dropout_fuser.py 2.5 KB
Newer Older
S
SunAhong1993 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
#   Copyright (c) 2020  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 numpy as np
S
SunAhong1993 已提交
16
from x2paddle.optimizer.pattern_matcher import FuseBase
S
SunAhong1993 已提交
17 18 19 20
from x2paddle.core.program import PaddleGraph, PaddleLayer
from x2paddle.core.util import *


S
renam  
SunAhong1993 已提交
21
class DygraphDropoutFuser(FuseBase):
S
SunAhong1993 已提交
22
    def __init__(self):
S
renam  
SunAhong1993 已提交
23
        super(DygraphDropoutFuser, self).__init__(graph_type="dygraph")
S
SunAhong1993 已提交
24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

    def build_pattern(self):
        """ 描述需要替换的constant图结构。
        constant层模式python实现代码示例:
            x3 = 10
            for _x70 in range(x3):
                ...
        """
        self.pattern.add_layer(
            "paddle.nn.Dropout",
            inputs={"input": "dropout-input-0"},
            outputs=["dropout0", "x1"])
        self.pattern.build(inputs={"input-0": "dropout-input-0"})
        self.pattern.outputs = ["dropout0", "x1"]

    def insert_new_layer(self, graph, parameters, matches):
        def replace_value(layer_connect, match_name, match_input):
            for k, v in layer_connect.inputs.items():
                if v == match_name:
                    layer_connect.inputs[k] = match_input
                    break
            if layer_connect.kernel == "prim.loop" or \
            layer_connect.kernel == "prim.if":
                for block in layer_connect.blocks:
                    for b_layer_id, b_layer in block.layers.items():
                        if block.edges_in.get(b_layer_id, 0) != 0 and  \
                        -1 in block.edges_in[b_layer_id]:
                            replace_value(b_layer, match_name, match_input)

        layer_id = list(matches.keys())[0]
        layer = list(matches.values())[0]
        layer_output_name = layer.outputs[1]
        layer_input = layer.inputs["input"]
        if graph.edges_out.get(layer_id, 0) != 0:
            for layer_id_out in graph.edges_out[layer_id]:
                layer_connect = graph.layers[layer_id_out]
                replace_value(layer_connect, layer_output_name, layer_input)