30 Pull requests merged by 13 people

• Cleanup build docker images

  #129273 merged Jun 21, 2024

• moving conda builds from builder to pytorch

  #129167 merged Jun 21, 2024

• [ROCm] Include hsa headers for rocm-triton whl

  #129235 merged Jun 21, 2024

• [custom ops] Switch out references from old landing page to new landi…

  #129237 merged Jun 21, 2024

• [docs] Redirect custom ops landing page to the correct place (#129177)

  #129236 merged Jun 21, 2024

• Re-enable py3.12 nightly wheel builds and add triton dependency for ROCm

  #129161 merged Jun 21, 2024

• [Release only] Temporary change to depend on pytorch-triton

  #129232 merged Jun 21, 2024

• [inductor][ci] Fix torchbench dependency issue with numpy

  #129074 merged Jun 21, 2024

• [ROCm] [Triton] - Include roctracer headers in triton whl

  #129227 merged Jun 21, 2024

• [Release 2.4] Release only changes for triton 3.0.x build

  #129143 merged Jun 20, 2024

• Revert "[Release 2.4] Release only changes - use pinned triton."

  #129139 merged Jun 20, 2024

• Remove leftover warning causing log spew

  #128837 merged Jun 19, 2024

• [Inductor] Fix the High Order Op layout issue (#128275)

  #128834 merged Jun 19, 2024

• [Port][Quant][Inductor] Bug fix: mutation nodes not handled correctly for QLinearPointwiseBinaryPT2E

  #128591 merged Jun 19, 2024

• [tp] refactor and fix PrepareModuleInput for DTensor inputs (#128431)

  #128719 merged Jun 19, 2024

• [inductor] fix compile time regression by caching get_gpu_type (#128363)

  #128717 merged Jun 19, 2024

• [Inductor] Update Intel GPU Triton commit pin. (#124842)

  #128615 merged Jun 19, 2024

• Revert "Make torch_geometric models compatible with export (#123403)"…

  #128511 merged Jun 19, 2024

• [custom_op] stop using nonlocals to store information (#128547)

  #128616 merged Jun 19, 2024

• Clean up xpu ut to make CI happy (#128383)

  #128614 merged Jun 19, 2024

• Change Dynamo's custom ops warning message to be less spammy (#128456)

  #128581 merged Jun 19, 2024

• [inductor] fix linear add bias pattern (#128473)

  #128577 merged Jun 19, 2024

• [ALI] Use lf runners for Lint

  #128978 merged Jun 19, 2024

• Revert "[cuDNN][SDPA] Remove `TORCH_CUDNN_SDPA_ENABLED=1`, enable cuDNN SDPA by default on H100 and 2nd on other archs >= sm80 (#125343)"

  #128539 merged Jun 19, 2024

• Revert "Deprecate `torch._utils.is_compiling()` and `torch._dynamo.external_utils.is_compiling()` (#127690)"

  #128542 merged Jun 19, 2024

• Revert "Set simdlen based on ATEN_CPU_CAPABILITY (#123514)"

  #128541 merged Jun 19, 2024

• [dynamo] Fix for #127696

  #128530 merged Jun 18, 2024

• Bump urllib3 from 2.2.1 to 2.2.2 in /tools/build/bazel

  #128908 merged Jun 18, 2024

• Add build-conda-images.yml in pytorch/pytorch (#128563)

  #128962 merged Jun 18, 2024

• [Inductor][FlexAttention] Tune backwards kernel block sizes

  #128767 merged Jun 17, 2024

190 Pull requests opened by 100 people

• [Dynamic Shapes] fixed dynamic shape inference

  #128807 opened Jun 17, 2024

• [cpp_extension][inductor] Fix sleef windows depends. (#128770)

  #128811 opened Jun 17, 2024

• [inductor] refine loop split logic

  #128812 opened Jun 17, 2024

• [ROCm] Tunableop record untuned

  #128813 opened Jun 17, 2024

• Fix negative value in profier dump table

  #128814 opened Jun 17, 2024

• [DOT NOT REVIEW] Update Intel Triton

  #128820 opened Jun 17, 2024

• [Inductor][CPP] Enable Quantized Linear GEMM Template with FP32 output

  #128825 opened Jun 17, 2024

• Scale XBLOCK in triton reduction configs to avoid hitting max grid

  #128826 opened Jun 17, 2024

• [caffe2][be] migrate global static initializer SingletonUndefinedTensor

  #128828 opened Jun 17, 2024

• [caffe2][be] migrate global static initializer - event_template

  #128829 opened Jun 17, 2024

• [caffe2][be] migrate global static initializer - version_map

  #128831 opened Jun 17, 2024

• [caffe2][be] [caffe2][be] migrate global static initializer - unused global initializer

  #128832 opened Jun 17, 2024

• [caffe2][be] migrate global static initializer - EventTable

  #128833 opened Jun 17, 2024

• [export] turn on runtime asserts by default

  #128839 opened Jun 17, 2024

• [not for commit] Random perf opts

  #128841 opened Jun 17, 2024

• Upload release cut source code to s3

  #128842 opened Jun 17, 2024

• [torchbind] fix bug of mutating FakeScriptObjects twice in aot_export

  #128844 opened Jun 17, 2024

• [export] experimental joint graph API.

  #128847 opened Jun 17, 2024

• [dynamo] Remove torchrec skips

  #128857 opened Jun 17, 2024

• [BE] enable UFMT for `torch/ao/nn/`

  #128861 opened Jun 17, 2024

• [BE] enable UFMT for `torch/ao/pruning/`

  #128862 opened Jun 17, 2024

• [BE] enable UFMT for `torch/ao/quantization/`

  #128863 opened Jun 17, 2024

• [BE] enable UFMT for `torch/ao/`

  #128864 opened Jun 17, 2024

• [BE][Easy] enable UFMT for `torch/nn/`

  #128865 opened Jun 17, 2024

• [WIP][inductor]fallback all view operations

  #128883 opened Jun 17, 2024

• [Traceable FSDP2] Fixes to preserve inplace ops in AOT joint graph and fwd graph

  #128886 opened Jun 17, 2024

• [dtensor][debug] fixing CommDebugMode module collective tracing

  #128887 opened Jun 17, 2024

• Updates to test packed layout

  #128888 opened Jun 17, 2024

• [aota] Needs autograd if an input requires_grad, agnostic to enable_grad

  #128890 opened Jun 17, 2024

• Forward fix to skip ROCm tests for #122836

  #128891 opened Jun 17, 2024

• [inductor] Separate Buffer and Operation into two concepts

  #128893 opened Jun 17, 2024

• [RFC][Not Aim for landing now] Add a dummy pp hang case for flight recorder

  #128897 opened Jun 17, 2024

• Grouped Query Attention

  #128898 opened Jun 17, 2024

• [Fix]: Internal test failures when testing the exportability

  #128900 opened Jun 17, 2024

• Fix mm pad regresion - more conservative estimation of plannable inputs

  #128909 opened Jun 17, 2024

• Delete unused line

  #128913 opened Jun 18, 2024

• Add Strided Input test for flex attention

  #128915 opened Jun 18, 2024

• __eq__ and __hash__ for SymNode

  #128916 opened Jun 18, 2024

• [reland][ROCm] TunableOp for gemm_and_bias

  #128919 opened Jun 18, 2024

• Always use high precision for SDPA math backend

  #128922 opened Jun 18, 2024

• [inductor] Constant folding for dynamic shape node before pattern matching

  #128937 opened Jun 18, 2024

• Update start_, end_ and retired only for the right entry when retire a work

  #128948 opened Jun 18, 2024

• test_jit: Replace plain assert by test assert

  #128950 opened Jun 18, 2024

• Set correct output dtype for dequantize op during convert_pt2e in decomposed mode

  #128953 opened Jun 18, 2024

• [BE] Do not crash weight_norm on empty tensors

  #128957 opened Jun 18, 2024

• Add weight_norm opinfo testing

  #128958 opened Jun 18, 2024

• [pipelining] Support arbitrary stage ordering on ranks

  #128976 opened Jun 18, 2024

• fix dynamo isinstance inlining for nn.Parameter + subclasses

  #128981 opened Jun 18, 2024

• [Pipelining] Support separate dw_runner for PipelineStage

  #128983 opened Jun 18, 2024

• temp run failing split build pull test

  #128984 opened Jun 18, 2024

• [Not to be committed][AOTI] Add an option to return cpp file only

  #128986 opened Jun 18, 2024

• pytorch slp

  #128990 opened Jun 18, 2024

• [dynamo][easy] Rename NotNNModuleSource to UnspecializedNNModuleSource

  #128992 opened Jun 18, 2024

• [After Rebase] Top of Traceable FSDP2 stack

  #128996 opened Jun 18, 2024

• [WIP] Test running canary jobs

  #129000 opened Jun 18, 2024

• [BE] update type annotations for basic utilities in `torch/__init__.py`

  #129001 opened Jun 18, 2024

• [experiment] run_test: Unset cpp stacktraces after reruns

  #129004 opened Jun 18, 2024

• [MPS] Fast math env var

  #129007 opened Jun 18, 2024

• [RFC] scaffolding of the new B2B_GEMM pass

  #129009 opened Jun 18, 2024

• Fix DEBUG=1 asserts with NJT ops

  #129014 opened Jun 18, 2024

• torch._inductor.config.joint_graph_constant_folding = False

  #129016 opened Jun 19, 2024

• [dtensor][debug] add operation tracing to comm_mode

  #129017 opened Jun 19, 2024

• Update script path in .github/workflows/build-conda-images.yml

  #129022 opened Jun 19, 2024

• Enable UFMT for numpy_test files, test_xnnpack_integration.py

  #129023 opened Jun 19, 2024

• [halide-backend] Dimension-based indexing

  #129026 opened Jun 19, 2024

• Back out "Remove circular import"

  #129031 opened Jun 19, 2024

• [halide-backend] Support scan kernels

  #129035 opened Jun 19, 2024

• [halide-backend] Enable bfloat16 support

  #129036 opened Jun 19, 2024

• [CI] Enable AOT inductor FP32 accuracy test for CPU

  #129040 opened Jun 19, 2024

• [wip][inductor] don't materialize the large sparse matrix in CE bwd

  #129043 opened Jun 19, 2024

• fix add decomposition for complex numbers

  #129044 opened Jun 19, 2024

• [Do Not Merge]include torch

  #129047 opened Jun 19, 2024

• [Inductor][CPP] Enable Quantized Linear GEMM Template with INT8 output and Unary Post Op

  #129048 opened Jun 19, 2024

• [Inductor][Quant] Change the schema of QLinear Binary

  #129049 opened Jun 19, 2024

• Provide a method to unregister privateuse1

  #129056 opened Jun 19, 2024

• Restore mixed dtypes GEMM auto-tuning for Ampere

  #129058 opened Jun 19, 2024

• Don't install remaining caffe2 python files

  #129067 opened Jun 19, 2024

• use shutil.which in check_compiler_ok_for_platform

  #129069 opened Jun 19, 2024

• [codemod][lowrisk] Remove extra semi colon from caffe2/aten/src/ATen/functorch/BatchRulesRandomness.cpp

  #129073 opened Jun 19, 2024

• [inductor] Remove comm-specific node attributes from scheduler

  #129084 opened Jun 19, 2024

• [Fix]: TSConverter errors on dynamic shapes

  #129087 opened Jun 19, 2024

• Relax constraints for creating a `GenericContextWrappingVariable`

  #129091 opened Jun 19, 2024

• Prototype for export_for_training

  #129092 opened Jun 19, 2024

• [ROCm] Use AOTriton as a dynamic library

  #129094 opened Jun 19, 2024

• re-export torch.optim._multi_tensor in torch/__init__.py

  #129095 opened Jun 19, 2024

• Fix scatter lowering when src is a Number

  #129096 opened Jun 19, 2024

• Fix rot90 decomposition for no rotation

  #129097 opened Jun 19, 2024

• [executorch hash update] update the pinned executorch hash

  #129099 opened Jun 20, 2024

• [Inductor][CPP] Enable Quantized Linear GEMM Template with Binary Fusion

  #129103 opened Jun 20, 2024

• [MPS] Generalize Fused optimizers

  #129105 opened Jun 20, 2024

• Fix max_pool2d decomposition for empty list and integer limits

  #129106 opened Jun 20, 2024

• [FSDP] Runtime Error on Checkpoint Loading for optimizer state

  #129110 opened Jun 20, 2024

• [PT2][Observability] Change the string to dict type

  #129112 opened Jun 20, 2024

• Refine the logic of device construction when only device index is given

  #129119 opened Jun 20, 2024

• [Inductor][CPP] Support more than one LocalBuffer

  #129121 opened Jun 20, 2024

• Fix typo in stack_module_state doc

  #129126 opened Jun 20, 2024

• Fix integer overflow in quantization

  #129127 opened Jun 20, 2024

• [aot] Keep backward mutations in backward

  #129130 opened Jun 20, 2024

• [BE] use relative backwards references in torch.optim._multi_tensor

  #129132 opened Jun 20, 2024

• [AOTI] Remove the epilogue for generating non-triggered kernels

  #129134 opened Jun 20, 2024

• [AOTI] Introduce DeferredCudaKernelLine for cuda cpp wrapper

  #129135 opened Jun 20, 2024

• Refine typing annotation for compile

  #129136 opened Jun 20, 2024

• Skip ao_sparsity TestComposability for missing FBGEMM

  #129137 opened Jun 20, 2024

• Move caffe2/serialize to torch/csrc/api

  #129141 opened Jun 20, 2024

• [inductor] switch CppCodeCache to new cpp_builder. (take 2)

  #129144 opened Jun 20, 2024

• Fixes T192448049

  #129146 opened Jun 20, 2024

• [C10D] Avoid lazily creating P2P communicators

  #129147 opened Jun 20, 2024

• Update README.md

  #129149 opened Jun 20, 2024

• Update test_torch.py

  #129151 opened Jun 20, 2024

• [nn-module] Use standard dict for _parameters, _modules and _buffers

  #129164 opened Jun 20, 2024

• [experiment] build

  #129170 opened Jun 20, 2024

• fix cpp compilation error

  #129173 opened Jun 20, 2024

• Back out "Remove circular import"

