pytorch / pytorch Public

7 Pull requests merged by 4 people

• Merge master into lazy_tensor_staging

  #74332 merged Mar 17, 2022

• Update lazy_bench.py to adapt to changes in TorchBench

  #74298 merged Mar 16, 2022

• [GHA][BE] Remove unneeded shellcheck suppressions

  #74308 merged Mar 16, 2022

• Add support for c10::Generator to lazy codegen and move bernoulli/random ops

  #73939 merged Mar 16, 2022

• [GHA][BE] Delete unused definitions

  #74270 merged Mar 16, 2022

• move shape to Node base

  #74071 merged Mar 15, 2022

• [LT] Skip tacotron2 and demucs in lazy_bench.py

  #74183 merged Mar 14, 2022

109 Pull requests opened by 65 people

• Testing builder changes

  #74141 opened Mar 12, 2022

• [quant][ns] Fix ns tool bug for mobilenetv2/v3

  #74149 opened Mar 13, 2022

• use gmock 1.10 instead of 1.8

  #74150 opened Mar 13, 2022

• Patch batch_norm/batch_norm_backward for perf potential measurement

  #74156 opened Mar 14, 2022

• set workspace size to 4M

  #74159 opened Mar 14, 2022

• [jiterator] log, log10 : complex

  #74160 opened Mar 14, 2022

• [jiterator] log2 : complex

  #74161 opened Mar 14, 2022

• Add type hints for a few random functions/classes

  #74171 opened Mar 14, 2022

• Implement C++-side basics of "dynamic shapes" API

  #74175 opened Mar 14, 2022

• Implement Python-side basics of "dynamic shapes" API

  #74176 opened Mar 14, 2022

• [JIT] Add autocasting to freezing pass & enable autocast pass by default

  #74178 opened Mar 14, 2022

• Add api for changing function executor settings, hook up execution with decomposition registry

  #74186 opened Mar 14, 2022

• [GHF] Add `Lint` to the list of mandatory checks

  #74193 opened Mar 14, 2022

• Add typing for torch.return_type

  #74199 opened Mar 14, 2022

• Add C++ implementation of histogramdd

  #74200 opened Mar 14, 2022

• Remove histogramdd functional wrapper

  #74201 opened Mar 14, 2022

• Restore TestTorchFunctionOverride

  #74202 opened Mar 14, 2022

• Add forward AD support for norm and related ops

  #74205 opened Mar 14, 2022

• Move to_sparse_csr into C++

  #74206 opened Mar 14, 2022

• Extend _save_for_mobile and _load_for_mobile to support flatbuffer format; Default format is pickle

  #74209 opened Mar 14, 2022

• [Easy][FSDP] Minor doc fixes

  #74214 opened Mar 15, 2022

• [FSDP] Add full optim state dict

  #74215 opened Mar 15, 2022

• docs: expose at::native::unfold

  #74224 opened Mar 15, 2022

• Adding gpuAtomicMin and gpuAtomicMax for floating types

  #74225 opened Mar 15, 2022

• Use ScatterGatherKernel for scatter_reduce

  #74226 opened Mar 15, 2022

• Adding gpuAtomicMin/Max/Mul support for integer types

  #74230 opened Mar 15, 2022

• Make addcmul and addcdiv support different dtypes

  #74234 opened Mar 15, 2022

• [WIP] fix: no duplicate final LayerNorm in Transformer

  #74237 opened Mar 15, 2022

• [Profiler] Store Input shapes, dtypes, and metadata into flat AppendOnlyList

  #74241 opened Mar 15, 2022

• Process inputs and outputs in fx interpreter

  #74242 opened Mar 15, 2022

• [ONNX] Adjust `is_train` flag for onnx pass deduplicate initializers

  #74247 opened Mar 15, 2022

• [SR] Avoid boxing inputs in DictConstruct/ListUnpack

  #74250 opened Mar 15, 2022

• Print reason for test skipped in CI

  #74254 opened Mar 15, 2022

• [Static Runtime] Fix a bug that `aten::full_like` reuses a tensor that does not match arguments

  #74255 opened Mar 15, 2022

• Use ONNX Runtime 1.11 in CI

  #74257 opened Mar 15, 2022

• [CUDACachingAlloc/GPUInference] Implement garbage collection without GPU sync

  #74261 opened Mar 15, 2022

• Use COO mul kernel for CSR

  #74266 opened Mar 15, 2022

• Parallelize `gatherTopK` on multiple blocks

  #74267 opened Mar 15, 2022

• Specify "Generic" BLAS library name.

  #74269 opened Mar 16, 2022

• [SR] Force split_and_squeeze usage via graph transformation

  #74274 opened Mar 16, 2022

• Fixes build of CUDAHooks.cpp

  #74275 opened Mar 16, 2022

• [AO][bugfix] Fixing FX QAT but for untraceable modules

  #74277 opened Mar 16, 2022

• Extends OpInfo architecture with reference inputs, adds them for elementwise binary operators

  #74280 opened Mar 16, 2022

• [torch::deploy] Replace ArrayRef with local implementation to torch::deploy

  #74284 opened Mar 16, 2022

• [torch::deploy] remove c10::irange from torch deploy

  #74285 opened Mar 16, 2022

• [follow-up] Update dispatch macros

  #74289 opened Mar 16, 2022

• add module infos for maxpool1d/3d

  #74292 opened Mar 16, 2022

• Move to_sparse_csr to C++

  #74294 opened Mar 16, 2022

• Workflow consolidation for GitHub actions

  #74306 opened Mar 16, 2022

• Undo progress bar off so we see what versions we install

  #74312 opened Mar 16, 2022

• Make test case creation code more compatible with meta tensors.

  #74313 opened Mar 16, 2022

• Add meta tensor support for squeeze

  #74314 opened Mar 16, 2022

• Make bitwise_left_shift/right_shift scalar variants composite

  #74317 opened Mar 16, 2022

• [FSDP] Override `named_parameters()` for clean names in `summon_full_params()`

  #74333 opened Mar 16, 2022

• [TEST] Asan shard fix

  #74335 opened Mar 16, 2022

• Dummy commit for CI testing

  #74336 opened Mar 16, 2022

• supporting complex with requires_grad in autodiff

  #74339 opened Mar 16, 2022

• Initial work on SSA without JIT Graphs

  #74340 opened Mar 16, 2022

• Automated submodule update: kineto

  #74343 opened Mar 16, 2022

• [nnc] Fuse loops that have variable bounds

  #74346 opened Mar 16, 2022

• temp

  #74348 opened Mar 16, 2022

• [ci] Pull some stragglers into the consolidated workflows

  #74351 opened Mar 16, 2022

• Reland "[pytorch][PR] Support dataclasses in TorchScript"

  #74353 opened Mar 17, 2022

• [Static Runtime] Codegen support for ops with float args

  #74356 opened Mar 17, 2022

• Throw python_error if the call returns nullptr.

