Description
We are welcome contributions, please check 14k github models test and debug env setup for how to reproduce these issues.
TorchDynamo (pass rate tracker)
- pytest ./generated/test_BlinkDL_RWKV_LM.py -k test_005 [Subclasses torch.jit.ScriptModule - unusual usage, not a priority]
- pytest ./generated/test_INVOKERer_DeepRFT.py -k test_010 [dynamo] Handle general tensor attributes with a getattr proxy node #91840 (@davidberard98)
- pytest ./generated/test_assassint2017_abdominal_multi_organ_segmentation.py -k test_000 [dynamo] support [tensor].type(torch.FloatTensor) #93043 (@davidberard98)
- pytest ./generated/test_facebookresearch_ClassyVision.py -k test_008
- pytest ./generated/test_facebookresearch_denoiser.py -k test_005 [fails eager]
- pytest ./generated/test_facebookresearch_pytext.py -k test_001 (shape_as_tensor handling)
- pytest ./generated/test_fangwei123456_spikingjelly.py -k test_016 @davidberard98 [dynamo] support custom __getattr__ on torch.nn.Modules #94658
- pytest ./generated/test_github_pengge_PyTorch_progressive_growing_of_gans.py -k test_002 [numpy.ndarray] (Can we rewrite numpy operators to pytorch operators? #93684 ?)
- pytest ./generated/test_guanfuchen_semseg.py -k test_034 [segfault in triton]
- pytest ./generated/test_ludwig_ai_ludwig.py -k test_018 (@yanboliang)
- pytest ./generated/test_nianticlabs_simplerecon.py -k test_001 [segfault in triton]
- pytest ./generated/test_pytorch_translate.py -k test_006 (InternalTorchDynamoError: 'int' object has no attribute 'view')
- pytest ./generated/test_easezyc_deep_transfer_learning.py -k test_004 (cc @ezyang @msaroufim @bdhirsh @anijain2305 @zou3519 @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @peterbell10 @ipiszy @yf225 @chenyang78 @kadeng @muchulee8 @aakhundov @ColinPeppler @davidberard98, this is regression caused by [dynamo] Handle general tensor attributes with a getattr proxy node #91840) [dynamo] fix general attr on tensor for user-provided attributes #94332
- pytest ./generated/test_ELEKTRONN_elektronn3.py -k test_018 ([Dynamo] Support context wrapping(e.g, torch.no_grad) on nested functions w/o closure #92922) @yanboliang
- pytest ./generated/test_HaSai666_rec_pangu.py -k test_017 ([Dynamo] Fix torch operators out variants bug if out=None #92926) @yanboliang
- pytest ./generated/test_Shmuma_ptan.py -k test_002 ([Dynamo] Support out variants of ops mutate the tensors out of the function frame #93177) @yanboliang
- pytest ./generated/test_awslabs_sockeye.py -k test_009 (FakeTensor should constant propagate through ops that allow numbers as scalars #94145) @yanboliang
- pytest ./generated/test_lorenmt_mtan.py -k test_005 (invalid combination of arguments) (python_arg_parser to allow fake tensor element in symint_list when in dynamo mode #95424 #97508) @lantiankaikai
- pytest ./generated/test_ustcml_RecStudio.py -k test_036 (invalid combination of arguments) (python_arg_parser to allow fake tensor element in symint_list when in dynamo mode #95424 #97508) @lantiankaikai
- pytest ./generated/test_PeterouZh_CIPS_3D.py -k test_003 (Fix strip_function_call in GuardBuilder #97810) @lantiankaikai
- pytest ./generated/test_IDEA_CCNL_Fengshenbang_LM.py -k test_008 ([Dynamo] Improve handling UnspecializedNNModuleVariable side effect #101141) @yanboliang
- pytest ./generated/test_VainF_pytorch_msssim.py -k test_000 @lyoka
- pytest ./generated/test_sony_sqvae.py -k test_007 (Tensor-likes are not close!) @chaekit
TorchInductor (pass rate tracker)
- pytest ./generated/test_baal_org_baal.py -k test_011 [inductor] fixes addmm pattern matcher to exclude non-conformant patt… #96634 @ngimel
- pytest ./generated/test_ZhaoJ9014_face_evoLVe.py -k test_000 (@yanboliang)
- pytest ./generated/test_ACheun9_Pytorch_implementation_of_Mobile_Former.py -k test_002 [addmm(inp, mat1, mat2) with scalar inp] @davidberard98
- pytest ./generated/test_BloodAxe_pytorch_toolbelt.py -k test_070 ([Inductor] Improve view/reshape on tensors with shape 0 #101051) @yanboliang
