# 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. from __future__ import print_function from __future__ import division 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 import sys import os import six import pickle import numpy as np class PaddleLayer(object): def __init__(self, id, kernel, inputs, outputs, scope_name="", **kwargs): assert isinstance( inputs, dict), "parameter 'inputs' for PaddleLayer should be type of dict" assert isinstance( outputs, 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" self.kernel = kernel self.inputs = inputs self.outputs = outputs self.scope_name = scope_name self.attrs = kwargs self.id = id self.blocks = list() def add_block(self, block): self.blocks.append(block) class PaddleGraph(object): def __init__(self, source_type=None, parent_layer=None, graph_type="static"): self.layers = OrderedDict() self.edges_out = dict() self.edges_in = dict() self.inputs = list() self.outputs = list() self.parameters = dict() self.parent_layer = parent_layer self.graph_type = graph_type self.source_type = source_type self.custom_code = None self.inputs_info = None def set_name(self, name): self.name = name.replace("-", "_").replace("/", "_") def set_parameters(self, parameters): self.parameters = parameters def set_custom(self, custom_code): self.custom_code = custom_code def set_inputs_info(self, inputs_info): self.inputs_info = inputs_info def set_script(self, script): self.script = script 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() def add_layer(self, kernel, inputs, outputs, scope_name="", **kwargs): 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, scope_name=scope_name, **kwargs) self.layers[layer_id] = layer return layer_id 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] 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) for output in layer.outputs: outputs_from_nodes[output] = layer_id # 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) # 处理子图 if len(layer.blocks) > 0: for block in layer.blocks: block.build(layer.inputs, layer.outputs) # 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) if self.graph_type == "dygraph": self.get_dygraph_inputs() if len(self.outputs) == 0: self.get_dygraph_outputs() 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) def gen_model(self, save_dir, jit_type=None): if not os.path.exists(save_dir): os.makedirs(save_dir) if self.graph_type == "static": self.gen_static_model(save_dir) else: self.gen_dygraph_model(save_dir, jit_type) def gen_static_model(self, save_dir): code_dir = os.path.join(save_dir, 'model_with_code') infer_dir = os.path.join(save_dir, 'inference_model') self.gen_static_code(code_dir) sys.path.append(code_dir) import x2paddle_model paddle.enable_static() scope = paddle.static.Scope() startup_program = paddle.static.Program() main_program = paddle.static.Program() with paddle.static.scope_guard(scope): with paddle.static.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) def gen_dygraph_model(self, save_dir, jit_type=None): if jit_type == "trace": from x2paddle.optimizer.code_optimizer import HierarchicalTree hierarchical_tree = HierarchicalTree(self) for layer_id, layer in self.layers.items(): hierarchical_tree.insert(layer) hierarchical_tree.save_source_files(save_dir) self.dump_dygraph_parameter(save_dir) else: self.gen_dygraph_code(save_dir) self.dump_dygraph_parameter(save_dir) input_shapes = list() input_types = list() for input_name in self.inputs: input_shapes.append(self.inputs_info[input_name][0]) input_types.append(self.inputs_info[input_name][1]) # 如果input_files非空，则导出推理模型；其值类似[[None, 3, 224, 224]] self.dygraph2static(save_dir, input_shapes, input_types) def gen_static_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') if not os.path.exists(code_dir): os.makedirs(code_dir) f = open(os.path.join(code_dir, 'x2paddle_model.py'), 'w') write_code( f, [ "from paddle.fluid.initializer import Constant", "from paddle.fluid.param_attr import ParamAttr", "import paddle.fluid as fluid", "import paddle", "import math", "", ], indent=0) if self.custom_code is not None: write_code( f, list(self.custom_code.values()), indent=0) write_code(f, ["", "def x2paddle_net():"], indent=0) write_code( f, [ "paddle.enable_static()" ], indent=1) 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: continue line = "" if len(layer.outputs) == 1: line = layer.outputs[0] else: for output in layer.outputs: line += "{}, ".format(output) line = line.strip(", ") if layer.kernel.startswith("custom_layer"): line += " = {}(".format(layer.kernel.split(":")[-1].lower() + "_layer") else: line += " = {}(".format(layer.kernel) for k, v in layer.inputs.items(): if