  #74357 opened Mar 17, 2022

• disable contiguity on cross dimensional overlapped tensor

  #74359 opened Mar 17, 2022

• [WIP] Added proxy tensor

  #74360 opened Mar 17, 2022

• [WIP][JIT] Optionally validate nvfuser outputs after execution

  #74361 opened Mar 17, 2022

• [Quant][fx] Refactor lowering code (part 2)

  #74362 opened Mar 17, 2022

• [quant] Remove assert for weight since it could be non-Tensor

  #74365 opened Mar 17, 2022

• Remove default parameter of ShufflerIterDataPipe

  #74370 opened Mar 17, 2022

• Link missing FSDP functions/classing in documentation

  #74371 opened Mar 17, 2022

• Passing explicit pretrained_backbone

  #74372 opened Mar 17, 2022

• [TEST] S3 stats sharding

  #74373 opened Mar 17, 2022

• Make test_foreach account for NotImplementedError in ref tests

  #74374 opened Mar 17, 2022

• Make test_foreach account for NotImplementedError in ref tests

  #74375 opened Mar 17, 2022

• Fix formatting of scalar tensors (don't call item)

  #74376 opened Mar 17, 2022

• [pytorch][PR] Added missing c10 namespace for C++ API

  #74377 opened Mar 17, 2022

• Update RELEASE.md with steps to prepare before cutting RC

  #74378 opened Mar 17, 2022

• [ROCm] enable HIP IPC

  #74383 opened Mar 17, 2022

• [Static Runtime] Add profile metric for ops directly supported from Static Runtime ((out variant count + native op count) / total op count)

  #74385 opened Mar 17, 2022

• move //tools/codegen:codegen into shared build structure

  #74386 opened Mar 17, 2022

• [4/5]Testing jit module in flatbuffer in Python.

  #74387 opened Mar 17, 2022

• Adds missing AT_CUDA_CHECK in CUDAGraph.cpp

  #74392 opened Mar 17, 2022

• Fix cusparse sync issue in bsrsv2 and bsrsm2 [v2]

  #74394 opened Mar 17, 2022

• [PyTorch] Make NestedTensorImpl::get_nested_size_tensor() const

  #74395 opened Mar 17, 2022

• Add list of supported ATen ops by ONNX converter into torch.onnx page

  #74397 opened Mar 18, 2022

• [GHF] Add remaining PyTorch Distributed Devs to GHF merge rules

  #74399 opened Mar 18, 2022

• Add Fujitsu SSL2 build support

  #74401 opened Mar 18, 2022

• [wip] try c10_mobile

  #74402 opened Mar 18, 2022

• [DO NOT REVIEW] draft showing code diff

  #74403 opened Mar 18, 2022

• initial test stats work

  #74406 opened Mar 18, 2022

• [quant][fx] Cleanup quantization_patterns.py

  #74407 opened Mar 18, 2022

• [quant][fx] Remove unused method from QuantizeHandler

  #74408 opened Mar 18, 2022

• add BFloat16 support for BatchNorm on CPU

  #74410 opened Mar 18, 2022

• tracer bullet to see if/which OSS CI jobs run test_lazy

  #74413 opened Mar 18, 2022

• [Doc] Clarify the functionality of forward_hook.

  #74414 opened Mar 18, 2022

• [Test] Test refactored ONNX shape inference with PyTorch ONNX exporter

  #74416 opened Mar 18, 2022

• [ROCm] enable foreach fastpath

  #74417 opened Mar 18, 2022

• Torchhub: automatically disambiguate ref if it's both a branch and a tag

  #74418 opened Mar 18, 2022

• Create jiterator cache dirs recursively

  #74425 opened Mar 19, 2022

• Add setup and teardown logic to torchhub tests. Also cleaned up uses of temp folders

  #74427 opened Mar 19, 2022

• Remove deprecated import_module hub function

  #74428 opened Mar 19, 2022

• Remove private _download_url_to_file

  #74429 opened Mar 19, 2022

• Properly catch warning in torchhub tests

  #74430 opened Mar 19, 2022

• Cosmetic changes to torchhub tests

  #74431 opened Mar 19, 2022

• [CI] Fix Android final artifact build step

  #74432 opened Mar 19, 2022

• Enable faster cuBLAS path for torch.linalg.lstsq for batch of small matrices

  #74434 opened Mar 19, 2022

• [TEST/Not for merge] demonstrate embedding failing

  #74435 opened Mar 19, 2022

45 Issues closed by 26 people

• bad_alloc after np.exp when creating a tensor

  #73356 closed Mar 19, 2022

• nn.Module.parameters() returns nothing with nn.DataParallel

  #74426 closed Mar 19, 2022

• [feature request] torch.left_shift, torch.right_shift (including inplace and out-variant) in addition to existing python magics

  #41624 closed Mar 18, 2022

• init_process_group constructor hangs when using gloo

  #74316 closed Mar 17, 2022

• GitHub Outage: No Github Actions workflows can be run

  #74304 closed Mar 17, 2022

• python setup.py develop fails

  #74352 closed Mar 17, 2022

• PyTorch 1.3: random "RuntimeError: CUDA error: unspecified launch failure"

  #27837 closed Mar 17, 2022

• TorchBench V1 performance signal on 20211106

  #68086 closed Mar 17, 2022

• How to Minimize Rounding Error in torch.autograd.functional.jacobian?

  #74288 closed Mar 17, 2022

• intermittent SIGSEGV segfault when plotting during training_step

  #74188 closed Mar 17, 2022

• Problem with beta and threshold parameters of nn.Softplus()

  #74366 closed Mar 17, 2022

• `torch.amax` and `torch.amin` returns random values, if `dim` arg is not provided

  #73905 closed Mar 16, 2022

• Lazy Tensor Core - Hardcoded `TsNode` instances

  #74037 closed Mar 16, 2022

• Investigate and update DDP tutorials

  #74246 closed Mar 16, 2022

• [proposal] Fuse mm+add for inputs > 2D

  #39661 closed Mar 16, 2022

• TestZeroRedundancyOptimizerDistributed.test_multiple_groups is failing intermittently

  #66059 closed Mar 16, 2022

• torch.linalg.solve updates are causing GPyTorch tests to fail

  #71406 closed Mar 15, 2022

• `torch.linalg.solve` INTERNAL ASSERT FAILED

  #71610 closed Mar 15, 2022

• AttributeError: 'NoneType' object has no attribute '_free_weak_ref'

  #74232 closed Mar 15, 2022

• Workflow changes can break PR CI

  #74249 closed Mar 15, 2022

• Re-order tests based on files in the PR

  #56557 closed Mar 15, 2022

• onnx export failed when output size are not factor of input size for adaptive_avg_pool2d

