025 • 0 new comments

• DISABLED test_inductor_reduce_scatter_tensor_coalesced (__main__.CompileTest)

  #147887 commented on Jul 7, 2025 • 0 new comments

• DISABLED test_dont_aggressively_write_assert (__main__.ReproTests)

  #156570 commented on Jul 7, 2025 • 0 new comments

• [CI] Need better way to detect OOMs especially on pet instances

  #157379 commented on Jul 7, 2025 • 0 new comments

• [FSDP2] figure out the contract for mp_policy and tensor subclass extention

  #157395 commented on Jul 7, 2025 • 0 new comments

• Perf drop when running with FSDP and torch.compile

  #156966 commented on Jul 8, 2025 • 0 new comments

• NCCL out of memory error after updating to PyTorch 2.7

  #152302 commented on Jul 8, 2025 • 0 new comments

• `torch.compile` fails on `torch.vdot` with complex tensors

  #157607 commented on Jul 8, 2025 • 0 new comments

• Running dispatch modes on compile-disabled regions of a compiled model

  #155825 commented on Jul 8, 2025 • 0 new comments

• DDP+TP composition does not work as expected

  #157445 commented on Jul 8, 2025 • 0 new comments

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

  #147913 commented on Jul 8, 2025 • 0 new comments

• General MPS op coverage tracking issue

  #77764 commented on Jul 8, 2025 • 0 new comments

• distributed/tensor/_op_schema has_symints does not check args_schema

  #151106 commented on Jul 8, 2025 • 0 new comments

• Shared `~/.cache/torch_extensions` needs to be pytorch version aware.

  #68905 commented on Jul 8, 2025 • 0 new comments

• [OSS tooling] pytorchbot fail to revert a PR

  #156607 commented on Jul 8, 2025 • 0 new comments

• RNN pseudocode wrong?

  #157457 commented on Jul 8, 2025 • 0 new comments

• Most requested ops for the MPS backend

  #154052 commented on Jul 8, 2025 • 0 new comments

• Is there some official method to extract the featuremap of each node in pt2 graph like the function torchvision.models.feature_extraction.create_feature_extractor()

  #157625 commented on Jul 8, 2025 • 0 new comments

• `version.txt` mismatch with tags in release branch

  #151425 commented on Jul 8, 2025 • 0 new comments

• FakeTensorUpdater doesn't support HOPs

  #156819 commented on Jul 8, 2025 • 0 new comments

• einops 0.6.1 x torch.compile broken in pytorch nightlies

  #157417 commented on Jul 8, 2025 • 0 new comments

• [Regression] The torchbench model resnet50_quantized_qat fail_to_run in Pytorch 2.8 but pass in PyTorch 2.7

  #157434 commented on Jul 8, 2025 • 0 new comments

• TorchInductor CPU Performance Dashboard

  #93531 commented on Jul 7, 2025 • 0 new comments

• Add the XPU item to pytorch.org/get-started

  #156810 commented on Jul 7, 2025 • 0 new comments

• functorch_maml_omniglot is a bad CPU performance smoketest model

  #156511 commented on Jul 7, 2025 • 0 new comments

• test_dtensor.py::test_dtensor_save_load_import conflicts with autoloader importing torch._dynamo

  #157545 commented on Jul 7, 2025 • 0 new comments

• Cannot copy data from one gpu to another using torch

  #157398 commented on Jul 7, 2025 • 0 new comments

• [WIP][RFC] Compilable flex_attention + Context Parallel

  #157015 commented on Jul 7, 2025 • 0 new comments

• [dynamo] using disable inside of compile always recompiles

  #157399 commented on Jul 7, 2025 • 0 new comments

• `torch.compile` fails on `prims.broadcast_in_dim` with alias annotation error

  #157610 commented on Jul 7, 2025 • 0 new comments

• Both DTensor TP and SP are missing the last collective in the backward pass

  #157606 commented on Jul 7, 2025 • 0 new comments

• `torch.compile` fails with `NotImplementedError: Unsupported for now if query, key, value are the same buffer.` in `flex_attention`

  #157612 commented on Jul 7, 2025 • 0 new comments

• `torch.export` ViT+flex attention: `Attempting to use FunctionalTensor on its own`