isinstance(v, list): line += "{}=[{}], ".format(k, ", ".join(v)) else: line += "{}={}, ".format(k, v) for k, v in layer.attrs.items(): line += "{}={}, ".format(k, v) line = line.strip(", ") line += ")" write_code(f, [line], indent=1) write_code( f, [ "return [{}], [{}]".format(", ".join(self.inputs), ", ".join(self.outputs)) ], indent=1) f.close() 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 == "paddle.to_tensor": data = layer.attrs["data"] if not data.startswith("params["): self.inputs.append(data) 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 i, output_name in enumerate(layer.outputs): if ("paddle.nn" in layer.kernel and "functional" not in layer.kernel) or \ (layer.kernel == "paddle.to_tensor" and layer.attrs["data"].startswith("params["))or \ "paddle.fluid.dygraph" in layer.kernel: if i == 0: continue if output_name not in self.outputs: self.outputs.append(output_name) 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(): if self.source_type == "caffe": custom_import = "from x2paddle.op_mapper.dygraph.caffe2paddle " + \ "import caffe_custom_layer as x2paddle_nn" else: custom_import = "" self.head = gen_codes( [ "from paddle.fluid.initializer import Constant", "from paddle.fluid.param_attr import ParamAttr", "import paddle", "import paddle.fluid as fluid", custom_import, "", "class {}(paddle.nn.Layer):".format(self.name), ], indent=0) input_data_name = ', '.join(self.inputs) self.init_func.extend( gen_codes( ["def __init__(self):"], 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 gen_main_code(code_dir): input_data_name = ', '.join(self.inputs) self.run_func = gen_codes( [ "", "def main({}):".format(input_data_name), ], indent=0) comment_list = list() comment_list.append("# 共{}个输入".format(len(self.inputs_info))) for k, v in self.inputs_info.items(): comment_list.append("# {}: 形状为{}，类型为{}。".format(k, v[0], v[1])) self.run_func.extend( gen_codes( comment_list, indent=1)) use_structured_name = False if self.source_type in ["tf", "onnx"] else True self.run_func.extend( gen_codes(["paddle.disable_static()", "params, _ = fluid.load_dygraph('{}/model')".format(code_dir), "model = {}()".format(self.name), "model.set_dict(params, use_structured_name={})".format(use_structured_name), "model.eval()", "out = model({})".format(input_data_name), "return out"], 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) for code_line in self.run_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 == "paddle.to_tensor" or \ "paddle.fluid.dygraph" in layer.kernel or \ layer.kernel.startswith("custom_layer"): line = "{}".format( layer.outputs[0] ) if layer.kernel == "paddle.to_tensor" and not layer.attrs[ "data"].startswith("params[") else "self.{}".format( layer.outputs[0]) if layer.kernel.startswith("custom_layer"): line += "= x2paddle_nn.{}(".format(layer.kernel.split(":")[-1]) else: line += " = {}(".format(layer.kernel) for k, v in layer.attrs.items(): line += "{}={}, ".format(k, v) line = line.strip(", ") line += ")" if layer.kernel == "paddle.to_tensor" and not layer.attrs[ "data"].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 == "paddle.to_tensor" and layer.attrs[ "data"].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.dygraph 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(): if isinstance(v, list): line += "{}=[{}], ".format(k, ", ".join(v)) else: line += "{}={}, ".format(k, v) for k, v in layer.attrs.items(): line += "{}={}, ".format(k, v) line = line.strip(", ") line += ")" if layer.kernel == "self.create_parameter": self.init_func.extend(gen_codes(["self." + line], indent=indent)) self.forward_func.extend(gen_codes(["{} = self.{}".format(layer.outputs[0], layer.outputs[0])], indent=indent)) else: self.forward_func.extend(gen_codes([line], indent=indent)) if indent == 2: gen_main_code(code_dir) 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() def dygraph2static(self, save_dir, input_shapes=[], input_types=[]): from paddle.fluid.dygraph.jit import declarative sepc_list = list() for i, name in enumerate(self.inputs): input_shapes[i][0] = -1 sepc_list.append( paddle.static.InputSpec( shape=input_shapes[i], name=name, dtype=input_types[i])) import sys path = osp.abspath(save_dir) sys.path.insert(0, save_dir) import x2paddle_code paddle.disable_static() restore, _ = fluid.load_dygraph(osp.join(save_dir, "model")) model = getattr(x2paddle_code, self.name)() if self.source_type in ["tf", "onnx"]: model.set_dict(restore, use_structured_name=False) else: model.set_dict(restore) model.eval() static_model = paddle.jit.to_static(model, input_spec=sepc_list) paddle.jit.save(static_model, osp.join(save_dir, "inference_model/model"))