  #64806 closed Mar 15, 2022

• Lazy Tensor Core - Autograd mechanics not working

  #74196 closed Mar 15, 2022

• Tests should be disabled in forks when they are expected to fail

  #72661 closed Mar 15, 2022

• Check mismatched number of parameters in DDP _verify_params_across_processes

  #73547 closed Mar 15, 2022

• Couldn't convert pytorch model to ONNX

  #74144 closed Mar 15, 2022

• `torch.kaiser_window` fails for meta tensors and `window_length >= 2`

  #72630 closed Mar 15, 2022

• Test `test_reduce_add_coalesced` failed

  #74015 closed Mar 15, 2022

• Libtorch dosen't has the function like python api (TORCH.SET_PRINTOPTIONS)

  #73911 closed Mar 15, 2022

• PyTorch-nightly can not me imported using python-3.7.0

  #74087 closed Mar 15, 2022

• OOM: tacotron2, densenet121, timm_nfnet, demucs

  #74182 closed Mar 14, 2022

• android prebuilt workflow builds wrong binary (I think)

  #74073 closed Mar 14, 2022

• I get an error when exporting to onnx but cannot find what it means

  #73129 closed Mar 14, 2022

• Typesetting in torch.nn.MultiheadAttention

  #74147 closed Mar 14, 2022

• torch.quantization.fuse_modules is not backward compatible in PyTorch 1.11

  #74028 closed Mar 14, 2022

• libtorch model predict cuda convert to cpu: C10::error at memory location

  #73912 closed Mar 14, 2022

• Does the Transformer need two layernorm layers?

  #51447 closed Mar 14, 2022

• Default `nn.Transformer` applies layer normalization twice at last

  #50086 closed Mar 14, 2022

• Preserve core dump as artifacts

  #67710 closed Mar 14, 2022

• torch.det behaviour for large batch sizes on cuda

  #73622 closed Mar 14, 2022

• Any bug triggers device-side assert

  #74079 closed Mar 14, 2022

• ci: Check if docker daemon is running and start it if it isn't

  #73859 closed Mar 14, 2022

• torch.no_grad() removes requires_grad from weight tensor while updating it by non-compound-operation

  #74158 closed Mar 14, 2022

• Have a `torch.tensor_like` creation op?

  #74026 closed Mar 14, 2022

• Pre trained faster rcnn model not deterministic. Run time error thrown when using torch.use_deterministic_algorithms(True)

  #74097 closed Mar 14, 2022

77 Issues opened by 56 people

• Bad formatting of code in documentation

  #74433 opened Mar 19, 2022

• Tensor "step" in optimizers causing major performance regression in PyTorch XLA after loading optimizer state dict

  #74424 opened Mar 19, 2022

• Embedding Pytorch in C++ using pybind fails on interpreter shutdown

  #74422 opened Mar 19, 2022

• can not build pytorch, failing due to missing _ctypes module

  #74421 opened Mar 18, 2022

• `torch.histogram` to support integer types with integer count tables.

  #74420 opened Mar 18, 2022

• Copying tensors sometimes doesn't copy last N elements when using cuda graphs

  #74419 opened Mar 18, 2022

• Warning: "Specified kernel cache directory could not be created"

  #74415 opened Mar 18, 2022

• 404 Not Found in Running PyTorch Models on Jetson Nano

  #74412 opened Mar 18, 2022

• UserWarning: size_average and reduce args will be deprecated, please use reduction='mean' instead.

  #74411 opened Mar 18, 2022

• Android nightly builds failing

  #74405 opened Mar 18, 2022

• Module parameterlist attr script error

  #74404 opened Mar 18, 2022

• Type promotion semantics not handled properly for `__torch_dispatch__`

  #74400 opened Mar 18, 2022

• _run_ninja_build failing with clang

  #74398 opened Mar 18, 2022

• Quantized cannot inference with cuda

  #74393 opened Mar 17, 2022

• Can this cudaDeviceSynchronize call be removed?

  #74391 opened Mar 17, 2022

• DISABLED test_shuffle (__main__.TestDataLoader2)

  #74390 opened Mar 17, 2022

• torch.package unpickling transforms: ModuleNotFoundError: No module named 'torch._C._linalg'; 'torch._C' is not a package

  #74389 opened Mar 17, 2022

• Deprecation warnings don't help the user identify their bad code

  #74388 opened Mar 17, 2022

• Package and Deploy: Generate External Python Registration files at runtime [WIP]

  #74384 opened Mar 17, 2022

• `dim` argument not checked for any of the fixed dimensional `fft` functions

  #74380 opened Mar 17, 2022

• stack(): argument 'tensors' (position 1) must be tuple of Tensors, not Proxy

  #74367 opened Mar 17, 2022

• Efficient very large (e.g., 31x31) depth-wise convolution

  #74363 opened Mar 17, 2022

• lshift and rshift are supported for non-integral inputs and shifts

  #74358 opened Mar 17, 2022

• ci: Add lint to check for trailing whitespaces in filenames

  #74341 opened Mar 16, 2022

• None returned from data loader causes debugging difficulty from collate function

  #74337 opened Mar 16, 2022

• SystemError: <class 'UserWarning'> returned a result with an error set

  #74334 opened Mar 16, 2022

• CompositeImplicitAutograd addr implementation does not have correct behavior for bool (meta gives wrong dtype)

  #74329 opened Mar 16, 2022

• instantiate_device_type_tests is misnamed

  #74328 opened Mar 16, 2022

• Lint rule that would fail for PRs containing subdmodule updates

  #74326 opened Mar 16, 2022

• torch.combinations requires unnecessary CUDA sync

  #74324 opened Mar 16, 2022

• [c10d] `gather`/`scatter` inconsistency behavior on non-dst/src rank

  #74323 opened Mar 16, 2022

• Make OpInfo repr more useful

  #74320 opened Mar 16, 2022

• [BUG] replicate/reflect padding fails for 2D tensor

  #74310 opened Mar 16, 2022

• NotImplementedError Description

  #74303 opened Mar 16, 2022

• Training MLM model XLM Roberta large on google machine specs not fast

  #74302 opened Mar 16, 2022

• Error in lobpcg when using largest=False

  #74301 opened Mar 16, 2022

• repeat_interleave is not meta friendly; not transform friendly either

  #74300 opened Mar 16, 2022

• Gradients tests are very time consuming

  #74299 opened Mar 16, 2022

• Time Gated Lstm

  #74296 opened Mar 16, 2022

• TypeError: new_empty(): argument 'size' (position 1) must be tuple of ints, not list

  #74295 opened Mar 16, 2022

• Loading of packaged model object fails with torch==1.11.0

  #74290 opened Mar 16, 2022

• RuntimeError: cuDNN error: CUDNN_STATUS_VERSION_MISMATCH for torchvision models

  #74282 opened Mar 16, 2022

• Move torchbench workflow to `workflow_dispatch`?

