9 Pull requests merged by 2 people

• Fix cuda 12.9 aarch64 GPU builds. Update CUDA_STABLE variable.

  #157641 merged Jul 4, 2025

• Remove +PTX from CUDA 12.8 builds

  #157634 merged Jul 4, 2025

• Cleanup leftover miniconda brew installation

  #157567 merged Jul 4, 2025

• Fix GITHUB_OUTPUT syntax in create_release.yml workflow

  #157539 merged Jul 4, 2025

• [aarch64] Add back NCCL lib to cuda arm wheel

  #157105 merged Jul 4, 2025

• [MPS] Revert cumsum/cumprod to MPSGraph implementation

  #157494 merged Jul 3, 2025

• [ez] Disable some failing periodic tests

  #157560 merged Jul 3, 2025

• Revert "Update triton version to 3.4"

  #157471 merged Jul 2, 2025

• [ROCm] Bump AOTriton to 0.10b

  #156845 merged Jun 30, 2025

154 Pull requests opened by 85 people

• [distributed] build enum for Backend class

  #157263 opened Jun 30, 2025

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

  #157264 opened Jun 30, 2025

• [inductor] fix tensor.to(uint8) error when tensor src type is float

  #157267 opened Jun 30, 2025

• Fix the Problems About Defining Static Variable in Inline Function

  #157269 opened Jun 30, 2025

• [inductor][templates] Finalize all registered hooks

  #157270 opened Jun 30, 2025

• Update docs dependencies

  #157287 opened Jun 30, 2025

• [nativert] add memory overlap debug assertion

  #157290 opened Jun 30, 2025

• adding the ability to record aten arg vals and types

  #157291 opened Jun 30, 2025

• Fixes typo in nccl_window_registration test

  #157293 opened Jun 30, 2025

• Enable `file_descriptor` strategy on Darwin

  #157295 opened Jun 30, 2025

• Test re-enabling ET test

  #157298 opened Jun 30, 2025

• [AOTI][experiment]

  #157301 opened Jun 30, 2025

• [dynamo] Fix source for lru_cache method

  #157308 opened Jun 30, 2025

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

  #157309 opened Jun 30, 2025

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

  #157317 opened Jun 30, 2025

• Making input dynamically adjust.

  #157324 opened Jun 30, 2025

• Add inductor lowerings for adaptive_avg_pool3d/adaptive_max_pool3d

  #157331 opened Jun 30, 2025

• [BE] Rename TorchVersion -> VersionString

  #157333 opened Jul 1, 2025

• Make the name assert actually do something, and reserve some more names

  #157342 opened Jul 1, 2025

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

  #157344 opened Jul 1, 2025

• Add Intel GPU info collection to the collect env script

  #157351 opened Jul 1, 2025

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

  #157353 opened Jul 1, 2025

• [BE] Update xpu driver repo for CD used almalinux 8.10

  #157356 opened Jul 1, 2025

• [BE] fix typo: inpt -> input

  #157361 opened Jul 1, 2025

• Fix diagnostic message for CUDA version mismatch in cuda.cmake

  #157370 opened Jul 1, 2025

• [HF][DCP] Upload local consolidated files to remote storage if needed

  #157371 opened Jul 1, 2025

• [submodule][cutlass] Update pin to b995f93 v4.0.0

  #157376 opened Jul 1, 2025

• [release/2.8] update Triton 3.4 pin to f81f19a7

  #157377 opened Jul 1, 2025

• 101385: Warning message when non-coo tensors are passed to `is_sparse`

  #157378 opened Jul 1, 2025

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

  #157380 opened Jul 1, 2025

• [multi-kernel][fix-comments] attempt-1

  #157384 opened Jul 1, 2025

• [CI] Fixes CI for CUDA Version > 12.9

  #157385 opened Jul 1, 2025

• Add explicit typing to nn.Module.__init__() parameters

  #157389 opened Jul 1, 2025

• [xpu] Correctly load RNG state during XPU checkpointing

  #157390 opened Jul 1, 2025

• [dynamic shapes] allocate fresh symbols for slice

  #157392 opened Jul 1, 2025

• Fix is_unaligned usage of statically_known_true

  #157400 opened Jul 1, 2025

• [SymmMem] Install NVSHMEM wheel in CI docker

  #157411 opened Jul 2, 2025

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

  #157418 opened Jul 2, 2025

• Preserve current stream in TestCuda::test_stream_compatibility

  #157421 opened Jul 2, 2025

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

  #157422 opened Jul 2, 2025

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

  #157423 opened Jul 2, 2025

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

  #157430 opened Jul 2, 2025

• [build] make SDist buildable: bootstrap git repo and submodules

  #157432 opened Jul 2, 2025

• [test] Yanbing/tf32 dev

  #157433 opened Jul 2, 2025

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

  #157437 opened Jul 2, 2025

• Fix FlexAttention int64 indexing for large tensors

  #157447 opened Jul 2, 2025

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

  #157454 opened Jul 2, 2025

• Add legacy note to autograd.profiler doc.

  #157459 opened Jul 2, 2025

• [PowerPC]: Fixed build issue that occur because of datatype f8 enablement for onednn in qlinear and prepack

  #157469 opened Jul 2, 2025

• [dynamo] Add an assertion in guards to fail early for non-sequence length checks

  #157478 opened Jul 2, 2025

• [wip] torch._dynamo.save/load() for saving and loading compiled models.

  #157481 opened Jul 2, 2025

• Fix typo: 'tracable' → 'traceable' in torch/_dynamo/variables/torch.py

  #157483 opened Jul 2, 2025

• [BE] rewrite `CacheBase` and `LocalCache` as generics

  #157493 opened Jul 2, 2025

• Add test for user-managed weights with load_state_dict

  #157496 opened Jul 2, 2025

• [EXPERIMENTL][dynamo] Remove `input_source_to_var`

  #157497 opened Jul 2, 2025

• Add `max_pool3d` backward pass for MPS

  #157498 opened Jul 2, 2025

• [EXPERIMENTAL] turn on `torch._dynamo.config.capture_scalar_outputs` by default

  #157499 opened Jul 2, 2025

• [EXPERIMENTAL] turn on `torch._dynamo.config.capture_dynamic_output_shape_ops` by default

  #157500 opened Jul 2, 2025

• [DeviceMesh] Use user set backend and pg option even for the global mesh

  #157501 opened Jul 2, 2025

• [autograd] Avoid creating and recording event when unnecessary

  #157503 opened Jul 2, 2025

• [1/N] cost coverage improvment

  #157504 opened Jul 2, 2025

• [refactor][dynamo] make BUILD_TUPLE instruction use inst.arg

  #157505 opened Jul 2, 2025

• [WIP][FSDP2] support dataclass args/kwargs and output

  #157506 opened Jul 2, 2025

• [DO NOT MERGE] Clone of PR #157309

  #157507 opened Jul 2, 2025

• [wip] inspect output code

  #157508 opened Jul 2, 2025

• [ONNX] Fix conversion of attention - 4D

  #157509 opened Jul 2, 2025

• [wip] merge async and progressive

  #157510 opened Jul 2, 2025

• [dynamo] fix infinite loop in computing all stack meta

  #157511 opened Jul 2, 2025

• [dynamo] Fix bug in dict(mapping_proxy)

  #157515 opened Jul 2, 2025

• [PGO] include module int attributes in PGO state

  #157518 opened Jul 3, 2025

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

  #157519 opened Jul 3, 2025

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

  #157520 opened Jul 3, 2025

• [c10d] support dynamic shapes for all_to_all_single_autograd

  #157521 opened Jul 3, 2025

• [DeviceMesh] Add error when users try to slice non contiguous flattened dim submesh

  #157523 opened Jul 3, 2025

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

  #157524 opened Jul 3, 2025

• [br][pc] consolidate attempt 1

  #157526 opened Jul 3, 2025

• [dynamo, docs] add dynamo programming model docs

  #157527 opened Jul 3, 2025

• [WIP] avoid unnecessary slices

  #157528 opened Jul 3, 2025

• [FSDP2] Use reduceOpSum for world size 1

  #157529 opened Jul 3, 2025

• Fix typo: 'reset_paramteres' → 'reset_parameters' in transformer.cpp

  #157536 opened Jul 3, 2025

• handling special case for pow(3) for GPU

  #157537 opened Jul 3, 2025

• Don't try installing missing cuda dependencies on s390x

  #157540 opened Jul 3, 2025

• S390x update test marks

  #157541 opened Jul 3, 2025

• [indcutor] pack linear for FP32 dynamic mode

  #157542 opened Jul 3, 2025

• Add is_hidden_event method to KinetoEvent Python interface

  #157546 opened Jul 3, 2025

• [BE][1/5] fix typos in aten/

  #157550 opened Jul 3, 2025

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

  #157551 opened Jul 3, 2025

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

  #157552 opened Jul 3, 2025

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

  #157553 opened Jul 3, 2025

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

  #157554 opened Jul 3, 2025

• Try adding sm_50-sm_70 arches for linux cuda 12.8 builds

  #157558 opened Jul 3, 2025

• Linux py 3.14 wheel builds

  #157559 opened Jul 3, 2025

• [PT2][memory] mutation size correctness

  #157562 opened Jul 3, 2025

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

  #157563 opened Jul 3, 2025

• [dynamo] [guard] Change the guard type of inside disable function to avoid unnecessary recompilation.

  #157566 opened Jul 3, 2025

• [MPS][DO NOT MERGE] CI signals for conv nan issue on macOS CPU

  #157568 opened Jul 3, 2025

• [simplefsdp auto-bucketing] ir node runtime estimation

  #157572 opened Jul 3, 2025

• Test case for nanogpt

  #157576 opened Jul 3, 2025

• Fixed the function to get the origin nodes of fused triton kernel.

