[ONNX] Treat CompositeImplicitAutograd ops as normal ops in decomp #136153
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
justinchuby
requested review from
shubhambhokare1,
titaiwangms and
wschin
as code owners
pytorch-bot
bot
added
the
release notes: onnx
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/136153
Note: Links to docs will display an error until the docs builds have been completed. ❌ 1 New FailureAs of commit ee32d85 with merge base 0aa41eb ( NEW FAILURE - The following job has failed:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
justinchuby
requested review from
shubhambhokare1,
titaiwangms,
tugsbayasgalan and
xadupre
and removed request for
shubhambhokare1,
titaiwangms and
wschin
justinchuby
added
module: onnx
xadupre
approved these changes
Sep 16, 2024
Collaborator
Author
|
@pytorchbot merge |
justinchuby
added
release notes: onnx
Collaborator
Merge startedYour change will be merged once all checks pass (ETA 0-4 Hours). Learn more about merging in the wiki. Questions? Feedback? Please reach out to the PyTorch DevX Team |
justinchuby
added a commit
to microsoft/onnxscript
that referenced
this pull request
Sep 16, 2024
Following pytorch/pytorch#136153, we remove stft, var, var_mean, std, std_mean ops. They were never used even before because the ops were always decomposed.
gramalingam
approved these changes
Sep 16, 2024
Collaborator
Merge failedReason: 1 mandatory check(s) failed. The first few are: Dig deeper by viewing the failures on hud |
Collaborator
Author
|
@pytorchbot merge -i |
Collaborator
Merge startedYour change will be merged while ignoring the following 1 checks: pull / linux-focal-cuda12.1-py3.10-gcc9 / test (default, 2, 5, lf.linux.4xlarge.nvidia.gpu) Learn more about merging in the wiki. Questions? Feedback? Please reach out to the PyTorch DevX Team |
Chao1Han
pushed a commit
to Chao1Han/pytorch
that referenced
this pull request
Sep 20, 2024
…ytorch#136153) Since pytorch#135080, the CompositeImplicitAutograd (CIA) ops are only decomposed when a decomp function is provided in a table. There is no longer a need to distinguish CIA ops like Upsample and preserve them explicitly. On the ONNX Script torchlib side I will unregister some ops from the following list to make sure some CIA ops are still decomposed. ``` <OpOverload(op='aten.__and__', overload='Scalar')>, <OpOverload(op='aten.__and__', overload='Tensor')>, <OpOverload(op='aten.__or__', overload='Scalar')>, <OpOverload(op='aten.__or__', overload='Tensor')>, <OpOverload(op='aten.__xor__', overload='Scalar')>, <OpOverload(op='aten.__xor__', overload='Tensor')>, <OpOverload(op='aten._add_batch_dim', overload='default')>, <OpOverload(op='aten._assert_tensor_metadata', overload='default')>, <OpOverload(op='aten._backward', overload='default')>, <OpOverload(op='aten._batch_norm_impl_index_backward', overload='default')>, <OpOverload(op='aten._cast_Byte', overload='default')>, <OpOverload(op='aten._cast_Char', overload='default')>, <OpOverload(op='aten._cast_Double', overload='default')>, <OpOverload(op='aten._cast_Float', overload='default')>, <OpOverload(op='aten._cast_Half', overload='default')>, <OpOverload(op='aten._cast_Int', overload='default')>, <OpOverload(op='aten._cast_Long', overload='default')>, <OpOverload(op='aten._cast_Short', overload='default')>, <OpOverload(op='aten._choose_qparams_per_tensor', overload='default')>, <OpOverload(op='aten._convolution', overload='deprecated')>, <OpOverload(op='aten._convolution_double_backward', overload='default')>, <OpOverload(op='aten._convolution_mode', overload='default')>, <OpOverload(op='aten._cufft_clear_plan_cache', overload='default')>, <OpOverload(op='aten._cufft_get_plan_cache_max_size', overload='default')>, <OpOverload(op='aten._cufft_get_plan_cache_size', overload='default')>, <OpOverload(op='aten._cufft_set_plan_cache_max_size', overload='default')>, <OpOverload(op='aten._debug_has_internal_overlap', overload='default')>, <OpOverload(op='aten._dim_arange', overload='default')>, <OpOverload(op='aten._embedding_bag_sparse_backward', overload='default')>, <OpOverload(op='aten._gather_sparse_backward', overload='default')>, <OpOverload(op='aten._grid_