program.py 21.1 KB
Newer Older
S
SunAhong1993 已提交
1
# -*- coding:UTF-8 -*-
S
SunAhong1993 已提交
2
#   Copyright (c) 2020  PaddlePaddle Authors. All Rights Reserved.
J
jiangjiajun 已提交
3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
#
# 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 __future__ import print_function
from __future__ import division
S
SunAhong1993 已提交
18 19 20 21 22 23
import paddle.fluid as fluid
import os.path as osp
import paddle
from paddle.fluid.proto import framework_pb2
from collections import OrderedDict
import numpy
J
jiangjiajun 已提交
24
import collections
S
SunAhong1993 已提交
25
import sys
J
jiangjiajun 已提交
26
import os
S
SunAhong1993 已提交
27 28
import six
import pickle
J
jiangjiajun 已提交
29 30 31


class PaddleLayer(object):
S
SunAhong1993 已提交
32
    def __init__(self, id, kernel, inputs, outputs, **kwargs):
J
jiangjiajun 已提交
33 34 35 36 37
        assert isinstance(
            inputs,
            dict), "parameter 'inputs' for PaddleLayer should be type of dict"
        assert isinstance(
            outputs,
S
SunAhong1993 已提交
38 39 40 41 42 43 44 45 46 47 48 49 50 51 52
            list), "parameter 'outputs' for PaddleLayer should be type of list"
        for k, v in inputs.items():
            if isinstance(v, list):
                for i in v:
                    assert isinstance(
                        i, six.string_types
                    ), "value in inputs should be type of string or list of string"
            else:
                assert isinstance(v, six.string_types) or isinstance(
                    v, list
                ), "value in inputs should be type of string or list of string"
        for v in outputs:
            assert isinstance(
                v, six.
                string_types), "elements in outputs should be type of string"
J
jiangjiajun 已提交
53 54 55 56
        self.kernel = kernel
        self.inputs = inputs
        self.outputs = outputs
        self.attrs = kwargs
S
SunAhong1993 已提交
57 58 59 60 61
        self.id = id
        self.blocks = list()

    def add_block(self, block):
        self.blocks.append(block)
J
jiangjiajun 已提交
62 63


S
SunAhong1993 已提交
64 65 66
class PaddleGraph(object):
    def __init__(self, parent_layer=None, graph_type="static"):
        self.layers = OrderedDict()
J
jiangjiajun 已提交
67 68 69 70 71
        self.edges_out = dict()
        self.edges_in = dict()
        self.inputs = list()
        self.outputs = list()
        self.parameters = dict()
S
SunAhong1993 已提交
72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91
        self.parent_layer = parent_layer
        self.graph_type = graph_type

    def set_name(self, name):
        self.name = name

    def set_parameters(self, parameters):
        self.parameters = parameters

    def clear(self):
        self.layers = OrderedDict()
        self.edges_out = dict()
        self.edges_in = dict()
        self.inputs = list()
        self.outputs = list()
        self.parameters = dict()

    def clear_edges(self):
        self.edges_out = dict()
        self.edges_in = dict()
J
jiangjiajun 已提交
92 93

    def add_layer(self, kernel, inputs, outputs, **kwargs):
S
SunAhong1993 已提交
94 95 96 97 98 99 100 101
        layer_id = str(len(self.layers))
        if self.parent_layer is not None:
            layer_id = "{}.{}.{}".format(self.parent_layer.id,
                                         len(self.parent_layer.blocks),
                                         layer_id)
        layer = PaddleLayer(layer_id, kernel, inputs, outputs, **kwargs)
        self.layers[layer_id] = layer
        return layer_id
J
jiangjiajun 已提交
102

J
jiangjiajun 已提交
103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149
    def del_layer(self, layer_id):
        layer = self.layers[layer_id]
        outputs = self.edges_out.get(layer_id, [])
        inputs = self.edges_in.get(layer_id, [])

        assert len(
            inputs) <= 1, "There should be 0 or 1 input for deleted layer."