  #157578 opened Jul 3, 2025

• [fbcode] switch to cutlass-4

  #157579 opened Jul 3, 2025

• allow user to pass in custom partitioner function

  #157580 opened Jul 3, 2025

• Fix typo: 'initalizer' → 'initializer' in test_reductions.cpp

  #157581 opened Jul 3, 2025

• allow _size_of to return individual element's size

  #157582 opened Jul 3, 2025

• correctly import torch.version

  #157584 opened Jul 3, 2025

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

  #157586 opened Jul 3, 2025

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

  #157588 opened Jul 3, 2025

• Add stack trace of exception to MultiProcContinousTest

  #157589 opened Jul 3, 2025

• Add master switch for aot_inductor.compile_standalone

  #157590 opened Jul 3, 2025

• [AOTI] Split aoti_runtime/model.h to prepare for model static linking

  #157592 opened Jul 3, 2025

• Fix einsum strategy shard dim > ndim

  #157593 opened Jul 3, 2025

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

  #157594 opened Jul 3, 2025

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

  #157595 opened Jul 3, 2025

• Fix typo: 'inital_grad' → 'initial_grad' in FSDP test

  #157596 opened Jul 3, 2025

• Fix einops x torch.compile interaction

  #157600 opened Jul 4, 2025

• [DRAFT] DDE-Free select with unbacked index.

  #157605 opened Jul 4, 2025

• [aot] add format_consts_to_cpp function for further development.

  #157608 opened Jul 4, 2025

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

  #157609 opened Jul 4, 2025

• [pruning] add more test cases for pruning

  #157613 opened Jul 4, 2025

• tlparse remove duplicate reasons

  #157618 opened Jul 4, 2025

• [pruning] Implement Taylor expansion unstructured pruning

  #157620 opened Jul 4, 2025

• [nativert] Move ModelRunnerBase to oss.

  #157633 opened Jul 4, 2025

• [BE][1/6] fix typos in test/

  #157635 opened Jul 4, 2025

• [BE][2/6] fix typos in test/ (test/test_*.py)

  #157636 opened Jul 4, 2025

• [BE][3/6] fix typos in test/

  #157637 opened Jul 4, 2025

• [BE][6/6] fix typos in test/ (test/distributed/)

  #157640 opened Jul 4, 2025

• Fix typo: 'initalization' → 'initialization' in profiler test comment

  #157645 opened Jul 4, 2025

• [MemoryViz] Add file selector button

  #157647 opened Jul 4, 2025

• Fix typo: 'occurance' → 'occurrence' in typing test

  #157649 opened Jul 4, 2025

• Fix typo: 'paramter' → 'parameter' in dynamo variable comment

  #157651 opened Jul 4, 2025

• [wip] async cancellation test

  #157652 opened Jul 4, 2025

• Fix typo: 'reset_paramteres' → 'reset_parameters' in transformer module comments

  #157656 opened Jul 4, 2025

• Fixes issue 157195 by adding error message

  #157658 opened Jul 5, 2025

• [wip] merge async and progressive compiles

  #157659 opened Jul 5, 2025

• Fix typo: 'occurance' → 'occurrence' in lazy extract_compiled_graph.py

  #157664 opened Jul 5, 2025

• Fix typo: 'initalizer' → 'initializer' in test_reductions.cpp

  #157667 opened Jul 5, 2025

• Fix 'dllimport attribute ignored on inline function'

  #157670 opened Jul 6, 2025

• Fix index_put propagate strategy arg unpack error

  #157671 opened Jul 6, 2025

• Fix torch._numpy advanced indexing to match NumPy when indices are separated

  #157676 opened Jul 6, 2025

• [pt2 event logging] add configurable prefix

  #157678 opened Jul 6, 2025

• installing requirements.txt fix

  #157681 opened Jul 6, 2025

• [dtensor] add support for fused optimizer with parameters across multiple meshes

  #157682 opened Jul 7, 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

• [canary] dedupe args + on by default

  #157690 opened Jul 7, 2025

• [canary] dedupe args + on by default

  #157691 opened Jul 7, 2025

• [BE][Easy] add `.editorconfig` setting for C/C++/CUDA/ObjC

  #157692 opened Jul 7, 2025

• [CI] Fix xpu ci test sccache issue

  #157693 opened Jul 7, 2025

• fix storage use_count

  #157694 opened Jul 7, 2025

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

  #157695 opened Jul 7, 2025

• Update slow tests

  #157696 opened Jul 7, 2025

84 Issues closed by 25 people

• DISABLED test_ranks_and_tag (__main__.CompileTest)

  #147974 closed Jul 7, 2025

• DISABLED test_dont_dce_rand (__main__.ReproTests)

  #156580 closed Jul 7, 2025

• DISABLED test_add_complex_conj (__main__.ReproTests)

  #156579 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_non_contiguous_input_mm_plus_mm (__main__.TestMaxAutotune)

  #126867 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

• test_ops.py extremely slow on cuda11.3

  #79528 closed Jul 5, 2025

• Torch.compile Dynamo failed to run FX node with fake tensors

  #157657 closed Jul 5, 2025

• Fix warning #177-D: variable "threshold" was declared but never referenced

  #157653 closed Jul 5, 2025

• DISABLED test_is_isnot (__main__.TestScript)

  #120694 closed Jul 4, 2025

• DISABLED test_sdpa_mask_fp16_L6_S17_NH23_HS121 (__main__.TestSDPA)

  #138905 closed Jul 4, 2025

• DISABLED test_foreach_reduce_large_input__foreach_max_w_empty_False_cuda_int32 (__main__.TestForeachCUDA)

  #156497 closed Jul 4, 2025

• DISABLED test_Linear_cuda_tf32 (__main__.TestNN)

  #155216 closed Jul 4, 2025

• DISABLED test_graph_partition_forward_backward (__main__.CudaGraphTreeTests)

  #157615 closed Jul 4, 2025

• Importing `torch` overwrites `typing.TypeIs` when `_running_with_deploy()` is true.

  #153942 closed Jul 4, 2025

• INTERNAL ASSERT FAILED in mse_loss when mixing CPU and CUDA tensors

  #154978 closed Jul 4, 2025

• DISABLED test_foreach_reduce_large_input__foreach_max_w_empty_False_cuda_int16 (__main__.TestForeachCUDA)

  #156430 closed Jul 4, 2025

• DISABLED test_graph_partition_dynamic_shapes (__main__.CudaGraphTreeTests)

  #157555 closed Jul 4, 2025

• [aot_compile]Explanation: Dynamo does not know how to trace the builtin `time.time.`

  #157352 closed Jul 4, 2025

• [inductor] `F.fractional_max_pool2d` throws `AssertionError` on Inductor when input `rank=3`

  #156682 closed Jul 4, 2025

• Set inplace operations are not updating the set inplace

  #153552 closed Jul 4, 2025

• dynamo cannot trace global op_set .__contains__

  #145761 closed Jul 4, 2025

• Why scale value of GradScaler sudden changed?

  #157436 closed Jul 4, 2025

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

  #157602 closed Jul 4, 2025

• A more flexible API for torch.compile fullgraph=True

  #144908 closed Jul 3, 2025

• Suggestion: integration of einops test suite

  #146782 closed Jul 3, 2025

• DISABLED test_set_stance_aot_eager_then_compile (__main__.DecoratorTests)

  #148644 closed Jul 3, 2025

• DISABLED test_graph_partition_custom_op_no_split (__main__.CudaGraphTreeTests)

  #157532 closed Jul 3, 2025

• Wrong vector shift results on PowerPC

  #109777 closed Jul 3, 2025

• Enhanced torch.chunk and torch.split

  #60531 closed Jul 3, 2025

• DISABLED test_foreach_reduce_large_input__foreach_max_w_empty_False_cuda_float64 (__main__.TestForeachCUDA)

  #153544 closed Jul 3, 2025

• DISABLED test_graph_partition_custom_op_mutation (__main__.CudaGraphTreeTests)

  #157448 closed Jul 3, 2025

• DISABLED [WORKFLOW_NAME] / [PLATFORM_NAME] / [JOB_NAME]

  #157530 closed Jul 3, 2025

• [CPU][flex attention] Llama 3 failed on CPU with PyTorch 2025-06-22 nightly wheel

  #156688 closed Jul 3, 2025

• an illegal memory access was encountered global exception

  #136407 closed Jul 3, 2025

• Bug with "make latexpdf"

  #135420 closed Jul 3, 2025

• pytorch 2.1.2+cu118, RTX 8000, backward show RuntimeError: Expected is_sm80 || is_sm90 to be true, but got false.

  #136979 closed Jul 3, 2025

• some tests regarding `torch.export` in `transformers` fail with `torch 2.8.0 rc` but pass with `torch 2.7.1`

  #157284 closed Jul 2, 2025

• Tiny Typo in Docs

  #157444 closed Jul 2, 2025

• [inductor] vision_maskrcnn dashboard failure on H100 (and MI300)

  #157316 closed Jul 2, 2025

• Profiler: Add hide metadata flag to skip events in key_averages() table

  #155213 closed Jul 2, 2025

• torch.compile triton kernel errors when there are """ docblocks

  #155006 closed Jul 2, 2025

• DISABLED test_graph_partition_custom_op_dynamoc_shapes (__main__.CudaGraphTreeTests)

  #157426 closed Jul 2, 2025

• DISABLED test_graph_partition_custom_op (__main__.CudaGraphTreeTests)

  #157412 closed Jul 2, 2025

• AttributeError: '_OpNamespace' 'aten' object has no attribute 'momentum'

  #145274 closed Jul 2, 2025

• DISABLED test_name_match (__main__.TestGuardSerialization)

  #156246 closed Jul 2, 2025

• DISABLED test_shape_env (__main__.TestGuardSerialization)

  #156264 closed Jul 2, 2025

• DISABLED test_graph_partition_cpu_tensor_symints (__main__.CudaGraphTreeTests)

  #157366 closed Jul 2, 2025

• torch.export produce stack_trace for output node that can fail decomposition

  #157183 closed Jul 1, 2025

• [MPS] `torch.compile` fails on `torch.linalg.cholesky` (possible memory layout issue?)

  #156658 closed Jul 1, 2025

• DISABLED test_foreach_reduce_large_input__foreach_max_w_empty_False_cuda_float32 (__main__.TestForeachCUDA)

  #153470 closed Jul 1, 2025

• test issue, ignore this

  #157151 closed Jul 1, 2025

• the example program using libtorch is not linked against torch_cuda even when USE_CUDA is defined

  #148770 closed Jul 1, 2025

• DISABLED test_graph_partition_cpu_scalar_mutation (__main__.CudaGraphTreeTests)

  #157358 closed Jul 1, 2025

• [ROCm] support torch._C._set_sm_carveout_experimental - Parity with Nvidia

  #149280 closed Jul 1, 2025

• DISABLED test_graph_partition_cpu_scalar4 (__main__.CudaGraphTreeTests)

  #157347 closed Jul 1, 2025

• DISABLED test_graph_partition_cpu_scalar3 (__main__.CudaGraphTreeTests)

  #157338 closed Jul 1, 2025

• [inductor] [triton backend] `Conv2d-unsqueeze-AdaptiveAvgPool3d` output incorrect results on inductor

  #157248 closed Jul 1, 2025

• DISABLED test_graph_partition_cpu_scalar2 (__main__.CudaGraphTreeTests)

  #157311 closed Jul 1, 2025

• avoid guarding on max() unnecessarily

  #149635 closed Jun 30, 2025

• [Upstream Triton] Support new host-side TMA API in user-defined triton kernels

  #155574 closed Jun 30, 2025

• [feature request][AOTI] Expand check input assertions to cover input guards created during compilation?