- pytest ./generated/test_adriansahlman_stylegan2_pytorch.py -k test_000 ([Inductor] Fix convolution lowering if stride or padding or dilation is 1 element list #98448) @chaekit
- pytest ./generated/test_HomebrewNLP_revlib.py -k test_004 (KeyError: 'val')
- pytest ./generated/test_ProGamerGov_neural_dream.py -k test_000 (LoweringException: AssertionError: target: aten.avg_pool2d.default) @yanboliang ([Inductor] aten.{avg_pool2d/max_pool2d_with_indices} arguments can be 1 element tuple #96727)
- pytest ./generated/test_NKI_AI_direct.py -k test_009 (Unhandled FakeTensor Device Propagation for aten.copy.default, found two different devices cpu, cuda:0) ([fake tensor] register copy, copy_, slice_scatter handling #102677) @davidberard98
- pytest ./generated/test_XuyangBai_PointDSC.py -k test_004 (RuntimeError: as_strided_scatter: sizes [4], strides [85], storage offset 256 and itemsize 4 requiring a storage size of 2048 are out of bounds for storage of size 1024) (Fix bug in check required output size in _as_strided_scatter_meta #98483) @lantiankaikai
- pytest ./generated/test_agrimgupta92_sgan.py -k test_000 (RuntimeError: Input and hidden tensors are not at the same device, found input tensor at cuda:0 and hidden tensor at cpu)
- pytest ./generated/test_hkchengrex_XMem.py -k test_013 (LoweringException: AssertionError: target: aten.index_put.default) @yanboliang ([Inductor] index_put - unsqueeze indices[0] if self and indices[0] are not broadcastable #97105)
- pytest ./generated/test_laddie132_Match_LSTM.py -k test_009 (RuntimeError: masked_fill only supports boolean masks, but got dtype Byte)
- pytest ./generated/test_nlp_uoregon_trankit.py -k test_021 (LoweringException: AssertionError: target: aten.cat.default) ([Inductor] Fix bug in lowering.slice_ when negative start out of range #98517) @yanboliang
- pytest ./generated/test_open_mmlab_mmclassification.py -k test_006 (LoweringException: RuntimeError: Tensor must have a last dimension with stride 1. target: aten.view_as_complex.default)
- pytest ./generated/test_HKUST_KnowComp_MnemonicReader.py -k test_003 (NotImplementedError: could not find kernel for aten._cudnn_rnn_flatten_weight.default at dispatch key DispatchKey.Meta)
- pytest ./generated/test_DKuan_MADDPG_torch.py -k test_002 (RuntimeError: backwards not supported on prim)
- [Inductor] Fix view/reshape on tensors with shape 0 in any dimension #99671 (@yanboliang)
- pytest ./generated/test_adapter_hub_adapter_transformers.py -k test_115 ([inductor] fix incorrect strides in copy() decomp, fix hf_LongFormer + hf_BigBird errors #100115) @bdhirsh (fixed once the PR is in)
- pytest ./generated/test_yizhou_wang_RODNet.py -k test_000 (Fix aten.copy device mismatch bug in FakeTensor #102664) @williamwen42
- pytest ./generated/test_shinianzhihou_ChangeDetection.py -k test_027 (AssertionError: Unexpected fake buffer _tensor_constant0 Enable TORCH_FAKE_TENSOR_DEBUG=1 to get creation stack traces on fake tensors.) (repro reported to nn.Parameter dispatch error in functionalization #102989) @williamwen4 90F1 2
- pytest ./generated/test_clovaai_lffont.py -k test_008 (triton codegen error) ([inductor] certain reductions cause triton compile error #103481, fixed by [inductor] Fix no-xdim reductions #103527) @williamwen42
- pytest ./generated/test_adriansahlman_stylegan2_pytorch.py -k test_000 (TypeError: unsupported operand type(s) for *: 'NoneType' and 'Tensor')
- pytest ./generated/test_alibaba_EasyCV.py -k test_054 (Please check that Nodes in the graph are topologically ordered) ([Inductor][FX passes] Pre grad pass modified graph should be topological sorted #103794) @yanboliang
- pytest ./generated/test_adriansahlman_stylegan2_pytorch.py -k test_000
cc @ezyang @soumith @msaroufim @wconstab @ngimel @bdhirsh @mlazos @voznesenskym @penguinwu @anijain2305 @EikanWang @jgong5 @Guobing-Chen @chunyuan-w @XiaobingSuper @zhuhaozhe @blzheng @Xia-Weiwen @wenzhe-nrv @jiayisunx @peterbell10 @desertfire