sampler_2d_cpu_fallback_backward', overload='default')>, <OpOverload(op='aten._has_compatible_shallow_copy_type', overload='default')>, <OpOverload(op='aten._is_zerotensor', overload='default')>, <OpOverload(op='aten._lu_with_info', overload='default')>, <OpOverload(op='aten._nnpack_available', overload='default')>, <OpOverload(op='aten._pack_padded_sequence_backward', overload='default')>, <OpOverload(op='aten._pad_circular', overload='default')>, <OpOverload(op='aten._pad_enum', overload='default')>, <OpOverload(op='aten._pad_packed_sequence', overload='default')>, <OpOverload(op='aten._propagate_xla_data', overload='default')>, <OpOverload(op='aten._remove_batch_dim', overload='default')>, <OpOverload(op='aten._reshape_from_tensor', overload='default')>, <OpOverload(op='aten._rowwise_prune', overload='default')>, <OpOverload(op='aten._saturate_weight_to_fp16', overload='default')>, <OpOverload(op='aten._scaled_dot_product_attention_math', overload='default')>, <OpOverload(op='aten._shape_as_tensor', overload='default')>, <OpOverload(op='aten._sobol_engine_draw', overload='default')>, <OpOverload(op='aten._sparse_bsc_tensor_unsafe', overload='default')>, <OpOverload(op='aten._sparse_bsr_tensor_unsafe', overload='default')>, <OpOverload(op='aten._sparse_compressed_tensor_unsafe', overload='default')>, <OpOverload(op='aten._sparse_coo_tensor_unsafe', overload='default')>, <OpOverload(op='aten._sparse_csc_tensor_unsafe', overload='default')>, <OpOverload(op='aten._sparse_csr_tensor_unsafe', overload='default')>, <OpOverload(op='aten._sparse_log_softmax', overload='Dimname')>, <OpOverload(op='aten._sparse_log_softmax', overload='int')>, <OpOverload(op='aten._sparse_mm', overload='default')>, <OpOverload(op='aten._sparse_mm', overload='reduce')>, <OpOverload(op='aten._sparse_softmax', overload='Dimname')>, <OpOverload(op='aten._sparse_softmax', overload='int')>, <OpOverload(op='aten._sparse_sum', overload='default')>, <OpOverload(op='aten._sparse_sum', overload='dim_dtype')>, <OpOverload(op='aten._sparse_sum', overload='dtype')>, <OpOverload(op='aten._test_ambiguous_defaults', overload='a')>, <OpOverload(op='aten._test_ambiguous_defaults', overload='b')>, <OpOverload(op='aten._test_autograd_multiple_dispatch', overload='ntonly')>, <OpOverload(op='aten._test_check_tensor', overload='default')>, <OpOverload(op='aten._test_serialization_subcmul', overload='default')>, <OpOverload(op='aten._test_string_default', overload='default')>, <OpOverload(op='aten._thnn_differentiable_gru_cell_backward', overload='default')>, <OpOverload(op='aten._thnn_differentiable_lstm_cell_backward', overload='default')>, <OpOverload(op='aten._thnn_fused_lstm_cell_backward', overload='default')>, <OpOverload(op='aten._to_cpu', overload='default')>, <OpOverload(op='aten._upsample_bicubic2d_aa', overload='vec')>, <OpOverload(op='aten._upsample_bilinear2d_aa', overload='vec')>, <OpOverload(op='aten._upsample_nearest_exact1d', overload='default')>, <OpOverload(op='aten._upsample_nearest_exact1d', overload='vec')>, <OpOverload(op='aten._upsample_nearest_exact2d', overload='default')>, <OpOverload(op='aten._upsample_nearest_exact2d', overload='vec')>, <OpOverload(op='aten._upsample_nearest_exact3d', overload='default')>, <OpOverload(op='aten._upsample_nearest_exact3d', overload='vec')>, <OpOverload(op='aten._use_cudnn_rnn_flatten_weight', overload='default')>, <OpOverload(op='aten._validate_sparse_bsc_tensor_args', overload='default')>, <OpOverload(op='aten._validate_sparse_bsr_tensor_args', overload='default')>, <OpOverload(op='aten._validate_sparse_compressed_tensor_args', overload='default')>, <OpOverload(op='aten._validate_sparse_coo_tensor_args', overload='default')>, <OpOverload(op='aten._validate_sparse_csc_tensor_args', overload='default')>, <OpOverload(op='aten._validate_sparse_csr_tensor_args', overload='default')>, <OpOverload(op='aten._version', overload='default')>, <OpOverload(op='aten._weight_norm', overload='default')>, <OpOverload(op='aten._weight_norm_differentiable_backward', overload='default')>, <OpOverload(op='aten.absolute', overload='default')>, <OpOverload(op='aten.adaptive_avg_pool1d', overload='default')>, <OpOverload