  #151925 closed Jun 30, 2025

• DISABLED test_lowering_to_x86 (__main__.TestQuantizePT2EX86Inductor)

  #153140 closed Jun 30, 2025

• aot inductor intermediate tensor debug printing (setting 2) not working

  #145425 closed Jun 30, 2025

• Certain MPS operations didn't properly check for data type

  #157303 closed Jun 30, 2025

• Missing MPS-compatible build for PyTorch 2.7.1 on Apple Silicon (M4)

  #157271 closed Jun 30, 2025

• Native BFloat16 Mixed BatchNorm Train gives incorrect gradients

  #156513 closed Jun 30, 2025

• [release] Make pytorch source distribution package respect pep-0517

  #150461 closed Jun 30, 2025

• DISABLED test_ddp_uneven_inputs (__main__.TestDistBackendWithSpawn)

  #75648 closed Jun 30, 2025

• DISABLED test_graph_partition_cpu_scalar1 (__main__.CudaGraphTreeTests)

  #157277 closed Jun 30, 2025

• Why is there such a big difference in size between the torch CUDA 12.6 whl and the cuda 12.9 whl? Why is only the 12.6 whl package placed on the PyPI source?

  #157265 closed Jun 30, 2025

• DISABLED test_quantize (__main__.TestOpenReg)

  #156089 closed Jun 30, 2025

• DISABLED test_jacobian_vectorize_raises_no_warnings_logging_tensor (__main__.TestAutogradFunctional)

  #153707 closed Jun 30, 2025

• DISABLED test_foreach_reduce_large_input__foreach_max_w_empty_False_cuda_float16 (__main__.TestForeachCUDA)

  #153379 closed Jun 30, 2025

• RuntimeError: d.is_cuda() INTERNAL ASSERT FAILED at "/pytorch/c10/cuda/impl/CUDAGuardImpl.h"

  #151486 closed Jun 30, 2025

• DISABLED test_reorder_peak_memory (__main__.TestOperatorReorderForPeakMemory)

  #145332 closed Jun 30, 2025

• DISABLED test_graph_partition_cpu_op_and_dynamic_shapes (__main__.CudaGraphTreeTests)

  #157257 closed Jun 30, 2025

127 Issues opened by 74 people

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

  #157683 opened Jul 7, 2025

• CONTRIBUTING.md install command incorrect

  #157680 opened Jul 6, 2025

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

  #157677 opened Jul 6, 2025

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

  #157675 opened Jul 6, 2025

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

  #157674 opened Jul 6, 2025

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

  #157673 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

• RuntimeError: operator torchvision::nms does not exist

  #157648 opened Jul 4, 2025

• DISABLED test_vmap_exhaustive_dot_cuda_float32 (__main__.TestVmapOperatorsOpInfoCUDA)

  #157644 opened Jul 4, 2025

• DISABLED test_graph_partition_forward_backward_not_called (__main__.CudaGraphTreeTests)

  #157643 opened Jul 4, 2025

• Einsum of 2 dtensors fails in inference mode

  #157631 opened Jul 4, 2025

• Regression: torch.distributed.gather_object segfaults

  #157627 opened Jul 4, 2025

• Segmentation faults in test_ops.py tests with gcc13 on AArch64 (v1)

  #157626 opened Jul 4, 2025

• 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 opened Jul 4, 2025

• file_name is not correctly read in here

  #157624 opened Jul 4, 2025

• `TORCH_DISTRIBUTED_DEBUG=DETAIL` causes DTensors to raise errors

  #157622 opened Jul 4, 2025

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

  #157621 opened Jul 4, 2025

• DISABLED test_vmap_exhaustive_addmv_cuda_float32 (__main__.TestVmapOperatorsOpInfoCUDA)

  #157617 opened Jul 4, 2025

• DISABLED test_graph_partition_forward_backward (__main__.CudaGraphTreeTests)

  #157616 opened Jul 4, 2025

• `torch.compile` fails with `NotImplementedError: Unsupported for now if query, key, value are the same buffer.` in `flex_attention`

  #157612 opened Jul 4, 2025

• `torch.compile` fails on `prims.broadcast_in_dim` with alias annotation error

  #157610 opened Jul 4, 2025

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

  #157607 opened Jul 4, 2025

• Both DTensor TP and SP are missing the last collective in the backward pass

  #157606 opened Jul 4, 2025

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

  #157603 opened Jul 4, 2025

• PyTorch 2.7.1 will probably break with einops 0.8.2 or 0.9.0

  #157601 opened Jul 4, 2025

• PT2E Quantization Migration Tracker

  #157591 opened Jul 3, 2025

• [DTensor] Better communication cost model for redistribute

  #157585 opened Jul 3, 2025

• [precompile] Precompile failure on nanogpt training

  #157577 opened Jul 3, 2025

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

  #157569 opened Jul 3, 2025

• DISABLED test_graph_partition_dynamic_shapes (__main__.CudaGraphTreeTests)

  #157556 opened Jul 3, 2025

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

  #157549 opened Jul 3, 2025

• Nightly cu128 aarch64 wheels haven't been built for weeks

  #157548 opened Jul 3, 2025

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

  #157547 opened Jul 3, 2025

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

  #157545 opened Jul 3, 2025

• Vmap error raised by mask_mod of FlexAttention

  #157543 opened Jul 3, 2025

• PyTorch fails to detect AVX through it's detected

  #157538 opened Jul 3, 2025

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

  #157535 opened Jul 3, 2025

• DISABLED test_vmap_exhaustive___rmatmul___cuda_float32 (__main__.TestVmapOperatorsOpInfoCUDA)

  #157534 opened Jul 3, 2025

• DISABLED test_graph_partition_custom_op_no_split (__main__.CudaGraphTreeTests)

  #157533 opened Jul 3, 2025

• pytorch

  #157531 opened Jul 3, 2025

• [release 2.9] Deprecate support for Maxwell, Pascal, and Volta architectures

  #157517 opened Jul 3, 2025

• Failure with cub::TransformInputIterator in 12.9 periodic CI test

  #157502 opened Jul 2, 2025

• [DTensor] Improve `tensor_metadata` and `redistribute_cost` coverage for op strategies.

  #157495 opened Jul 2, 2025

• Quantized version of Gather layer

  #157490 opened Jul 2, 2025

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

  #157485 opened Jul 2, 2025

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

  #157479 opened Jul 2, 2025

• torch 2.6 and torchvision 0.21.0 incompatibility?

  #157476 opened Jul 2, 2025

• [AOTI] Unit test for testing load_state_dict and

  #157474 opened Jul 2, 2025

• Nightly NCCL builds are missing optional features from NCCL

  #157465 opened Jul 2, 2025

• vLLM tests failing in torch 2.8rc but passing with torch 2.7

  #157461 opened Jul 2, 2025

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

  #157458 opened Jul 2, 2025

• RNN pseudocode wrong?

  #157457 opened Jul 2, 2025

• Deprecation of CUTLASS Python interface

  #157456 opened Jul 2, 2025

• we should graph break on nn.Parameter constructors

  #157452 opened Jul 2, 2025

• Dynamo's einops version check is bogus

  #157451 opened Jul 2, 2025

• DISABLED test_op_has_batch_rule_vdot_cuda_float32 (__main__.TestVmapOperatorsOpInfoCUDA)

  #157450 opened Jul 2, 2025

• DISABLED test_graph_partition_custom_op_mutation (__main__.CudaGraphTreeTests)

  #157449 opened Jul 2, 2025

• FlexAttention + int64 indexing

  #157446 opened Jul 2, 2025

• DDP+TP composition does not work as expected

  #157445 opened Jul 2, 2025

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

  #157434 opened Jul 2, 2025

• ``torch.quantile`` edge case

  #157431 opened Jul 2, 2025

• DISABLED test_graph_partition_custom_op_dynamoc_shapes (__main__.CudaGraphTreeTests)

  #157428 opened Jul 2, 2025

• DISABLED test_addmm_relu_cuda_float32 (__main__.TestLinalgCUDA)

  #157427 opened Jul 2, 2025

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

  #157425 opened Jul 2, 2025

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

  #157420 opened Jul 2, 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 opened Jul 2, 2025

• einops 0.6.1 x torch.compile broken in pytorch nightlies

  #157417 opened Jul 2, 2025

• DISABLED test_old_cholesky_batched_upper_cuda_float32 (__main__.TestLinalgCUDA)

  #157415 opened Jul 2, 2025

• DISABLED test_addmm_mv_transpose_a_True_transpose_b_True_alpha_0_2_beta_1_0_cuda_float32 (__main__.TestLinalgCUDA)

  #157414 opened Jul 2, 2025

• DISABLED test_graph_partition_custom_op (__main__.CudaGraphTreeTests)

  #157413 opened Jul 2, 2025

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

  #157409 opened Jul 2, 2025

• [autograd] Slowdown in backward after #151079

  #157407 opened Jul 1, 2025

• Calling unbind on 2D NestedTensor throws RuntimeError

  #157404 opened Jul 1, 2025

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

  #157401 opened Jul 1, 2025

• [dynamo] using disable inside of compile always recompiles

  #157399 opened Jul 1, 2025

• Cannot copy data from one gpu to another using torch

  #157398 opened Jul 1, 2025

• [dynamo] non-strict trace'd functions cannot return constants

  #157397 opened Jul 1, 2025

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

  #157395 opened Jul 1, 2025

• How to compose HSDP with CP?