        if len(inputs) == 0:
            for out in outputs:
                while layer_id in self.edges_in[out]:
                    index = self.edges_in[out].index(layer_id)
                    del self.edges_in[out][index]

                input_keys = list(self.layers[out].inputs.keys())
                for k in input_keys:
                    if self.layers[out].inputs[k] == layer.outputs[0]:
                        del self.layers[out].inputs[k]

            del self.layers[layer_id]
            if layer_id in self.edges_in:
                del self.edges_in[layer_id]
            if layer_id in self.edges_out:
                del self.edges_out[layer_id]
            return

        # 将所有输出layer的输入layer进行替换
        for out in outputs:
            for i in range(len(self.edges_in[out])):
                if self.edges_in[out][i] == layer_id:
                    self.edges_in[out][i] = inputs[0]

        # 将输出layer赋给输入layer的输出
        replace_index = self.edges_out[inputs[0]].index(layer_id)
        del self.edges_out[inputs[0]][replace_index]
        for i, out in enumerate(outputs):
            self.edges_out[inputs[0]].insert(replace_index + i, out)
            for k, v in self.layers[out].inputs.items():
                if v == layer.outputs[0]:
                    self.layers[out].inputs[k] = list(layer.inputs.values())[0]

        del self.layers[layer_id]
        if layer_id in self.edges_out:
            del self.edges_out[layer_id]
        if layer_id in self.edges_in:
            del self.edges_in[layer_id]

S
SunAhong1993 已提交
150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175
    def build(self, inputs=None, outputs=None):
        self.clear_edges()
        outputs_from_nodes = dict()
        for layer_id, layer in self.layers.items():
            for input_key, input_var in layer.inputs.items():
                vs = input_var
                if not isinstance(vs, list):
                    vs = [vs]
                for v in vs:
                    assert v in outputs_from_nodes or (
                        inputs is not None and v in list(inputs.values())
                    ) or (
                        outputs is not None and v in outputs
                    ), "Couldn't find {} in previous layers, the layers should be make by topological sort".format(
                        v)
                    if v in outputs_from_nodes:
                        in_layer_id = outputs_from_nodes[v]
                    else:
                        in_layer_id = -1
                    if in_layer_id not in self.edges_out:
                        self.edges_out[in_layer_id] = list()
                    self.edges_out[in_layer_id].append(layer_id)

                    if layer_id not in self.edges_in:
                        self.edges_in[layer_id] = list()
                    self.edges_in[layer_id].append(in_layer_id)
J
jiangjiajun 已提交
176
            for output in layer.outputs:
S
SunAhong1993 已提交
177
                outputs_from_nodes[output] = layer_id
J
jiangjiajun 已提交
178

S
SunAhong1993 已提交
179 180 181 182 183 184
            # 将block的输出用于父图
            if inputs is not None and outputs is not None and set(
                    layer.outputs).issubset(outputs):
                if layer_id not in self.edges_out:
                    self.edges_out[layer_id] = list()
                self.edges_out[layer_id].append(-1)
J
jiangjiajun 已提交
185

S
SunAhong1993 已提交
186 187 188 189
            # 处理子图
            if len(layer.blocks) > 0:
                for block in layer.blocks:
                    block.build(layer.inputs, layer.outputs)
J
jiangjiajun 已提交
190

S
SunAhong1993 已提交
191 192 193 194 195 196 197 198 199 200 201
        # 删除不必要的节点
        invalid_list = list()
        for layer_id, layer in self.layers.items():
            if len(self.layers) > 1:
                if self.edges_in.get(layer_id, 0) == 0 and self.edges_out.get(
                        layer_id, 0) == 0 and layer.kernel != "prim.assert" \
                        and layer.kernel != "prim.exception" \
                        and layer.kernel != "prim.warnings":
                    invalid_list.append(layer_id)
        for layer_id in invalid_list:
            self.layers.pop(layer_id)
J
jiangjiajun 已提交
202

S
SunAhong1993 已提交
203 204 205 206
        if self.graph_type == "dygraph":
            self.get_dygraph_inputs()
            if len(self.outputs) == 0:
                self.get_dygraph_outputs()
J
jiangjiajun 已提交
207