  #129180 opened Jun 20, 2024

• typing proxy_tensor.py

  #129182 opened Jun 20, 2024

• [AOTI] Fix test_cond_non_tensor_predicates

  #129183 opened Jun 20, 2024

• [CUDAGraph Trees] Enable input mutation support in OSS

  #129184 opened Jun 20, 2024

• [3.13, WIP] directly use frame localsplus in guards

  #129185 opened Jun 20, 2024

• Proof-of-concept: manage registered communication buffers with Inductor

  #129186 opened Jun 20, 2024

• Add lowering for updated _scaled_mm

  #129187 opened Jun 21, 2024

• [bazel] fix --config=shell

  #129194 opened Jun 21, 2024

• [RFC] Add JSON logging

  #129196 opened Jun 21, 2024

• Log whenever we sleep

  #129197 opened Jun 21, 2024

• Pianpwk/dedup2

  #129199 opened Jun 21, 2024

• [wip] merge_csv tool

  #129202 opened Jun 21, 2024

• Add xpu to getAccelerator

  #129205 opened Jun 21, 2024

• [Inductor] Draft version of block sparse mask for flex attention

  #129216 opened Jun 21, 2024

• Remove more ONNX Caffe2 code

  #129218 opened Jun 21, 2024

• Fix license metadata in setup.py

  #129219 opened Jun 21, 2024

• [Inductor][CPP] Enable Quantized Linear with AMX MicroGEMM

  #129220 opened Jun 21, 2024

• [Inductor][CPP] Pass weight dtype explicitly for cpp gemm template

  #129221 opened Jun 21, 2024

• [inductor][cpp] support nested kernel with indirect indexing

  #129223 opened Jun 21, 2024

• [easy][DCP] make BroadcastingTorchSaveReader device generic

  #129231 opened Jun 21, 2024

• [pipelining] Support W action for schedules

  #129233 opened Jun 21, 2024

• [sparse][bfloat16] bmm_sparse_cuda

  #129234 opened Jun 21, 2024

• Add warning for weights_only

  #129239 opened Jun 21, 2024

• [codemod][lowrisk] Remove unused exception parameter from pytorch/genai-stable-import-prototype/torch/csrc/distributed/c10d/TCPStoreLibUvBackend.cpp

  #129240 opened Jun 21, 2024

• [FSDP2] Fixed `unshard` without lazy init

  #129241 opened Jun 21, 2024

• Enable dynamic rollout for pull workflow

  #129243 opened Jun 21, 2024

• Fix allowlisting of builtins for weights_only unpickler

  #129244 opened Jun 21, 2024

• [BE] Runner determinator: Expect usernames to be prefixed with '@'

  #129246 opened Jun 21, 2024

• Make run_decomp work

  #129249 opened Jun 21, 2024

• Have torch_key hash entire torch directory

  #129250 opened Jun 21, 2024

• Allow BUILD/NEWOBJ instruction for items added via torch.serialization.add_safe_globals

  #129251 opened Jun 21, 2024

• [DSD] Correctly handle shared parameters for optimizer state_dict (#1…

  #129252 opened Jun 21, 2024

• Support HSDP + Monolith Checkpointing (#128446)

  #129254 opened Jun 21, 2024

• [DSD] Add unittest to verify HSDP1 + broadcast_from_rank0 (#128755)

  #129255 opened Jun 21, 2024

• [inductor] Fix TORCHINDUCTOR_FORCE_DISABLE_CACHES

  #129257 opened Jun 21, 2024

• [FSDP2] Used multi-grad hook when no inputs require grad

  #129259 opened Jun 21, 2024

• [export] Rewrite exportdb formatting.

  #129260 opened Jun 21, 2024

• TCPStore: retry on validate errors

  #129261 opened Jun 21, 2024

• Allow SAC policy_fn to return bool for backward compatibility

  #129262 opened Jun 21, 2024

• [Pipelining] Add to/from CSV format and improved __repr__

  #129264 opened Jun 21, 2024

• Pianpwk/dedup3

  #129265 opened Jun 21, 2024

• Documentations for XPU functionality to PyTorch

  #129266 opened Jun 21, 2024

• [AOTI][refactor] Move generate_user_defined_triton_kernel

  #129267 opened Jun 21, 2024

• [AOTI] Introduce DeferredCudaGridLine

  #129268 opened Jun 21, 2024

• TunableOp hotfix

  #129281 opened Jun 21, 2024

• [C10D] Make new_group eager when used with comm_split

  #129284 opened Jun 21, 2024

• [dynamo][compile-time][inlining-inbuilt-nn-modules] Manually implement nn.Module._call_impl

  #129285 opened Jun 21, 2024

• [FSDP2] Added `set_reduce_scatter_divide_factor`

  #129286 opened Jun 21, 2024

• Preserve _numeric_debug_handle throguh deepcopy

  #129287 opened Jun 22, 2024

• Inductor to fail gracefully on Voltas for bf16 tensors

  #129288 opened Jun 22, 2024

• Implement operator for micro-pipelined all-gather -> _scaled_mm

  #129289 opened Jun 22, 2024

• 3d Composability

  #129290 opened Jun 22, 2024

• Skip signals from older runs of the same workflows

  #129291 opened Jun 22, 2024

• [Fix]: TSConverter handles call ops with multiple outputs

  #129294 opened Jun 22, 2024

• [Reopen #114036] Allow "must recompute" in torch.compile + selective checkpointing (SAC)

  #129295 opened Jun 22, 2024

• Support tensor stride

  #129297 opened Jun 22, 2024

• Add one more shard for CPU jobs

  #129299 opened Jun 22, 2024

• [2/N] Fix clang-tidy warnings in torch/csrc/jit/serialization

  #129300 opened Jun 22, 2024

• [aotinductor][UserDefinedTritonKernel] use appropriate expr printer when printing args

  #129301 opened Jun 22, 2024

• Added host-side associative scan function

  #129307 opened Jun 22, 2024

• [halide-backend] Random number generation

  #129314 opened Jun 22, 2024

• [dynamo][compile-time] Manually implement nn.Module.__getattr__ to reduce compile time

  #129315 opened Jun 22, 2024

• [docs] fix incorrect example in `convert_conv3d_weight_memory_format`

  #129318 opened Jun 22, 2024

• [halide-backend] Disable split reductions for Halide

  #129320 opened Jun 23, 2024

• [halide-backend] Support manual schedules

  #129321 opened Jun 23, 2024

• fake_tensor - flatten keys

  #129323 opened Jun 23, 2024

• WIP: fake tensor SymInt support, try 2

  #129324 opened Jun 23, 2024

• [inductor] Make UserDefinedTritonKernel a multi-output operation

  #129325 opened Jun 23, 2024

• Fix build error on s390x

  #129326 opened Jun 23, 2024

• [aotinductor] Only autotune at compile time when enabled via config

  #129335 opened Jun 23, 2024

• [Inductor][Intel GPU] Support reduction split. (#129120)

  #129337 opened Jun 24, 2024

• [Easy][Traceable FSDP2] Skip rocm for the E2E tests

  #129339 opened Jun 24, 2024

• Remove test_mps_allocator_module XFAIL

  #129340 opened Jun 24, 2024

• [AOTI] Switch the CUDA codegen to one-pass

  #129342 opened Jun 24, 2024

• [inductor] Add FileCheck to flex attention epilogue test

  #129343 opened Jun 24, 2024

• [inductor] Use multiple outputs for flex-attention

  #129344 opened Jun 24, 2024

• [AOTI][not for review] Test cpp_wrapper mode

  #129345 opened Jun 24, 2024

• [inductor] Kill mark_node_as_mutating

  #129346 opened Jun 24, 2024

136 Issues closed by 43 people

• dynamo eval the subfunction of a skiped frame with callback, bad performance and more error

  #128928 closed Jun 24, 2024

• DISABLED test_post_localSGD_optimizer_step_reload (__main__.TestDistBackendWithSpawn)

  #84886 closed Jun 24, 2024

• UNSTABLE inductor-A100-perf-nightly / cuda12.1-py3.10-gcc9-sm80 / test (inductor_torchbench_perf)

  #128846 closed Jun 23, 2024

• UNSTABLE pull / linux-focal-cuda12.4-py3.10-gcc9-sm86 / build

  #127104 closed Jun 23, 2024

• UNSTABLE rocm / linux-focal-rocm6.1-py3.8 / test (default)

  #129208 closed Jun 23, 2024

• UNSTABLE periodic / linux-focal-rocm6.1-py3.8 / test (distributed)

  #129209 closed Jun 23, 2024

• UNSTABLE trunk / linux-focal-rocm6.1-py3.8 / test (default)

  #129210 closed Jun 23, 2024

• UNSTABLE trunk / linux-focal-rocm6.1-py3.8 / test (distributed)

  #129211 closed Jun 23, 2024

• UNSTABLE inductor-periodic / cuda12.1-py3.10-gcc9-sm86-periodic-dynamo-benchmarks / test (dynamo_eager_torchbench)

  #128932 closed Jun 23, 2024

• UNSTABLE inductor-periodic / cuda12.1-py3.10-gcc9-sm86-periodic-dynamo-benchmarks / test (dynamic_aot_eager_torchbench)

  #128931 closed Jun 23, 2024

• UNSTABLE inductor-periodic / cuda12.1-py3.10-gcc9-sm86-periodic-dynamo-benchmarks / test (aot_eager_torchbench)

  #128929 closed Jun 23, 2024

• UNSTABLE inductor / cuda12.1-py3.10-gcc9-sm86 / test (inductor_torchbench)

  #128901 closed Jun 23, 2024

• UNSTABLE inductor / cuda12.1-py3.10-gcc9-sm86 / test (dynamic_inductor_torchbench)

  #128902 closed Jun 23, 2024

• UNSTABLE inductor / cuda12.1-py3.10-gcc9-sm86 / test (aot_inductor_torchbench)

  #128903 closed Jun 23, 2024

• DISABLED test_comprehensive_nn_functional_huber_loss_cuda_float16 (__main__.TestInductorOpInfoCUDA)

  #129238 closed Jun 22, 2024

• [ONNX] Migrate OpSignature to ONNX Script

  #129278 closed Jun 22, 2024

• Illegal instruction (core dumped): PyTorch 2.3.0+rocm6.0

  #125310 closed Jun 21, 2024

• nn.Linear outputs differ on the same input tensor #129029 answer does not match

  #129111 closed Jun 21, 2024

• Saved variable packing unpacking incorrect aliases version counter

  #128611 closed Jun 21, 2024

• Forward hooks not called when fast path is used in TransformerEncoderLayer

  #128413 closed Jun 21, 2024

• [ONNX] Provide an option to not generate `report_dynamo_export.sarif`

  #109137 closed Jun 21, 2024

• Torch compile does not work on python 3.12

  #120233 closed Jun 21, 2024

• Gumbel Vector Quantizer produces NaN when using with torch.compile

  #127749 closed Jun 21, 2024

• Huge solibs in Linux wheel for torch 2.3.1+rocm6.0

  #129165 closed Jun 21, 2024

• Fix docstring errors in nadam.py, radam.py, sgd.py, anomaly_mode.py, rprop.py, __init__.py, swa_utils.py, rmsprop.py, optimizer.py, lr_scheduler.py

  #112593 closed Jun 21, 2024

• Slow DataLoader in new version when num_workers>0 / objects in collate_fn slow down batching

  #123439 closed Jun 21, 2024

• The function name `SGD()` should be `GD()` if it's still (classic) GD,

  #129190 closed Jun 21, 2024

• ¿Cómo llamar a United desde Estados Unidos?{1*844*499*2050} CALL NOW !!

  #129212 closed Jun 21, 2024

• ¿Cómo puedo hablar con una persona en Delta? [꜍D꜉🅔꜍L꜉🅣꜍A꜉ AiRlInEs]

  #129214 closed Jun 21, 2024

• [TorchAO] fail to do fake_tensor_prop with freezing pass

  #123522 closed Jun 21, 2024

• [dynamo] 'torch._C.ScriptFunction' object has no attribute '__defaults__'

  #93698 closed Jun 21, 2024

• Is it possble to add a registerable api at the beginning of torch.save

  #117840 closed Jun 21, 2024

• frombuffer() → "The given buffer is not writable" warning, tensor has some NaNs

  #129077 closed Jun 21, 2024

• Incompatability between torch>=2.3 and torchdatasets==0.2.0

  #129060 closed Jun 21, 2024

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

  #129195 closed Jun 21, 2024

• Ignore this

  #129128 closed Jun 21, 2024

• UNSTABLE periodic / win-vs2019-cuda11.8-py3 / test (default)

  #129064 closed Jun 21, 2024

• [Feature Request] switch amx isa detection in onednn to cpuinfo

  #127368 closed Jun 21, 2024

• Failure with setup-ssh on Amazon Linux 2023 runners

  #129152 closed Jun 21, 2024

• Outdated ncclResult code

  #128756 closed Jun 20, 2024

• Using PyTorch with Transformers to run inference with 'MPS' backend causes poor results.