  #157393 opened Jul 1, 2025

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

  #157391 opened Jul 1, 2025

• Torch is unusable when cuda-12.4 is installed locally

  #157381 opened Jul 1, 2025

• [CI] M2Pro MacOS-15 tests are unstable again

  #157379 opened Jul 1, 2025

• DISABLED test_addmm_mv_transpose_a_True_transpose_b_True_alpha_0_2_beta_0_5_cuda_float32 (__main__.TestLinalgCUDA)

  #157369 opened Jul 1, 2025

• DISABLED test_op_has_batch_rule_nn_functional_conv_transpose2d_cuda_float32 (__main__.TestVmapOperatorsOpInfoCUDA)

  #157368 opened Jul 1, 2025

• DISABLED test_graph_partition_cpu_tensor_symints (__main__.CudaGraphTreeTests)

  #157367 opened Jul 1, 2025

• [MPS] test_linalg_cholesky fails on M4

  #157364 opened Jul 1, 2025

• torch.Tensor.addmm_ The calculation result is inconsistent with the formula calculation result

  #157360 opened Jul 1, 2025

• DISABLED test_graph_partition_cpu_scalar_mutation (__main__.CudaGraphTreeTests)

  #157359 opened Jul 1, 2025

• 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 opened Jul 1, 2025

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

  #157354 opened Jul 1, 2025

• DISABLED test_graph_partition_cpu_scalar4 (__main__.CudaGraphTreeTests)

  #157350 opened Jul 1, 2025

• DISABLED test_matmul_small_brute_force_3d_Nd_cuda_float32 (__main__.TestLinalgCUDA)

  #157349 opened Jul 1, 2025

• DISABLED test_addmm_mv_transpose_a_True_transpose_b_False_alpha_0_2_beta_1_0_cuda_float32 (__main__.TestLinalgCUDA)

  #157348 opened Jul 1, 2025

• DISABLED test_conv2d_api (__main__.TestQuantizedFunctionalOps)

  #157346 opened Jul 1, 2025

• nn.rmsnorm is super slower than nn.layernorm

  #157345 opened Jul 1, 2025

• ImportError: cannot import name 'scaled_mm_configs' from 'torch._inductor.kernel.mm_common

  #157343 opened Jul 1, 2025

• DISABLED test_graph_partition_cpu_scalar3 (__main__.CudaGraphTreeTests)

  #157339 opened Jul 1, 2025

• DISABLED test_matmul_small_brute_force_2d_Nd_cuda_float32 (__main__.TestLinalgCUDA)

  #157337 opened Jul 1, 2025

• DISABLED test_addmm_mv_transpose_a_True_transpose_b_False_alpha_0_2_beta_0_5_cuda_float32 (__main__.TestLinalgCUDA)

  #157336 opened Jul 1, 2025

• DISABLED test_op_has_batch_rule_nn_functional_conv2d_strided_padding_dilation_with_bias_cuda_float32 (__main__.TestVmapOperatorsOpInfoCUDA)

  #157335 opened Jul 1, 2025

• Inefficient 2D convolution compared to JAX

  #157334 opened Jul 1, 2025

• [inductor][dynamic shapes] hugging face models fail while creating error guard

  #157330 opened Jun 30, 2025

• Regression in llama2 model export

  #157323 opened Jun 30, 2025

• Symmetric memory test failed with TORCH_SYMMMEM=NVSHMEM

  #157321 opened Jun 30, 2025

• Torch Elastic Wait timeout increase

  #157318 opened Jun 30, 2025

• DISABLED test_op_has_batch_rule_nn_functional_conv2d_strided_padding_dilation_no_bias_cuda_float32 (__main__.TestVmapOperatorsOpInfoCUDA)

  #157315 opened Jun 30, 2025

• DISABLED test_linalg_solve_triangular_cuda_float32 (__main__.TestLinalgCUDA)

  #157314 opened Jun 30, 2025

• DISABLED test_addmm_mv_transpose_a_False_transpose_b_True_alpha_0_2_beta_1_0_cuda_float32 (__main__.TestLinalgCUDA)

  #157313 opened Jun 30, 2025

• DISABLED test_graph_partition_cpu_scalar2 (__main__.CudaGraphTreeTests)

  #157312 opened Jun 30, 2025

• PyTorch Tutorial Audit - ONNX

  #157300 opened Jun 30, 2025

• DISABLED test_tensordot_cuda (__main__.TestLinalgCUDA)

  #157297 opened Jun 30, 2025

• DISABLED test_conv1d_api (__main__.TestQuantizedFunctionalOps)

  #157296 opened Jun 30, 2025

• [export] run_decompositions generates inefficient operations

  #157289 opened Jun 30, 2025

• DISABLED test_linalg_matrix_exp_compare_with_taylor_cuda_float32 (__main__.TestLinalgCUDA)

  #157282 opened Jun 30, 2025

• DISABLED test_addmm_mv_transpose_a_False_transpose_b_True_alpha_0_2_beta_0_0_cuda_float32 (__main__.TestLinalgCUDA)

  #157281 opened Jun 30, 2025

• DISABLED test_graph_partition_cpu_scalar1 (__main__.CudaGraphTreeTests)

  #157280 opened Jun 30, 2025

• DISABLED test_addmm_dynamic_shapes_cuda (__main__.DynamicShapesGPUTests)

  #157279 opened Jun 30, 2025

• DISABLED test_op_has_batch_rule_nn_functional_conv2d_cuda_float32 (__main__.TestVmapOperatorsOpInfoCUDA)

  #157278 opened Jun 30, 2025

• several `transformers` tests fail with `torch 2.8 RC` but pass with `torch 2.7.1` on `T4` (but both pass on `A10`)

  #157276 opened Jun 30, 2025

• `torch 2.8 RC` gives 10000 larger output difference in some `transformers` tests

  #157274 opened Jun 30, 2025

• `test_can_compile_fast_image_processor` in `transformers` pass with `torch 2.7` but fail with `torch 2.8 RC`

  #157273 opened Jun 30, 2025

• `torch.reciprocal` and `torch.divide` for Complex `inf` Incorrectly Returns `NaN` Only for Tensors with >= 4 Elements on CPU

  #157272 opened Jun 30, 2025

• Better typechecking of `int` only-operators `|`, `^`, `&`, `<<`, `>>`, `~` and `@`

  #157266 opened Jun 30, 2025

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

  #157363 opened Jun 30, 2025

418 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.

• [ONNX] remove unnecessary slices before converting into onnx

  #157192 commented on Jul 3, 2025 • 23 new comments

• [AOTI] codegen for static linkage

  #157129 commented on Jul 4, 2025 • 16 new comments

• [dynamo] Add FakeProcessGroup support for fx_graph_runnable with distributed collectives

  #157162 commented on Jul 4, 2025 • 9 new comments

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

  #155734 commented on Jul 1, 2025 • 9 new comments

• [AOTI][Intel GPU] Add XPU quantization ops to AOT Inductor.

  #156572 commented on Jul 3, 2025 • 8 new comments

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

  #157140 commented on Jul 2, 2025 • 8 new comments

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

  #156851 commented on Jun 30, 2025 • 7 new comments

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

  #150218 commented on Jul 4, 2025 • 5 new comments

• Added philox based RNG context for HPU device in Dtensor scenarios

  #156581 commented on Jul 3, 2025 • 5 new comments

• Optimize scatter/gather kernel for ARM.

  #156161 commented on Jul 4, 2025 • 4 new comments

• [WIP] Automatically load and save dynamo entries via caching_precompile

  #155913 commented on Jul 3, 2025 • 4 new comments

• Fused RMSNorm implementation

  #153666 commented on Jul 2, 2025 • 4 new comments

• multi-kernel matmuls based on varying hint sizes

  #156628 commented on Jul 3, 2025 • 4 new comments

• [ZENDNN] Integrate ZenDNN library, implement Linear op, add unit-tests

  #156599 commented on Jul 2, 2025 • 4 new comments

• [dynamo] Avoid recompiling over unused objects

  #156891 commented on Jul 3, 2025 • 3 new comments

• [ROCm] logsumexp on ROCm needs scaling back to natural base.

  #156903 commented on Jul 2, 2025 • 3 new comments

• [ci][cutlass backend] Add ci for cutlass backend tests

  #156626 commented on Jul 3, 2025 • 3 new comments

• Adapting pipeline parallelism test cases to be device agnostic

  #155108 commented on Jul 2, 2025 • 3 new comments

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

  #153343 commented on Jul 4, 2025 • 3 new comments

• Fix slice op redistribute_cost compute

  #157178 commented on Jul 3, 2025 • 3 new comments

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

  #155422 commented on Jul 2, 2025 • 3 new comments

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

  #150762 commented on Jul 2, 2025 • 3 new comments

• Add cascade sum support for Inductor CPP backend

  #156296 commented on Jul 3, 2025 • 2 new comments

• Fix: fallback in deserialize_torch_artifact for ScriptObject using weights_only=FalseFix: fallback in deserialize_torch_artifact for ScriptObject using we…

  #154333 commented on Jul 1, 2025 • 2 new comments

• [BE] add a linter to check consistency for cmake minimum version in requirements

  #156961 commented on Jul 3, 2025 • 2 new comments

• [TEST] triton Update 3.4 - 2

  #156664 commented on Jul 4, 2025 • 2 new comments

• Update _torch_docs.py to Fix torch.bernoulli()

  #152104 commented on Jul 2, 2025 • 2 new comments

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

  #156049 commented on Jul 4, 2025 • 2 new comments

• [generator] Close all open generators in compile_subgraph

  #157149 commented on Jul 2, 2025 • 2 new comments

• [CUDA] Use runtime driver API for cuStreamWriteValue32

  #156097 commented on Jul 3, 2025 • 1 new comment

• Enhance testing infrastructure to add half-precision support for `histc` on XPU