(op='aten.adaptive_avg_pool2d', overload='default')>, <OpOverload(op='aten.adaptive_avg_pool3d', overload='default')>, <OpOverload(op='aten.adaptive_max_pool1d', overload='default')>, <OpOverload(op='aten.affine_grid_generator_backward', overload='default')>, <OpOverload(op='aten.align_as', overload='default')>, <OpOverload(op='aten.align_tensors', overload='default')>, <OpOverload(op='aten.all', overload='dimname')>, <OpOverload(op='aten.any', overload='dimname')>, <OpOverload(op='aten.arccos', overload='default')>, <OpOverload(op='aten.arccosh', overload='default')>, <OpOverload(op='aten.arcsin', overload='default')>, <OpOverload(op='aten.arcsinh', overload='default')>, <OpOverload(op='aten.arctan', overload='default')>, <OpOverload(op='aten.arctan2', overload='default')>, <OpOverload(op='aten.arctanh', overload='default')>, <OpOverload(op='aten.argsort', overload='default')>, <OpOverload(op='aten.argsort', overload='dimname')>, <OpOverload(op='aten.argsort', overload='stable')>, <OpOverload(op='aten.argwhere', overload='default')>, <OpOverload(op='aten.atleast_1d', overload='Sequence')>, <OpOverload(op='aten.atleast_2d', overload='Sequence')>, <OpOverload(op='aten.atleast_3d', overload='Sequence')>, <OpOverload(op='aten.avg_pool1d', overload='default')>, <OpOverload(op='aten.bilinear', overload='default')>, <OpOverload(op='aten.broadcast_tensors', overload='default')>, <OpOverload(op='aten.can_cast', overload='default')>, <OpOverload(op='aten.cat', overload='names')>, <OpOverload(op='aten.cdist', overload='default')>, <OpOverload(op='aten.chain_matmul', overload='default')>, <OpOverload(op='aten.chalf', overload='default')>, <OpOverload(op='aten.choose_qparams_optimized', overload='default')>, <OpOverload(op='aten.clip', overload='Tensor')>, <OpOverload(op='aten.clip', overload='default')>, <OpOverload(op='aten.column_stack', overload='default')>, <OpOverload(op='aten.combinations', overload='default')>, <OpOverload(op='aten.concat', overload='default')>, <OpOverload(op='aten.concat', overload='names')>, <OpOverload(op='aten.concatenate', overload='default')>, <OpOverload(op='aten.concatenate', overload='names')>, <OpOverload(op='aten.conv1d', overload='default')>, <OpOverload(op='aten.conv1d', overload='padding')>, <OpOverload(op='aten.conv2d', overload='default')>, <OpOverload(op='aten.conv2d', overload='padding')>, <OpOverload(op='aten.conv3d', overload='default')>, <OpOverload(op='aten.conv3d', overload='padding')>, <OpOverload(op='aten.conv_tbc_backward', overload='default')>, <OpOverload(op='aten.conv_transpose1d', overload='default')>, <OpOverload(op='aten.conv_transpose2d', overload='input')>, <OpOverload(op='aten.conv_transpose3d', overload='input')>, <OpOverload(op='aten.corrcoef', overload='default')>, <OpOverload(op='aten.cosine_embedding_loss', overload='default')>, <OpOverload(op='aten.cosine_similarity', overload='default')>, <OpOverload(op='aten.cov', overload='default')>, <OpOverload(op='aten.cross', overload='default')>, <OpOverload(op='aten.cross_entropy_loss', overload='default')>, <OpOverload(op='aten.ctc_loss', overload='IntList')>, <OpOverload(op='aten.ctc_loss', overload='Tensor')>, <OpOverload(op='aten.cudnn_is_acceptable', overload='default')>, <OpOverload(op='aten.cummax', overload='dimname')>, <OpOverload(op='aten.cummaxmin_backward', overload='default')>, <OpOverload(op='aten.cummin', overload='dimname')>, <OpOverload(op='aten.cumprod', overload='dimname')>, <OpOverload(op='aten.cumprod_backward', overload='default')>, <OpOverload(op='aten.cumsum', overload='dimname')>, <OpOverload(op='aten.cumulative_trapezoid', overload='dx')>, <OpOverload(op='aten.cumulative_trapezoid', overload='x')>, <OpOverload(op='aten.data', overload='default')>, <OpOverload(op='aten.det', overload='default')>, <OpOverload(op='aten.diag', overload='default')>, <OpOverload(op='aten.diagflat', overload='default')>, <OpOverload(op='aten.diff', overload='default')>, <OpOverload(op='aten.divide', overload='Scalar')>, <OpOverload(op='aten.divide', overload='Scalar_mode')>, <OpOverload(op='aten.divide', overload='Tensor')>, <OpOverload(op='aten.divide', overload='Tensor_mode')>, <OpOverload(op='aten.dstack', overload='default')>, <OpOverload(op='aten.einsum', overload='default')>, <OpOverload(op='aten.embedding_backward', overload='default')>, <OpOverload(op='aten.embedding_bag', overload='default')>, <OpOverload(op='aten.embedding_bag', overload='padding_idx')>, <OpOverload(op='aten.embedding_sparse_backward', overload='default')>, <OpOverload(op='aten.fake_quantize_per_channel_affine', overload='default')>, <OpOverload(op='aten.fake_quantize_per_channel_affine_cachemask_backward', overload='default')>, <OpOverload(op='aten.fake_quantize_per_tensor_affine', overload='default')>, <OpOverload(op='aten.fake_quantize_per_tensor_affine', overload='tensor_qparams')>, <OpOverload(op='aten.fake_quantize_per_tensor_affine_cachemask_backward', overload='default')>, <OpOverload(op='aten.fbgemm_linear_fp16_weight', overload='default')>, <OpOverload(op='aten.fbgemm_linear_fp16_weight_fp32_activation', overload='default')>, <OpOverload(op='aten.fbgemm_linear_int8_weight', overload='default')>, <OpOverload(op='aten.fbgemm_linear_int8_weight_fp32_activation', overload='default')>, <OpOverload(op='aten.fbgemm_linear_quantize_weight', overload='default')>, <OpOverload(op='aten.fbgemm_pack_gemm_matrix_fp16', overload='default')>, <OpOverload(op='aten.fbgemm_pack_quantized_matrix', overload='KN')>, <OpOverload(op='aten.fbgemm_pack_quantized_matrix', overload='default')>, <OpOverload(op='aten.fft_fft', overload='default')>, <OpOverload(op='aten.fft_fft2', overload='default')>, <OpOverload(op='aten.fft_fftn', overload='default')>, <OpOverload(op='aten.fft_fftshift', overload='default')>, <OpOverload(op='aten.fft_hfft', overload='default')>, <OpOverload(op='aten.fft_hfft2', overload='default')>, <OpOverload(op='aten.fft_hfftn', overload='default')>, <OpOverload(op='aten.fft_ifft', overload='default')>, <OpOverload(op='aten.fft_ifft2', overload='default')>, <OpOverload(op='aten.fft_ifftn', overload='default')>, <OpOverload(op='aten.fft_ifftshift', overload='default')>, <OpOverload(op='aten.fft_ihfft', overload='default')>, <OpOverload(op='aten.fft_ihfft2', overload='default')>, <OpOverload(op='aten.fft_ihfftn', overload='default')>, <OpOverload(op='aten.fft_irfft', overload='default')>, <OpOverload(op='aten.fft_irfft2', overload='default')>, <OpOverload(op='aten.fft_irfftn', overload='default')>, <OpOverload(op='aten.fft_rfft', overload='default')>, <OpOverload(op='aten.fft_rfft2', overload='default')>, <OpOverload(op='aten.fft_rfftn', overload='default')>, <OpOverload(op='aten.fix', overload='default')>, <OpOverload(op='aten.flatten_dense_tensors', overload='default')>, <OpOverload(op='aten.fliplr', overload='default')>, <OpOverload(op='aten.flipud', overload='default')>, <OpOverload(op='aten.float_power', overload='Scalar')>, <OpOverload(op='aten.float_power', overload='Tensor_Scalar')>, <OpOverload(op='aten.float_power', overload='Tensor_Tensor')>, <OpOverload(op='aten.frobenius_norm', overload='dim')>, <OpOverload(op='aten.gather', overload='dimname')>, <OpOverload(op='aten.gather_backward', overload='default')>, <OpOverload(op='aten.ger', overload='default')>, <OpOverload(op='aten.gradient', overload='array')>, <OpOverload(op='aten.gradient', overload='scalararray')>, <OpOverload(op='aten.gradient', overload='scalarint')>, <OpOverload(op='aten.gradient', overload='scalarrayarray')>, <OpOverload(op='aten.gradient', overload='scalarrayint')>, <OpOverload(op='aten.gradient', overload='tensorarray')>, <OpOverload(op='aten.gradient', overload='tensorarrayint')>, <OpOverload(op='aten.greater', overload='Scalar')>, <OpOverload(op='aten.greater', overload='Tensor')>, <OpOverload(op='aten.greater_equal', overload='Scalar')>, <OpOverload(op='aten.greater_equal', overload='Tensor')>, <OpOverload(op='aten.grid_sampler', overload='default')>, <OpOverload(op='aten.group_norm', overload='default')>, <OpOverload(op='aten.gru', overload='data')>, <OpOverload(op='aten.gru', overload='input')>, <OpOverload(op='aten.gru_cell', overload='default')>, <OpOverload(op='aten.hinge_embedding_loss', overload='default')>, <OpOverload(op='aten.histogramdd', overload='TensorList_bins')>, <OpOverload(op='aten.histogramdd', overload='default')>, <OpOverload(op='