  #74281 opened Mar 16, 2022

• pow, float_power, xlogy, xlog1py are all skipping TestGradients

  #74279 opened Mar 16, 2022

• [PyTorch] Lightweight dispatch

  #74272 opened Mar 16, 2022

• Can't download torch vision models for older versions via torch.hub.load: "multiple choices"

  #74368 opened Mar 15, 2022

• Create secure credential storage for metrics credentials and associated documentation on how to regenerate them if needed

  #74256 opened Mar 15, 2022

• dir() on torch.ops.aten doesn't work

  #74248 opened Mar 15, 2022

• FX QAT Untraceable Modules are still converted to qat versions if possible

  #74240 opened Mar 15, 2022

• Hugging face code with Pytorch throwing error when running using vertex-ai

  #74238 opened Mar 15, 2022

• "histogram_cpu" not implemented for 'Int'

  #74236 opened Mar 15, 2022

• RuntimeError: CUDA error: unspecified launch failure

  #74235 opened Mar 15, 2022

• DISABLED test_exception_all (__main__.SpawnTest)

  #74223 opened Mar 15, 2022

• Autograd API to get saved for backwards tensors for an autograd graph

  #74222 opened Mar 15, 2022

• `package.PackageExporter` does not actually appear to have a `file_structure` method

  #74221 opened Mar 15, 2022

• Cannot get op through FX when using nn.Sequential

  #74218 opened Mar 15, 2022

• DISABLED test_trace_buffer_slice (__main__.TestFX)

  #74208 opened Mar 14, 2022

• General policy for supported Python version

  #74203 opened Mar 14, 2022

• [docker] test_corrcoef_cpu_complex64 fails on CPU build

  #74187 opened Mar 14, 2022

• [DDP] Parallelize initialization collectives

  #74185 opened Mar 14, 2022

• DISABLED test_init_rpc_twice (__main__.TensorPipeRpcTest)

  #74180 opened Mar 14, 2022

• get_dropout_state, seed and reproducibility

  #74177 opened Mar 14, 2022

• Specify partial inverses using torch.linalg.inv

  #74168 opened Mar 14, 2022

• Slower performance of `torch.mm` method with sparse CSR tensor

  #74167 opened Mar 14, 2022

• [FSDP] Investigate summon_full_params with CPU offloading

  #74166 opened Mar 14, 2022

• [FSDP] Generalize `summon_full_params()` to non-FSDP root module

  #74165 opened Mar 14, 2022

• consume_prefix_in_state_dict_if_present can not remove prefix of _metadata.

  #74162 opened Mar 14, 2022

• post local sgd decreases accuracy

  #74157 opened Mar 14, 2022

• [Android Studio] DefaultCPUAllocator: not enough memory: you tried to allocate 280166432 bytes

  #74155 opened Mar 14, 2022

• DISABLED test_pdist_norm_large_cuda (__main__.TestTorchDeviceTypeCUDA)

  #74154 opened Mar 14, 2022

• [BE][Docs][FSDP] Clarify microbatching support and tradeoffs

  #74153 opened Mar 13, 2022

• All-caps names for pages are confusing

  #74152 opened Mar 13, 2022

• powersgd can not get linear growth due to extra 2 times allreduce

  #74148 opened Mar 13, 2022

• RuntimeError: Connection reset by peer when backened by NCCL

  #74146 opened Mar 13, 2022

• conv2d with large kernel uses all the memory on M1 mac

  #74145 opened Mar 13, 2022

• Tracing model parameter shapes without instantiating the model parameters

  #74143 opened Mar 13, 2022

• [Bug][ONNX] Specification Inconsistency in Flatten

  #74142 opened Mar 13, 2022

250 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.

• Adds cudaMallocAsync as an alternative backend for the CUDA allocator

  #65365 commented on Mar 18, 2022 • 32 new comments

• Land remaining parts of Torchscript Lazy Tensor backend

  #74111 commented on Mar 18, 2022 • 25 new comments

• [shard] Add ReplicatedTensor

  #73529 commented on Mar 17, 2022 • 20 new comments

• ASAN shards are very unbalanced

  #72368 commented on Mar 17, 2022 • 15 new comments

• Add non-eager registration to dispatch autogen

  #73973 commented on Mar 18, 2022 • 15 new comments

• [shard] Sharder and ShardingPlan prototype

  #73873 commented on Mar 15, 2022 • 13 new comments

• Fix ONNX ATen fallback for non-caffe2 engines

  #73954 commented on Mar 18, 2022 • 12 new comments

• grid_sampler_3d_backward aborts due to heap corruption

  #73187 commented on Mar 16, 2022 • 11 new comments

• DOC: Initialize numpy compatibility note

  #71688 commented on Mar 15, 2022 • 10 new comments

• [jiterator] reduce kernel code duplication

  #73908 commented on Mar 17, 2022 • 9 new comments

• BoxedInt take 2 [WIP]

  #73774 commented on Mar 16, 2022 • 8 new comments

• [ONNX] update default opset_version to 13

  #73898 commented on Mar 16, 2022 • 8 new comments

• Remove deprecated torch.qr

  #70989 commented on Mar 15, 2022 • 7 new comments

• remove need for torch in torch.package

  #72961 commented on Mar 16, 2022 • 7 new comments

• [WIP] Implement complex support in DDP

  #74039 commented on Mar 14, 2022 • 7 new comments

• [feature request] Jagged / padding version of torch.stack / torch.cat

  #65156 commented on Mar 14, 2022 • 6 new comments

• `torch.set_printoptions` overwrites settings of its own previous calls

  #74138 commented on Mar 14, 2022 • 6 new comments

• Changes to support input sequence ID tracking

  #70264 commented on Mar 18, 2022 • 6 new comments

• [Dynamic RPC] Allow for optional world_size argument in init_rpc

  #73372 commented on Mar 16, 2022 • 6 new comments

• expanded weights: embedding faster rule

  #73693 commented on Mar 17, 2022 • 6 new comments

• [ROCm] Update the magma commit

  #73906 commented on Mar 17, 2022 • 6 new comments

• [torch::deploy] replace c10::optional with generic optional of boost

  #74002 commented on Mar 15, 2022 • 6 new comments

• Support dataclasses in TorchScript

  #72901 commented on Mar 17, 2022 • 5 new comments

• AttributeError: 'NoneType' object has no attribute '_free_weak_ref'

  #74016 commented on Mar 17, 2022 • 5 new comments

• [shard] Add basic math ops to ShardedTensor and add ReplicatedTensor inter-op

  #73703 commented on Mar 17, 2022 • 5 new comments

• Running `tensorboard_profiler_tutorial.py` intermittently crashes with Pytorch-11.1

  #74139 commented on Mar 15, 2022 • 4 new comments

• Two consecutive nn.LayerNorm are used in transformer model when norm_first is False

  #74092 commented on Mar 15, 2022 • 4 new comments

• undefined symbol curandCreateGenerator for torch extensions

  #69666 commented on Mar 18, 2022 • 4 new comments

• Add BFloat16 support for multinomial and poisson on CPU

  #63215 commented on Mar 16, 2022 • 4 new comments

• Support masked sum on sparse COO tensors.