  #128435 closed Jun 20, 2024

• auto_functionalized doesn't work with non-Tensor returns

  #120490 closed Jun 20, 2024

• torch_dispatch mode silent incorrectness with torch.compile

  #115653 closed Jun 20, 2024

• torch.compile hang/crashes with worker_start_method=spawn

  #126311 closed Jun 20, 2024

• DISABLED test_quantization_doc_ptsq (__main__.TestQuantizationDocs)

  #125669 closed Jun 20, 2024

• DISABLED test_creation_with_zeros_cuda_float8_e5m2 (__main__.TestFloat8DtypeCUDA)

  #124474 closed Jun 20, 2024

• DISABLED test_graph_concurrent_replay (__main__.TestCuda)

  #104055 closed Jun 20, 2024

• DISABLED test_tensor_subclasses (__main__.TestScript)

  #119949 closed Jun 20, 2024

• DISABLED test_quantization_doc_custom (__main__.TestQuantizationDocs)

  #125668 closed Jun 20, 2024

• DISABLED test_is_isnot (__main__.TestScript)

  #120694 closed Jun 20, 2024

• DISABLED test_index (__main__.TestPythonBuiltinOP)

  #119160 closed Jun 20, 2024

• DISABLED test_quantization_doc_ptdq (__main__.TestQuantizationDocs)

  #125667 closed Jun 20, 2024

• DISABLED test_add_loggers_conv_bn_relu_fusion_quant (__main__.TestFXNumericSuiteNShadows)

  #127814 closed Jun 20, 2024

• DISABLED test_quantization_doc_fx (__main__.TestQuantizationDocs)

  #125670 closed Jun 20, 2024

• DISABLED test_quantization_doc_qat (__main__.TestQuantizationDocs)

  #128118 closed Jun 20, 2024

• DISABLED test_comprehensive_special_bessel_y1_cuda_int32 (__main__.TestInductorOpInfoCUDA)

  #127080 closed Jun 20, 2024

• DISABLED test_cusparse_multiple_threads_same_device (__main__.TestCuda)

  #127536 closed Jun 20, 2024

• [nnc][perf] CPU fuser needs to support intra-op parallelism

  #50853 closed Jun 20, 2024

• [Dynamo x torch_function] methods on torch_function objects require id_match guards, causing recompiles

  #128964 closed Jun 20, 2024

• Torchrun / torch.distributed.run throws RendezvousConnectionError / DistNetworkError (Connection reset by peer)

  #128970 closed Jun 19, 2024

• ONNX export for gelu at version 20

  #128772 closed Jun 19, 2024

• UNSTABLE periodic / win-vs2019-cuda11.8-py3 / test (default)

  #129065 closed Jun 19, 2024

• nn.Linear outputs differ on the same input tensor

  #129029 closed Jun 19, 2024

• torch.Tensor.tolist() cancels torch.round()

  #128943 closed Jun 19, 2024

• torch.compile crash - Aborted exit code 134

  #125804 closed Jun 19, 2024

• [xpu] ERROR: Failed building wheel for triton when USE_XPU=1 make triton

  #129042 closed Jun 19, 2024

• torch.gather can be slow on AMD with duplicated index

  #128631 closed Jun 19, 2024

• [dynamo] Slow compile times for optimizers due to for loops

  #110506 closed Jun 18, 2024

• [Inductor] [Distributed] DDP torch.compile model hangs on exit (python 3.8/3.9)

  #125235 closed Jun 18, 2024

• torch.compile() bug in AOTAutograd or Dynamo

  #103727 closed Jun 18, 2024

• Request: flag to know model is compiled after torch.compile()

  #103553 closed Jun 18, 2024

• Dynamo trouble shooting dead link

  #103276 closed Jun 18, 2024

• nondeterminism in torch.compile + custom op

  #127995 closed Jun 18, 2024

• torch.compile + constructing an nn.Parameter + mutating it can give wrong results

  #125284 closed Jun 18, 2024

• SyntaxError: unterminated string literal (detected at line 1) (<unknown>, line 1)

  #127637 closed Jun 18, 2024

• torch._dynamo.export segfaults when calling nn.Parameter

  #126109 closed Jun 18, 2024

• fullgraph=True doesn't actually raise error when you don't manage full graph inside DDP

  #107639 closed Jun 18, 2024

• Tied Weight Embeddings Models Fail to Load on Torch 2.4 Nightly

  #128011 closed Jun 18, 2024

• distributed.gather shape constraints

  #103305 closed Jun 18, 2024

• DISABLED test_optimizer_parameters_sgd (__main__.TestTorchTidyProfiler)

  #123624 closed Jun 18, 2024

• DISABLED test_isinstance (__main__.TestScript)

  #123832 closed Jun 18, 2024

• DISABLED test_tensor_number_math (__main__.TestScript)

  #123701 closed Jun 18, 2024

• DISABLED test_math_ops (__main__.TestScript)

  #123693 closed Jun 18, 2024

• DISABLED test_index (__main__.TestScript)

  #123635 closed Jun 18, 2024

• DISABLED test_number_math (__main__.TestScript)

  #123660 closed Jun 18, 2024

• [export] Export warnings as no-ops

  #113792 closed Jun 18, 2024

• Tracing per-param sharding FSDP: Dynamo tracing weakrefs

  #114288 closed Jun 18, 2024

• `pytest test/dynamo/test_ctx_manager.py -v -k "test_autocast_graph_break_method"` fails locally

  #117000 closed Jun 18, 2024

• Dynamo'ing Rprop, RMSprop, and Adadelta misses incrementing step due to skipping _init_group

  #115679 closed Jun 18, 2024

• Performance impact of pre-division.

  #128918 closed Jun 18, 2024

• [dynamo][eval frame] frame->f_locals is empty after call_callback

  #118068 closed Jun 18, 2024

• [Dynamo] Better support DTensor

  #117670 closed Jun 18, 2024

• Power and multiple multiplication don't give the same gradient

  #128836 closed Jun 18, 2024

• Parameters out of sync over different ranks due to unused parameters

  #128949 closed Jun 18, 2024

• compiling profiler with ExecutionTraceObserver breaks

  #124500 closed Jun 18, 2024

• torchinductor error in torchao tests

  #128263 closed Jun 18, 2024

• Inlining nn modules and FSDP

  #128154 closed Jun 18, 2024

• address TODO: model is somehow not being freed when z3 is available

  #127444 closed Jun 18, 2024

• Tracing through __getitem__ -> __len__ for ModuleList fails.

  #126445 closed Jun 18, 2024

• UNSTABLE pull / win-vs2019-cpu-py3 / build

  #103729 closed Jun 18, 2024

• UNSTABLE trunk / win-vs2019-cpu-py3 / build

  #103732 closed Jun 18, 2024

• UNSTABLE trunk / win-vs2019-cuda11.8-py3 / build

  #103733 closed Jun 18, 2024

• UNSTABLE periodic / win-vs2019-cuda11.8-py3 / build

  #128855 closed Jun 18, 2024

• taking upper triangular of "-inf" matrix results in nan values

  #128429 closed Jun 18, 2024

• heartbeatMonitor error after run script multiple times

  #128680 closed Jun 18, 2024

• [inductor][cpu] AMP models static/dynamic default/CPP wrapper accuracy/performance crash in 2024-06-08 nightly release

  #128507 closed Jun 18, 2024

• PT2 custom ops does not work with future annotations

  #105157 closed Jun 18, 2024

• ImportError: cannot import name 'OrderedDict' from partially initialized module 'collections'

  #128838 closed Jun 18, 2024

• OptimizedModule should call _orig_mod's load_state_dict()/state_dict() methods.

  #123625 closed Jun 17, 2024

• [dynamo] Are we over guarding on `__defaults__`?

  #123490 closed Jun 17, 2024

• Document the torch.cuda.profiler.profile function

  #127901 closed Jun 17, 2024

• Can't compile torchaudio.transforms.Spectrogram

  #121718 closed Jun 17, 2024

• [Dynamo] nn.Module forward hook ends up with a separate graph

  #121695 closed Jun 17, 2024

• Using torch.compile, batch training step time takes long to converge when adding a LR Scheduler

  #120934 closed Jun 17, 2024

• TypeError: unhashable type 'dict'

  #120932 closed Jun 17, 2024

• InternalTorchDynamoError on KL Divergences

  #120497 closed Jun 17, 2024

• inductor_torchbench_perf jobs are broken due to numpy 2.0 update

  #128845 closed Jun 17, 2024

• `InternalTorchDynamoError: source code not available` when using multiple modules in ipynb

  #119225 closed Jun 17, 2024

• Dynamo incorrectly classifies bound methods when used in a closure

  #118988 closed Jun 17, 2024

• torch.compile() AssertionError: target must be of GPUTarget type

  #128357 closed Jun 17, 2024

• Dynamo fails due to `TorchRuntimeError: slice step cannot be zero` with `dynamic=True`

  #128827 closed Jun 17, 2024

• onnx.dynamo_export() fails on torch.numel()

  #128882 closed Jun 17, 2024

• CuDNN Attention Kernel _scaled_dot_product_cudnn_attention unable to run.

  #122695 closed Jun 17, 2024

• Error linking(name collision) to libtorch on Windows with OneAPI and Visual Studio

  #128823 closed Jun 17, 2024

• HSDP + `set_optimizer_state_dict` errors with monolithic checkpointing

  #128444 closed Jun 17, 2024

• UNSTABLE inductor-A100-perf-nightly / cuda12.1-py3.10-gcc9-sm80 / test (inductor_torchbench_perf)

  #128848 closed Jun 17, 2024

• UNSTABLE inductor / cuda12.1-py3.10-gcc9-sm86 / test

  #120841 closed Jun 17, 2024

• Document the torch.cuda.cudart function

  #127908 closed Jun 17, 2024

• Document the torch.nn.parallel.scatter_gather.gather function

  #127899 closed Jun 17, 2024

• Does torch have any features or future plans to improve performance on ARM?

  #128817 closed Jun 17, 2024

• The running will become slower and slower with epoch continuing, but without error when the mlpmixer model is used

  #128604 closed Jun 17, 2024

• Cannot access data pointer of Tensor that doesn't have storage when using torch.func.vmap with c++/cuda extension

  #122706 closed Jun 17, 2024

• Link https://pytorch.org/docs/stable/nn.html#torch.nn.EmbeddingBag may not exist anymore

  #128774 closed Jun 17, 2024

• Tensor.new_empty type annotation does not accept SymInt

  #115456 closed Jun 17, 2024

• [dynamo] Trace through invalid bool tensor operations properly

  #127003 closed Jun 17, 2024

• DISABLED test_pointwise_bessel_y1_cuda (__main__.GPUTests)

  #127756 closed Jun 17, 2024

113 Issues opened by 81 people

• Error when loading a module file by calling torch::jit::load(..)

  #129347 opened Jun 24, 2024

• NCCL Blocking Send/Recv are Non-blocking in practice

  #129341 opened Jun 24, 2024

• `linspace()` can also use `complex` and `bool` type for `start` and `end` argument against the doc

  #129338 opened Jun 24, 2024

• pytorch-nightly export KINETO_USE_DAEMON=1 Cannot initialize CUDA without ATen_cuda library

  #129336 opened Jun 23, 2024

• `end`, `start` and `step` argument of `arange()` work with a 0D tensor against error messages

  #129334 opened Jun 23, 2024

• `start` and `step` of `arange()` should be optional on the doc

  #129333 opened Jun 23, 2024

• torch.Tensor.register_hook() source link does not work

  #129332 opened Jun 23, 2024

• Exporting the operator 'aten::fft_fft' to ONNX opset version 12 is not supported.

  #129331 opened Jun 23, 2024

• Fuyou Training Framework Integration for PyTorch

  #129330 opened Jun 23, 2024

• _foreach_addc_

  #129329 opened Jun 23, 2024

• Torch dynamo deep dive and overview discrepancy

  #129328 opened Jun 23, 2024

• [export/dynamo] torch._check fails at compile time when the condition evaluates to False

  #129327 opened Jun 23, 2024

• `repeat_interleave()` without `repeats` argument and `input` keyword works

  #129322 opened Jun 23, 2024

• `int` type for `dims` of `tile()` without `dims=` works with a tensor against the doc

  #129319 opened Jun 22, 2024

• [TP+FSDP2] model weights become fully shard again after calling model.unshard() followed by dcp get_model_state_dict

  #129313 opened Jun 22, 2024

• `msort()` can use the 0D tensor of a complex type value against error message

  #129312 opened Jun 22, 2024

• The unexpected behavior of `argsort()`

  #129311 opened Jun 22, 2024

• Upgrade dependencies MKL and Intel OpenMP to 2024.2.0

  #129310 opened Jun 22, 2024

• `argsort()` can use the 0D tensor of a complex type value against error message

  #129309 opened Jun 22, 2024

• [ONNX][low pri] Move old (non-public) implementation into legacy/ and schedule for deprecation

  #129308 opened Jun 22, 2024

• [ExecutionTraceObserver] Tracer gets stuck using Pytorch 2.2 versions for some models using torch.compile

  #129306 opened Jun 22, 2024

• C++ API: add torch::manual_seed run error failed(-1073741819)

  #129305 opened Jun 22, 2024

• `python3 setup.py bdist_wheel` tries to write to /usr/local/... during build

  #129304 opened Jun 22, 2024

• Incorrect index from torch.mode

  #129303 opened Jun 22, 2024

• (Refactor) Change default nonlinearity and bound calculation in kaiming_uniform_ & kaiming_normal_ and change kaiming_uniform call in reset_parameters in conv.py & linear.py to avoid sqrt(5) confusion (and maybe change numerical bound val in kaiming_uniform_ and numerical std val in kaiming_normal?)

  #129302 opened Jun 22, 2024

• The unexpected behavior of `sort()`

  #129298 opened Jun 22, 2024

• `sort()` can use the 0D tensor of a `complex` type value against error message

  #129296 opened Jun 22, 2024

• forward_ad ignores checkpoints

  #129293 opened Jun 22, 2024

• DISABLED test_dummy_mha_with_nt_cuda (__main__.TestNestedTensorSubclassCUDA)

  #129292 opened Jun 22, 2024

• Support for torch.Generator with JIT

  #129282 opened Jun 21, 2024

• [ONNX] Create a new compiler in torchbench to start measuring torch-onnx

  #129280 opened Jun 21, 2024

• [ONNX] Create unit tests for the new export path by adapting all existing tests

  #129279 opened Jun 21, 2024

• [ONNX] Migrate logic from torch-onnx to torch.onnx

  #129277 opened Jun 21, 2024

• [ONNX] Full support of dynamic axes

  #129276 opened Jun 21, 2024

• [ONNX] Missing operator tracker

  #129275 opened Jun 21, 2024

• [ONNX] Exporter improvement tasks

  #129274 opened Jun 21, 2024

• RecursionError for MaskedTensor.where

  #129272 opened Jun 21, 2024

• Move Memory Allocation for Autotuning out of the critical path

  #129258 opened Jun 21, 2024

• UNSTABLE pull / linux-focal-py3.12-clang10-experimental-split-build / test (dynamo)

  #129256 opened Jun 21, 2024

• UNSTABLE pull / linux-focal-py3.12-clang10-experimental-split-build / test (default)

  #129248 opened Jun 21, 2024

• RuntimeError: get_parameter is not supported on ScriptModules

  #129247 opened Jun 21, 2024

• Release corronspoding CUDA deps of `pytorch-nightly::pytorch-cuda` along with `pytorch-nightly::pytorch`

  #129230 opened Jun 21, 2024

• Incorrect behavior of dtensor full_tensor for TP+FSDP2

  #129229 opened Jun 21, 2024

• CVE-2024-5480 reported by security analyzers

  #129228 opened Jun 21, 2024

• model.generate(..) slow and huge GPU memory consumption

  #129226 opened Jun 21, 2024

• Why need to transpose when collate a sequence data in dataloader?