aten.histogramdd', overload='int_bins')>, <OpOverload(op='aten.hstack', overload='default')>, <OpOverload(op='aten.index_add', overload='dimname')>, <OpOverload(op='aten.index_copy', overload='dimname')>, <OpOverload(op='aten.index_fill', overload='Dimname_Scalar')>, <OpOverload(op='aten.index_fill', overload='Dimname_Tensor')>, <OpOverload(op='aten.index_select', overload='dimname')>, <OpOverload(op='aten.index_select_backward', overload='default')>, <OpOverload(op='aten.infinitely_differentiable_gelu_backward', overload='default')>, <OpOverload(op='aten.inner', overload='default')>, <OpOverload(op='aten.instance_norm', overload='default')>, <OpOverload(op='aten.inverse', overload='default')>, <OpOverload(op='aten.is_complex', overload='default')>, <OpOverload(op='aten.is_conj', overload='default')>, <OpOverload(op='aten.is_distributed', overload='default')>, <OpOverload(op='aten.is_floating_point', overload='default')>, <OpOverload(op='aten.is_inference', overload='default')>, <OpOverload(op='aten.is_leaf', overload='default')>, <OpOverload(op='aten.is_neg', overload='default')>, <OpOverload(op='aten.is_nonzero', overload='default')>, <OpOverload(op='aten.is_signed', overload='default')>, <OpOverload(op='aten.is_vulkan_available', overload='default')>, <OpOverload(op='aten.isclose', overload='default')>, <OpOverload(op='aten.isfinite', overload='default')>, <OpOverload(op='aten.isreal', overload='default')>, <OpOverload(op='aten.istft', overload='default')>, <OpOverload(op='aten.item', overload='default')>, <OpOverload(op='aten.kl_div', overload='default')>, <OpOverload(op='aten.kron', overload='default')>, <OpOverload(op='aten.kthvalue', overload='dimname')>, <OpOverload(op='aten.l1_loss', overload='default')>, <OpOverload(op='aten.layer_norm', overload='default')>, <OpOverload(op='aten.ldexp', overload='Tensor')>, <OpOverload(op='aten.less', overload='Scalar')>, <OpOverload(op='aten.less', overload='Tensor')>, <OpOverload(op='aten.less_equal', overload='Scalar')>, <OpOverload(op='aten.less_equal', overload='Tensor')>, <OpOverload(op='aten.linalg_cholesky', overload='default')>, <OpOverload(op='aten.linalg_cond', overload='default')>, <OpOverload(op='aten.linalg_cond', overload='p_str')>, <OpOverload(op='aten.linalg_det', overload='default')>, <OpOverload(op='aten.linalg_eigh', overload='default')>, <OpOverload(op='aten.linalg_eigvals', overload='default')>, <OpOverload(op='aten.linalg_eigvalsh', overload='default')>, <OpOverload(op='aten.linalg_inv', overload='default')>, <OpOverload(op='aten.linalg_ldl_factor', overload='default')>, <OpOverload(op='aten.linalg_lu_factor', overload='default')>, <OpOverload(op='aten.linalg_matmul', overload='default')>, <OpOverload(op='aten.linalg_matrix_norm', overload='default')>, <OpOverload(op='aten.linalg_matrix_norm', overload='str_ord')>, <OpOverload(op='aten.linalg_matrix_power', overload='default')>, <OpOverload(op='aten.linalg_matrix_rank', overload='atol_rtol_float')>, <OpOverload(op='aten.linalg_matrix_rank', overload='atol_rtol_tensor')>, <OpOverload(op='aten.linalg_matrix_rank', overload='default')>, <OpOverload(op='aten.linalg_matrix_rank', overload='tol_tensor')>, <OpOverload(op='aten.linalg_multi_dot', overload='default')>, <OpOverload(op='aten.linalg_norm', overload='default')>, <OpOverload(op='aten.linalg_norm', overload='ord_str')>, <OpOverload(op='aten.linalg_pinv', overload='atol_rtol_float')>, <OpOverload(op='aten.linalg_pinv', overload='default')>, <OpOverload(op='aten.linalg_pinv', overload='rcond_tensor')>, <OpOverload(op='aten.linalg_slogdet', overload='default')>, <OpOverload(op='aten.linalg_solve', overload='default')>, <OpOverload(op='aten.linalg_solve_ex', overload='default')>, <OpOverload(op='aten.linalg_svd', overload='default')>, <OpOverload(op='aten.linalg_svdvals', overload='default')>, <OpOverload(op='aten.linalg_tensorinv', overload='default')>, <OpOverload(op='aten.linalg_tensorsolve', overload='default')>, <OpOverload(op='aten.linalg_vander', overload='default')>, <OpOverload(op='aten.linalg_vecdot', overload='default')>, <OpOverload(op='aten.linear', overload='default')>, <OpOverload(op='aten.log_sigmoid', overload='default')>, <OpOverload(op='aten.log_softmax', overload='Dimname')>, <OpOverload(op='aten.log_softmax', overload='int')>, <OpOverload(op='aten.logcumsumexp', overload='dimname')>, <OpOverload(op='aten.logdet', overload='default')>, <OpOverload(op='aten.logsumexp', overload='names')>, <OpOverload(op='aten.lstm', overload='data')>, <OpOverload(op='aten.lstm', overload='input')>, <OpOverload(op='aten.lstm_cell', overload='default')>, <OpOverload(op='aten.lu_solve', overload='default')>, <OpOverload(op='aten.margin_ranking_loss', overload='default')>, <OpOverload(op='aten.masked_select_backward', overload='default')>, <OpOverload(op='aten.matmul', overload='default')>, <OpOverload(op='aten.matrix_exp', overload='default')>, <OpOverload(op='aten.matrix_exp_backward', overload='default')>, <OpOverload(op='aten.matrix_power', overload='default')>, <OpOverload(op='aten.max', overload='names_dim')>, <OpOverload(op='aten.max', overload='other')>, <OpOverload(op='aten.max_pool1d', overload='default')>, <OpOverload(op='aten.max_pool1d_with_indices', overload='default')>, <OpOverload(op='aten.max_pool2d', overload='default')>, <OpOverload(op='aten.max_pool3d', overload='default')>, <OpOverload(op='aten.mean', overload='names_dim')>, <OpOverload(op='aten.median', overload='names_dim')>, <OpOverload(op='aten.meshgrid', overload='default')>, <OpOverload(op='aten.meshgrid', overload='indexing')>, <OpOverload(op='aten.min', overload='names_dim')>, <OpOverload(op='aten.min', overload='other')>, <OpOverload(op='aten.mish_backward', overload='default')>, <OpOverload(op='aten.mode', overload='dimname')>, <OpOverload(op='aten.msort', overload='default')>, <OpOverload(op='aten.multilabel_margin_loss', overload='default')>, <OpOverload(op='aten.multiply', overload='Scalar')>, <OpOverload(op='aten.multiply', overload='Tensor')>, <OpOverload(op='aten.nanmean', overload='default')>, <OpOverload(op='aten.nanmedian', overload='names_dim')>, <OpOverload(op='aten.nanquantile', overload='default')>, <OpOverload(op='aten.nanquantile', overload='scalar')>, <OpOverload(op='aten.native_channel_shuffle', overload='default')>, <OpOverload(op='aten.negative', overload='default')>, <OpOverload(op='aten.nested_to_padded_tensor', overload='default')>, <OpOverload(op='aten.nll_loss', overload='default')>, <OpOverload(op='aten.nll_loss2d', overload='default')>, <OpOverload(op='aten.nll_loss_nd', overload='default')>, <OpOverload(op='aten.nonzero_numpy', overload='default')>, <OpOverload(op='aten.norm', overload='names_ScalarOpt_dim')>, <OpOverload(op='aten.norm', overload='names_ScalarOpt_dim_dtype')>, <OpOverload(op='aten.norm_except_dim', overload='default')>, <OpOverload(op='aten.not_equal', overload='Scalar')>, <OpOverload(op='aten.not_equal', overload='Tensor')>, <OpOverload(op='aten.nuclear_norm', overload='default')>, <OpOverload(op='aten.nuclear_norm', overload='dim')>, <OpOverload(op='aten.one_hot', overload='default')>, <OpOverload(op='aten.orgqr', overload='default')>, <OpOverload(op='aten.outer', overload='default')>, <OpOverload(op='aten.output_nr', overload='default')>, <OpOverload(op='aten.pad', overload='default')>, <OpOverload(op='aten.pad_sequence', overload='default')>, <OpOverload(op='aten.pairwise_distance', overload='default')>, <OpOverload(op='aten.pdist', overload='default')>, <OpOverload(op='aten.pinverse', overload='default')>, <OpOverload(op='aten.poisson_nll_loss', overload='default')>, <OpOverload(op='aten.prelu', overload='default')>, <OpOverload(op='aten.prod', overload='dim_Dimname')>, <OpOverload(op='aten.promote_types', overload='default')>, <OpOverload(op='aten.qr', overload='default')>, <OpOverload(op='aten.quantile', overload='default')>, <OpOverload(op='aten.quantile', overload='scalar')>, <OpOverload(op='aten.quantized_gru_cell', overload='default')>, <OpOverload(op='aten.quantized_lstm_cell', overload='default')>, <OpOverload(op='aten.quantized_rnn_relu_cell', overload='default')>, <OpOverload(op='aten.quantized_rnn_tanh_cell', overload='default')>, <OpOverload(op='aten.relu6', overload='default')>, <OpOverload(op='aten.repeat_interleave', overload='self_Tensor')>, <OpOverload(op='aten.repeat_interleave', overload='self_int')>, <