  #154339 commented on Jul 3, 2025 • 1 new comment

• ROCm OCP Micro-scaling Format (mx-fp8/mx-fp4) Support

  #151360 commented on Jul 1, 2025 • 1 new comment

• [oss] Add version to metadata

  #155343 commented on Jul 2, 2025 • 1 new comment

• [TEST] Triton 3.4.0 pin update

  #156186 commented on Jul 4, 2025 • 1 new comment

• Use CMake wholearchive group

  #156393 commented on Jul 7, 2025 • 1 new comment

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

  #157262 commented on Jul 2, 2025 • 0 new comments

• [BE]: Update CUTLASS submodule to 4.0.0

  #153541 commented on Jul 4, 2025 • 0 new comments

• implement MKLGenerator

  #154199 commented on Jul 3, 2025 • 0 new comments

• Upgrade MKL in CI

  #154198 commented on Jul 2, 2025 • 0 new comments

• [BE]: Update pybind11 submodule to 3.0.0rc

  #154115 commented on Jul 4, 2025 • 0 new comments

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

  #155649 commented on Jul 3, 2025 • 0 new comments

• [pytorch_146643] fixed max triton generation

  #154056 commented on Jul 2, 2025 • 0 new comments

• [pytorch][triton] Enabling TMA for flex-attention for supported device types

  #153662 commented on Jul 3, 2025 • 0 new comments

• Add MPS implementation of CTC Loss based on CUDA version

  #154044 commented on Jul 2, 2025 • 0 new comments

• [dict] Raise TypeError in dict methods

  #154003 commented on Jul 5, 2025 • 0 new comments

• [list] Implement list.count

  #153969 commented on Jul 5, 2025 • 0 new comments

• [dict] Implement dict subclass `fromkeys` classmethod

  #155608 commented on Jul 5, 2025 • 0 new comments

• [OrderedDict] Set the correct dict class in UserDefinedDictVariable

  #155502 commented on Jul 5, 2025 • 0 new comments

• [OrderedDict] Add `bool(OrderedDict)`

  #155503 commented on Jul 5, 2025 • 0 new comments

• FractionalMaxPool3d add kernel_size check

  #155549 commented on Jul 4, 2025 • 0 new comments

• Fix conversion of values in libtorch agnostic tests

  #155115 commented on Jul 2, 2025 • 0 new comments

• Fixes #154982: add missing to_result_dtype in vector_norm

  #155111 commented on Jul 1, 2025 • 0 new comments

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

  #155152 commented on Jul 5, 2025 • 0 new comments

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

  #155072 commented on Jul 5, 2025 • 0 new comments

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

  #155153 commented on Jul 5, 2025 • 0 new comments

• [Intel GPU] Refactor Matmul integration: Modularize bias handling and memory creation

  #154977 commented on Jul 3, 2025 • 0 new comments

• [dict] Implement `__eq__` for dict_items

  #155154 commented on Jul 5, 2025 • 0 new comments

• update the baseline for nightly max_autotune tests

  #154973 commented on Jul 1, 2025 • 0 new comments

• [OrderedDict] Implement explicit OrderedDict dunder method call

  #154943 commented on Jul 5, 2025 • 0 new comments

• [dict] Implement dict.__eq__ and dict.__ne__

  #154942 commented on Jul 5, 2025 • 0 new comments

• [BE]: Try to enable LTO

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

• [dict] Allow Dynamo to trace through explicit dict dunder method call

  #154794 commented on Jul 5, 2025 • 0 new comments

• [dict] Add dict.popitem

  #154793 commented on Jul 5, 2025 • 0 new comments

• [vision hash update] update the pinned vision hash

  #154694 commented on Jul 7, 2025 • 0 new comments

• Use official CUDAToolkit module in CMake

  #154595 commented on Jul 2, 2025 • 0 new comments

• Fix MKL error: Inconsistent configuration parameters

  #154585 commented on Jul 3, 2025 • 0 new comments

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

  #155501 commented on Jul 5, 2025 • 0 new comments

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

  #154551 commented on Jul 4, 2025 • 0 new comments

• [Dynamo] Guard serialization for BUILTIN_MATCH

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

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

  #152682 commented on Jul 3, 2025 • 0 new comments

• Raise error when no record on extra_files

  #152664 commented on Jul 2, 2025 • 0 new comments

• Add assert_fp8_close helper for FP8 tensor comparisons

  #152651 commented on Jul 5, 2025 • 0 new comments

• [BE]remove vulkan test

  #152643 commented on Jul 1, 2025 • 0 new comments

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

  #152624 commented on Jul 3, 2025 • 0 new comments

• Parameterized CUDA Graph Launch

  #152622 commented on Jul 1, 2025 • 0 new comments

• Update padding_mode type annotation to use Literal type (PaddingMode)

  #152610 commented on Jul 1, 2025 • 0 new comments

• [Testing] Is FindCUDA.cmake from `Modules_CUDA_fix` called at all?

  #152604 commented on Jun 30, 2025 • 0 new comments

• [BE] Delete `Module_CUDA_fix`

  #152603 commented on Jul 1, 2025 • 0 new comments

• [BE] Update numba versions

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

• [compile async] [cache] testing

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

• [inductor] [compile async] Don't compile in eager

  #152507 commented on Jul 5, 2025 • 0 new comments

• fix: Update padding_mode to use Literal for type checking

  #152458 commented on Jul 1, 2025 • 0 new comments

• Add epoch to fake tensor cache key

  #152453 commented on Jul 1, 2025 • 0 new comments

• fix: outdated contents in dynamo overview

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

• Updates to build on Noble (Ubuntu24.04) and py3.12

  #152240 commented on Jul 4, 2025 • 0 new comments

• IGNORE: Testing OIDC

  #152181 commented on Jun 30, 2025 • 0 new comments

• Extend compute_global_tensor_shape to multi dimension sharding

  #152166 commented on Jul 2, 2025 • 0 new comments

• Add dynamo config to HOP-ify context managers

  #152159 commented on Jul 2, 2025 • 0 new comments

• Add standard Python source distribution generation to (pre-)release workflow

  #152098 commented on Jul 3, 2025 • 0 new comments

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

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

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

  #151996 commented on Jul 6, 2025 • 0 new comments

• Skip fuse attention on fp32 if not tf32

  #151924 commented on Jul 4, 2025 • 0 new comments

• Idea: Add SBOM Generation (and optional vuln scan) for better supply chain insight

  #156085 commented on Jul 2, 2025 • 0 new comments

• [CUDA] Allow cuDNN or flash attn in `test_activation_checkpointing` pattern match check

  #153272 commented on Jul 4, 2025 • 0 new comments

• fix dtensor and tensor inconsistent compute mesh

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

• Adding XPU support to DTensor examples

  #153213 commented on Jul 1, 2025 • 0 new comments

• [TESTING] Triton pin (Jul 1) f81f19a7f6cb7f905fde3195014c1bf51659642f

  #153117 commented on Jul 2, 2025 • 0 new comments

• Add CUDA support for Adagrad(fused=True)

  #153038 commented on Jul 1, 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

• [ROCm] Ck gemm architecture guard

  #152951 commented on Jun 30, 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 overall tensor similarity comparison (#152647)

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

• Clarify wrap_triton doc about optional triton_op usage

  #152874 commented on Jul 5, 2025 • 0 new comments

• ci: Remove conda-env-macOS-ARM64, prefer pip

  #152843 commented on Jul 5, 2025 • 0 new comments

• [MSVC] Enable updated lambda processor by setting compiler flag /Zc:lambda globally

  #152828 commented on Jul 5, 2025 • 0 new comments

• another try

  #152808 commented on Jul 4, 2025 • 0 new comments

• wip

  #152807 commented on Jul 4, 2025 • 0 new comments

• Update CMakeLists.txt

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

• added short integer for repeat_interleave_cpu, Fixes #151311

  #152762 commented on Jul 5, 2025 • 0 new comments

• Allow ATen ops overloading

  #152759 commented on Jul 4, 2025 • 0 new comments

• Handle less functions than number of segments

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

• Conditionally support experimental filesystem include in jit_opt_limit

  #152748 commented on Jul 5, 2025 • 0 new comments

• [BE][Cleanup][Dynamo] Stop logging entire_frame_compile_time_s

  #152738 commented on Jul 5, 2025 • 0 new comments

• docs: fix dead link in torch.compile docs

  #152734 commented on Jul 5, 2025 • 0 new comments

• [BE]: Update NCCL to 2.27.5

  #157108 commented on Jul 4, 2025 • 0 new comments

• [Quant][CPU] Enable fp8 qconv

  #157076 commented on Jul 7, 2025 • 0 new comments

• Build CPP Extensions with COLOR

  #157051 commented on Jun 30, 2025 • 0 new comments

• Use std::string_view in torchgen

  #157050 commented on Jul 2, 2025 • 0 new comments

• [a2av] Make test input more random

  #157029 commented on Jul 3, 2025 • 0 new comments

• [EXPERIMENTAL][dynamo] Avoid potential graph breaks by relaxing `handle_traced_output` checks

  #157013 commented on Jul 2, 2025 • 0 new comments

• [itertools] Add CPython tests for itertools

  #156981 commented on Jul 2, 2025 • 0 new comments

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

  #156980 commented on Jun 30, 2025 • 0 new comments

• [TESTING] test new xpu runner

  #156917 commented on Jun 30, 2025 • 0 new comments

• Track monitor

  #156907 commented on Jul 1, 2025 • 0 new comments

• Add cuda 12.9 periodic tests

  #156900 commented on Jul 3, 2025 • 0 new comments

• ci: Add ability to test images for build-triton-wheel

  #156894 commented on Jul 1, 2025 • 0 new comments

• [refactor][dynamo] extract a helper function create_resume_fn from create_call_resume_at

  #156869 commented on Jul 3, 2025 • 0 new comments

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

  #156841 commented on Jul 2, 2025 • 0 new comments

• [logging] [redo] dynamo_timed for CachingAutotuner.coordinate_descent_tuning

  #156840 commented on Jul 3, 2025 • 0 new comments

• [gtest][listing] Enable gtest json listing for the fbcode/caffe2 project

  #156816 commented on Jul 7, 2025 • 0 new comments

• add device generalization support for distributed tests

  #156796 commented on Jul 4, 2025 • 0 new comments

• [inductor] initial triton static config lookup table

  #156785 commented on Jun 30, 2025 • 0 new comments

• [cherry-pick] revert #156552

  #156767 commented on Jul 4, 2025 • 0 new comments

• add tests for Thunk utility function

  #156759 commented on Jun 30, 2025 • 0 new comments

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

  #156753 commented on Jul 6, 2025 • 0 new comments

• WIP `fast_autotune`: Add lookup table and ML model to filter triton matmul configs

  #156683 commented on Jul 1, 2025 • 0 new comments

• Enable set SDPA backend by torch.nn.attention.sdpa_kernel on XPU

  #156669 commented on Jul 4, 2025 • 0 new comments

• [invoke_subgraph] make same subgraph share get_attr target

  #157253 commented on Jun 30, 2025 • 0 new comments

• [cc][pac] attempt 1.1

  #157250 commented on Jun 30, 2025 • 0 new comments

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

  #157241 commented on Jul 2, 2025 • 0 new comments

• remove allow-untyped-defs from torch/ao/pruning/_experimental/pruner/parametrization.py

  #157235 commented on Jul 1, 2025 • 0 new comments

• remove allow-untyped-defs from torch/ao/nn/quantized/modules/rnn.py

  #157234 commented on Jul 5, 2025 • 0 new comments

• remove allow-untyped-defs from torch/backends/mkl/__init__.py

  #157233 commented on Jul 2, 2025 • 0 new comments

• remove allow-untyped-defs from torch/backends/cusparselt/__init__.py

  #157232 commented on Jul 5, 2025 • 0 new comments

• remove allow-untyped-defs from torch/_classes.py

  #157231 commented on Jul 2, 2025 • 0 new comments

• remove allow-untyped-defs from torch/utils/data/_utils/fetch.py

  #157230 commented on Jul 1, 2025 • 0 new comments

• remove allow-untyped-defs from torch/_lazy/__init__.py

  #157228 commented on Jul 2, 2025 • 0 new comments

• Updating default value of eps in RMSNorm documentation

  #157223 commented on Jul 2, 2025 • 0 new comments

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

  #157216 commented on Jul 7, 2025 • 0 new comments

• fix type hints for interpolation functions

  #157202 commented on Jul 6, 2025 • 0 new comments

• Adding bias argument to NN normalization methods

  #157198 commented on Jul 2, 2025 • 0 new comments

• [DO NOT MERGE] Test new MI300X capacity.