  #129225 opened Jun 21, 2024

• _If pin_memory_thread is alive but dataqueue is empty, it will fall into a dead loop,

  #129224 opened Jun 21, 2024

• ignore all target for CrossEntropyLoss 2d，return nan but 1d return 0

  #129222 opened Jun 21, 2024

• `Conv1d` with out_channels > 65536 gives wrong result in MPS

  #129207 opened Jun 21, 2024

• Bug in calling full_tensor() when model is sharded with tensor parallel and FSDP-2

  #129206 opened Jun 21, 2024

• Export model using onnx.dynamo_export has bug of "torch._dynamo.exc.Unsupported: call_method TupleVariable() size [ConstantVariable(int)] {}"

  #129200 opened Jun 21, 2024

• RuntimeError when using torch.ops.aten._jagged_to_padded_dense_forward with large jagged tensors

  #129191 opened Jun 21, 2024

• extending forward-mode AD docs should really point to an example

  #129176 opened Jun 20, 2024

• [custom_op] support str as default values

  #129175 opened Jun 20, 2024

• [custom_op] support dtype as default values

  #129174 opened Jun 20, 2024

• TORCHINDUCTOR_FORCE_DISABLE_CACHES=1 doesn't appear to clear file cache

  #129159 opened Jun 20, 2024

• Example usage for `convert_conv3d_weight_memory_format` does not work anymore

  #129158 opened Jun 20, 2024

• DISABLED test_nn_sequential_invocation (__main__.MiscTests)

  #129156 opened Jun 20, 2024

• DISABLED test_metadata_parsing_with_layer_split (__main__.TestSerialize)

  #129155 opened Jun 20, 2024

• [RFC][C10D] Avoid creating new nccl communicator for each P2P pair

  #129140 opened Jun 20, 2024

• Torch compile initialises CUDA context, even compiling CPU functions

  #129131 opened Jun 20, 2024

• Questions about CVE-2024-31583 and CVE-2024-31580

  #129122 opened Jun 20, 2024

• interpolate nearest get values zero when outputs over 4G elements

  #129118 opened Jun 20, 2024

• [inductor][perf] Inductor/Triton softmax kernel is slower than eager

  #129104 opened Jun 20, 2024

• [inductor][perf] Suboptimal codegen for horizontally fused softmax

  #129102 opened Jun 20, 2024

• Torch Threading causes Seg Fault in pygame.

  #129100 opened Jun 20, 2024

• take_along_dim or gather unstable results on cpu with stride 1

  #129093 opened Jun 19, 2024

• Adding betainc

  #129085 opened Jun 19, 2024

• UNSTABLE pull / linux-focal-cuda12.1-py3.10-gcc9-experimental-split-build / test (default)

  #129080 opened Jun 19, 2024

• Regression in loading optimizer learning rate

  #129079 opened Jun 19, 2024

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

  #129053 opened Jun 19, 2024

• Add comment for label_smoothing parameter in torch.nn.CrossEntropyLoss

  #129050 opened Jun 19, 2024

• [PT2E Quantization] Graph with concatenation of the same node will raise RecursionError when prepare_pt2e

  #129038 opened Jun 19, 2024

• torch parallel Broadcast inconsistency

  #129032 opened Jun 19, 2024

• Extract some public APIs from torch::cuda::initModule(module) to torch::initModule()

  #129027 opened Jun 19, 2024

• Expand Tag Set

  #129020 opened Jun 19, 2024

• Zluda Support

  #129019 opened Jun 19, 2024

• Spurious "socket cannot be initialized" error messages

  #128998 opened Jun 18, 2024

• Look up tensor device member inside Tensor is_pinned() implementation instead of accepting an outside input

  #128988 opened Jun 18, 2024

• [RFC][Pipelining] Support separate dW/dInput in Schedule and Stage

  #128974 opened Jun 18, 2024

• Trying to use forward AD with _scaled_dot_product_flash_attention that does not support it because it has not been implemented yet.

  #128971 opened Jun 18, 2024

• RuntimeError: NCCL error: invalid usage when deploy LLM model by vllm. (torch version: 2.3.0+cu118)

  #128963 opened Jun 18, 2024

• `torch.compile` fails with `fullgraph=True` when accessing `getitem` of a `Tensor` subclass

  #128961 opened Jun 18, 2024

• [Dynamo x torch_function] magic methods and methods with regular Tensors don't seem to work

  #128960 opened Jun 18, 2024

• caffe2 removal

  #128959 opened Jun 18, 2024

• [Dynamo] Maybe we shouldn't attempt to recursively compile inside of a frame if the frame hit cache limit

  #128954 opened Jun 18, 2024

• Question about torch.lstsq and torch.linalg.lstsq

  #128952 opened Jun 18, 2024

• Memory consumption of conv3d grows too quickly with certain input shapes.

  #128947 opened Jun 18, 2024

• [MAC] Convolution with kernel size 3 yields different results depending on whether gradient is enabled or not.

  #128945 opened Jun 18, 2024

• torch.compile graph break due to unsupported builtin filter function

  #128944 opened Jun 18, 2024

• torch.compile graph break with unsupported LOAD_BUILD_CLASS

  #128942 opened Jun 18, 2024

• Inquiry Regarding PyTorch Data Mirroring and Proxy Services

  #128940 opened Jun 18, 2024

• Unable to install PyTorch on M1 Macos with Python 3.10.14

  #128939 opened Jun 18, 2024

• Performance degradation for certain input using Conv2D

  #128936 opened Jun 18, 2024

• version inquiry

  #128934 opened Jun 18, 2024

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

  #128933 opened Jun 18, 2024

• onnx.export() fails on aten::embedding_bag with padding_idx

  #128930 opened Jun 18, 2024

• xpu: set of not implemented aten ops affecting huggingface tests

  #128914 opened Jun 18, 2024

• [pipelining] Free memory in stage after use

  #128910 opened Jun 17, 2024

• Unable to export Phi-3-vision model to exported program

  #128906 opened Jun 17, 2024

• TypeError: Cannot convert symbols to int

  #128895 opened Jun 17, 2024

• Improve concat fusion with matmuls when autotuning

  #128889 opened Jun 17, 2024

• UNSTABLE inductor / rocm6.1-py3.8-inductor / test (inductor)

  #128871 opened Jun 17, 2024

• Update PyTorch CI to numpy 2.0

  #128860 opened Jun 17, 2024

• UNSTABLE inductor-cu124 / cuda12.4-py3.10-gcc9-sm86 / test (inductor_torchbench)

  #128851 opened Jun 17, 2024

• UNSTABLE inductor-cu124 / cuda12.4-py3.10-gcc9-sm86 / test (dynamic_inductor_torchbench)

  #128850 opened Jun 17, 2024

• UNSTABLE inductor-cu124 / cuda12.4-py3.10-gcc9-sm86 / test (aot_inductor_torchbench)

  #128849 opened Jun 17, 2024

• Windows builds with VS2022

  #128835 opened Jun 17, 2024

• TORCHDYNAMO_REPRO_AFTER=aot produces invalid repro code

  #128830 opened Jun 17, 2024

• DISABLED test_nn_sequential_invocation_reposition_indices_inline_inbuilt_nn_modules (__main__.InlineInbuiltNNModulesMiscTests)

  #128822 opened Jun 17, 2024

• Questions about TCPStoreLibUV

  #128821 opened Jun 17, 2024

• Environment gating of CUDA_VISIBLE_DEVICES returns a CUDA initialization error.

  #128819 opened Jun 17, 2024

• ONNX Dynamo Export - Unsupported FX nodes: {'call_function': ['aten._upsample_bilinear2d_aa.default']}.

  #128818 opened Jun 17, 2024

502 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.

• [runtime asserts] deduplicate runtime asserts & CSE

  #128599 commented on Jun 22, 2024 • 38 new comments

• Don't decompose functional composite ops in export inference IR

  #128077 commented on Jun 22, 2024 • 31 new comments

• Fix device propagation for checkpointing

  #128671 commented on Jun 24, 2024 • 23 new comments

• [vision hash update] update the pinned vision hash

  #125806 commented on Jun 24, 2024 • 21 new comments

• [Inductor][ROCm] Composable Kernel backend for Inductor

  #125453 commented on Jun 23, 2024 • 20 new comments

• [cuDNN] Graph-capturable cuDNN CTCLoss

  #128271 commented on Jun 22, 2024 • 20 new comments

• [inductor] Add lowering and codegen for aten.sort

  #128458 commented on Jun 22, 2024 • 19 new comments

• General MPS op coverage tracking issue

  #77764 commented on Jun 23, 2024 • 17 new comments

• [v.2.4.0] Release Tracker

  #128436 commented on Jun 21, 2024 • 15 new comments

• [BE] enable UFMT for `torch/nn/functional.py`

  #128592 commented on Jun 23, 2024 • 13 new comments

• [Inductor][CPP] Enable Local Buffer for Outer loop fusion

  #126967 commented on Jun 20, 2024 • 13 new comments

• Add decompositions for copy variants of view ops

  #128416 commented on Jun 21, 2024 • 12 new comments

• Created docs for make_fx in torch.fx.experimental.proxy_tensor

  #128441 commented on Jun 21, 2024 • 10 new comments

• [Nested Tensor]fix sdpa backward for the special case with ragged second batch dim and constant length

  #128349 commented on Jun 18, 2024 • 9 new comments

• Add `torch.put_along_dim` and `torch.put_along_dim_` like `np.put_along_axis`

  #125601 commented on Jun 20, 2024 • 9 new comments

• skip test_graph_capture_oom for jetson

  #128661 commented on Jun 19, 2024 • 9 new comments

• Fix tensor subclass + dynamic shapes in torch.compile + aot autograd

  #125941 commented on Jun 20, 2024 • 8 new comments

• [inductor] switch CppCodeCache to new cpp_builder.

  #128303 commented on Jun 21, 2024 • 8 new comments

• TorchInductor CPU Performance Dashboard

  #93531 commented on Jun 18, 2024 • 8 new comments

• Support allowlisted modules and op overloads in AOTAutogradCache

  #128329 commented on Jun 21, 2024 • 8 new comments

• Add aten._unsafe_masked_index

  #116491 commented on Jun 20, 2024 • 8 new comments

• Modularize aten parameter parser and checker

  #125308 commented on Jun 20, 2024 • 7 new comments

• Change index_put on GPU to accept FP8 inputs

  #128758 commented on Jun 22, 2024 • 7 new comments

• [ROCm] Check supported archs before setting preferred blas backend to hipblasLT

  #128753 commented on Jun 22, 2024 • 7 new comments

• [NT] Implementing Multi-Head Attention with NestedTensors

  #125214 commented on Jun 23, 2024 • 6 new comments

• [DO NOT MERGE] Testing cuDNN SDPA on sm80+

  #128571 commented on Jun 19, 2024 • 6 new comments

• [CI] add xpu test in periodic workflow

  #126410 commented on Jun 21, 2024 • 6 new comments

• Reduce all tensors to their metadata in AOTAutogradCache; add tests

  #128583 commented on Jun 21, 2024 • 6 new comments

• [ROCm] hipSPARSELt Integration

  #124320 commented on Jun 21, 2024 • 6 new comments

• [CI] Enable amp accuracy check for inductor cpu

  #127758 commented on Jun 24, 2024 • 6 new comments

• Write dynamo benchmarks performance result to csv when throw exceptions

  #126764 commented on Jun 17, 2024 • 6 new comments

• [RFC] Add support for device extension autoloading

  #127074 commented on Jun 21, 2024 • 6 new comments

• [ROCm] Unskip scaled_dot_product_attention tests on ROCm

  #127966 commented on Jun 20, 2024 • 6 new comments

• [1/N] Enable unused variable warnings on torch_cpu and fix some violations

  #128670 commented on Jun 23, 2024 • 6 new comments

• ROCm: `fatal error: aotriton/flash.h: No such file or directory` when building with `USE_ROCM=1`

  #125230 commented on Jun 22, 2024 • 5 new comments

• [AOTAutograd] Micro-optimize runtime_wrapper

  #128188 commented on Jun 20, 2024 • 5 new comments

• Feat: Updated torch.nn.Modules.set_submodules()

  #127714 commented on Jun 21, 2024 • 5 new comments

• [RFC][pipelining] PipelineStage should let user control send/recv endpoints

  #128665 commented on Jun 20, 2024 • 5 new comments

• fix torch.prod vectorized path for bool

  #128009 commented on Jun 22, 2024 • 5 new comments

• [3/N] Non-Tensor: Support string parameter for aten operations

  #125831 commented on Jun 20, 2024 • 5 new comments

• [halide-backend] Add GPU support

  #127506 commented on Jun 23, 2024 • 5 new comments

• sdp::SDPBackend::flash_attention support PrivateUse1

  #126392 commented on Jun 20, 2024 • 5 new comments

• 2.6.0 Released a second time on the same version breaking production customers

  #128653 commented on Jun 21, 2024 • 4 new comments

• [RFC] Per-Parameter-Sharding FSDP

  #114299 commented on Jun 18, 2024 • 4 new comments

• Nested tensor subclass support

  #127431 commented on Jun 21, 2024 • 4 new comments

• [Inductor][Intel GPU] Support codegen empty_strided_xpu, align with #118255.

  #126678 commented on Jun 21, 2024 • 4 new comments

• Custom attention recompilations

  #121504 commented on Jun 18, 2024 • 4 new comments

• crash@sleef_tryVXE2 () while trying to run torch.compile() BERT model

  #128503 commented on Jun 19, 2024 • 4 new comments

• Fp8 support for item() with cuda, index_select, and fill_ with cpu

  #128780 commented on Jun 18, 2024 • 4 new comments

• Doc (nn): improve doc-string of class Linear.

  #128792 commented on Jun 22, 2024 • 4 new comments

• [sparse] Add cuSPARSELt as a backend

  #128534 commented on Jun 20, 2024 • 4 new comments

• Errors when 0-dim tensor of complex or bool type passed to aminmax.

  #128404 commented on Jun 22, 2024 • 4 new comments

• [docs] Urls changed => forum links would become invalid

  #39007 commented on Jun 17, 2024 • 4 new comments

• Make `torch.autograd.Function` support `vmap`

  #128020 commented on Jun 18, 2024 • 3 new comments

• [xla hash update] update the pinned xla hash

  #126672 commented on Jun 17, 2024 • 3 new comments

• Fixed CUDA randint generation for large ranges.