  #71239 commented on Mar 17, 2022 • 4 new comments

• Enable TE fuser to support user defined operator

  #73073 commented on Mar 19, 2022 • 4 new comments

• Add prelu op and module for quantized CPU backend

  #73491 commented on Mar 16, 2022 • 4 new comments

• [quant][fx] Enable input and output idx support dtype

  #73816 commented on Mar 16, 2022 • 4 new comments

• CrossEntropyLoss triggers floating point exception

  #73837 commented on Mar 18, 2022 • 4 new comments

• [JIT] Make aot autograd decompositions usable in JIT, add script for serializing the decompositions

  #73938 commented on Mar 17, 2022 • 4 new comments

• Introduce function-local settings for executor, expose in c++

  #74012 commented on Mar 17, 2022 • 4 new comments

• Add checks for public and private API

  #74051 commented on Mar 16, 2022 • 4 new comments

• Linear bias fusion for ND input

  #74070 commented on Mar 18, 2022 • 4 new comments

• ARM Mac 16-core Neural Engine

  #47688 commented on Mar 14, 2022 • 3 new comments

• Possible issue with memory allocation.

  #73883 commented on Mar 14, 2022 • 3 new comments

• `torch.{addcmul,addcdiv}` triggers INTERNAL ASSERT FAILED

  #70486 commented on Mar 15, 2022 • 3 new comments

• [torch.onnx] ONNX export failed on adaptive_avg_pool2d because input size not accessible not supported

  #74034 commented on Mar 15, 2022 • 3 new comments

• torch.nn.Module.__init__ does not call super().__init__

  #74036 commented on Mar 16, 2022 • 3 new comments

• [feature request] Add native dict support for nn.Sequential

  #49530 commented on Mar 16, 2022 • 3 new comments

• return_index option for torch.unique

  #36748 commented on Mar 18, 2022 • 3 new comments

• CUSOLVER_STATUS_EXECUTION_FAILED when using the torch.logdet()

  #70669 commented on Mar 18, 2022 • 3 new comments

• Make ProcessGroupNCCL load torch_ucc.so when TORCH_UCC_LIBRARY_PATH is set

  #69552 commented on Mar 18, 2022 • 3 new comments

• Add upgrader related logic to flatbuffer

  #71451 commented on Mar 18, 2022 • 3 new comments

• Moving Remove Tensor Type Specializations to after custom passes

  #71748 commented on Mar 14, 2022 • 3 new comments

• [shard] use scatter in shard_parameter API

  #72160 commented on Mar 17, 2022 • 3 new comments

• [pytorch profiler] enable iteation tracking for kineto

  #72292 commented on Mar 17, 2022 • 3 new comments

• Fix SequentialLR initialization

  #72856 commented on Mar 15, 2022 • 3 new comments

• [tensorexpr] Add support for aten::stack

  #73801 commented on Mar 16, 2022 • 3 new comments

• Clean up profiling mode and profiling executor strategy

  #73875 commented on Mar 17, 2022 • 3 new comments

• Remove bailout logic

  #73876 commented on Mar 17, 2022 • 3 new comments

• [WIP] Add AdaptiveAvgPool to test lightweight dispatch

  #73948 commented on Mar 17, 2022 • 3 new comments

• Inplace forward AD formula optimization

  #74017 commented on Mar 19, 2022 • 3 new comments

• Add forward AD support for clamp.Tensor

  #74042 commented on Mar 19, 2022 • 3 new comments

• [c10d] Use c10::Synchronized in c10d::SequenceNum

  #74049 commented on Mar 19, 2022 • 3 new comments

• GPU acceleration for Apple's M1 chip?

  #47702 commented on Mar 13, 2022 • 2 new comments

• Inconsistent numpy indexing

  #73236 commented on Mar 14, 2022 • 2 new comments

• sparse.mm: CUDA error: internal error when calling `cusparseSpGEMM_workEstimation [...]`

  #68323 commented on Mar 14, 2022 • 2 new comments

• Enable squid proxy to use TLS + work with HTTPS

  #73812 commented on Mar 15, 2022 • 2 new comments

• Ambiguous docstring on `register_module_forward_hook`

  #74024 commented on Mar 15, 2022 • 2 new comments

• processes hang when executing cross-machine asynchronous P2P communication on NCCL backend

  #73960 commented on Mar 16, 2022 • 2 new comments

• Wrong onnx export for v1.10. LTS was correct

  #69605 commented on Mar 16, 2022 • 2 new comments

• `CrossEntropyLoss` triggers floating point exception

  #73165 commented on Mar 16, 2022 • 2 new comments

• `torch.combinations` will allocate large memory when `r` is greater than the length of input

  #71082 commented on Mar 16, 2022 • 2 new comments

• torch.where leads to unexpected gradients

  #52248 commented on Mar 17, 2022 • 2 new comments

• Inconsistent implementation on SWA

  #74022 commented on Mar 19, 2022 • 2 new comments

• Add JIT graph fuser for oneDNN Graph API (Preview4)

  #68111 commented on Mar 14, 2022 • 2 new comments

• add autocast cpu doc

  #68567 commented on Mar 16, 2022 • 2 new comments

• set_dir expanding "~"

  #69763 commented on Mar 19, 2022 • 2 new comments

• Remove deprecated torch.lstsq

  #70980 commented on Mar 14, 2022 • 2 new comments

• Remove deprecated torch.solve

  #70986 commented on Mar 14, 2022 • 2 new comments

• Remove deprecated torch.symeig

  #70988 commented on Mar 14, 2022 • 2 new comments

• rocblas alt impl during backward pass only

  #71881 commented on Mar 15, 2022 • 2 new comments

• numpy-compat: Add an offset to `torch.eye`

  #72443 commented on Mar 18, 2022 • 2 new comments

• Nvfuser code bump 030122

  #73627 commented on Mar 14, 2022 • 2 new comments

• expanded weights: group norm faster rule

  #73696 commented on Mar 17, 2022 • 2 new comments

• [jiterator] kaiser_window

  #73734 commented on Mar 14, 2022 • 2 new comments

• Fix `asarray` docs + add test case.