S
SunAhong1993 已提交
208 209 210 211 212 213 214 215 216 217 218 219
    def get_global_layers(self):
        # 该全局layers的信息是按照拓扑排序组成的
        def update(layers):
            global_layers = dict()
            for layer_id, layer in layers.items():
                global_layers[layer_id] = layer
                for block in layer.blocks:
                    block_global_layers = update(block.layers)
                    global_layers.update(block_global_layers)
            return global_layers

        return update(self.layers)
J
jiangjiajun 已提交
220 221 222 223 224 225 226 227 228 229

    def gen_code(self, code_dir):
        def write_code(f, code_list, indent=0):
            indent_blank = "    " * indent
            for code_line in code_list:
                if code_line.strip() == "":
                    f.write('\n')
                else:
                    f.write(indent_blank + code_line + '\n')

S
SunAhong1993 已提交
230 231 232
        if not os.path.exists(code_dir):
            os.makedirs(code_dir)
        f = open(os.path.join(code_dir, 'x2paddle_model.py'), 'w')
J
jiangjiajun 已提交
233 234 235 236 237

        write_code(
            f, [
                "from paddle.fluid.initializer import Constant",
                "from paddle.fluid.param_attr import ParamAttr",
S
SunAhong1993 已提交
238 239
                "import paddle.fluid as fluid", "import math", "",
                "def x2paddle_net():"
J
jiangjiajun 已提交
240 241
            ],
            indent=0)
S
SunAhong1993 已提交
242 243 244 245
        for layer_id, layer in self.layers.items():
            edges_in = self.edges_in.get(layer_id, [])
            edges_out = self.edges_out.get(layer_id, [])
            if len(edges_in) == 0 and len(edges_out) == 0:
J
jiangjiajun 已提交
246 247 248 249 250 251 252 253 254 255 256 257 258
                continue

            line = ""

            if len(layer.outputs) == 1:
                line = layer.outputs[0]
            else:
                for output in layer.outputs:
                    line += "{}, ".format(output)
                line = line.strip(", ")

            line += " = {}(".format(layer.kernel)
            for k, v in layer.inputs.items():
S
SunAhong1993 已提交
259 260 261 262
                if isinstance(v, list):
                    line += "{}=[{}], ".format(k, ", ".join(v))
                else:
                    line += "{}={}, ".format(k, v)
J
jiangjiajun 已提交
263 264 265 266 267
            for k, v in layer.attrs.items():
                line += "{}={}, ".format(k, v)
            line = line.strip(", ")
            line += ")"
            write_code(f, [line], indent=1)
S
SunAhong1993 已提交
268 269 270 271 272 273 274

        write_code(
            f, [
                "return [{}], [{}]".format(", ".join(self.inputs),
                                           ", ".join(self.outputs))
            ],
            indent=1)
J
jiangjiajun 已提交
275 276
        f.close()

S
SunAhong1993 已提交
277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514
    def gen_model(self, save_dir, input_shapes=None):
        if not os.path.exists(save_dir):
            os.makedirs(save_dir)
        if self.graph_type == "static":
            code_dir = os.path.join(save_dir, 'model_with_code')
            infer_dir = os.path.join(save_dir, 'inference_model')
            self.gen_code(code_dir)
            sys.path.append(code_dir)
            import x2paddle_model
            scope = fluid.Scope()
            startup_program = fluid.Program()
            main_program = fluid.Program()
            with fluid.scope_guard(scope):
                with fluid.program_guard(main_program, startup_program):
                    inputs, outputs = x2paddle_model.x2paddle_net()
                    exe = fluid.Executor(fluid.CPUPlace())
                    exe.run(startup_program)

                    param_dir = os.path.join(code_dir, 'weights')
                    for k, v in self.parameters.items():
                        if scope.find_var(k):
                            self.dump_parameter(k, v, param_dir)

                    def if_exist(var):
                        b = os.path.exists(
                            os.path.join(os.path.join(param_dir, var.name)))
                        return b

                    fluid.io.load_vars(
                        exe, param_dir, main_program, predicate=if_exist)
                    fluid.io.save_inference_model(
                        dirname=infer_dir,
                        feeded_var_names=[i.name for i in inputs],
                        target_vars=outputs,
                        executor=exe)
        else:
            self.gen_dygraph_code(save_dir)
            self.dump_dygraph_parameter(save_dir)
            if input_shapes is not None:
                # 如果input_shapes非空,则导出推理模型;其值类似[[None, 3, 224, 224]]
                self.dygraph2static(save_dir, input_shapes)