  #126066 commented on Jun 20, 2024 • 3 new comments

• Remove ProcessGroupCudaP2P and change async-TP to use SymmetricMemory

  #128762 commented on Jun 22, 2024 • 3 new comments

• [ONNX] Skip assertion nodes

  #126889 commented on Jun 22, 2024 • 3 new comments

• Remove unused type traits in torch/csrc/utils

  #128799 commented on Jun 21, 2024 • 3 new comments

• [pytree] update treespec `children_specs` access

  #116374 commented on Jun 22, 2024 • 3 new comments

• [WIP] mark NestedInts as symints instead of symfloats

  #127587 commented on Jun 17, 2024 • 3 new comments

• adjust thresholds for gluon_inception_v3, beit_base_patch16_224, phli…

  #127664 commented on Jun 19, 2024 • 3 new comments

• [inline-inbuilt-nn-modules] Torch compile with DDP errors on parameterized modules

  #113415 commented on Jun 23, 2024 • 3 new comments

• Separate AOTI Eager utils as a single file

  #125819 commented on Jun 20, 2024 • 3 new comments

• [CI] Add inductor cpu accuracy test running on AVX2 runners

  #128682 commented on Jun 18, 2024 • 3 new comments

• [cuDNN][Quantization] Don't print when plan finalization fails in cuDNN quantization backend

  #128177 commented on Jun 19, 2024 • 3 new comments

• [CI][BE] Update retry action to v3.0.0

  #119403 commented on Jun 18, 2024 • 3 new comments

• torch.compile not compatible with multiprocessing pool

  #97992 commented on Jun 17, 2024 • 3 new comments

• [Profiler] Add TSC Clock Callback to CUPTI

  #125036 commented on Jun 22, 2024 • 3 new comments

• `torch.special.gammainc` backward pass with respect to the first argument

  #80025 commented on Jun 17, 2024 • 3 new comments

• Let dynamo inline functional_call

  #128646 commented on Jun 20, 2024 • 3 new comments

• Support for expandable segments with cuda graph trees

  #128068 commented on Jun 19, 2024 • 3 new comments

• Add support for XPU accumulate type

  #128579 commented on Jun 21, 2024 • 3 new comments

• dynamo: use equality guards instead of id guards for Placement/DeviceMesh

  #124401 commented on Jun 22, 2024 • 3 new comments

• [cuDNN][SDPA] Remove `TORCH_CUDNN_SDPA_ENABLED=1`, enable cuDNN SDPA by default on H100 and 2nd on other archs >= sm80

  #125343 commented on Jun 21, 2024 • 3 new comments

• [Split Build][no commit] Test CI with builder changes

  #127958 commented on Jun 19, 2024 • 3 new comments

• `__getitem__` fails to vmap for one dimensional tensors

  #124423 commented on Jun 18, 2024 • 3 new comments

• Flex Decoding

  #128678 commented on Jun 21, 2024 • 3 new comments

• [WIP] Warn on future divergent behavior for conditional views

  #126129 commented on Jun 22, 2024 • 2 new comments

• 2 Dynamo test are failing with "'int' object has no attribute 'wrapped'".

  #120650 commented on Jun 18, 2024 • 2 new comments

• Improve debugability of warnings/errors "Triggered internally at"

  #128064 commented on Jun 18, 2024 • 2 new comments

• Support "symmetric" reflection padding

  #46240 commented on Jun 21, 2024 • 2 new comments

• ```FlopCounterMode``` returns 0 when inference mode is on during forwardpropagation.

  #126268 commented on Jun 23, 2024 • 2 new comments

• Dynamo should prune non-live captured variables

  #127350 commented on Jun 18, 2024 • 2 new comments

• Don't run addruntimeassertion pass

  #125948 commented on Jun 18, 2024 • 2 new comments

• 2nd compile of deepcopy(model) fails on multiple ubuntu-pc (fatal error: Python.h: file not found)

  #128121 commented on Jun 18, 2024 • 2 new comments

• Expected grad_output types don't match available grad_output types when using tensor parallelism with DTensors

  #128636 commented on Jun 18, 2024 • 2 new comments

• Spectral Normalization can not be applied to Conv{1,2,3}d

  #99149 commented on Jun 24, 2024 • 2 new comments

• Pytorch ROCM windows builds

  #109204 commented on Jun 21, 2024 • 2 new comments

• Failed to compile: null in call to `__builtin_memmove(__result, __first, sizeof(_Tp) * _Num);` Debian 12, ppc64le, gcc 12.2

  #112089 commented on Jun 22, 2024 • 2 new comments

• Dynamo export: limited support in Torch.cond

  #123972 commented on Jun 21, 2024 • 2 new comments

• cudnn not found

  #15167 commented on Jun 21, 2024 • 2 new comments

• Bug in `torch.compile` with standard type checking tools beartype

  #122093 commented on Jun 18, 2024 • 2 new comments

• `torch.cuda.memory_summary()` can give `KeyError`

  #117130 commented on Jun 19, 2024 • 2 new comments

• [dtensor][test] test case suite for comm_mode features

  #128729 commented on Jun 21, 2024 • 2 new comments

• [cond] inlining into one of the branches when pred is a python constant

  #128709 commented on Jun 20, 2024 • 2 new comments

• [c10d][simple] increase the default heartbeat timeout to be larger

  #128751 commented on Jun 21, 2024 • 2 new comments

• [dynamo] Fakify result of delegate

  #128752 commented on Jun 18, 2024 • 2 new comments

• New example for preserve_node_meta

  #128681 commented on Jun 18, 2024 • 2 new comments

• [Fix] torch.numel() in TSCovnerter

  #128761 commented on Jun 22, 2024 • 2 new comments

• [Bug] The cuDNN version is too old!

  #128207 commented on Jun 17, 2024 • 2 new comments

• GradType: a subset of dtype that is differentiable, containing all float and complex dtypes

  #128793 commented on Jun 17, 2024 • 2 new comments

• partitioner doesn't appear to respect SAC region

  #128730 commented on Jun 20, 2024 • 2 new comments

• autograd with `is_grads_batched=True` fails on GroupNorm

  #128703 commented on Jun 17, 2024 • 2 new comments

• [DDP] DDP bucket memory release during fwd step

  #128696 commented on Jun 17, 2024 • 2 new comments

• Add MaskedTensor support to _is_any_true

  #128574 commented on Jun 18, 2024 • 2 new comments

• [C10d]: Work state in dump trace file is not deterministic.

  #128805 commented on Jun 19, 2024 • 2 new comments

• Unable to assign `nn.Parameter(DTensor)` (created outside of compile region) to an nn.Module param attribute during Dynamo tracing

  #128742 commented on Jun 18, 2024 • 2 new comments

• Tracing per-param sharding FSDP

  #114286 commented on Jun 18, 2024 • 2 new comments

• [inductor] use same method to handle exception with aten

  #127868 commented on Jun 19, 2024 • 2 new comments

• [Intel GPU] xpu-ops codegen via backend whitelist

  #127865 commented on Jun 19, 2024 • 2 new comments

• [Dynamo] Fix refleak in 3.12+ and Dynamic Shapes test_parameter_free

  #124302 commented on Jun 18, 2024 • 2 new comments

• Added hpu backend support in fsdp utils

  #127757 commented on Jun 18, 2024 • 2 new comments

• Long compilation time for hf_T5_generate inference cause timeout

  #121989 commented on Jun 17, 2024 • 2 new comments

• [caffe2][be][2/n] migrate gloabl static initializer

  #127620 commented on Jun 23, 2024 • 2 new comments

• Save quantization_tag in export graph serialization

  #127473 commented on Jun 20, 2024 • 2 new comments

• Using autograd.Functions defined in torch/ cause graph breaks

  #118334 commented on Jun 18, 2024 • 2 new comments

• [triton hash update] update the pinned triton hash

  #115529 commented on Jun 17, 2024 • 2 new comments

• [dynamo] Automatically convert loop bodies to function calls

  #113538 commented on Jun 21, 2024 • 2 new comments

• torch.cumprod will silently cast the output data type to int64

  #128294 commented on Jun 18, 2024 • 2 new comments

• [PT2D] Make the speedup benchmark works with DDP + CompiledAutograd

  #121315 commented on Jun 23, 2024 • 1 new comment

• turned on matrix-multiplication => matrix-vector multiplication always on if reduction-dim is contiguous

  #120954 commented on Jun 19, 2024 • 1 new comment

• Increase riscv implementation in DepthwiseConvKernel

  #127867 commented on Jun 18, 2024 • 1 new comment

• PyPy support

  #17835 commented on Jun 23, 2024 • 1 new comment

• `RuntimeError: invalid dtype for bias` when use compile + autocast

  #124901 commented on Jun 23, 2024 • 1 new comment

• [Dynamo] Check for __bool__ attribute before accessing it

  #120943 commented on Jun 18, 2024 • 1 new comment

• [FSDP] Removed clamp to `NO_SHARD` for world size 1

  #120334 commented on Jun 23, 2024 • 1 new comment

• [WIP]Intel GPU oneDNN upstreaming: Conv pointwise fusion

  #118064 commented on Jun 18, 2024 • 1 new comment

• [WIP]Intel GPU oneDNN upstreaming: Linear pointwise fusion

  #117824 commented on Jun 18, 2024 • 1 new comment

• [pytree] support PyStructSequence types for Python pytree

  #113258 commented on Jun 22, 2024 • 1 new comment

• SummaryWriter reports encoding error

  #73909 commented on Jun 20, 2024 • 1 new comment

• [RFC] A device-agnostic Python runtime API design for stream-based accelerators

  #128403 commented on Jun 20, 2024 • 1 new comment

• `torch.compile` with `reduce-overhead`: very long compile time + GPU memory continuously to grow

  #128424 commented on Jun 20, 2024 • 1 new comment

• does FSDP support AMSP (a new DP shard strategy)

  #128706 commented on Jun 20, 2024 • 1 new comment

• Fan out calculation broken for group (depthwise) convolution

  #23854 commented on Jun 20, 2024 • 1 new comment

• ROCm loses some supported GPUs by requiring hipblaslt

  #119081 commented on Jun 20, 2024 • 1 new comment

• PyTorch trunk is frequently broken

  #128180 commented on Jun 20, 2024 • 1 new comment

• DISABLED test_arange_dynamic_cuda (__main__.TestInductorDynamicCUDA)

  #127067 commented on Jun 20, 2024 • 1 new comment

• custom ops should have needs_fixed_stride_order by default

  #124647 commented on Jun 20, 2024 • 1 new comment

• Dynamo silently ignores TorchDispatchMode

  #105929 commented on Jun 20, 2024 • 1 new comment

• [compile] output does not match eager mode

  #100075 commented on Jun 20, 2024 • 1 new comment

• Torch.compile Error: RuntimeError: aten::_conj() Expected a value of type 'Tensor' for argument 'self' but instead found type 'complex'.

  #105290 commented on Jun 20, 2024 • 1 new comment

• torch.compile incorrect when imperative autograd APIs are used

  #91468 commented on Jun 20, 2024 • 1 new comment

• [RFC] Support reinplaceble ops for custom ops in Inductor

  #124933 commented on Jun 20, 2024 • 1 new comment

• Inductor generates unnecessary allocation + copy operations for custom ops with mutable inputs

  #127660 commented on Jun 20, 2024 • 1 new comment

• Add `int` type to `device` parameter of torch.set_default_device() on the doc

  #126646 commented on Jun 20, 2024 • 1 new comment

• custom ops with needs_fixed_stride_order doesn't work with auto_functionalized

  #128084 commented on Jun 20, 2024 • 1 new comment

• version libcudnn_ops_infer.so.8 not defined in file libcudnn_ops_infer.so.8 with link time reference

  #104591 commented on Jun 21, 2024 • 1 new comment

• Pytorch dataloader not loading first-available data with multiple workers

  #105203 commented on Jun 21, 2024 • 1 new comment

• custom_op API follow-ups

  #101191 commented on Jun 21, 2024 • 1 new comment

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

  #109445 commented on Jun 21, 2024 • 1 new comment

• Pytorch build from source failed with GCC 12.3

  #127920 commented on Jun 22, 2024 • 1 new comment

• compilation fails `error: invalid argument '-std=c++17' not allowed with 'C'`

  #103222 commented on Jun 22, 2024 • 1 new comment

• Dead link in `torch.compile` docs

  #119272 commented on Jun 22, 2024 • 1 new comment

• UserWarning: The TorchScript type system doesn't support instance-level annotations on empty non-base types in `__init__`.

  #89064 commented on Jun 22, 2024 • 1 new comment

• [dynamo] Dynamo traces through __torch_dispatch__ on custom tensor subclasses

  #128160 commented on Jun 22, 2024 • 1 new comment

• [feature request] New function `torch.slice(...)` mirroring TorchScript op signature or add step argument to `torch.narrow`

  #41625 commented on Jun 22, 2024 • 1 new comment

• RuntimeError: Expected to have finished reduction in the prior iteration before starting a new one.