  #140400 commented on Jul 8, 2025 • 0 new comments

• Support reductions in FlexAttention's score_mod/mask_mod

  #141627 commented on Jul 8, 2025 • 0 new comments

• DISABLED test_cat_max_autotune_triton (__main__.TestMaxAutotune)

  #145830 commented on Jul 8, 2025 • 0 new comments

• DISABLED test_is_isnot (__main__.TestScript)

  #120694 commented on Jul 8, 2025 • 0 new comments

• DISABLED test_get_parameter_dtype (__main__.ReproTests)

  #156598 commented on Jul 8, 2025 • 0 new comments

• DISABLED test_add_sub_alpha_out (__main__.ReproTests)

  #156597 commented on Jul 8, 2025 • 0 new comments

• DISABLED test_ranks_and_tag (__main__.CompileTest)

  #147974 commented on Jul 8, 2025 • 0 new comments

• DISABLED test_graph_partition_custom_op (__main__.CudaGraphTreeTests)

  #157413 commented on Jul 8, 2025 • 0 new comments

• DISABLED test_remove_noop_slice1_cuda (__main__.GPUTests)

  #151381 commented on Jul 11, 2025 • 0 new comments

• DISABLED test_execution_into_recording (__main__.CudaGraphTreeTests)

  #156838 commented on Jul 11, 2025 • 0 new comments

• [release 2.8-2.9] Delete support for Maxwell, Pascal, and Volta architectures for CUDA 12.8 and 12.9 builds

  #157517 commented on Jul 11, 2025 • 0 new comments

• vLLM tests failing in torch 2.8rc but passing with torch 2.7

  #157461 commented on Jul 11, 2025 • 0 new comments

• Inductor output code source nodes is missing nodes for backwards graphs

  #130147 commented on Jul 11, 2025 • 0 new comments

• [Feature Request] Experimental support to Moore Threads GPU MUSA

  #151303 commented on Jul 11, 2025 • 0 new comments

• [RFC] PT2-Friendly Traceable, Functional Collective Communication APIs

  #93173 commented on Jul 11, 2025 • 0 new comments

• document functional_collectives

  #113669 commented on Jul 11, 2025 • 0 new comments

• Updated Scaled_mm to support more scaling formats via CuBlas

  #153555 commented on Jul 11, 2025 • 0 new comments

• support side effects in HOPs?

  #124866 commented on Jul 11, 2025 • 0 new comments

• non-negative least squares solver feature request

  #48972 commented on Jul 11, 2025 • 0 new comments

• a log_softmax kernel get much worse perf with padding

  #122840 commented on Jul 11, 2025 • 0 new comments

• `torch.compile` doesn't consider the alias tensor created by `tensor[:]`

  #94773 commented on Jul 11, 2025 • 0 new comments

• libTorch cpp docs missing for Tensor::item()

  #41213 commented on Jul 11, 2025 • 0 new comments

• compile PixArt-sigma error

  #128012 commented on Jul 11, 2025 • 0 new comments

• Add capturable Adagrad implementation

  #118715 commented on Jul 11, 2025 • 0 new comments

• Lowering after pointwise cat can lead to uncontiguous memory accesses

  #124002 commented on Jul 11, 2025 • 0 new comments

• [torch.compile]: Enhanced Error Reporting and Performance Canary Mode

  #126644 commented on Jul 10, 2025 • 0 new comments

• torch.view_copy(x, dtype) diverges from eager when the destiny dtype has less bytes than the origin

  #129966 commented on Jul 10, 2025 • 0 new comments

• torch.compile x custom ops: op that accepts float also accepts Tensor in eager-mode

  #123470 commented on Jul 10, 2025 • 0 new comments

• [RFC] Emit better Telemetry in PyTorch

  #103173 commented on Jul 10, 2025 • 0 new comments

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

  #146769 commented on Jul 11, 2025 • 0 new comments

• Process never ends when sending tensors through multiprocessing queues in Python 3.12+ with filesystem strategy

  #153050 commented on Jul 11, 2025 • 0 new comments

• [inductor] Incorrect handle of `autocast` results in type mismatch

  #121631 commented on Jul 11, 2025 • 0 new comments

• switch more test cases to use MultithreadTestCase

  #108744 commented on Jul 11, 2025 • 0 new comments

• Change `automatic_dynamic_shapes` to trigger on `cache_size_limit` recompiles but not `accumulated_cache_size_limit` recompiles.