OpOverload(op='aten.result_type', overload='Scalar')>, <OpOverload(op='aten.result_type', overload='Scalar_Scalar')>, <OpOverload(op='aten.result_type', overload='Scalar_Tensor')>, <OpOverload(op='aten.result_type', overload='Tensor')>, <OpOverload(op='aten.retains_grad', overload='default')>, <OpOverload(op='aten.rms_norm', overload='default')>, <OpOverload(op='aten.rnn_relu', overload='data')>, <OpOverload(op='aten.rnn_relu', overload='input')>, <OpOverload(op='aten.rnn_relu_cell', overload='default')>, <OpOverload(op='aten.rnn_tanh', overload='data')>, <OpOverload(op='aten.rnn_tanh', overload='input')>, <OpOverload(op='aten.rnn_tanh_cell', overload='default')>, <OpOverload(op='aten.row_stack', overload='default')>, <OpOverload(op='aten.rrelu', overload='default')>, <OpOverload(op='aten.scaled_dot_product_attention', overload='default')>, <OpOverload(op='aten.scatter', overload='dimname_src')>, <OpOverload(op='aten.scatter', overload='dimname_value')>, <OpOverload(op='aten.scatter_add', overload='dimname')>, <OpOverload(op='aten.selu', overload='default')>, <OpOverload(op='aten.silu_backward', overload='default')>, <OpOverload(op='aten.size', overload='Dimname')>, <OpOverload(op='aten.size', overload='int')>, <OpOverload(op='aten.slogdet', overload='default')>, <OpOverload(op='aten.slow_conv3d', overload='default')>, <OpOverload(op='aten.smm', overload='default')>, <OpOverload(op='aten.softmax', overload='Dimname')>, <OpOverload(op='aten.softmax', overload='int')>, <OpOverload(op='aten.sort', overload='dimname')>, <OpOverload(op='aten.sort', overload='dimname_stable')>, <OpOverload(op='aten.sparse_bsc_tensor', overload='ccol_row_value')>, <OpOverload(op='aten.sparse_bsc_tensor', overload='ccol_row_value_size')>, <OpOverload(op='aten.sparse_bsr_tensor', overload='crow_col_value')>, <OpOverload(op='aten.sparse_bsr_tensor', overload='crow_col_value_size')>, <OpOverload(op='aten.sparse_coo_tensor', overload='indices')>, <OpOverload(op='aten.sparse_coo_tensor', overload='indices_size')>, <OpOverload(op='aten.sparse_csc_tensor', overload='ccol_row_value')>, <OpOverload(op='aten.sparse_csc_tensor', overload='ccol_row_value_size')>, <OpOverload(op='aten.sparse_csr_tensor', overload='crow_col_value')>, <OpOverload(op='aten.sparse_csr_tensor', overload='crow_col_value_size')>, <OpOverload(op='aten.special_digamma', overload='default')>, <OpOverload(op='aten.special_erf', overload='default')>, <OpOverload(op='aten.special_erfc', overload='default')>, <OpOverload(op='aten.special_erfinv', overload='default')>, <OpOverload(op='aten.special_exp2', overload='default')>, <OpOverload(op='aten.special_expit', overload='default')>, <OpOverload(op='aten.special_expm1', overload='default')>, <OpOverload(op='aten.special_gammainc', overload='default')>, <OpOverload(op='aten.special_gammaincc', overload='default')>, <OpOverload(op='aten.special_gammaln', overload='default')>, <OpOverload(op='aten.special_i0', overload='default')>, <OpOverload(op='aten.special_log1p', overload='default')>, <OpOverload(op='aten.special_log_softmax', overload='default')>, <OpOverload(op='aten.special_logit', overload='default')>, <OpOverload(op='aten.special_logsumexp', overload='default')>, <OpOverload(op='aten.special_multigammaln', overload='default')>, <OpOverload(op='aten.special_ndtr', overload='default')>, <OpOverload(op='aten.special_polygamma', overload='default')>, <OpOverload(op='aten.special_psi', overload='default')>, <OpOverload(op='aten.special_round', overload='default')>, <OpOverload(op='aten.special_sinc', overload='default')>, <OpOverload(op='aten.special_softmax', overload='default')>, <OpOverload(op='aten.special_xlogy', overload='default')>, <OpOverload(op='aten.special_xlogy', overload='other_scalar')>, <OpOverload(op='aten.special_xlogy', overload='self_scalar')>, <OpOverload(op='aten.square', overload='default')>, <OpOverload(op='aten.sspaddmm', overload='default')>, <OpOverload(op='aten.std', overload='correction_names')>, <OpOverload(op='aten.std', overload='default')>, <OpOverload(op='aten.std', overload='dim')>, <OpOverload(op='aten.std', overload='names_dim')>, <OpOverload(op='aten.std_mean', overload='correction_names