  #43259 commented on Jun 22, 2024 • 1 new comment

• Add Swiglu activation function

  #128712 commented on Jun 23, 2024 • 1 new comment

• [FSDP2] Allowed `List[nn.Module]` as arg

  #127786 commented on Jun 22, 2024 • 1 new comment

• Supervisor as a torchrun rendezvous impl

  #127515 commented on Jun 20, 2024 • 1 new comment

• Implement a generic function scheduler

  #127200 commented on Jun 21, 2024 • 1 new comment

• [Split Build] Test split build in CI

  #126699 commented on Jun 19, 2024 • 1 new comment

• Use float data type for Half var_sum in batchnorm stats updating on CPU

  #126525 commented on Jun 24, 2024 • 1 new comment

• Re-implement pin_memory to be device-agnostic by leveraging the Accelerator concept

  #126376 commented on Jun 19, 2024 • 1 new comment

• [4/N] Non-Tensor: Support layout, device and dtype for aten operations

  #125897 commented on Jun 19, 2024 • 1 new comment

• bool inherited from number

  #125577 commented on Jun 24, 2024 • 1 new comment

• Add raise_last_usage memory optimization pass to Inductor

  #125559 commented on Jun 19, 2024 • 1 new comment

• add uuid in cudaDeviceProperties

  #125083 commented on Jun 19, 2024 • 1 new comment

• [rfc]: vendor in open-telemetry

  #124800 commented on Jun 24, 2024 • 1 new comment

• Allow device tensors that use numpy for serialization to use weights_only unpickler

  #124763 commented on Jun 23, 2024 • 1 new comment

• Fix issue 112919

  #124746 commented on Jun 23, 2024 • 1 new comment

• Prevent cuda:0 context initialization when working on another cuda device

  #124722 commented on Jun 23, 2024 • 1 new comment

• [pytorch] Add a c10::Bfloat16 ctor in OSS repo which takes __hip_bfloat16

  #124713 commented on Jun 22, 2024 • 1 new comment

• Skip `deepspeed` and `triton` in dynamo

  #124273 commented on Jun 18, 2024 • 1 new comment

• [Inductor] support masked vectorization for the tail_loop for integer datatypes and bool datatype

  #128802 commented on Jun 17, 2024 • 1 new comment

• [caffe2][be] migrate gloabl static initializer

  #128784 commented on Jun 21, 2024 • 1 new comment

• AutogradMeta is nullptr for non-differentiable tensors on creation

  #128746 commented on Jun 19, 2024 • 1 new comment

• Kineto profiler: collecting observer traces from C++ child threads

  #128743 commented on Jun 18, 2024 • 1 new comment

• Raise exception if torch.func.* calls torch.compile functions

  #128736 commented on Jun 18, 2024 • 1 new comment

• fix the decomposition of aten.threshold

  #128707 commented on Jun 18, 2024 • 1 new comment

• Remove caffe2 namespace from c10/macros/Macros.h

  #128672 commented on Jun 18, 2024 • 1 new comment

• [POC] Split before autograd allow in graph

  #128647 commented on Jun 20, 2024 • 1 new comment

• Allow get attributes on DDP similar to FSDP

  #128620 commented on Jun 18, 2024 • 1 new comment

• Add hooks for execution on intel gaudi devices - 1

  #128584 commented on Jun 24, 2024 • 1 new comment

• [ROCm] Enable F8 Inductor Unit tests

  #128353 commented on Jun 21, 2024 • 1 new comment

• Ignore functional tensor wrapper when caching

  #128335 commented on Jun 21, 2024 • 1 new comment

• update call map to allow multiple input parameters

  #128282 commented on Jun 20, 2024 • 1 new comment

• Wrap output with FakeTensor if input FakeTensor is not preserved

  #128206 commented on Jun 24, 2024 • 1 new comment

• Make Tensor's __dlpack__ and __dlpack_device__ account for XLA.

  #128176 commented on Jun 21, 2024 • 1 new comment

• [pipelining] enable inputs for all model stages

  #128115 commented on Jun 17, 2024 • 1 new comment

• [inductor] custom do_bench_gpu with smart cache flushing

  #127953 commented on Jun 22, 2024 • 1 new comment

• [sym_shapes][perf] Optimize bound_sympy avoiding sympy equals

  #124211 commented on Jun 22, 2024 • 1 new comment

• Remove seq support check in process group

  #124138 commented on Jun 18, 2024 • 1 new comment

• Add realize after pointwise lowering

  #124118 commented on Jun 22, 2024 • 1 new comment

• Hacks to work around that ScriptMethod does not have code/signature

  #124115 commented on Jun 18, 2024 • 1 new comment

• Update README.md

  #124028 commented on Jun 19, 2024 • 1 new comment

• [BE]: Update NCCL submodule to 2.21.5

  #124014 commented on Jun 23, 2024 • 1 new comment

• [FSDP] Move the flattened tensors back to GPU to prevent CPU OOM

  #124008 commented on Jun 19, 2024 • 1 new comment

• Aarch64 cd upgrade

  #123747 commented on Jun 17, 2024 • 1 new comment

• Fix numerical instability in vector_norm when receiving large size tensor

  #123416 commented on Jun 22, 2024 • 1 new comment

• Reenable dim for python 3.12

  #123384 commented on Jun 23, 2024 • 1 new comment

• prototype for graph transform observer

  #123361 commented on Jun 20, 2024 • 1 new comment

• Dynamo: support proxying tensor subclass constructors, including with non-fx types

  #123350 commented on Jun 17, 2024 • 1 new comment

• DTensor: avoiding crashing on dynamic shapes in a few places

  #123349 commented on Jun 17, 2024 • 1 new comment

• Add Gaudi support to benchmarks/dynamo/* benchmark.

  #122960 commented on Jun 23, 2024 • 1 new comment

• [Dynamic Shapes] Fix error handling for indirectly fully constrained dynamic dimensions

  #122913 commented on Jun 22, 2024 • 1 new comment

• [Quant][PT2E] enable qlinear post op fusion for dynamic quant & qat

  #122667 commented on Jun 20, 2024 • 1 new comment

• [Not for review] Collect cpp_wrapper dashboard status

  #124691 commented on Jun 22, 2024 • 1 new comment

• [WIP}[FSDP] Switch to more memory efficient impl of _sync_module_states

  #124679 commented on Jun 22, 2024 • 1 new comment

• [TESTING] Don't clamp upper to 2

  #124631 commented on Jun 22, 2024 • 1 new comment

• chore(quantization): Enable PT2E symmetric dynamic quantization

  #124615 commented on Jun 22, 2024 • 1 new comment

• Make the CI failures less noicy

  #124558 commented on Jun 20, 2024 • 1 new comment

• [testing] ... int(True) != 1 ??

  #124539 commented on Jun 19, 2024 • 1 new comment

• [do not review] Add API for setting backward stream

  #124538 commented on Jun 19, 2024 • 1 new comment

• modified documentation torch.histogramdd ISSUE#124435

  #124537 commented on Jun 21, 2024 • 1 new comment

• Update README.md

  #124514 commented on Jun 18, 2024 • 1 new comment

• [Environment Variable][2/N] Use thread-safe setenv wrapper

  #124485 commented on Jun 22, 2024 • 1 new comment

• Use thread-safe getenv wrapper

  #124478 commented on Jun 22, 2024 • 1 new comment

• [sym_shapes][perf] Skip repetitive check_is_size on same expr

  #124471 commented on Jun 18, 2024 • 1 new comment

• [comm] Ensure graceful shutdown by waiting watchdog thread to finish

  #124467 commented on Jun 18, 2024 • 1 new comment

• Pianpwk/dynamo qualname

  #124434 commented on Jun 18, 2024 • 1 new comment

• Add trace_via_export option and allow exporting funcs

  #124431 commented on Jun 18, 2024 • 1 new comment

• Update _custom_ops.py to accomodate renaming of impl_abstract

  #124410 commented on Jun 18, 2024 • 1 new comment

• [DONOTREVIEW][DTenosr][Test] DTensor 2D sharding

  #124339 commented on Jun 22, 2024 • 1 new comment

• DISABLED test_deepcopy_after_parametrization_swap_True (__main__.TestNNParametrization)

  #127738 commented on Jun 18, 2024 • 1 new comment

• RuntimeError using nested tensor in Apple M1 device MPS

  #127743 commented on Jun 18, 2024 • 1 new comment

• `CompileProfiler` reports graph breaks while `dynamo.explain` reports no graph breaks

  #113443 commented on Jun 18, 2024 • 1 new comment

• [pt2d] module register_pre_forward_hook and register_forward_hook triggered graph break when it's root module

  #117584 commented on Jun 18, 2024 • 1 new comment

• [RFC] PyTorch next wheel build platform: manylinux-2.28

  #123649 commented on Jun 18, 2024 • 1 new comment

• CUDA error in torch.cdist with compute_mode=donot_use_mm_for_euclid_dist

  #128791 commented on Jun 18, 2024 • 1 new comment

• MultiheadAttention returns NaNs when need_weights=False for long sequences with a mask that ignores old tokens

  #127055 commented on Jun 18, 2024 • 1 new comment

• Dynamo ignores frame using yield

  #126360 commented on Jun 18, 2024 • 1 new comment

• The "step unsupported" graph break will make dynamo can't completely trace code after break

  #125141 commented on Jun 18, 2024 • 1 new comment

• Change the type hint for nn.Module.__call__ to be friendly to overrides.

  #74746 commented on Jun 18, 2024 • 1 new comment

• upstream `apex.normalization.FusedRMSNorm`

  #72643 commented on Jun 18, 2024 • 1 new comment

• Support Exceptions in Pytorch Export

  #123499 commented on Jun 17, 2024 • 1 new comment

• masked_index_add

  #122092 commented on Jun 17, 2024 • 1 new comment

• ImportError: libcudnn.so.8: cannot open shared object file: No such file or directory

  #104259 commented on Jun 17, 2024 • 1 new comment

• Dynamo doesn't generate resume calls after graph breaking on log calls

  #120375 commented on Jun 17, 2024 • 1 new comment

• 31 Dynamo test are failing with "'NoneType' object has no attribute 'profiler'".

  #119783 commented on Jun 17, 2024 • 1 new comment

• [feature request] `torch.to(obj, device)` supporting recursive lists/dicts/tuples of tensors probably by uplifting/promoting `torch.distributed.utils._recursive_to`

  #69431 commented on Jun 17, 2024 • 1 new comment

• [dynamo] Refactor switch statements to improve compile times

  #119128 commented on Jun 17, 2024 • 1 new comment

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

  #127008 commented on Jun 17, 2024 • 1 new comment

• lr_scheduler()

  #127884 commented on Jun 17, 2024 • 1 new comment

• [Inductor] Generate triton block pointers for discontiguous strided tensors

  #125077 commented on Jun 17, 2024 • 1 new comment

• Conformal Prediction framework to enhance reliability in risk sensitive industrial applications

  #128380 commented on Jun 17, 2024 • 1 new comment

• No factory functions for strided quantized tensors

  #74540 commented on Jun 17, 2024 • 1 new comment

• Cannot get deterministic Mask RCNN without running out of CUDA memory

  #120240 commented on Jun 17, 2024 • 1 new comment

• False INTERNAL ASSERT FAILED bug whilst training Neural Network

  #128778 commented on Jun 17, 2024 • 1 new comment

• Support for one-hot of dtypes besides torch.int64

  #53785 commented on Jun 17, 2024 • 1 new comment

• jacrev and jacfwd yield different results if one uses torch.no_grad blocks in module

  #128600 commented on Jun 17, 2024 • 1 new comment

• [export] `nn.GRU` fails to `torch.export` due to unimplemented operator

  #120626 commented on Jun 17, 2024 • 1 new comment

• binaries/dump_operator_names.cc missing iostream include

  #125134 commented on Jun 17, 2024 • 1 new comment

• Add RMS Norm layer

  #128713 commented on Jun 17, 2024 • 1 new comment

• `torch.sparse.sum` does not support boolean and int when summing over dense dimensions

  #122711 commented on Jun 18, 2024 • 1 new comment

• Don't populate f_locals to check guards

  #93753 commented on Jun 18, 2024 • 1 new comment

• Dynamo Export: Support for mutating module attributes

  #123971 commented on Jun 18, 2024 • 1 new comment

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

  #64947 commented on Jun 19, 2024 • 1 new comment

• Import Error: cannot import name 'XNNPACKQuantizer' from 'torch.ao.quantization.quantizer'

  #128114 commented on Jun 18, 2024 • 1 new comment

• Backward pass over torch.nn.functional.pad is extremely slow with half tensors

  #13058 commented on Jun 19, 2024 • 1 new comment

• torch.triu() may returns wrong values using MPS

  #100005 commented on Jun 19, 2024 • 1 new comment

• Support AMD Ryzen Unified Memory Architecture (UMA)

  #107605 commented on Jun 19, 2024 • 1 new comment

• Dark mode please 🙏🏻

  #120407 commented on Jun 20, 2024 • 1 new comment

• RuntimeError: false INTERNAL ASSERT FAILED at "C:\\actions-runner\\_work\\pytorch\\pytorch\\builder\\windows\\pytorch\\aten\\src\\ATen\\native\\BatchLinearAlgebra.cpp":1538, please report a bug to PyTorch. torch.linalg.lstsq: (Batch element 0): Argument 6 has illegal value. Most certainly there is a bug in the implementation calling the backend library.

  #125892 commented on Jun 20, 2024 • 1 new comment

• How to enable XNNPACK instead of NNPACK/MKLDNN in Windows?

  #128414 commented on Jun 20, 2024 • 1 new comment

• [Reland2] Update NVTX to NVTX3

  #109843 commented on Jun 23, 2024 • 0 new comments

• [BE] enable UFMT for `torch/storage.py`

  #127706 commented on Jun 19, 2024 • 0 new comments

• inspect.signature.bind is not supported

  #93760 commented on Jun 18, 2024 • 0 new comments

• [DONT MERGE][dynamo] Turn on inlining of inbuilt nn modules

  #128148 commented on Jun 23, 2024 • 0 new comments

• torch.Tensor.random_ causes invalid syntax in InternalTorchDynamoError

  #121621 commented on Jun 17, 2024 • 0 new comments

• Attempting to copy from device cpu to device meta, but cross-device copies are not allowed!

  #121619 commented on Jun 17, 2024 • 0 new comments

• [autograd] Support GradientEdge as output for torch.autograd.grad

  #127766 commented on Jun 19, 2024 • 0 new comments

• [PT-D] Relaxed `contract` to allow `Sequence[nn.Module]`

  #127773 commented on Jun 22, 2024 • 0 new comments

• [Dynamo] einsum `ConstantVariable(str: 'i').has_unpack_var_sequence(tx)` returns True

  #121551 commented on Jun 17, 2024 • 0 new comments

• [experiment] batch files

  #127787 commented on Jun 17, 2024 • 0 new comments

• [Intel GPU] Dispatch Stub support

  #127860 commented on Jun 20, 2024 • 0 new comments

• torch.compile + ring attention

  #121386 commented on Jun 17, 2024 • 0 new comments

• Compiling lumiere-pytorch results in ~600 recompiles and cache size exceeded

  #121369 commented on Jun 17, 2024 • 0 new comments

• [AudioLM] Graph break: 'skip function zip_longest

  #121348 commented on Jun 17, 2024 • 0 new comments

• Adds support for accelerated sorting with x86-simd-sort

  #127936 commented on Jun 17, 2024 • 0 new comments

• [AudioLM] Graph break: call_method UserDefinedObjectVariable(dict) get [TorchVariable(<class 'torch.Tensor'>), ConstantVariable(NoneType)]

  #121345 commented on Jun 17, 2024 • 0 new comments

• 2 Dynamo test are failing with "Global state changed while dynamo tracing, please report a bug".