  #114516 commented on Jul 11, 2025 • 0 new comments

• [Torch Inductor] Torch Inductor Better Support for GNN workload and Inductor Sparse Compiler

  #113232 commented on Jul 11, 2025 • 0 new comments

• [Feature] Taylor expansion pruning

  #157218 commented on Jul 11, 2025 • 0 new comments

• [rocm] HIP Graph (on AMD GPU) capture does not raise `operation not permitted` for illegal operation whereas CUDA Graph (Nvidia GPU) does

  #155684 commented on Jul 11, 2025 • 0 new comments

• DISABLED test_comprehensive_nn_functional_conv_transpose3d_cuda_float32 (__main__.TestInductorOpInfoCUDA)

  #148853 commented on Jul 11, 2025 • 0 new comments

• Documentation Clarification Needed for Clamping of Scale Coefficient in clip_grads_with_norm_

  #151554 commented on Jul 11, 2025 • 0 new comments

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

  #148315 commented on Jul 11, 2025 • 0 new comments

• Torch RPC examples from docs say usage is deprecated.

  #149393 commented on Jul 11, 2025 • 0 new comments

• Using Inductor always throws a warning

  #154160 commented on Jul 11, 2025 • 0 new comments

• [discussion] Analyzing a list of tensors stored as intermediate values / saved_for_backward in autograd graph

  #91692 commented on Jul 11, 2025 • 0 new comments

• DISABLED test_remove_noop_slice_cuda (__main__.GPUTests)

  #151383 commented on Jul 11, 2025 • 0 new comments

• DISABLED test_expanded_inputs (__main__.CudaGraphTreeTests)

  #156886 commented on Jul 11, 2025 • 0 new comments

• DISABLED test_inductor_all_gather_into_tensor_coalesced (__main__.CompileTest)

  #146806 commented on Jul 11, 2025 • 0 new comments

• DISABLED test_fallback_to_eager_if_recompiling_too_many_times (__main__.CudaGraphTreeTests)

  #130749 commented on Jul 11, 2025 • 0 new comments

• DISABLED test_op_has_batch_rule___rmatmul___cuda_float32 (__main__.TestVmapOperatorsOpInfoCUDA)

  #157003 commented on Jul 11, 2025 • 0 new comments

• DISABLED test_remove_noop_slice_scatter_cuda (__main__.GPUTests)

  #151378 commented on Jul 11, 2025 • 0 new comments

• DISABLED test_against_reference_multi_input_jacfwd_cuda (__main__.TestJacCUDA)

  #156998 commented on Jul 11, 2025 • 0 new comments

• DISABLED test_remove_noop_slice1_cpu (__main__.CpuTests)

  #151379 commented on Jul 11, 2025 • 0 new comments

• DISABLED test_inplace_on_view_makes_base_require_grad_cpu (__main__.TestAutogradDeviceTypeCPU)

  #156209 commented on Jul 11, 2025 • 0 new comments

• DISABLED test_remove_noop_slice_scatter_cpu (__main__.CpuTests)

  #151382 commented on Jul 11, 2025 • 0 new comments

• DISABLED test_remove_noop_slice_cpu (__main__.CpuTests)

  #151384 commented on Jul 11, 2025 • 0 new comments

• Move loop ordering after fusion

  #126255 commented on Jul 11, 2025 • 0 new comments

• [CD] Windows Wheel builds CUDA 12.9.1 Stack Overflow during build

  #156181 commented on Jul 11, 2025 • 0 new comments

• Triton kernel doing more work uses less registers

  #126463 commented on Jul 11, 2025 • 0 new comments

• [dynamo] Investigate interop issues with torch_scatter/torch_sparse/pyg_lib

  #111223 commented on Jul 11, 2025 • 0 new comments

• [DTensor] Improve `tensor_metadata` and `redistribute_cost` coverage for op strategies.