    def dump_parameter(self, param_name, param, save_dir):
        if not os.path.exists(save_dir):
            os.makedirs(save_dir)
        dtype_map = {
            "int16": [framework_pb2.VarType.INT16, 'h'],
            "int32": [framework_pb2.VarType.INT32, 'i'],
            "int64": [framework_pb2.VarType.INT64, 'q'],
            "float16": [framework_pb2.VarType.FP16, 'e'],
            "float32": [framework_pb2.VarType.FP32, 'f'],
            "float64": [framework_pb2.VarType.FP64, 'd'],
            "bool": [framework_pb2.VarType.BOOL, None]
        }
        shape = param.shape
        if str(param.dtype) in ['uint8', 'uint_8', 'bool']:
            param = param.astype('int64')
        if len(shape) == 0:
            assert param.size == 1, "Unexpected situation happend!"
            shape = [1]
        assert str(
            param.dtype) in dtype_map, "Unknown dtype {} of params: {}.".format(
                str(param.dtype), param_name)
        fp = open(os.path.join(save_dir, param_name), 'wb')
        numpy.array([0], dtype='int32').tofile(fp)
        numpy.array([0], dtype='int64').tofile(fp)
        numpy.array([0], dtype='int32').tofile(fp)
        tensor_desc = framework_pb2.VarType.TensorDesc()
        tensor_desc.data_type = dtype_map[str(param.dtype)][0]
        tensor_desc.dims.extend(shape)
        desc_size = tensor_desc.ByteSize()
        numpy.array([desc_size], dtype='int32').tofile(fp)
        fp.write(tensor_desc.SerializeToString())
        param.tofile(fp)
        fp.close()

    def get_dygraph_inputs(self):
        def update(layers):
            for layer_id, layer in layers.items():
                if self.edges_in.get(layer_id, 0) == 0 and self.edges_out.get(
                        layer_id, 0) == 0:
                    continue
                if layer.kernel == "fluid.dygraph.base.to_variable":
                    value = layer.attrs["value"]
                    if not value.startswith("params["):
                        self.inputs.append(value)
                if len(layer.blocks) > 0:
                    for block in layer.blocks:
                        block.get_dygraph_inputs()
                        self.inputs.extend(block.inputs)

        update(self.layers)
        self.inputs = list(set(self.inputs))
        if self.inputs is not None:
            self.inputs.sort()

    def get_dygraph_outputs(self):
        for layer_id, layer in self.layers.items():
            if self.edges_in.get(layer_id, 0) == 0 and self.edges_out.get(
                    layer_id, 0) == 0:
                continue
            if self.edges_out.get(layer_id, 0) == 0:
                for output_name in layer.outputs:
                    if not output_name.startswith("x"):
                        continue
                    self.outputs.append(output_name)
        self.outputs = list(set(self.outputs))

    def gen_dygraph_code(self, code_dir=None, indent=2):
        def gen_codes(code_list, indent=0):
            indent_blank = "    " * indent
            codes = []
            for code_line in code_list:
                if code_line.strip() == "":
                    codes.append('\n')
                else:
                    codes.append(indent_blank + code_line + '\n')
            return codes

        def gen_head():
            self.head = gen_codes(
                [
                    "from paddle.fluid.initializer import Constant",
                    "from paddle.fluid.param_attr import ParamAttr",
                    "import paddle",
                    "import paddle.fluid as fluid",
                    "",
                    "class {}(fluid.dygraph.Layer):".format(self.name),
                ],
                indent=0)
            input_data_name = ', '.join(self.inputs)
            self.init_func.extend(
                gen_codes(
                    ["def __init__(self, params):"], indent=1))
            self.init_func.extend(
                gen_codes(
                    ["super({}, self).__init__()".format(self.name)], indent=2))
            self.forward_func.extend(
                gen_codes(
                    ["def forward(self, {}):".format(input_data_name)],
                    indent=1))

        def write_code(code_dir):
            f = open(os.path.join(code_dir, 'x2paddle_code.py'), 'w')
            for code_line in self.head:
                f.write(code_line)
            init_writen_codes = []
            for code_line in self.init_func:
                if code_line in init_writen_codes:
                    continue
                f.write(code_line)
                init_writen_codes.append(code_line)
            f.write("\n")
            return_code = "return {}".format(", ".join(self.outputs))
            self.forward_func.extend(gen_codes([return_code], indent=2))
            for code_line in self.forward_func:
                f.write(code_line)
            f.close()