  #157191 commented on Jul 2, 2025 • 0 new comments

• [Do Not Merge] moved pytorch mi300 worfklows to test scale sets

  #157190 commented on Jul 3, 2025 • 0 new comments

• [dynamo] auto-rewrite data-dependent if into torch.cond

  #157161 commented on Jul 3, 2025 • 0 new comments

• [dynamo] remove dead object from keepalive

  #157159 commented on Jul 2, 2025 • 0 new comments

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

  #157153 commented on Jul 2, 2025 • 0 new comments

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

  #157152 commented on Jul 2, 2025 • 0 new comments

• [nativert] libtorch kernel registry

  #157150 commented on Jul 7, 2025 • 0 new comments

• [contextlib] Fixes for CPython contextlib tests

  #157148 commented on Jul 2, 2025 • 0 new comments

• [Bugfix][Inductor] Fix dependency list merged incorrectly for a custom op with multiple mutated inputs and None return type.

  #157133 commented on Jul 1, 2025 • 0 new comments

• ReplaceWithCopy graph pass

  #156666 commented on Jul 1, 2025 • 0 new comments

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

  #156175 commented on Jul 1, 2025 • 0 new comments

• Implementation of a ScannedModule

  #156172 commented on Jul 1, 2025 • 0 new comments

• [WIP] Deprecate some functions in CUDAAllocatorConfig, use AcceleratorAllocatorConfig instead

  #156165 commented on Jul 2, 2025 • 0 new comments

• [list] Raise exception in invalid list method call

  #156148 commented on Jul 5, 2025 • 0 new comments

• [executorch hash update] update the pinned executorch hash

  #156141 commented on Jul 7, 2025 • 0 new comments

• [cuDNN][64-bit indexing] update conv depthwise 64bit indexing dispatch condition to match native kernel

  #156140 commented on Jul 3, 2025 • 0 new comments

• Convert to markdown: jit.rst

  #156094 commented on Jun 30, 2025 • 0 new comments

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

  #156042 commented on Jul 1, 2025 • 0 new comments

• [BE][Easy] set end-of-line for `.bat` file to CRLF in `.editorconfig`

  #156032 commented on Jul 7, 2025 • 0 new comments

• [build] modernize build-frontend: `python setup.py develop/install` -> `[uv ]pip install --no-build-isolation [-e ].`

  #156027 commented on Jul 5, 2025 • 0 new comments

• [BE] add a minimal linter to check `pyproject.toml` consistency

  #156017 commented on Jul 5, 2025 • 0 new comments

• Handling overflow for long int overflow for the product of kernel_hei…

  #155989 commented on Jul 3, 2025 • 0 new comments

• [CI][cpp_wrapper] Fix selection of CPU OpInfo tests

  #155967 commented on Jul 2, 2025 • 0 new comments

• [FSDP2] Fix issue with set_reduce_scatter_divide_factor errors and MixedPrecisionPolicy

  #155964 commented on Jul 4, 2025 • 0 new comments

• HF loads dcp - don't do a full deserialize on every file

  #155942 commented on Jul 1, 2025 • 0 new comments

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

  #155928 commented on Jul 4, 2025 • 0 new comments

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

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

• [NOT FOR MERGE] Exploratory work on AOTInductor training

  #155877 commented on Jul 4, 2025 • 0 new comments

• [einops] Ensure Dynamo can trace through explicit set dunder method call

  #155842 commented on Jul 2, 2025 • 0 new comments

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

  #155834 commented on Jul 3, 2025 • 0 new comments

• [DONT MERGE][TESTING][1/2] xpu test runner

  #155793 commented on Jun 30, 2025 • 0 new comments

• add sfdp pattern

  #155792 commented on Jul 2, 2025 • 0 new comments

• [Misc] handle sys exit caused by skip_if_lt_x_gpu in test_composabili…

  #155665 commented on Jul 2, 2025 • 0 new comments

• [C10d][Gloo] Enable complex datatype support in ProcessGroupGloo

  #156633 commented on Jun 30, 2025 • 0 new comments

• [BE] fix typo in torch/distributed/tensor/: childs -> children

  #156609 commented on Jul 6, 2025 • 0 new comments

• [BE] fix typo in torch/_numpy/_normalizations.py: parm -> param

  #156608 commented on Jul 6, 2025 • 0 new comments

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

  #156605 commented on Jul 6, 2025 • 0 new comments

• docstring_linter: Fix #151692 and other issues

  #156596 commented on Jul 4, 2025 • 0 new comments

• [Inductor Dashboard] Enable deterministic algorithms for some models

  #156592 commented on Jun 30, 2025 • 0 new comments

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

  #156590 commented on Jun 30, 2025 • 0 new comments

• [CPU] Fix memory access for sbgemm bf16

  #156585 commented on Jul 7, 2025 • 0 new comments

• [xla hash update] update the pinned xla hash

  #156584 commented on Jul 7, 2025 • 0 new comments

• Enable target-determination (TD) for ROCm CI

  #156545 commented on Jul 5, 2025 • 0 new comments

• [dynamo] Guard eagerly on list objects to avoid guard on getitem index

  #156531 commented on Jul 1, 2025 • 0 new comments

• [DO NOT MERGE] Update trunk.yml to change the runner that the job runs-on

  #156491 commented on Jul 4, 2025 • 0 new comments

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

  #156486 commented on Jul 2, 2025 • 0 new comments

• [DONT MERGE][TESTING][2/2] test new xpu runner

  #156410 commented on Jun 30, 2025 • 0 new comments

• [list] Add list.__delitem__

  #156339 commented on Jul 5, 2025 • 0 new comments

• [BE][2/16] fix typos in torch/ (torch/_*/)

  #156312 commented on Jul 6, 2025 • 0 new comments

• [BE][1/16] fix typos in torch/

  #156311 commented on Jul 6, 2025 • 0 new comments

• [list] Add list.__mul__ and list.__imul__

  #156271 commented on Jul 5, 2025 • 0 new comments

• Implement list.__add__ and list.__iadd__

  #156270 commented on Jul 5, 2025 • 0 new comments

• Add fallback-aware device checking for MPS operations

  #156267 commented on Jul 1, 2025 • 0 new comments

• [list] Implement `list.remove`

  #156242 commented on Jul 5, 2025 • 0 new comments

• [Native][CPU][TopK] Improve perf by reducing swap operations

  #156183 commented on Jul 1, 2025 • 0 new comments

• [NVIDIA] Refactor Family Blackwell Support codegen

  #156176 commented on Jul 2, 2025 • 0 new comments

• PyTorch CPP Extensions fail when same kernel is compiled more than once on ROCm servers

  #155344 commented on Jul 2, 2025 • 0 new comments

• SourcelessBuilder.create does not know how to wrap <class '__main__.InFlexData'>

  #154009 commented on Jul 2, 2025 • 0 new comments

• `torch.linalg.solve` does not raise an error for singular matrix on CPU.

  #154842 commented on Jul 2, 2025 • 0 new comments

• TORCH_COMPILE_DEBUG=1 does not consistently generate debug logs

  #152374 commented on Jul 2, 2025 • 0 new comments

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

  #64947 commented on Jul 2, 2025 • 0 new comments

• [RFC] Remove the FSDP data copy from compute stream critical path

  #157027 commented on Jul 2, 2025 • 0 new comments

• DISABLED test_op_has_batch_rule_tensordot_cuda_float32 (__main__.TestVmapOperatorsOpInfoCUDA)

  #142769 commented on Jul 2, 2025 • 0 new comments

• DISABLED test_graph_partition_cpu_op_and_dynamic_shapes (__main__.CudaGraphTreeTests)

  #157258 commented on Jul 2, 2025 • 0 new comments

• DISABLED test_op_has_batch_rule_nn_functional_conv_transpose3d_cuda_float32 (__main__.TestVmapOperatorsOpInfoCUDA)

  #82340 commented on Jul 2, 2025 • 0 new comments

• DISABLED test_slice_scatter_reinplace_cuda (__main__.GPUTests)

  #145189 commented on Jul 2, 2025 • 0 new comments

• DISABLED test_module_and_optimizer_ids (__main__.TestTorchTidyProfiler)

  #87581 commented on Jul 2, 2025 • 0 new comments

• torch compile does not support SyncBatchNorm with fullgraph=True

  #156680 commented on Jul 2, 2025 • 0 new comments

• [RFC][API-Unstable] Intel GPU distributed Backend integration in `torch-xpu-ops`and registeration in PyTorch

  #141741 commented on Jul 2, 2025 • 0 new comments

• `torch.compile` creates a CUDA context even for CPU based code

  #150622 commented on Jul 1, 2025 • 0 new comments

• Support for Bazel workspace function or Bazel module

  #112903 commented on Jul 1, 2025 • 0 new comments

• Export + autocast is eating the exception

  #153202 commented on Jul 1, 2025 • 0 new comments

• TorchInductor CPU Performance Dashboard

  #93531 commented on Jul 1, 2025 • 0 new comments

• FSDP offload doesn't prefetch param to GPU

  #157209 commented on Jul 1, 2025 • 0 new comments

• Upgrade AWS lambda functions from version 2.x to 3.x of the AWS SDK for JavaScript