  #73736 commented on Mar 14, 2022 • 2 new comments

• Add CPU slow_test job

  #73748 commented on Mar 18, 2022 • 2 new comments

• Add warning when importing caffe2 on build without BUILD_CAFFE2=1

  #73770 commented on Mar 18, 2022 • 2 new comments

• [quant][core][gpu][refactor] Refactored auxiliary functions in cudnn Conv.cpp to an utilities file

  #73957 commented on Mar 19, 2022 • 2 new comments

• Refactor error input tests in test_torch.py to OpInfos

  #73981 commented on Mar 14, 2022 • 2 new comments

• Extend Graph Export to NNC, extend script to support CPU

  #74076 commented on Mar 17, 2022 • 2 new comments

• [tensorexpr] Enabled aten::stack in the fuser pass with static shapes

  #74077 commented on Mar 16, 2022 • 2 new comments

• Split up torch.distributions docs into multiple pages

  #72545 commented on Mar 13, 2022 • 1 new comment

• [Caffe2] Missing CMAKE_CUDA_COMPILE_WHOLE_COMPILATION

  #18524 commented on Mar 14, 2022 • 1 new comment

• C++ Context::setDeterministicAlgorithms default 2nd arg not defined in header

  #74101 commented on Mar 14, 2022 • 1 new comment

• at::unfold is not present in the C++ docs even though it's an available function

  #74091 commented on Mar 14, 2022 • 1 new comment

• Got error about invalid header or archive is corrupted

  #73168 commented on Mar 14, 2022 • 1 new comment

• Feature Request: Hausdorff distance

  #73778 commented on Mar 14, 2022 • 1 new comment

• Tensor.new_tensor now preserves input tensor device, making it inconsistent with documentation

  #73838 commented on Mar 14, 2022 • 1 new comment

• Quantizable LSTMCell does not work correctly.

  #55945 commented on Mar 14, 2022 • 1 new comment

• torch::jit::pickle_load is unable to read some files saved by torch.save

  #67902 commented on Mar 14, 2022 • 1 new comment

• [Tensorboard] Problem with subfolders from SummaryWriter

  #32651 commented on Mar 14, 2022 • 1 new comment

• Enable fft support for mobile builds

  #62821 commented on Mar 14, 2022 • 1 new comment

• when converting scripted model to coremltools, Unknown type bool encountered in graph lowering. This type is not supported in ONNX export

  #52889 commented on Mar 14, 2022 • 1 new comment

• No module named 'pygame': soft_actor_critic

  #74134 commented on Mar 14, 2022 • 1 new comment

• LBFGS always give nan results, why

  #5953 commented on Mar 14, 2022 • 1 new comment

• PyTorch CUDA-11.5 readiness

  #69302 commented on Mar 14, 2022 • 1 new comment

• RuntimeError: Global alloc not supported yet in TorchScript

  #69078 commented on Mar 14, 2022 • 1 new comment

• universal binaries for libtorch on mac (x86_64+arm)

  #63558 commented on Mar 14, 2022 • 1 new comment

• ONNX export failed on ATen operator rfft because torch.onnx.symbolic_opset9.rfft does not exist

  #44822 commented on Mar 15, 2022 • 1 new comment

• Conversion Error in pytorch mobile with metal

  #72110 commented on Mar 15, 2022 • 1 new comment

• Add and Mul torch tensors on Metal (IOS)

  #72330 commented on Mar 15, 2022 • 1 new comment

• Soft mandatory labeling for release notes

  #68459 commented on Mar 15, 2022 • 1 new comment

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

  #67146 commented on Mar 15, 2022 • 1 new comment

• [bug] the LTS torch==1.8.2 pip package is incomplete

  #69689 commented on Mar 15, 2022 • 1 new comment

• Create complete release matrix and validation script

  #74102 commented on Mar 15, 2022 • 1 new comment

• INTERNAL ASSERT FAILED at "../aten/src/ATen/MapAllocator.cpp":263

  #67864 commented on Mar 15, 2022 • 1 new comment

• Slowdown due to thread specific model caching

  #27902 commented on Mar 15, 2022 • 1 new comment

• Mixed precision causes NaN loss

  #40497 commented on Mar 16, 2022 • 1 new comment

• Matrix multiplication broken on PyTorch 1.8.1 with CUDA 11.1 and Nvidia GTX 1080 Ti

  #56747 commented on Mar 16, 2022 • 1 new comment

• [feature request] "Batched" index_select (i.e. simplified torch.gather with not specifying full index)

  #64208 commented on Mar 16, 2022 • 1 new comment

• Gradient of nansum and nanmean wrongly produces `nan`

  #67180 commented on Mar 16, 2022 • 1 new comment

• [bug] `parametrize` can't be used together with `swa_utils.AveragedModel`

  #69413 commented on Mar 16, 2022 • 1 new comment

• Gradient of clamp is nan for inf inputs

  #10729 commented on Mar 16, 2022 • 1 new comment

• x.grad should be 0 but get NaN after x/0

  #4132 commented on Mar 16, 2022 • 1 new comment

• Clarify hook support in DDP

  #66229 commented on Mar 16, 2022 • 1 new comment

• download speed

  #17023 commented on Mar 17, 2022 • 1 new comment

• Segmentation fault when using torch.profiler

  #56008 commented on Mar 17, 2022 • 1 new comment

• Unable to build and use libtorch function via pybind11: undefined symbol error upon import

  #73016 commented on Mar 17, 2022 • 1 new comment

• INTERNAL ASSERT FAILED at "..\\aten\\src\\ATen\\MapAllocator.cpp":135

  #67844 commented on Mar 17, 2022 • 1 new comment

• operation not supported crash when initializing RPC tensorpipe

  #73885 commented on Mar 17, 2022 • 1 new comment

• Run libtorch examples, export error "undefined reference to xxx"

  #73829 commented on Mar 17, 2022 • 1 new comment

• RFC: multi-part `torch.load`/`torch.save` to support huge models and/or low CPU memory

  #64327 commented on Mar 17, 2022 • 1 new comment

• Safe way of loading only weights from *.pt file

  #52181 commented on Mar 17, 2022 • 1 new comment

• Export torch.maximum() op in ONNX format

  #72790 commented on Mar 17, 2022 • 1 new comment

• Uncleared memory use after torch.load()

  #64043 commented on Mar 18, 2022 • 1 new comment

• [Runtime Error in ddp_pipeline.py] RuntimeError: unsupported operation: some elements of the input tensor and the written-to tensor refer to a single memory location. Please clone() the tensor before performing the operation.