        self.init_func = []
        self.forward_func = []
        if indent == 2 and code_dir is not None:
            gen_head()

        for layer_id, layer in self.layers.items():
            if ("paddle.nn" in layer.kernel and "functional" not in layer.kernel
                ) or layer.kernel == "fluid.dygraph.base.to_variable" or \
               "paddle.fluid.dygraph" in layer.kernel:
                line = "{}".format(
                    layer.outputs[0]
                ) if layer.kernel == "fluid.dygraph.base.to_variable" and not layer.attrs[
                    "value"].startswith("params[") else "self.{}".format(
                        layer.outputs[0])
                line += " = {}(".format(layer.kernel)
                for k, v in layer.attrs.items():
                    line += "{}={}, ".format(k, v)
                line = line.strip(", ")
                line += ")"

                if layer.kernel == "fluid.dygraph.base.to_variable" and not layer.attrs[
                        "value"].startswith("params["):
                    self.forward_func.extend(gen_codes([line], indent=indent))
                    continue
                else:
                    self.init_func.extend(gen_codes([line], indent=2))

                if len(layer.outputs) == 1:
                    line = layer.outputs[0]
                elif len(layer.outputs) == 2:
                    line = layer.outputs[1]
                else:
                    line = ','.join(layer.outputs[1:])
                if layer.kernel == "fluid.dygraph.base.to_variable" and layer.attrs[
                        "value"].startswith("params["):
                    line += " = self.{}".format(layer.outputs[0])
                else:
                    line += " = self.{}(".format(layer.outputs[0])
                    for k, v in layer.inputs.items():
                        line += "{}, ".format(v)
                    line = line.strip(", ")
                    line += ")"
                self.forward_func.extend(gen_codes([line], indent=indent))
            elif "prim" in layer.kernel:
                func_name = layer.kernel.replace(".", "_")
                from x2paddle.op_mapper.pytorch2paddle import prim2code
                if hasattr(prim2code, func_name):
                    func = getattr(prim2code, func_name)
                    func(
                        layer,
                        indent=indent,
                        init_func=self.init_func,
                        forward_func=self.forward_func)
                else:
                    raise Exception(
                        "The kind {} in paddle model is not supported yet.".
                        format(layer.kernel))
            else:
                if len(layer.outputs) == 1:
                    line = layer.outputs[0]
                else:
                    line = ','.join(layer.outputs)
                line += " = {}(".format(layer.kernel)
                for k, v in layer.inputs.items():
                    line += "{}={}, ".format(k, v)
                for k, v in layer.attrs.items():
                    line += "{}={}, ".format(k, v)
                line = line.strip(", ")
                line += ")"
                self.forward_func.extend(gen_codes([line], indent=indent))
        if indent == 2:
            write_code(code_dir)
        else:
            return self.init_func, self.forward_func

    def dump_dygraph_parameter(self, code_dir):
        params_output = open(os.path.join(code_dir, 'model.pdparams'), 'wb')
        pickle.dump(self.parameters, params_output)
        params_output.close()
J
jiangjiajun 已提交
515

S
SunAhong1993 已提交
516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535
    def dygraph2static(self, save_dir, input_shapes=[]):
        from paddle.fluid.dygraph.jit import declarative
        sepc_list = list()
        for i, name in enumerate(self.inputs):
            sepc_list.append(
                paddle.static.InputSpec(
                    shape=input_shapes[i], name=name))
        import sys
        path = osp.abspath(save_dir)
        sys.path.insert(0, save_dir)
        import x2paddle_code
        place = fluid.CPUPlace()
        with fluid.dygraph.guard(place):
            restore, _ = fluid.load_dygraph(osp.join(save_dir, "model"))
            model = getattr(x2paddle_code, self.name)(restore)
            model.set_dict(restore)
            model.eval()
            model.forward = declarative(model.forward, sepc_list)
        fluid.dygraph.jit.save(
            layer=model, model_path=osp.join(save_dir, "inference"))