  #120648 commented on Jun 17, 2024 • 0 new comments

• [AudioLM] Graph break: call_method UserDefinedObjectVariable(_lru_cache_wrapper)

  #121344 commented on Jun 17, 2024 • 0 new comments

• [Traceable FSDP2] Top of Traceable FSDP2 stack

  #128103 commented on Jun 17, 2024 • 0 new comments

• [autograd] Do not detach when unpacking tensors that do not require grad

  #127959 commented on Jun 22, 2024 • 0 new comments

• Dynamo cannot work with non-classmethod torch_function implementation

  #120799 commented on Jun 17, 2024 • 0 new comments

• [AudioLM] Graph break: const method call float.is_integer

  #121334 commented on Jun 17, 2024 • 0 new comments

• Accuracy mismatch with torch.compile(backend="eager") for float16

  #121238 commented on Jun 17, 2024 • 0 new comments

• DO NOT MERGE: Test ALI runner

  #128024 commented on Jun 18, 2024 • 0 new comments

• Support sum() forward and backward for NJT

  #128031 commented on Jun 21, 2024 • 0 new comments

• fake_tensor.py: annotate types

  #128041 commented on Jun 23, 2024 • 0 new comments

• operator.eq(Tensor, non-tensor-scalar) not handled correctly

  #120907 commented on Jun 17, 2024 • 0 new comments

• dont prune unused symint graphargs from inner subclass tensors

  #128045 commented on Jun 19, 2024 • 0 new comments

• Umbrella issue for PyTorch test suite failures from torch.* returned non-Tensor output unimplemented

  #93479 commented on Jun 18, 2024 • 0 new comments

• [torch.compile] torch._dynamo.exc.TorchRuntimeError: Failed running call_function <method 'numpy' of 'torch._C.TensorBase' objects>(*(FakeTensor(..., size=(32, 3, 64, 64)),), **{})

  #124247 commented on Jun 17, 2024 • 0 new comments

• Support `dynamic=True` in torch._dynamo.explain

  #124163 commented on Jun 17, 2024 • 0 new comments

• Dynamo-based ONNX Export: Failed to produce a graph during tracing as no tensor operations were found.

  #123973 commented on Jun 17, 2024 • 0 new comments

• [Inductor] support masked vectorization for the tail_loop

  #126526 commented on Jun 17, 2024 • 0 new comments

• Use return_and_correct_aliasing() for NJT + compatible storage setting

  #126552 commented on Jun 18, 2024 • 0 new comments

• torch.compiler.disable doesn't disable nested functions (also doesn't work as a context manager)

  #123771 commented on Jun 17, 2024 • 0 new comments

• Dynamo unsupported: Dynamic slicing on data-dependent value is not supported

  #123592 commented on Jun 17, 2024 • 0 new comments

• torch.compile dynamo fails indexing into array from internal mutable state

  #123535 commented on Jun 17, 2024 • 0 new comments

• support setattr of arbitrary user provided types in tracing

  #93511 commented on Jun 18, 2024 • 0 new comments

• Add decomposition for upsample_bicubic2d_backward

  #126815 commented on Jun 18, 2024 • 0 new comments

• dynamo/fx doesn't honor 'non-persistent' buffers

  #123411 commented on Jun 17, 2024 • 0 new comments

• [dynamo] incorrect error traceback for runtime errors when executing dynamo codegen

  #123374 commented on Jun 17, 2024 • 0 new comments

• `@functools.wraps` graph breaks in many cases where we should be able to handle it

  #123365 commented on Jun 17, 2024 • 0 new comments

• [ONNX] Use ExportedProgram in dynamo_exporter 1/n

  #127096 commented on Jun 21, 2024 • 0 new comments

• FakeTensor support of pin_memory

  #123252 commented on Jun 17, 2024 • 0 new comments

• UFMT format on test_fake_tesnor.py test_futures.py test_fx.py

  #127369 commented on Jun 24, 2024 • 0 new comments

• [dynamo] Unsupported calling 'getattr' + 'getitem' on custom class

  #122649 commented on Jun 17, 2024 • 0 new comments

• Recompiles and cache_size_limit from detectron2 CycleBatchNormList

  #122578 commented on Jun 17, 2024 • 0 new comments

• AOTInductor Does Not Recompile when Saving at Same Path Even if Model Definition Changes

  #122487 commented on Jun 17, 2024 • 0 new comments

• [Intel GPU]Enable fp64 double GEMM

  #127508 commented on Jun 19, 2024 • 0 new comments

• [PT2][DTensor] crash during compiling 1D TP or SP on MLP models

  #122447 commented on Jun 17, 2024 • 0 new comments

• `torch.compile` should result in an optimized module where `module.training` is the same as in the unoptimized module

  #122414 commented on Jun 17, 2024 • 0 new comments

• WIP: fake tensor SymInt support

  #127596 commented on Jun 19, 2024 • 0 new comments

• torch.export: Unsupported: call_function args: UserDefinedObjectVariable(BatchEncoding) on Gemma

  #122340 commented on Jun 17, 2024 • 0 new comments

• Deprecate `torch._utils.is_compiling()` and `torch._dynamo.external_utils.is_compiling()`

  #127690 commented on Jun 18, 2024 • 0 new comments

• [BE] sort imports in `torch.utils.data`

  #127704 commented on Jun 19, 2024 • 0 new comments

• [BE] enable UFMT in `torch.utils.data`

  #127705 commented on Jun 19, 2024 • 0 new comments

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

  #120643 commented on Jun 17, 2024 • 0 new comments

• Do dynamic rollout for the pull workflow

  #128597 commented on Jun 17, 2024 • 0 new comments

• [cuDNN][cuDNN V8 API] cuDNN Flash-Attention Upstreaming RFC/tracking issue

  #113713 commented on Jun 17, 2024 • 0 new comments

• NotImplementedError: Operator aten.native_layer_norm_backward.default does not have a sharding strategy registered.

  #128699 commented on Jun 17, 2024 • 0 new comments

• Resolve circular dependence between `torch.autograd` and `torch.nn.parameter`

  #128633 commented on Jun 19, 2024 • 0 new comments

• [ONNX][dynamo_export] ONNX::Celu Half unsupported but export passed w/ invalid model when opmath disabled

  #113808 commented on Jun 20, 2024 • 0 new comments

• Enable UFMT on all files in PyTorch

  #123062 commented on Jun 17, 2024 • 0 new comments

• [ts migration] Support aten::tensor, prim::Enter, prim::Exit

  #128660 commented on Jun 17, 2024 • 0 new comments

• Refactor c10::DataPtr by subclassing from c10::detail::UniqueVoidPtr

  #128669 commented on Jun 21, 2024 • 0 new comments

• torch.save and torch.load is slow. Slower than numpy. Slower even than pickle.

  #124195 commented on Jun 17, 2024 • 0 new comments

• [RFC] Add new CPP builder for inductor on pytorch Windows

  #124245 commented on Jun 20, 2024 • 0 new comments

• put split memory block in the front of memory block set when stream and size equal.

  #128674 commented on Jun 17, 2024 • 0 new comments

• CUDA nightly docker actually includes CPU build of torch

  #125879 commented on Jun 17, 2024 • 0 new comments

• Pytorch nightly docker image invalidated layers

  #125862 commented on Jun 17, 2024 • 0 new comments

• xpu: a set of foreach ops not implemented for XPU backend affecting Huggingface examples

  #127931 commented on Jun 17, 2024 • 0 new comments

• Fix typo when using _check_tensor_list

  #128697 commented on Jun 17, 2024 • 0 new comments

• xpu: implement grid_sample op for XPU (fallback to CPU not possible for fp16 and bf16)

  #127002 commented on Jun 17, 2024 • 0 new comments

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

  #24422 commented on Jun 17, 2024 • 0 new comments

• add x/0 gradient behaviour to documentation

  #128796 commented on Jun 17, 2024 • 0 new comments

• Any plans for a "torch.minmax" (min-max normalization) function?

  #128785 commented on Jun 17, 2024 • 0 new comments

• Assign `torch.Generator` in APIs like `torch.randn_like()`

  #128786 commented on Jun 17, 2024 • 0 new comments

• Flaky test page should include retry runs

  #128735 commented on Jun 17, 2024 • 0 new comments

• The name of the function `nn.L1Loss()` should be `nn.MAE()` or the name of the function `MSELoss()` should be `nn.L2Loss()`

  #128779 commented on Jun 17, 2024 • 0 new comments

• Automatically bind all DispatchKey to Python

  #124083 commented on Jun 17, 2024 • 0 new comments

• [inductor][cpu]hf_BigBird AMP multiple thread static/dynamic shape default/CPP wrapper performance regression

  #128513 commented on Jun 19, 2024 • 0 new comments

• Build clang18 image for ASAN tests

  #128763 commented on Jun 17, 2024 • 0 new comments

• [Inductor] matmuls in `test_cuda_cpp_wrapper.py` appear broken on A16/A2

  #121562 commented on Jun 17, 2024 • 0 new comments

• xpu: gradient checkpointing wrongly hits cuda path running on non-cuda devices

  #128478 commented on Jun 17, 2024 • 0 new comments

• Add hash function of std::string_view to torch/csrc/lazy/core/hash.h

  #128800 commented on Jun 20, 2024 • 0 new comments

• Ban or change behavior of TensorVariable.size

  #120568 commented on Jun 17, 2024 • 0 new comments

• botorch dynamo errors

  #93633 commented on Jun 18, 2024 • 0 new comments

• Link with MKL::MKL instead of MKL_LIBRARIES

  #128195 commented on Jun 17, 2024 • 0 new comments

• Dynamo support dataclasses with default_factory=list

  #120108 commented on Jun 17, 2024 • 0 new comments

• Set seed per sample for OpInfo tests + support for restricting to a single sample input

  #128238 commented on Jun 22, 2024 • 0 new comments

• 6 Dynamo test are failing with "torch.utils.checkpoint: trying to save more tensors during recomputation than during the original forward pass.".

  #119794 commented on Jun 17, 2024 • 0 new comments

• 9 Dynamo test are failing with "Failed running call_function <function interpolate at 0xDEADBEEF".

  #119790 commented on Jun 17, 2024 • 0 new comments

• [1/N] Change #include <c10/util/Optional.h> to #include <optional>

  #128301 commented on Jun 18, 2024 • 0 new comments

• 10 Dynamo test are failing with "GetAttrVariable(NumpyVariable(), __name__) is not a constant".

  #119789 commented on Jun 17, 2024 • 0 new comments

• 17 Dynamo test are failing with "Failed running call_function <function embedding_bag at 0xDEADBEEF".

  #119786 commented on Jun 17, 2024 • 0 new comments

• [Don't merge] Try to restructure code

  #128330 commented on Jun 19, 2024 • 0 new comments

• add TORCH_FORCE_SYNCHRONOUS_COLLECTIVES to force functional collectives to be synchronous

  #128331 commented on Jun 18, 2024 • 0 new comments

• 63 Dynamo test are failing with "'QuantizationConfig' object has no attribute '__bool__'".

  #119782 commented on Jun 17, 2024 • 0 new comments

• 78 Dynamo test are failing with "somehow causing hanging during python shutdown".

  #119781 commented on Jun 17, 2024 • 0 new comments

• 110 Dynamo test are failing with "Failed running call_function <built-in method sparse_coo_tensor of type object at 0xDEADBEEF".

  #119780 commented on Jun 17, 2024 • 0 new comments

• [56+] Graph-break if we try to Fakeify an "unknown" Tensor with no data_ptr.

  #119695 commented on Jun 17, 2024 • 0 new comments

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

  #120219 commented on Jun 17, 2024 • 0 new comments

• Add warpSize to Device properties

  #128449 commented on Jun 19, 2024 • 0 new comments

• Deprecate unsupported types in operator registration

  #124863 commented on Jun 20, 2024 • 0 new comments

• [pipelining] lazy shape inference for manual

  #128527 commented on Jun 17, 2024 • 0 new comments

• Report sizes/strides of input argument that raised an error

  #119396 commented on Jun 17, 2024 • 0 new comments

• Preserve storage size when generating functional tensor

  #128546 commented on Jun 17, 2024 • 0 new comments

• [traced-graph][sparse] propagate compressed sparsity (WIP)

  #128549 commented on Jun 18, 2024 • 0 new comments

• Dynamo does not support user-defined objects that define custom __new__

  #119203 commented on Jun 17, 2024 • 0 new comments

• Fast path detach()/alias() in FakeTensor

  #128281 commented on Jun 20, 2024 • 0 new comments

• dynamo graph breaks on DTensor.to_local(grad_placements=grad_placements)

  #119023 commented on Jun 17, 2024 • 0 new comments

• [hierarchical compilation] A way to designate a portion of torch.compile as a noinline block that is compiled/guarded separately, but less disruptive than a graph break (e.g., for loops)

  #118966 commented on Jun 17, 2024 • 0 new comments

• vmap fails to call torch.compiled function

  #128711 commented on Jun 17, 2024 • 0 new comments

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

  #113257 commented on Jun 22, 2024 • 0 new comments

• [dynamo] we do not instantiate guards for ambient autocast mode

  #112260 commented on Jun 18, 2024 • 0 new comments

• Dynamo Compile samples should record file/line that raised exception

  #111674 commented on Jun 18, 2024 • 0 new comments

• [pytree] traverse `dict` in sorted key ordering

  #114947 commented on Jun 22, 2024 • 0 new comments

• [POC][pytree] test flattening dict in sorted order

  #115014 commented on Jun 22, 2024 • 0 new comments

• Automated submodule update: FBGEMM

  #115316 commented on Jun 24, 2024 • 0 new comments

• Custom `ModuleDict.__getitem__(key: tuple)` produces a graph break

  #111551 commented on Jun 18, 2024 • 0 new comments

• [pytree] update treespec dict keys access

  #116372 commented on Jun 22, 2024 • 0 new comments

• torch.compile support for SeamlessExpressivity/SeamlessM4T in fairseq2

  #114373 commented on Jun 18, 2024 • 0 new comments

• [pytree] make `context` and `children_specs` as private implementation details

  #116375 commented on Jun 22, 2024 • 0 new comments

• [PT2] Compile Cold Start - Async JIT compile with Eager fallback

  #114346 commented on Jun 18, 2024 • 0 new comments

• [1/N] Elimates c10::to_string and other STL string workarounds

  #116571 commented on Jun 22, 2024 • 0 new comments

• torch._dynamo.exc.Unsupported: call_method UserDefinedObjectVariable(FrozenDict) __contains__ [ConstantVariable(str)] {}

