paddlepaddle--paddle/test/dygraph_to_static/test_tensor_attr_consistency.py

# Copyright (c) 2024 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

import unittest

import paddle
from paddle.jit.utils import OrderedSet

DYGRAPH_ONLY_TENSOR_ATTRS_ALLOW_LIST = OrderedSet(
    [
        '__array__',
        '__array_ufunc__',
        '__deepcopy__',
        '__index__',
        '__len__',
        '__nonzero__',
        '__dict__',
        'apply_',
        'backward',
        'clear_grad',
        'coalesce',
        'col_indices',
        'cols',
        'copy_',
        'crow_indices',
        'crows',
        'data',
        'data_ptr',
        'detach_',
        'fill_',
        'fill_diagonal_',
        'fill_diagonal_tensor',
        'fill_diagonal_tensor_',
        'get_map_tensor',
        'get_selected_rows',
        'get_strides',
        'get_tensor',
        'grad',
        'grad_',
        'grad_fn',
        'gradient',
        'inplace_version',
        'is_dense',
        'is_dist',
        'is_leaf',
        'is_same_shape',
        'is_selected_rows',
        'is_sparse',
        'is_sparse_coo',
        'is_sparse_csr',
        'layout',
        'nnz',
        'num_shard',
        'offset',
        'pin_memory',
        'placements',
        'process_mesh',
        'reconstruct_from_',
        'retain_grads',
        'rows',
        'set_string_list',
        'set_value',
        'set_tensor',
        'set_vocab',
        'strides',
        'to_sparse',
        'to_sparse_coo',
        'to_sparse_csr',
        'value',
        'zero_',
        "__cuda_array_interface__",
        '__dlpack__',
        "__dlpack_device__",
        "__tvm_ffi_env_stream__",
        "__c_dlpack_exchange_api__",
        "__dlpack_c_exchange_api__",
        "retain_grad",
        "sparse_mask",
    ]
)
STATIC_ONLY_TENSOR_ATTRS_ALLOW_LIST = OrderedSet(
    [
        '__dict__',
        'all_used_ops',
        'append',
        'first_use',
        'index',
        'get_defining_op',
        'has_one_use',
        'has_name',
        'hash',
        'id',
        'initialized',
        'is_dense_tensor_array_type',
        'is_dense_tensor_type',
        'is_same',
        'is_selected_row_type',
        'pop',
        'replace_all_uses_with',
        'set_shape',
        'set_type',
        'use_empty',
        'is_dist_dense_tensor_type',
        'dist_attr',
        'place_attr',
        'update_dist_attr',
        'is_combine',
        'value_assign',
        'replace_grad_users_with',
        'do_model_average',
        'is_distributed',
        'is_parameter',
        'need_clip',
        'optimize_attr',
        'regularizer',
        'trainable',
        'is_sparse_coo_tensor_type',
        'is_sparse_csr_tensor_type',
        'all_used_ops_in_same_block',
    ]
)


def is_magic_method(attr):
    return attr.startswith('__') and attr.endswith('__')


def get_public_attributes(cls):
    return [
        attr
        for attr in dir(cls)
        if not attr.startswith('_') or is_magic_method(attr)
    ]


class TestTensorAttrConsistency(unittest.TestCase):
    def get_tensor_dygraph_and_static_attrs(self):
        dygraph_tensor_cls = paddle.Tensor
        static_tensor_cls = paddle.pir.Value
        dygraph_tensor_attrs = get_public_attributes(dygraph_tensor_cls)
        static_tensor_attrs = get_public_attributes(static_tensor_cls)
        return OrderedSet(dygraph_tensor_attrs), OrderedSet(static_tensor_attrs)

    def test_dygraph_tensor_attr_consistency_check(self):
        (
            dygraph_tensor_attrs,
            static_tensor_attrs,
        ) = self.get_tensor_dygraph_and_static_attrs()
        dygraph_only_attrs = dygraph_tensor_attrs - static_tensor_attrs
        not_allow_dygraph_only_attrs = (
            dygraph_only_attrs - DYGRAPH_ONLY_TENSOR_ATTRS_ALLOW_LIST
        )

        self.assertEqual(
            len(not_allow_dygraph_only_attrs),
            0,
            f"Value should have same attributes as Tensor, but found dygraph only tensor attributes: {not_allow_dygraph_only_attrs}.\n"
            + "If you believe these attributes are not supported in static graph, please add them to DYGRAPH_ONLY_TENSOR_ATTRS_ALLOW_LIST in test_dygraph_to_static/test_tensor_attr_consistency.py.",
        )

    def test_static_tensor_attr_consistency_check(self):
        (
            dygraph_tensor_attrs,
            static_tensor_attrs,
        ) = self.get_tensor_dygraph_and_static_attrs()
        static_only_attrs = static_tensor_attrs - dygraph_tensor_attrs
        not_allow_static_only_attrs = (
            static_only_attrs - STATIC_ONLY_TENSOR_ATTRS_ALLOW_LIST
        )

        self.assertEqual(
            len(not_allow_static_only_attrs),
            0,
            f"Tensor should have same attributes as Value, but found static only tensor attributes: {not_allow_static_only_attrs}.\n"
            + "If you believe these attributes are not supported in dygraph, please add them to STATIC_ONLY_TENSOR_ATTRS_ALLOW_LIST in test_dygraph_to_static/test_tensor_attr_consistency.py.",
        )


if __name__ == '__main__':
    unittest.main()