  #137228 commented on Jul 1, 2025 • 0 new comments

• cd: Migrate binary builds off of Jinja

  #149660 commented on Jul 1, 2025 • 0 new comments

• Add Python NoGil support in CI

  #156854 commented on Jul 1, 2025 • 0 new comments

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

  #156810 commented on Jul 1, 2025 • 0 new comments

• Vendored wheels on PyTorch pip repository are outdated (e.g., `cmake`, `certifi`)

  #156694 commented on Jul 1, 2025 • 0 new comments

• [OSS tooling] pytorchbot fail to revert a PR

  #156607 commented on Jul 1, 2025 • 0 new comments

• Add explicit typing to nn.Module __init__()

  #156740 commented on Jul 1, 2025 • 0 new comments

• Adafactor foreach impl performance tracker

  #133367 commented on Jul 3, 2025 • 0 new comments

• [RFC] Experimental Wheel Variant Support

  #155141 commented on Jul 3, 2025 • 0 new comments

• Triton pin update for PyTorch 2.8 / Triton 3.4

  #154206 commented on Jul 3, 2025 • 0 new comments

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

  #153863 commented on Jul 3, 2025 • 0 new comments

• Add batched torch.combinations

  #40375 commented on Jul 3, 2025 • 0 new comments

• Python 3.14 support for PyTorch

  #156856 commented on Jul 3, 2025 • 0 new comments

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

  #89438 commented on Jul 3, 2025 • 0 new comments

• DISABLED test_host_memory_stats (__main__.TestCuda)

  #148607 commented on Jul 3, 2025 • 0 new comments

• DISABLED test_matrix_rank_basic_cuda_float32 (__main__.TestLinalgCUDA)

  #150406 commented on Jul 3, 2025 • 0 new comments

• DISABLED test_inductor_all_gather_into_tensor_single (__main__.CompileTest)

  #147707 commented on Jul 3, 2025 • 0 new comments

• DISABLED test_per_sample_api_compute_batch_size_not_pytreeable_cpu (__main__.TestExpandedWeightModuleCPU)

  #146972 commented on Jul 3, 2025 • 0 new comments

• MPS Memory Leak

  #154329 commented on Jul 3, 2025 • 0 new comments

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

  #148315 commented on Jul 3, 2025 • 0 new comments

• DISABLED test_fake_crossref_backward_no_amp_cholesky_solve_cuda_float32 (__main__.TestFakeTensorCUDA)

  #156419 commented on Jul 3, 2025 • 0 new comments

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

  #147913 commented on Jul 3, 2025 • 0 new comments

• Most requested ops for the MPS backend

  #154052 commented on Jul 3, 2025 • 0 new comments

• [cudagraph] simplify usage of how cudagraph dumps debug file

  #126753 commented on Jul 3, 2025 • 0 new comments

• [MPS] Performance regression and visual bug with ComfyUI Flux dev since nightly 20250510

  #155797 commented on Jul 2, 2025 • 0 new comments

• [compile][transformers] Recompilation with mark_static_address with cudagraphs

  #156377 commented on Jul 2, 2025 • 0 new comments

• Broadcasting matmul is much slower than corresponding einsum

  #110858 commented on Jul 2, 2025 • 0 new comments

• torch._dynamo.mark_static_address refuses to work with nn.Parameter

  #157221 commented on Jul 2, 2025 • 0 new comments

• [Upstream Triton] persistent mm + tma accuracy failures

  #156028 commented on Jul 2, 2025 • 0 new comments

• Registering function that takes `const SymInt&` to op that accepts `SymInt` leads to cryptic error

  #124645 commented on Jul 2, 2025 • 0 new comments

• PyTorch Memory Management in GPU-to-CPU Transfers issue

  #124487 commented on Jul 2, 2025 • 0 new comments

• Gross mismatch in PDF between CUDA and CPU for multivariate Gaussian mixture models

  #156959 commented on Jul 2, 2025 • 0 new comments

• Tensorboard `add_video()` broken for `moviepy>=2.0`

  #147317 commented on Jul 2, 2025 • 0 new comments

• Export Huggingface models with StaticCache

  #155862 commented on Jun 30, 2025 • 0 new comments

• [ONNX] torch.nn.functional.interpolate \w antialias=True isn't op.Resize compatible

  #157220 commented on Jun 30, 2025 • 0 new comments

• `torch.ldexp` goes out of range when `2**other` is out of range

  #153069 commented on Jun 30, 2025 • 0 new comments

• Is compilation caching for NumPy operators not supported in PyTorch 2.7.1?

  #156943 commented on Jun 30, 2025 • 0 new comments

• Compilation issues with ROCm 6.4.1 on Debian 12

  #155794 commented on Jun 30, 2025 • 0 new comments

• Update epsilon logic to improve numerical stability

  #151110 commented on Jun 30, 2025 • 0 new comments

• DISABLED test_forward_generation (__main__.CudaGraphTreeTests)

  #157058 commented on Jun 30, 2025 • 0 new comments

• Windows Source Build Fails with OSError: [WinError 126] on aoti_custom_ops.dll for RTX 5080 (sm_120), Pre-built PyTorch Works

  #157128 commented on Jun 30, 2025 • 0 new comments

• QAT support for conv2d with groups > 1

  #157222 commented on Jun 30, 2025 • 0 new comments

• NVFp4 Cublas Error

  #157054 commented on Jun 30, 2025 • 0 new comments

• Setting up for development

  #157141 commented on Jun 30, 2025 • 0 new comments

• [CUDA][CUTLASS] test_cutlass_backend.py unit test failures on SM90+

  #155888 commented on Jun 30, 2025 • 0 new comments

• [feature request] Native checkpointing to/from `s3://`

  #155992 commented on Jun 30, 2025 • 0 new comments

• CMake improperly configures pybind11. 3 different versions of pybind11 in use at the sametime.

  #156725 commented on Jun 30, 2025 • 0 new comments

• Preload CUDA fails if CUDA libs in different PYTHONPATH

  #147001 commented on Jun 30, 2025 • 0 new comments

• [RFC] Integrate NCCL scalable init API

  #136539 commented on Jun 30, 2025 • 0 new comments

• Pypi Support for Windows arm64

  #154260 commented on Jun 30, 2025 • 0 new comments

• [Feature] Taylor expansion pruning

  #157218 commented on Jun 30, 2025 • 0 new comments

• `TorchScript` does not allow accessing methods of nested tensors

  #156544 commented on Jun 30, 2025 • 0 new comments

• Set dependencies lower bound

  #156587 commented on Jun 30, 2025 • 0 new comments

• [CD] Windows Wheel builds CUDA 12.9.1 Stack Overflow during build

  #156181 commented on Jun 30, 2025 • 0 new comments

• DISABLED test_forward_backward_not_called_backend_inductor (__main__.CudaGraphTreeTests)

  #157035 commented on Jun 30, 2025 • 0 new comments

• DISABLED test_remove_noop_view_dtype_cuda (__main__.GPUTests)

  #151541 commented on Jun 30, 2025 • 0 new comments

• DISABLED test_hessian_vectorize_raises_no_warnings_logging_tensor (__main__.TestAutogradFunctional)

  #153644 commented on Jun 30, 2025 • 0 new comments

• Unable to compile

  #156915 commented on Jun 30, 2025 • 0 new comments

• Libtorch segfault when used with libqpOASES

  #33890 commented on Jun 30, 2025 • 0 new comments

• Improve debug message for metadata guard failure

  #157075 commented on Jul 1, 2025 • 0 new comments

• [dynamo] torch.randint_like on DTensor does not work with compile

  #156649 commented on Jul 1, 2025 • 0 new comments

• [dynamic shapes] translation validation failure under `fake_tensor_propagate_real_tensors`

  #156251 commented on Jul 1, 2025 • 0 new comments

• FakeTensorUpdater doesn't support HOPs

  #156819 commented on Jul 1, 2025 • 0 new comments

• Convolution NN for complex numbers and more special functions

  #116414 commented on Jul 1, 2025 • 0 new comments

• torch.Tensor.is_sparse returns false for non-COO sparse tensors

  #101385 commented on Jul 1, 2025 • 0 new comments

• test_tensor_with_grad_to_scalar_warning failure

  #157252 commented on Jul 1, 2025 • 0 new comments

• DISABLED test_graph_partition (__main__.CudaGraphTreeTests)

  #157173 commented on Jul 1, 2025 • 0 new comments

• DISABLED test_function_compiled_multiple_times (__main__.CudaGraphTreeTests)

  #157143 commented on Jul 1, 2025 • 0 new comments

• DISABLED test_matmul_small_brute_force_tunableop_cuda_float32 (__main__.TestLinalgCUDA)

  #141635 commented on Jul 1, 2025 • 0 new comments

• [Windows] pytorch >= 2.5

  #140875 commented on Jul 1, 2025 • 0 new comments

• [ued][gemma3] HF + torch.compile - torch.compile on Gemma3

  #149574 commented on Jul 1, 2025 • 0 new comments

• DISABLED test_op_has_batch_rule_nn_functional_conv_transpose1d_cuda_float32 (__main__.TestVmapOperatorsOpInfoCUDA)

  #142566 commented on Jul 1, 2025 • 0 new comments

• DISABLED test_frozen_fn (__main__.CudaGraphTreeTests)

  #157112 commented on Jul 1, 2025 • 0 new comments

• CTCLoss gradient is incorrect

  #52241 commented on Jul 1, 2025 • 0 new comments

• UR Error when calling grid_sample

  #153996 commented on Jul 1, 2025 • 0 new comments

• [inductor][cpu] pyhpc_isoneutral_mixing, lennard_jones and pyhpc_equation_of_state performance regression in 2025-06-23 nightly release

  #157077 commented on Jul 1, 2025 • 0 new comments

• Torch RPC examples from docs say usage is deprecated.