  #157495 commented on Jul 11, 2025 • 0 new comments

• FSDP offload doesn't prefetch param to GPU

  #157209 commented on Jul 12, 2025 • 0 new comments

• Tune whether to use mm or bmm for matmul in inductor max-autotune

  #118774 commented on Jul 11, 2025 • 0 new comments

• Torch compile fx graph is not removing constant propagation

  #120057 commented on Jul 11, 2025 • 0 new comments

• Use sys.settrace or torch function mode to compute how much of a model was not covered by Dynamo

  #120079 commented on Jul 11, 2025 • 0 new comments

• torch.compile Conv1d AssertionError

  #123242 commented on Jul 11, 2025 • 0 new comments

• Use Incremental Fake Tensor Updater more uniformly across torch.compile compilation

  #120116 commented on Jul 11, 2025 • 0 new comments

• [Inductor] Generate triton block pointers for discontiguous strided tensors

  #125077 commented on Jul 11, 2025 • 0 new comments

• opcheck should support TorchBind custom classes

  #121162 commented on Jul 11, 2025 • 0 new comments

• Match HuggingFace T5 SDPA pattern in Inductor

  #121371 commented on Jul 11, 2025 • 0 new comments

• Triton kernel unexpectedly gets 1.35x slower by more specializaiton

  #120667 commented on Jul 11, 2025 • 0 new comments

• [export] 14k models: AssertionError: graph-captured input # 2, of type <class 'torch.nn.parameter.Parameter'>, is not among original inputs of types

  #111693 commented on Jul 11, 2025 • 0 new comments

• [Inductor] Freezing Add support for Caching Parameter Conversions

  #103990 commented on Jul 11, 2025 • 0 new comments

• Rework Dynamic Benchmarks To Actually Vary Shapes

  #113063 commented on Jul 11, 2025 • 0 new comments

• Minifier doesn't work with dynamic shapes

  #114296 commented on Jul 11, 2025 • 0 new comments

• [pt2] Unable to trace LSTM with dynamic sequence length

  #115092 commented on Jul 11, 2025 • 0 new comments

• 2 Dynamo test are failing with "Global state changed while dynamo tracing, please report a bug".

  #120648 commented on Jul 11, 2025 • 0 new comments

• Higher peak memory with torch.compile

  #122512 commented on Jul 11, 2025 • 0 new comments

• torch.compile doesn't convert all input scalar types to symbolic values

  #119778 commented on Jul 11, 2025 • 0 new comments

• Inconsistent Behavior of `torch.dsplit` with torch.compile

  #118741 commented on Jul 11, 2025 • 0 new comments

• Triton Kernel Rejects NamedTupleVariable Arguments

  #148289 commented on Jul 10, 2025 • 0 new comments

• Inconsistent export behavior for nonzero+grid_sample between CUDA and CPU/MPS backends

  #152791 commented on Jul 10, 2025 • 0 new comments

• DISABLED test_mm_plus_mm (__main__.TestPatternMatcher)

  #145335 commented on Jul 10, 2025 • 0 new comments

• [RFC] Offload collectives to NVSwitch when possible

  #136567 commented on Jul 10, 2025 • 0 new comments

• [Tracker] AutoParallel's feature request to DTensor

  #156217 commented on Jul 10, 2025 • 0 new comments

• SourcelessBuilder.create does not know how to wrap <class '__main__.InFlexData'>

  #154009 commented on Jul 10, 2025 • 0 new comments

• Is compilation caching for NumPy operators not supported in PyTorch 2.7.1?

  #156943 commented on Jul 10, 2025 • 0 new comments

• DISABLED test_end_recording_early (__main__.CudaGraphTreeTests)

  #156778 commented on Jul 10, 2025 • 0 new comments

• DISABLED test_inductor_inplace_op_on_view (__main__.CompileTest)

  #147852 commented on Jul 10, 2025 • 0 new comments

• DISABLED test_reentrant_parent_error_on_cpu_cuda (__main__.TestAutogradDeviceTypeCUDA)

  #86735 commented on Jul 10, 2025 • 0 new comments

• DISABLED test_dataclass_init_with_default_factory_with_inputs (__main__.ReproTests)