  #68407 commented on Mar 18, 2022 • 1 new comment

• ONNX and grid_sample layer

  #27212 commented on Mar 18, 2022 • 1 new comment

• Error during training: falseINTERNAL ASSERT FAILED

  #72782 commented on Mar 18, 2022 • 1 new comment

• [feature request] quantized operators on CUDA

  #69364 commented on Mar 18, 2022 • 1 new comment

• [feature request] BitTensors and bit packed tensors utilities and binary neural nets

  #32867 commented on Mar 18, 2022 • 1 new comment

• ONNX fails to export Fold / col2im

  #41423 commented on Mar 18, 2022 • 1 new comment

• Calling super().__torch_dispatch__ with arguments list causes segfault

  #73933 commented on Mar 19, 2022 • 1 new comment

• [discussion] Comparison operator chaining

  #41323 commented on Mar 19, 2022 • 1 new comment

• Support matrix operations between complex and real tensors

  #61909 commented on Mar 19, 2022 • 1 new comment

• Add step of measure and path in add_hparam

  #26804 commented on Mar 16, 2022 • 1 new comment

• File renamed/replaced whitelist with allowlist

  #42070 commented on Mar 18, 2022 • 1 new comment

• Improve the performance of repeat_interleave

  #51869 commented on Mar 16, 2022 • 1 new comment

• Add BFloat16 support for GLU, and randperm operators on CPU

  #61944 commented on Mar 15, 2022 • 1 new comment

• Add BFloat16 support for smooth_l1_loss on CPU

  #62558 commented on Mar 15, 2022 • 1 new comment

• Add BFloat16 support for logsigmoid, hardsigmoid, hardshrink, softshrink, hardswish and softplus on CPU

  #63134 commented on Mar 15, 2022 • 1 new comment

• bug fix for D32374003

  #69266 commented on Mar 17, 2022 • 1 new comment

• Add torch.linalg.ldl_factor_ex and torch.linalg.ldl_solve

  #69828 commented on Mar 14, 2022 • 1 new comment

• added handling for r=0 edge case in torch.combinations(tensor, r)

  #70270 commented on Mar 16, 2022 • 1 new comment

• Remove deprecated torch.chain_matmul

  #70978 commented on Mar 14, 2022 • 1 new comment

• Remove deprecated torch.cholesky

  #70979 commented on Mar 14, 2022 • 1 new comment

• Remove deprecated torch.matrix_rank

  #70981 commented on Mar 14, 2022 • 1 new comment

• Remove deprecated torch.eig

  #70982 commented on Mar 14, 2022 • 1 new comment

• [fix] Contiguity of `torch.ravel`!

  #71771 commented on Mar 17, 2022 • 1 new comment

• Implement numpy-like function `torch.block`

  #71859 commented on Mar 15, 2022 • 1 new comment

• [ROCm] Enable topk operator for bfloat16 dtype

  #71913 commented on Mar 16, 2022 • 1 new comment

• [pkg] add generic ZipFile Reader/Writer

  #72237 commented on Mar 16, 2022 • 1 new comment

• [ONNX] Make Non-Float Op Exportation Compatible to Avoid Invalid ONNX Models

  #72401 commented on Mar 14, 2022 • 1 new comment

• [JIT] document issues with eager autocast & scripting

  #72760 commented on Mar 14, 2022 • 1 new comment

• [nnc] Strides to Tensor

  #72962 commented on Mar 17, 2022 • 1 new comment

• Class name

  #73201 commented on Mar 15, 2022 • 1 new comment

• [ONNX] Support optional type (#68793)

  #73284 commented on Mar 17, 2022 • 1 new comment

• skip HPU tensor in TensorIterator

  #73343 commented on Mar 16, 2022 • 1 new comment

• [Dynamic RPC] Allow newly joined ranks to communicate with existing ranks

  #73373 commented on Mar 16, 2022 • 1 new comment

• decouple arrayRef from deploy

  #73526 commented on Mar 14, 2022 • 1 new comment

• [ROCM] Navi21 Enablement 6: Tensor kernels

  #73546 commented on Mar 18, 2022 • 1 new comment

• adjust conditions to enable jit decomposition pass only for GPU device

  #73637 commented on Mar 16, 2022 • 1 new comment

• Get rid of DefaultNumBailouts

  #73689 commented on Mar 14, 2022 • 1 new comment

• expanded weights: layer norm faster rule

  #73694 commented on Mar 17, 2022 • 1 new comment

• Remove outdated logic

  #74043 commented on Mar 14, 2022 • 1 new comment

• Write to rockset test stats

  #74123 commented on Mar 12, 2022 • 1 new comment

• [ROCm] revert cat operator performance work-around

  #74129 commented on Mar 18, 2022 • 1 new comment

• [LT] Make lazy_bench.py regconize detectron2 outputs

  #74133 commented on Mar 16, 2022 • 1 new comment

• Add nan-safe einsum and bilinear

  #73143 commented on Mar 14, 2022 • 0 new comments

• torch.cuda.amp: Remove SPMD DDP doc portion

  #73924 commented on Mar 14, 2022 • 0 new comments

• max_pool1d() returns when given invalid large `kernel_size` inputs

  #73870 commented on Mar 14, 2022 • 0 new comments

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

  #42654 commented on Mar 14, 2022 • 0 new comments

• Override tests in test/test_overrides.py failing for some functions

  #74122 commented on Mar 14, 2022 • 0 new comments

• Add ZeroTensor fastpath for torch.addmm

  #72345 commented on Mar 14, 2022 • 0 new comments

• libtorch 1.11.0 libiomp5.dylib contains erroneous link to /DLC/torch instead of using @rpath

  #74120 commented on Mar 14, 2022 • 0 new comments

• Add Python-3.10 support

  #66424 commented on Mar 14, 2022 • 0 new comments

• Lazy Tensor: TorchBench correctness issues in the noopt mode

  #73962 commented on Mar 14, 2022 • 0 new comments

• TestSparseCSRCUDA.test_block_triangular_solve_cuda_complex128 fails on A100

  #71297 commented on Mar 15, 2022 • 0 new comments

• Clarify test dependencies (e.g., into a test-requirements.txt file)

  #72556 commented on Mar 15, 2022 • 0 new comments

• DistributedDataParallel high peak memory usage with find_unused_parameters=True

  #74115 commented on Mar 15, 2022 • 0 new comments

• Ensure that the message for disabled tests is printed for disabled tests in CI

  #69380 commented on Mar 16, 2022 • 0 new comments

• Error on MacOS related to custom class registration in torchscript

  #65000 commented on Mar 16, 2022 • 0 new comments

• [Meta] Change default branch name to `main` for repos in `pytorch` project

  #62540 commented on Mar 16, 2022 • 0 new comments

• Migrate master to main: https://github.com/pytorch/pytorch

  #71806 commented on Mar 16, 2022 • 0 new comments

• [DataLoader] Implement Map-style functional DataPipe

  #57031 commented on Mar 16, 2022 • 0 new comments

• DISABLED test_ddp_comparison_uneven_inputs (__main__.TensorPipeDdpComparisonTest)

