未验证 提交 3a159ca0 编写于 作者: W WJJ1995 提交者: GitHub

add div2scale pass for huggingface (#813)

上级 2050891f
......@@ -42,3 +42,5 @@ from .onnx_layernorm_fuser import LayerNormFuser
from .onnx_layernorm_fuse_pass import LayerNormFusePass
from .onnx_gelu_fuser import GeluFuser
from .onnx_gelu_fuse_pass import GeluFusePass
from .replace_div_to_scale import Div2Scale
from .replace_div_to_scale_pass import Div2ScalePass
# Copyright (c) 2022 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 copy
import numpy as np
from collections import OrderedDict
from x2paddle.optimizer.pattern_matcher import FuseBase
from x2paddle.core.program import PaddleGraph, PaddleLayer
from x2paddle.core.util import *
class Div2Scale(FuseBase):
def __init__(self):
super(Div2Scale, self).__init__()
def build_pattern(self):
"""
code describe:
x2paddle_296 = paddle.full(dtype='float32', shape=[1], fill_value=8.0)
x2paddle_293 = paddle.transpose(x=x2paddle_292, perm=[0, 2, 1, 3])
x2paddle_294 = paddle.transpose(x=x2paddle_260, perm=[0, 2, 3, 1])
x2paddle_295 = paddle.matmul(x=x2paddle_293, y=x2paddle_294)
x2paddle_297 = paddle.divide(x=x2paddle_295, y=x2paddle_296)
"""
def gen_name(id):
return "x" + str(id)
self.pattern.add_layer(
"paddle.full",
inputs={},
outputs=[gen_name(0)],
shape=[1],
fill_value=8)
self.pattern.add_layer(
"paddle.transpose",
inputs={"x": "div2scale-input-0"},
outputs=[gen_name(1)],
perm=[0, 2, 1, 3])
self.pattern.add_layer(
"paddle.transpose",
inputs={"x": "div2scale-input-1"},
outputs=[gen_name(2)],
perm=[0, 2, 1, 3])
self.pattern.add_layer(
"paddle.matmul",
inputs={"x": gen_name(1),
"y": gen_name(2)},
outputs=[gen_name(3)])
self.pattern.add_layer(
"paddle.divide",
inputs={"x": gen_name(3),
"y": gen_name(0)},
outputs=[gen_name(4)])
self.pattern.build(inputs={
"input-0": "div2scale-input-0",
"input-1": "div2scale-input-1",
})
def insert_new_layer(self, graph, parameters, matches):
new_layer, new_layer_id = self.gen_new_layer(parameters, matches)
graph.layers[new_layer_id] = new_layer
matches_copy = copy.deepcopy(matches)
for layer_id, layer in matches_copy.items():
if layer.kernel in ["paddle.transpose", "paddle.matmul"]:
matches.pop(layer_id)
matches.pop(new_layer_id)
def gen_new_layer(self, parameters, matches):
layer_id_list = list(matches.keys())
layer_id_list.sort(key=int)
layer_inputs = list()
layer_inputs_ids = list()
fill_value = 0
for layer_id, layer in matches.items():
if layer.kernel == "paddle.full":
fill_value = layer.attrs["fill_value"]
if layer.kernel == "paddle.divide":
layer_inputs.append(layer.inputs["x"])
layer_inputs_ids.append(layer_id)
output_name = layer.outputs[0]
new_layer = PaddleLayer(
layer_id_list[0],
"paddle.scale",
inputs={"x": layer_inputs[0]},
outputs=[output_name],
scale=1 / fill_value)
return new_layer, layer_inputs_ids[0]
# Copyright (c) 2022 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.
from x2paddle.optimizer.pass_ import Pass
from x2paddle.optimizer.fusion import Div2Scale
from x2paddle.optimizer.pass_manager import pass_register
@pass_register
class Div2ScalePass(Pass):
name = "replace_div_to_scale_pass"
def __init__(self):
Pass.__init__(self)
def apply(self, graph):
fuser = Div2Scale()
fuser.operate(graph, match_kind="edge")
# register huggingface div2scale pass
replace_div_to_scale_pass = Div2ScalePass()
......@@ -37,7 +37,11 @@ class GraphOptimizer(object):
"prelu_fuse_pass", "transpose_eliminate_pass"
]
elif source_frame == "onnx":
self.passes = ["onnx_layernorm_fuse_pass", "onnx_gelu_fuse_pass"]
self.passes = [
"onnx_layernorm_fuse_pass",
"onnx_gelu_fuse_pass",
"replace_div_to_scale_pass",
]
else:
self.passes = []
......@@ -54,7 +58,7 @@ class GraphOptimizer(object):
before_len = len(graph.layers)
pass_.apply(graph)
after_len = len(graph.layers)
if after_len < before_len:
if after_len <= before_len:
show_pass_log = True
if before_len == after_len:
break
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册