  #156799 commented on Jul 10, 2025 • 0 new comments

• FlopCounterMode doesn't support HOP

  #134385 commented on Jul 10, 2025 • 0 new comments

• Batched multi_dot / chain_matmul + let it accept a tensor instead of tuple

  #55261 commented on Jul 10, 2025 • 0 new comments

• `__getitem__` fails to vmap for one dimensional tensors

  #124423 commented on Jul 10, 2025 • 0 new comments

• DISABLED test_error_on_dealloc_use (__main__.CudaGraphTreeTests)

  #156801 commented on Jul 10, 2025 • 0 new comments

• DISABLED test_comprehensive_pca_lowrank_cuda_float32 (__main__.TestInductorOpInfoCUDA)

  #139828 commented on Jul 10, 2025 • 0 new comments

• DISABLED test_inductor_reduce_scatter_tensor_single (__main__.CompileTest)

  #147911 commented on Jul 10, 2025 • 0 new comments

• Outdated install commands

  #152213 commented on Jul 10, 2025 • 0 new comments

• PyTorch 2.6 License Issues

  #150118 commented on Jul 9, 2025 • 0 new comments

• [XPU User Empathy Day][whisper][Arc770][Win]XPU performance is worse than CPU

  #151985 commented on Jul 9, 2025 • 0 new comments

• xpu: missing aten ops needed to support Huggingface quanto

  #132947 commented on Jul 9, 2025 • 0 new comments

• Preload CUDA fails if CUDA libs in different PYTHONPATH

  #147001 commented on Jul 9, 2025 • 0 new comments

• Install pytorch from pypi using local CUDA build

  #150742 commented on Jul 9, 2025 • 0 new comments

• ☂️ Update submodule dependencies to supported version of Cmake

  #150328 commented on Jul 9, 2025 • 0 new comments

• Remove Direct Arm Compute Libray (ACL) Integration for Quantized Matmuls: `qlinear`/`qlinear_dynamic`

  #148902 commented on Jul 9, 2025 • 0 new comments

• Multihead Attention does not work with jagged tensors due to __torch_function__

  #153472 commented on Jul 9, 2025 • 0 new comments

• [RFC] Experimental Wheel Variant Support

  #155141 commented on Jul 9, 2025 • 0 new comments

• Gross mismatch in PDF between CUDA and CPU for multivariate Gaussian mixture models

  #156959 commented on Jul 9, 2025 • 0 new comments

• [ONNX] Create a tutorial for exporting hf transformers model

  #156258 commented on Jul 9, 2025 • 0 new comments

• `bytes(...)` support of torch tensor does not match numpy + it would be nice to support tensor.tobytes() as alias

  #108565 commented on Jul 9, 2025 • 0 new comments

• cpp_extension.py expects an integer on CUDA_ARCH, failing with Grace Hopper.

  #144037 commented on Jul 9, 2025 • 0 new comments

• torch.nn.InstanceNorm2d throws "mixed dtype" error with track_running_stats set to True

  #139140 commented on Jul 9, 2025 • 0 new comments

• Failure with cub::TransformInputIterator in 12.9 periodic CI test

  #157502 commented on Jul 9, 2025 • 0 new comments

• Matmul with int32 parameters on Intel GPU leads to errors

  #144766 commented on Jul 9, 2025 • 0 new comments

• Dump bytecode of resumption frames in tlparse

  #136038 commented on Jul 9, 2025 • 0 new comments

• Regression in llama2 model export

  #157323 commented on Jul 10, 2025 • 0 new comments

• ROCm, 7900 XTX: Pytorch FLASH_ATTENTION SDPA is 2.5x slower than MATH (fp16, head_dim 256, seqlen 4360, 12 heads)

  #152595 commented on Jul 10, 2025 • 0 new comments

• MPS Memory Leak

  #154329 commented on Jul 10, 2025 • 0 new comments

• torch._dynamo.mark_static_address refuses to work with nn.Parameter

  #157221 commented on Jul 10, 2025 • 0 new comments

• PyTorch source code build failed on some Windows 11 environment caused by C++ protocol buffer compiler

  #143795 commented on Jul 10, 2025 • 0 new comments

• [inductor][dynamic shapes] hugging face models fail while creating error guard

  #157330 commented on Jul 10, 2025 • 0 new comments

• An error occurs when ‘max_split_size_mb ’and ‘expandable_segments ’ are enabled at the same time.