  #149393 commented on Jul 1, 2025 • 0 new comments

• Documentation Clarification Needed for Clamping of Scale Coefficient in clip_grads_with_norm_

  #151554 commented on Jul 1, 2025 • 0 new comments

• torch.compile with mode = "max-autotune" breaks when starting from inference_mode

  #135892 commented on Jul 1, 2025 • 0 new comments

• UNSTABLE pull / cuda12.8-py3.10-gcc9-sm75 / test (pr_time_benchmarks)

  #153987 commented on Jul 1, 2025 • 0 new comments

• DISABLED test_forward_with_skipped_cudagraphed_backward (__main__.CudaGraphTreeTests)

  #157086 commented on Jul 1, 2025 • 0 new comments

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

  #156882 commented on Jul 1, 2025 • 0 new comments

• Perf drop when running with FSDP and torch.compile

  #156966 commented on Jul 1, 2025 • 0 new comments

• DeviceMesh's `_set_mesh_dim_group_options` ineffective for 1-dim meshes

  #156593 commented on Jun 30, 2025 • 0 new comments

• [user triton] on-device TMA + AOTI causes IMA with pytorch 2.8 branch

  #157240 commented on Jun 30, 2025 • 0 new comments

• Fix `SequentialLR` deprecate warning about invoke `step(epoch)`

  #149392 commented on Jul 4, 2025 • 0 new comments

• NUMA Binding Integration with torchrun

  #149334 commented on Jul 2, 2025 • 0 new comments

• [cuDNN][SDPA] cuDNN SDPA refactor/cleanup, nested tensor backward, test priority bump for `sm90`, `sm100`

  #149282 commented on Jul 2, 2025 • 0 new comments

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

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

• [test] test for keep going

  #149003 commented on Jul 1, 2025 • 0 new comments

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

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

• C++ support to print symbolic tensors as `Symbolic tensor: size=(...)`

  #148846 commented on Jul 3, 2025 • 0 new comments

• Trunk workflow for Windows Arm64

  #148753 commented on Jul 1, 2025 • 0 new comments

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

  #148616 commented on Jul 1, 2025 • 0 new comments

• [BE][pytree] cleanup parameterized pytree tests

  #148569 commented on Jul 3, 2025 • 0 new comments

• [triton hash update] update the pinned triton hash

  #148492 commented on Jul 7, 2025 • 0 new comments

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

  #148484 commented on Jul 3, 2025 • 0 new comments

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

  #148474 commented on Jul 3, 2025 • 0 new comments

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

  #148328 commented on Jul 3, 2025 • 0 new comments

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

  #148180 commented on Jul 3, 2025 • 0 new comments

• Support `contextlib.suppress`

  #147990 commented on Jul 2, 2025 • 0 new comments

• Update triton_heuristics.py

  #147690 commented on Jul 2, 2025 • 0 new comments

• removed zero dim cpu logic from fake_tensor.py

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

• Deprecate DataLoader pin_memory_device param

  #146821 commented on Jul 4, 2025 • 0 new comments

• Support contextlib.ExitStack

  #146506 commented on Jul 2, 2025 • 0 new comments

• Update quantile doc

  #146485 commented on Jul 1, 2025 • 0 new comments

• [dcp] Minor improvements to filesystem writer

  #146273 commented on Jul 5, 2025 • 0 new comments

• docs: change log to ln in Softplus function and class

  #146199 commented on Jul 1, 2025 • 0 new comments

• Avoid data-dependent errors by runtime assert substitution.

  #145681 commented on Jul 1, 2025 • 0 new comments

• Fix full_like decomposition to preserve strides

  #144765 commented on Jul 2, 2025 • 0 new comments

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

  #144557 commented on Jul 5, 2025 • 0 new comments

• Deprecated pkg_resources and use distributions instead

  #151915 commented on Jun 30, 2025 • 0 new comments

• [reland][ROCm] remove caffe2 from hipify

  #151845 commented on Jul 1, 2025 • 0 new comments

• Horizontal

  #151780 commented on Jul 4, 2025 • 0 new comments

• enable windows inductor UT in CI

  #151777 commented on Jul 7, 2025 • 0 new comments

• Add adaptive_avg_pool2d input and output_size check

  #151769 commented on Jul 1, 2025 • 0 new comments

• Implement avg_pool3d for MPS backend

  #151742 commented on Jul 5, 2025 • 0 new comments

• Update OpenBLAS commit

  #151547 commented on Jul 2, 2025 • 0 new comments

• Implement fast exp for AVX2 and AVX512 for the flash attention

  #151441 commented on Jul 7, 2025 • 0 new comments

• Use Allocator API raw_allocate & raw_dealloc in CUDAAllocator

  #151305 commented on Jul 5, 2025 • 0 new comments

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

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

• Implement MKLGenerator

  #151218 commented on Jul 4, 2025 • 0 new comments

• Fix `MaskedTensor` to device ignored mask

  #151205 commented on Jul 4, 2025 • 0 new comments

• TESTING: IGNORE

  #151116 commented on Jun 30, 2025 • 0 new comments

• [export] add runtime assert messages to python torch checks

  #150719 commented on Jul 5, 2025 • 0 new comments

• Make LazyModuleMixin materialize after load_state_dict

  #150593 commented on Jul 1, 2025 • 0 new comments

• Refactor CUDAAllocatorConfig to reuse AcceleratorAllocatorConfig

  #150312 commented on Jul 1, 2025 • 0 new comments

• Add differentiable ops hint message in Module docs

  #150291 commented on Jul 5, 2025 • 0 new comments

• softmax: add device check for xpu with half_to_float

  #150278 commented on Jul 3, 2025 • 0 new comments

• Add cmake variable USE_ROCM_CK

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

• [WIP][dynamic shapes] rewrite should_swap with guard_or_false

  #150164 commented on Jul 1, 2025 • 0 new comments

• AOTI freezing: fix test issues and enable by default

  #149961 commented on Jul 2, 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

• Add SWA with a cyclical scheduler example

  #149847 commented on Jul 1, 2025 • 0 new comments

• Inductor logging + analysis of torch.profile

  #149697 commented on Jul 7, 2025 • 0 new comments

• Introduce AcceleratorAllocatorConfig as the common class

  #149601 commented on Jul 7, 2025 • 0 new comments

• [test] sccache docker build

  #149536 commented on Jul 6, 2025 • 0 new comments

• ROCm+gcc 15 asserts

  #145608 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

• 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

• RendezvousConnectionError when use C10d on multi nodes

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

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

  #157195 commented on Jul 5, 2025 • 0 new comments

• Run existing eager DTensor tests under torch.compile

  #127772 commented on Jul 5, 2025 • 0 new comments

• Trying to build from source with use_flash_attention fails on windows due to fatal error C1189

  #134854 commented on Jul 5, 2025 • 0 new comments

• Dead link in `torch.compile` docs

  #119272 commented on Jul 5, 2025 • 0 new comments

• [v.2.8.0] Release Tracker

  #156745 commented on Jul 4, 2025 • 0 new comments

• torch.compile does not work with Flash attention 3

  #144540 commented on Jul 4, 2025 • 0 new comments

• DISABLED test_simple_multi_arch_embed_kernel_binary_True_cuda (__main__.AOTInductorTestABICompatibleGpu)

  #156930 commented on Jul 4, 2025 • 0 new comments

• DISABLED test_assigning_back_deleter_fns_to_tensor (__main__.TestBlockStateAbsorption)

  #134810 commented on Jul 4, 2025 • 0 new comments

• DISABLED test_wait_tensor (__main__.CompileTest)

  #148014 commented on Jul 4, 2025 • 0 new comments

• DISABLED test_index (__main__.TestPythonBuiltinOP)

  #119160 commented on Jul 4, 2025 • 0 new comments

• Process never ends when sending tensors through multiprocessing queues in Python 3.12+ with filesystem strategy

  #153050 commented on Jul 4, 2025 • 0 new comments

• [dynamo] dynamo is unable to enter `except RuntimeError` while eager can

  #157217 commented on Jul 4, 2025 • 0 new comments

• DISABLED test_mempool_limited_memory_with_allocator (__main__.TestMemPool)

  #157256 commented on Jul 4, 2025 • 0 new comments

• Device Error on vmap

  #151591 commented on Jul 4, 2025 • 0 new comments

• Segmentation fault in torch.repeat_interleave

  #157097 commented on Jul 4, 2025 • 0 new comments

• Floating point exception in torch.nn.functional.conv_transpose3d

  #157098 commented on Jul 4, 2025 • 0 new comments

• [FSDP2] allow different dtypes for the model params with gradients

  #156784 commented on Jul 4, 2025 • 0 new comments

• flex_attention + dynamic=True with large batch or heads causes Triton Error [CUDA]: invalid argument

  #157018 commented on Jul 3, 2025 • 0 new comments

• Documentation: explaining the STFT formula

  #153531 commented on Jul 3, 2025 • 0 new comments

• [Tracker] AutoParallel's feature request to DTensor

  #156217 commented on Jul 3, 2025 • 0 new comments

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

  #157237 commented on Jul 3, 2025 • 0 new comments

• `RuntimeError: UR error` with XPU

  #149953 commented on Jul 3, 2025 • 0 new comments

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

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

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

  #144552 commented on Jul 5, 2025 • 0 new comments

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

  #144516 commented on Jul 2, 2025 • 0 new comments

• Add where_ ops

  #143636 commented on Jul 5, 2025 • 0 new comments

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

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

• Fix `USE_STATIC_MKL` lost functionality

  #138996 commented on Jul 1, 2025 • 0 new comments

• Always produce XML

  #138513 commented on Jul 5, 2025 • 0 new comments

• Add DeviceAllocator as the base device allocator

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

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

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

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

  #137400 commented on Jul 3, 2025 • 0 new comments

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

  #123020 commented on Jul 4, 2025 • 0 new comments

• Automated submodule update: FBGEMM

  #115316 commented on Jul 4, 2025 • 0 new comments

• [pytree] support PyStructSequence types for Python pytree

  #113258 commented on Jul 3, 2025 • 0 new comments

• Automated submodule update: kineto

  #106149 commented on Jul 2, 2025 • 0 new comments

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

  #157015 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

• DISABLED test_inductor_reduce_scatter_tensor_coalesced (__main__.CompileTest)

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

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

  #151740 commented on Jul 6, 2025 • 0 new comments

• [ONNX] Create a tutorial for exporting hf transformers model

  #156258 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

• MPS operator coverage tracking issue (2.6+ version)

  #141287 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

• Migrating existing backend-MAIA integration toward PrivateUse1 / openReg

  #155864 commented on Jul 6, 2025 • 0 new comments

• Flex Attention is incompatible with selective AC

  #147879 commented on Jul 6, 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

• General MPS op coverage tracking issue

  #77764 commented on Jul 6, 2025 • 0 new comments