  #72891 commented on Mar 16, 2022 • 0 new comments

• Enable clang-tidy on all of master

  #62011 commented on Mar 16, 2022 • 0 new comments

• Refactor/Cleanup Lazy Tensor Core

  #73710 commented on Mar 16, 2022 • 0 new comments

• PyTorch CI TTS Tracker

  #68543 commented on Mar 17, 2022 • 0 new comments

• Model parallel with DDP get `Socket Timeout` error when using NCCL, while GLOO works fine

  #25767 commented on Mar 17, 2022 • 0 new comments

• Improve autograd codegened automatic inplace formulas

  #71987 commented on Mar 17, 2022 • 0 new comments

• [c10d] Work objects should have a general operator<<

  #59745 commented on Mar 17, 2022 • 0 new comments

• Print details of why a test has been skipped in CI

  #69014 commented on Mar 17, 2022 • 0 new comments

• Rollup: forward-mode AD operator coverage

  #71117 commented on Mar 17, 2022 • 0 new comments

• Add Python Version to Torch.Package metadata

  #74068 commented on Mar 17, 2022 • 0 new comments

• [DO NOT DELETE OR MERGE] Nightly release

  #26921 commented on Mar 19, 2022 • 0 new comments

• [DO NOT DELETE OR MERGE] Postnightly release

  #27167 commented on Mar 19, 2022 • 0 new comments

• add channels last support for PixelShuffle and PixelUnshuffle

  #50573 commented on Mar 14, 2022 • 0 new comments

• add channels last (2d) support for mkldnn_convolution

  #55584 commented on Mar 14, 2022 • 0 new comments

• enable BFloat16 mkldnn_convolution on both contiguous and channels last memory format

  #55864 commented on Mar 14, 2022 • 0 new comments

• [ROCm] libtorch nightly builds

  #57506 commented on Mar 18, 2022 • 0 new comments

• add BFloat16 support for UpSample on CPU

  #58297 commented on Mar 18, 2022 • 0 new comments

• opitimze ConvTransposedND with mkldnn float32 and bfloat16 on CPU

  #58348 commented on Mar 14, 2022 • 0 new comments

• add BFloat16 support for BatchNorm2d on CPU

  #61015 commented on Mar 14, 2022 • 0 new comments

• Add python-3.10 CI

  #66530 commented on Mar 13, 2022 • 0 new comments

• add channels last support for thnn_conv2d (non-dilated)

  #68101 commented on Mar 14, 2022 • 0 new comments

• improve group_norm channels last performance on CPU

  #69067 commented on Mar 18, 2022 • 0 new comments

• expanded weights: instance norm faster rules

  #70141 commented on Mar 17, 2022 • 0 new comments

• Add fast path of qmean/qstd for quantized CPU

  #70172 commented on Mar 16, 2022 • 0 new comments

• Add native impl for group norm on quantized CPU for channels-last inputs

  #70520 commented on Mar 16, 2022 • 0 new comments

• add channels last support for slow_conv_dilated2d

  #70665 commented on Mar 14, 2022 • 0 new comments

• add channels last support for slow_conv_transpose2d

  #70897 commented on Mar 14, 2022 • 0 new comments

• [DO NOT MERGE!] Verify the performance impact of #67977 with TorchBench CI

  #71145 commented on Mar 14, 2022 • 0 new comments

• CUDA Kernels: Use per-operator headers (2/4)

  #71213 commented on Mar 17, 2022 • 0 new comments

• CUDA Kernels: Use per-operator headers (3/4)

  #71214 commented on Mar 17, 2022 • 0 new comments

• CUDA Kernels: Use per-operator headers (4/4)

  #71215 commented on Mar 17, 2022 • 0 new comments

• cuDNN/miopen: Use per-operator headers

  #71216 commented on Mar 17, 2022 • 0 new comments

• QuantizedCPU: Use per-operator headers

  #71217 commented on Mar 18, 2022 • 0 new comments

• improve LayerNorm bfloat16 performance on CPU

  #71376 commented on Mar 18, 2022 • 0 new comments

• Add support for backend to register reducer timer

  #71700 commented on Mar 17, 2022 • 0 new comments

• torch.hub security improvement

  #72060 commented on Mar 19, 2022 • 0 new comments

• Allow specifying tags for aten operators in native_functions.yaml

  #72549 commented on Mar 16, 2022 • 0 new comments

• Upgrade CI to ROCm5.0

  #73105 commented on Mar 15, 2022 • 0 new comments

• [PyTorch] Existing MHA: fuse the attn_mask addition

  #73219 commented on Mar 16, 2022 • 0 new comments

• [Pytorch] Existing MHA: do in-place division on q

  #73249 commented on Mar 16, 2022 • 0 new comments

• [special] Add special.gammaincinv and special.gammainccinv

  #73307 commented on Mar 16, 2022 • 0 new comments

• [JIT] Enable NVFuser tests in CI

  #73322 commented on Mar 14, 2022 • 0 new comments

• [PyTorchEdge] Extend _save_for_mobile to support flatbuffer format

  #73403 commented on Mar 17, 2022 • 0 new comments

• Implement _pad_circular in ATen

  #73410 commented on Mar 19, 2022 • 0 new comments

• Implement F.pad in ATen

  #73431 commented on Mar 19, 2022 • 0 new comments

• stft: Implement center padding in ATen

  #73432 commented on Mar 19, 2022 • 0 new comments

• Remove F.pad python implementation

  #73433 commented on Mar 19, 2022 • 0 new comments

• stft: remove non-center overload and python functional wrapper

  #73434 commented on Mar 19, 2022 • 0 new comments

• Support for tensor subclasses as parameters

  #73459 commented on Mar 17, 2022 • 0 new comments

• torch.mm(dense, sparse)

  #73686 commented on Mar 15, 2022 • 0 new comments

• Refactor plain tensor parameters to set a flag instead of using _make_subclass()

  #73690 commented on Mar 17, 2022 • 0 new comments

• add mixed data type mode for LayerNorm forward path

  #73844 commented on Mar 18, 2022 • 0 new comments

• add native kernel for weight_norm on CPU

  #73845 commented on Mar 18, 2022 • 0 new comments

• add simd horizantal reduce to improve log_softmax and softmax performance on CPU

  #73953 commented on Mar 18, 2022 • 0 new comments

• Reuse Lazy IR node when possible

  #73968 commented on Mar 18, 2022 • 0 new comments

• enable channels last 3d for Conv3d and ConvTranspose3d on mkldnn path

  #74023 commented on Mar 14, 2022 • 0 new comments

• [Dynamic RPC] Allow existing ranks to communicate with newly joined ranks

  #74035 commented on Mar 16, 2022 • 0 new comments

• [Tracer][ONNX] Enable numel tracing

  #74081 commented on Mar 14, 2022 • 0 new comments

• Add hipify_torch submodule to replace pytorch hipify

  #74103 commented on Mar 16, 2022 • 0 new comments