  #123548 commented on Jul 10, 2025 • 0 new comments

• Segmentation faults in test_ops.py tests with gcc13 on AArch64 (v1)

  #157626 commented on Jul 10, 2025 • 0 new comments

• [feature request] Native checkpointing to/from `s3://`

  #155992 commented on Jul 10, 2025 • 0 new comments

• ImportError: cannot import name 'scaled_mm_configs' from 'torch._inductor.kernel.mm_common

  #157343 commented on Jul 10, 2025 • 0 new comments

• [inductor][cpu]mobilenet_v2_quantized_qat float32 single thread static/dynamic shape CPP/default wrapper performance regression in 2024-04-28 nightly release

  #125672 commented on Jul 10, 2025 • 0 new comments

• torch.compile not compatible with multiprocessing pool

  #97992 commented on Jul 10, 2025 • 0 new comments

• Higher train loss and worse evaluation metrics when using `torch.compile()`

  #113180 commented on Jul 10, 2025 • 0 new comments

• Different behaviors in `torch.nn.functional.hinge_embedding_loss` between eagermode and torch.compile

  #118175 commented on Jul 10, 2025 • 0 new comments

• CompiledFxGraph.current_callable is not thread-safe

  #138961 commented on Jul 10, 2025 • 0 new comments

• In Inductor-wrapped tests, reset() before and after each test and turn off suppress_errors=True

  #122804 commented on Jul 10, 2025 • 0 new comments

• Out of bounds error with `nn.MultiMarginLoss`

  #105597 commented on Jul 10, 2025 • 0 new comments

• Undefined symbol: cuOccupancyMaxActiveClusters

  #115075 commented on Jul 10, 2025 • 0 new comments

• DISABLED test_inductor_reuse_buffer_after_inplace_collective (__main__.CompileTest)

  #147950 commented on Jul 10, 2025 • 0 new comments

• DISABLED test_error_on_dealloc_use2 (__main__.CudaGraphTreeTests)

  #156808 commented on Jul 10, 2025 • 0 new comments

• DISABLED test_deferred_runtime_asserts (__main__.ReproTests)

  #156817 commented on Jul 10, 2025 • 0 new comments

• [torch.compile] tighten FX graph restrictions post-functionalization

  #133250 commented on Jul 10, 2025 • 0 new comments

• [proposal] "Name" string attribute for modules, parameters, buffers, tensors for more pleasant debugging (especially for graph printouts / export / studying compiled generated code)

  #104247 commented on Jul 10, 2025 • 0 new comments

• empty_cache does not work for CUDAPluggableAllocator + MemPool

  #145168 commented on Jul 10, 2025 • 0 new comments

• [ONNX] Flip `dynamo` default to True in torch.onnx.export

  #151693 commented on Jul 10, 2025 • 0 new comments

• [MPS] Migrate torch.sort to Metal shader

  #155560 commented on Jul 10, 2025 • 0 new comments

• StrideAPI caused regression in channels-last logic

  #141836 commented on Jul 10, 2025 • 0 new comments

• MPS Performance regressions on Sonoma 14.0

  #111517 commented on Jul 10, 2025 • 0 new comments

• MaxPool2D memory leakage on device MPS

  #125217 commented on Jul 10, 2025 • 0 new comments

• [MPS] BatchNorm2D produces incorrect results for column first tensors

  #134580 commented on Jul 10, 2025 • 0 new comments

• Inefficient 2D convolution compared to JAX

  #157334 commented on Jul 10, 2025 • 0 new comments

• Fix broken linalg unittests on ARM platform

  #125438 commented on Jul 10, 2025 • 0 new comments

• torch.compile bug when using resize

  #155209 commented on Jul 10, 2025 • 0 new comments

• Compilation issues with ROCm 6.4.1 on Debian 12

  #155794 commented on Jul 10, 2025 • 0 new comments

• PyTorch CPP Extensions fail when same kernel is compiled more than once on ROCm servers

  #155344 commented on Jul 10, 2025 • 0 new comments

• DISABLED test_slice_scatter_reinplace_cuda (__main__.GPUTests)

  #145189 commented on Jul 10, 2025 • 0 new comments