  #114202 commented on Jun 18, 2024 • 0 new comments

• Unexpected `None` value for stream with dynamo

  #114105 commented on Jun 18, 2024 • 0 new comments

• WIP Add 3D channels last tensor iterator support

  #118377 commented on Jun 23, 2024 • 0 new comments

• [MPS] Add SDPA implentation

  #119200 commented on Jun 19, 2024 • 0 new comments

• [ONNX] stft export fails with dynamo_export

  #113067 commented on Jun 21, 2024 • 0 new comments

• Implement Variable Tracker for Dataclasses

  #113670 commented on Jun 18, 2024 • 0 new comments

• [FSDP2] Eager-Mode Execution Tracker

  #120003 commented on Jun 21, 2024 • 0 new comments

• [dynamo,torch_function] __torch_function__ does not respect kwargs

  #117971 commented on Jun 18, 2024 • 0 new comments

• [dynamo] Assigning result of Tensor in-place op destroys mutation tracking

  #113271 commented on Jun 18, 2024 • 0 new comments

• (WIP) to_padded_tensor() triton kernel for NJT

  #121947 commented on Jun 18, 2024 • 0 new comments

• [draft] python 3.13 test

  #121979 commented on Jun 21, 2024 • 0 new comments

• [dynamo] self-assigning operation causes `TensorVariable` to lose `mutable_local`, thus causing its attribute mutations to be untracked

  #113160 commented on Jun 18, 2024 • 0 new comments

• torch._dynamo.exc.InternalTorchDynamoError: DeviceMeshVariable() has no type

  #117042 commented on Jun 18, 2024 • 0 new comments

• torch._dynamo.exc.InternalTorchDynamoError: ListIteratorVariable() has no type

  #117026 commented on Jun 18, 2024 • 0 new comments

• Decompositions for upsample linear backward

  #123222 commented on Jun 18, 2024 • 0 new comments

• torch.compile fullgraph=True is failing for GPTJ model for toy_backend

  #116835 commented on Jun 18, 2024 • 0 new comments

• [Dynamo][DeepSpeed] torch._dynamo.exc.InternalTorchDynamoError: NestedUserFunctionVariable() has no type

  #116766 commented on Jun 18, 2024 • 0 new comments

• [dtensor][compile] assertion on placements causing trouble with torch.compile

  #116712 commented on Jun 18, 2024 • 0 new comments

• [ONNX] None as input to `aten::index_put` unsupported

  #119363 commented on Jun 21, 2024 • 0 new comments

• [ONNX] Support Fake Tensor Mode on new Dynamo based ONNX exporter

  #105464 commented on Jun 21, 2024 • 0 new comments

• [dynamo] Diffusers - Graph break on OrderedDict

  #102878 commented on Jun 18, 2024 • 0 new comments

• ONNX export fails for aten::full_like op when exporting UDOP model from transformers

  #122898 commented on Jun 21, 2024 • 0 new comments

• [ONNX] export() with dynamic shapes fails where dynamo_export(dynamic_shapes=True) succeeds

  #126607 commented on Jun 21, 2024 • 0 new comments

• [ONNX] beartype discovers previously undiscovered type annotation errors

  #123203 commented on Jun 21, 2024 • 0 new comments

• Dynamo should only unroll loops by a preset factor (unless otherwise explicitly instructed)

  #102839 commented on Jun 18, 2024 • 0 new comments

• [inlined-inbuilt-nn-modules][dynamo][BE] Revisit call_method of NNModuleVariable

  #102063 commented on Jun 18, 2024 • 0 new comments

• [Dynamo] Can't inline functions under torch.nn.parallel

  #101609 commented on Jun 18, 2024 • 0 new comments

• [Dynamo] TB hf_Reformer graph breaks

  #101154 commented on Jun 18, 2024 • 0 new comments

• Stop importing HuggingFace transformers in DataClassVariable

  #100386 commented on Jun 18, 2024 • 0 new comments

• [dynamo] Investigate interop issues with torch_scatter/torch_sparse/pyg_lib

  #111223 commented on Jun 18, 2024 • 0 new comments

• [dynamo] Add asserts to prevent user defined objects/classes from going into ConstantVariable

  #110871 commented on Jun 18, 2024 • 0 new comments

• torch._dynamo.exc.Unsupported: unexpected sourceless type bases: (<class 'torchrec.streamable.Pipelineable'>,)

  #110315 commented on Jun 18, 2024 • 0 new comments

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

  #109895 commented on Jun 18, 2024 • 0 new comments

• [DDP + Dynamo] Tracing DDP AllReduce (Compiled DDP)

  #109774 commented on Jun 18, 2024 • 0 new comments

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

  #109504 commented on Jun 18, 2024 • 0 new comments

• Support the `ExitStack` context manager (or a simplified version)

  #109309 commented on Jun 18, 2024 • 0 new comments

• 'make html' will print 'duplicate object description' warnings when there are 1~5 CPUs in the running machine

  #128495 commented on Jun 22, 2024 • 0 new comments

• [RFC] Add third-party malloc library to improve pytorch memory performance on Windows

  #102534 commented on Jun 22, 2024 • 0 new comments

• Dynamo Swallowing Exception In Lambda

  #108798 commented on Jun 18, 2024 • 0 new comments

• ONNX Export - miscompilation for complex-valued operators

  #113444 commented on Jun 21, 2024 • 0 new comments

• __torch_dispatch__ + compile: extra guards

  #114405 commented on Jun 18, 2024 • 0 new comments

• `torch.cuda.is_bf16_compatible()` output inconsistent with with TorchInductor support

  #118122 commented on Jun 18, 2024 • 0 new comments

• [dynamo] Implement enumerate fallback as polyfill

  #112794 commented on Jun 18, 2024 • 0 new comments

• torch._dynamo.export raises Unexpected type in sourceless builder <class 'nemo.core.neural_types.elements.VoidType'> for torchaudio model

  #112745 commented on Jun 18, 2024 • 0 new comments

• [dynamo] Implement iter fallback (and possibly all iters/generators) as polyfill

  #112727 commented on Jun 18, 2024 • 0 new comments

• [Tracking] Follow ups for itertools infinite iterators

  #112532 commented on Jun 18, 2024 • 0 new comments

• [test/dynamo] BE: cleanup `test_misc.py`

  #112344 commented on Jun 18, 2024 • 0 new comments

• `set` of enums produces a graph break (no repro)

  #112338 commented on Jun 18, 2024 • 0 new comments

• Automated submodule update: kineto

  #106149 commented on Jun 21, 2024 • 0 new comments

• 'FakeRootModule' object has no attribute 'self___aot_engines_0_short_term_memories_list_0_0_0'

  #128251 commented on Jun 18, 2024 • 0 new comments

• torch.compile Jamba: long compilation time with backend="eager"

  #128153 commented on Jun 18, 2024 • 0 new comments

• torch.compile with Custom tensor subclass doesn't inline the tensor subclass methods

  #128149 commented on Jun 18, 2024 • 0 new comments

• [inductor] Graph breaks in CohereForAI/aya-23-8b

  #128095 commented on Jun 18, 2024 • 0 new comments

• Verify that guards are well formed before concluding that Dynamo complication has succeeded

  #128090 commented on Jun 18, 2024 • 0 new comments

• [User Empathy Day 2] non-deterministic recompiles for ChatTTS model

  #128074 commented on Jun 18, 2024 • 0 new comments

• Map with multiple arguments not supported in Dynamo and causes graph breaks

  #128072 commented on Jun 18, 2024 • 0 new comments

• [user empathy day 2][based] torch.compile issues

  #128071 commented on Jun 18, 2024 • 0 new comments

• Dynamo Graph break in Unsupported: call_method ConstDictVariable()

  #128067 commented on Jun 18, 2024 • 0 new comments

• Add MaskedTensor passthrough: unfold, F.Unfold, F.Fold, stack

  #125262 commented on Jun 17, 2024 • 0 new comments

• Invalidate StorageImpl instances when tensor is overwritten with cudagraphs

  #125264 commented on Jun 18, 2024 • 0 new comments

• [dynamo] DAC: 'AudioSignal' object has no attribute 'sample_rate'

  #128065 commented on Jun 18, 2024 • 0 new comments

• [Dynamo] torch.cuda.device context manager doesn't work

  #128059 commented on Jun 18, 2024 • 0 new comments

• Add line number to ` _warn_capture_scalar_outputs():`

  #127667 commented on Jun 18, 2024 • 0 new comments

• [FX] Refactor immutable collections implementation

  #125470 commented on Jun 22, 2024 • 0 new comments

• [inline-inbuilt-nn-modules] tensordict functional calls with nn.Module silently gives the wrong (non-functional) result

  #127173 commented on Jun 18, 2024 • 0 new comments

• Requesting dynamo support for fraction.Fraction

  #126917 commented on Jun 18, 2024 • 0 new comments

• [dynamo] Handle inplace op aliasing errors

  #126474 commented on Jun 18, 2024 • 0 new comments

• [AOTI][not for review] Test cpp_wrapper mode

  #125733 commented on Jun 20, 2024 • 0 new comments

• torch.export 'inline in skipfiles: Signature.bind | bind /usr/lib/python3.10/inspect.py, skipped according trace_rules.lookup SKIP_DIRS'

  #126242 commented on Jun 18, 2024 • 0 new comments

• [inline-inbuilt-nn-modules] dynamo recompiles identical layers when they have (identical) hooks

  #125836 commented on Jun 18, 2024 • 0 new comments

• [Dynamo] Support tracing through _get_current_dispatch_mode_stack

  #125694 commented on Jun 18, 2024 • 0 new comments

• [inductor] Enable FX graph caching in OSS by default

  #125863 commented on Jun 21, 2024 • 0 new comments

• [autograd.Function] freevar lifting is too aggressive?

  #106894 commented on Jun 18, 2024 • 0 new comments

• torch._dynamo.allow_in_graph seems to silently no-op on staticmethods

  #124735 commented on Jun 18, 2024 • 0 new comments

• NJT <-> padded dense conversions

  #125947 commented on Jun 19, 2024 • 0 new comments

• inductor creates unnecessary buffers

  #124653 commented on Jun 18, 2024 • 0 new comments

• [dynamo][inlining-inbuilt-nn-modules] decide how dynamo/export should handle parametrizations

  #124524 commented on Jun 18, 2024 • 0 new comments

• Dynamo handling for all methods of torch.Generator

  #88576 commented on Jun 18, 2024 • 0 new comments

• Enable UFMT format on test/quantization

  #126152 commented on Jun 24, 2024 • 0 new comments

• [wip][inductor] move loop ordering after fusion

  #126254 commented on Jun 18, 2024 • 0 new comments

• Dynamo fails to track dataclass

  #116264 commented on Jun 18, 2024 • 0 new comments

• torch.compile(fullgraph=True): can't pass lambdas to hooks?

  #116220 commented on Jun 18, 2024 • 0 new comments

• Add auto-tuning for sparse semi-structured MM operator

  #123742 commented on Jun 23, 2024 • 0 new comments

• [Dynamo] bytecode transformed by Dynamo is not serializable by marshal

  #116013 commented on Jun 18, 2024 • 0 new comments

• torch._dynamo.exc.Unsupported: SETUP_WITH UserDefinedObjectVariable(TorchAutocast)

  #115520 commented on Jun 18, 2024 • 0 new comments

• torch.compile() breaks when using DeepSpeed ZeRO Level 3 sharding

  #115484 commented on Jun 18, 2024 • 0 new comments

• [Tracker] Move nested tensors to beta

  #112398 commented on Jun 18, 2024 • 0 new comments

• [dynamo] Format string with __class__

  #118675 commented on Jun 18, 2024 • 0 new comments

• [dynamo][recompilation] test_set_get_descriptor

  #118563 commented on Jun 18, 2024 • 0 new comments

• Dynamo x autograd.Function: graph breaks on all the staticmethods on autograd.Function

  #118397 commented on Jun 18, 2024 • 0 new comments

• Make it more obvious when Dynamo is triggering on unexpected frames

  #118262 commented on Jun 18, 2024 • 0 new comments

• TorchDynamo mistranslates end of tensor slice

  #118227 commented on Jun 18, 2024 • 0 new comments

• streams x torch.compile: stream is treated as None sometimes

  #118204 commented on Jun 18, 2024 • 0 new comments

• Dynamo CI Shard naming proposal

  #118127 commented on Jun 18, 2024 • 0 new comments

• Dynamo: assert "source" in options and options["source"] is not None for default_generator.set_state call

  #118072 commented on Jun 18, 2024 • 0 new comments

• Supporting custom attributes with `__torch_function__` tensor subclasses

  #117806 commented on Jun 18, 2024 • 0 new comments

• Can't call allow_in_graph inside of a function being torch.compile'd

  #103615 commented on Jun 18, 2024 • 0 new comments

• Pt2 - Discussion around user defined type->behavior dispatching

  #117321 commented on Jun 18, 2024 • 0 new comments

• [dynamo][inline-inbuilt-nn-modules]torch.compile silently incorrect with full_backward_pre_hook

  #117265 commented on Jun 18, 2024 • 0 new comments

• [dynamo] AssertionError for custom iterable nn.Module

  #103831 commented on Jun 18, 2024 • 0 new comments

• xpu: set of unimplemented ops affect huggingface examples performance

  #127941 commented on Jun 18, 2024 • 0 new comments

• tts_angular: fail_to_run, torch._dynamo.exc.Unsupported: call_method NNModuleVariable() flatten_parameters [] {}

  #105532 commented on Jun 18, 2024 • 0 new comments

• [dynamo] calling __torch_function__ with dynamically created subclass of torch.Tensor fails compilation

  #107143 commented on Jun 18, 2024 • 0 new comments

• Extend dict and by extension __dict__ modeling in dynamo to support `setdefault`, `get`

  #107054 commented on Jun 18, 2024 • 0 new comments

• AssertionError: <class 'torch._dynamo.variables.torch.TorchInGraphFunctionVariable'> when compiling `torch.nn.functional.layer_norm`

  #128797 commented on Jun 18, 2024 • 0 new comments

• Dynamo: contextlib.contextmanager doesn't work

  #128651 commented on Jun 18, 2024 • 0 new comments

• Extend Dynamo support for arbitrary context managers

  #128650 commented on Jun 18, 2024 • 0 new comments

• Migrate linux-jammy-py3-clang12-mobile-build to ARC

  #124605 commented on Jun 17, 2024 • 0 new comments

• Migrate linux-jammy-cuda-11_8-cudnn8-py3_8-clang12-build to ARC

  #124606 commented on Jun 17, 2024 • 0 new comments

• torch._dynamo.exc.Unsupported: call_method GetAttrVariable(UnspecializedNNModuleVariable(CenterCrop), _transformed_types) __iter__ () {}

  #128417 commented on Jun 18, 2024 • 0 new comments

• [dynamo] Recompilation on a counter-like attribute of nn module

  #128319 commented on Jun 18, 2024 • 0 new comments