')>, <OpOverload(op='aten.std_mean', overload='default')>, <OpOverload(op='aten.std_mean', overload='dim')>, <OpOverload(op='aten.std_mean', overload='names_dim')>, <OpOverload(op='aten.stft', overload='center')>, <OpOverload(op='aten.stft', overload='default')>, <OpOverload(op='aten.stride', overload='Dimname')>, <OpOverload(op='aten.stride', overload='int')>, <OpOverload(op='aten.subtract', overload='Scalar')>, <OpOverload(op='aten.subtract', overload='Tensor')>, <OpOverload(op='aten.sum', overload='dim_DimnameList')>, <OpOverload(op='aten.sum_to_size', overload='default')>, <OpOverload(op='aten.svd', overload='default')>, <OpOverload(op='aten.sym_size', overload='int')>, <OpOverload(op='aten.sym_stride', overload='int')>, <OpOverload(op='aten.take_along_dim', overload='default')>, <OpOverload(op='aten.tensordot', overload='default')>, <OpOverload(op='aten.thnn_conv2d', overload='default')>, <OpOverload(op='aten.tile', overload='default')>, <OpOverload(op='aten.to_dense', overload='default')>, <OpOverload(op='aten.to_dense_backward', overload='default')>, <OpOverload(op='aten.to_mkldnn_backward', overload='default')>, <OpOverload(op='aten.to_sparse', overload='default')>, <OpOverload(op='aten.to_sparse', overload='sparse_dim')>, <OpOverload(op='aten.to_sparse_bsc', overload='default')>, <OpOverload(op='aten.to_sparse_bsr', overload='default')>, <OpOverload(op='aten.to_sparse_csc', overload='default')>, <OpOverload(op='aten.to_sparse_csr', overload='default')>, <OpOverload(op='aten.trace_backward', overload='default')>, <OpOverload(op='aten.trapezoid', overload='dx')>, <OpOverload(op='aten.trapezoid', overload='x')>, <OpOverload(op='aten.trapz', overload='dx')>, <OpOverload(op='aten.trapz', overload='x')>, <OpOverload(op='aten.triplet_margin_loss', overload='default')>, <OpOverload(op='aten.true_divide', overload='Scalar')>, <OpOverload(op='aten.true_divide', overload='Tensor')>, <OpOverload(op='aten.type_as', overload='default')>, <OpOverload(op='aten.unflatten_dense_tensors', overload='default')>, <OpOverload(op='aten.upsample_bicubic2d', overload='vec')>, <OpOverload(op='aten.upsample_bilinear2d', overload='vec')>, <OpOverload(op='aten.upsample_linear1d', overload='vec')>, <OpOverload(op='aten.upsample_nearest1d', overload='default')>, <OpOverload(op='aten.upsample_nearest1d', overload='vec')>, <OpOverload(op='aten.upsample_nearest2d', overload='default')>, <OpOverload(op='aten.upsample_nearest2d', overload='vec')>, <OpOverload(op='aten.upsample_nearest3d', overload='default')>, <OpOverload(op='aten.upsample_nearest3d', overload='vec')>, <OpOverload(op='aten.upsample_trilinear3d', overload='vec')>, <OpOverload(op='aten.value_selecting_reduction_backward', overload='default')>, <OpOverload(op='aten.vander', overload='default')>, <OpOverload(op='aten.var', overload='correction_names')>, <OpOverload(op='aten.var', overload='default')>, <OpOverload(op='aten.var', overload='dim')>, <OpOverload(op='aten.var', overload='names_dim')>, <OpOverload(op='aten.var_mean', overload='correction_names')>, <OpOverload(op='aten.var_mean', overload='default')>, <OpOverload(op='aten.var_mean', overload='dim')>, <OpOverload(op='aten.var_mean', overload='names_dim')>, <OpOverload(op='aten.vstack', overload='default')>, <OpOverload(op='aten.where', overload='Scalar')>, <OpOverload(op='aten.where', overload='ScalarOther')>, <OpOverload(op='aten.where', overload='ScalarSelf')>, <OpOverload(op='aten.where', overload='default')>, <OpOverload(op='aten.wrapped_linear_prepack', overload='default')>, <OpOverload(op='aten.wrapped_quantized_linear_prepacked', overload='default')> ``` Pull Request resolved: pytorch#136153 Approved by: https://github.com/xadupre, https://github.com/gramalingam
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
5 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Since #135080, the CompositeImplicitAutograd (CIA) ops are only decomposed when a decomp function is provided in a table. There is no longer a need to distinguish CIA ops like Upsample and preserve them explicitly. On the ONNX Script torchlib side I will unregister some ops from the following list to make sure some CIA ops are still decomposed.