#coding:utf-8 # Copyright (c) 2019 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 absolute_import from __future__ import division from __future__ import print_function import os import time import sys import functools from shutil import copyfile import paddle import paddle.fluid as fluid from paddlehub.common import utils from paddlehub.common import paddle_helper from paddlehub.common.logger import logger from paddlehub.common.lock import lock from paddlehub.common.downloader import default_downloader from paddlehub.module import module_desc_pb2 from paddlehub.common.dir import CONF_HOME from paddlehub.module import check_info_pb2 from paddlehub.module.signature import Signature, create_signature from paddlehub.module.checker import ModuleChecker from paddlehub.module.manager import default_module_manager from paddlehub.module.base_processor import BaseProcessor from paddlehub.io.parser import yaml_parser from paddlehub import version __all__ = ['Module', 'create_module'] # PaddleHub module dir name ASSETS_DIRNAME = "assets" MODEL_DIRNAME = "model" MODULE_DESC_PBNAME = "module_desc.pb" PYTHON_DIR = "python" PROCESSOR_NAME = "processor" # PaddleHub var prefix HUB_VAR_PREFIX = "@HUB_%s@" def create_module(sign_arr, module_dir, processor=None, assets=None, module_info=None, exe=None, extra_info=None): sign_arr = utils.to_list(sign_arr) module = Module( signatures=sign_arr, processor=processor, assets=assets, module_info=module_info, extra_info=extra_info) module.serialize_to_path(path=module_dir, exe=exe) class ModuleHelper(object): def __init__(self, module_dir): self.module_dir = module_dir def module_desc_path(self): return os.path.join(self.module_dir, MODULE_DESC_PBNAME) def model_path(self): return os.path.join(self.module_dir, MODEL_DIRNAME) def processor_path(self): return os.path.join(self.module_dir, PYTHON_DIR) def processor_name(self): return PROCESSOR_NAME def assets_path(self): return os.path.join(self.module_dir, ASSETS_DIRNAME) class Module(object): def __init__(self, name=None, module_dir=None, signatures=None, module_info=None, assets=None, processor=None, extra_info=None, version=None): self.desc = module_desc_pb2.ModuleDesc() self.program = None self.assets = [] self.helper = None self.signatures = {} self.default_signature = None self.module_info = None self.processor = None self.extra_info = {} if extra_info is None else extra_info if not isinstance(self.extra_info, dict): raise TypeError( "The extra_info should be an instance of python dict") # cache data self.last_call_name = None self.cache_feed_dict = None self.cache_fetch_dict = None self.cache_program = None fp_lock = open(os.path.join(CONF_HOME, 'config.json')) lock.flock(fp_lock, lock.LOCK_EX) if name: self._init_with_name(name=name, version=version) lock.flock(fp_lock, lock.LOCK_UN) elif module_dir: self._init_with_module_file(module_dir=module_dir[0]) lock.flock(fp_lock, lock.LOCK_UN) elif signatures: if processor: if not issubclass(processor, BaseProcessor): raise TypeError( "Processor shoule be an instance of paddlehub.BaseProcessor" ) if assets: self.assets = utils.to_list(assets) # for asset in assets: # utils.check_path(assets) self.processor = processor self._generate_module_info(module_info) self._init_with_signature(signatures=signatures) lock.flock(fp_lock, lock.LOCK_UN) else: lock.flock(fp_lock, lock.LOCK_UN) raise ValueError("Module initialized parameter is empty") def _init_with_name(self, name, version=None): log_msg = "Installing %s module" % name if version: log_msg += "-%s" % version logger.info(log_msg) extra = {"command": "install"} result, tips, module_dir = default_module_manager.install_module( module_name=name, module_version=version, extra=extra) if not result: logger.error(tips) raise RuntimeError(tips) else: logger.info(tips) self._init_with_module_file(module_dir[0]) def _init_with_url(self, url): utils.check_url(url) result, tips, module_dir = default_downloader.download_file_and_uncompress( url, save_path=".") if not result: logger.error(tips) raise RuntimeError(tips) else: self._init_with_module_file(module_dir) def _dump_processor(self): import inspect pymodule = inspect.getmodule(self.processor) pycode = inspect.getsource(pymodule) processor_path = self.helper.processor_path() processor_md5 = utils.md5(pycode) processor_md5 += str(time.time()) processor_name = utils.md5(processor_md5) output_file = os.path.join(processor_path, processor_name + ".py") utils.mkdir(processor_path) with open(output_file, "w") as file: file.write(pycode) utils.from_pyobj_to_module_attr( processor_name, self.desc.attr.map.data['processor_info']) def _load_processor(self): processor_path = self.helper.processor_path() if os.path.exists(processor_path): sys.path.append(processor_path) processor_name = utils.from_module_attr_to_pyobj( self.desc.attr.map.data['processor_info']) self.processor = __import__(processor_name).Processor(module=self) else: self.processor = None def _dump_assets(self): utils.mkdir(self.helper.assets_path()) for asset in self.assets: filename = os.path.basename(asset) newfile = os.path.join(self.helper.assets_path(), filename) copyfile(asset, newfile) def _load_assets(self): assets_path = self.helper.assets_path() self.assets = [] for file in os.listdir(assets_path): filepath = os.path.join(self.helper.assets_path(), file) self.assets.append(filepath) def _init_with_module_file(self, module_dir): checker = ModuleChecker(module_dir) checker.check() self.helper = ModuleHelper(module_dir) with open(self.helper.module_desc_path(), "rb") as fi: self.desc.ParseFromString(fi.read()) exe = fluid.Executor(fluid.CPUPlace()) self.program, _, _ = fluid.io.load_inference_model( self.helper.model_path(), executor=exe) for block in self.program.blocks: for op in block.ops: if "op_callstack" in op.all_attrs(): op._set_attr("op_callstack", [""]) self._load_processor() self._load_assets() self._recover_from_desc() self._generate_sign_attr() self._generate_extra_info() self._restore_parameter(self.program) self._recover_variable_info(self.program) def _init_with_signature(self, signatures): self.name_prefix = HUB_VAR_PREFIX % self.name self._process_signatures(signatures) self._check_signatures() self._generate_desc() self._generate_sign_attr() self._generate_extra_info() def _init_with_program(self, program): pass def _process_signatures(self, signatures): self.signatures = {} self.program = signatures[0].inputs[0].block.program for sign in signatures: if sign.name in self.signatures: raise ValueError( "Error! Signature array contains duplicated signatrues %s" % sign) if self.default_signature is None and sign.for_predict: self.default_signature = sign self.signatures[sign.name] = sign def _restore_parameter(self, program): global_block = program.global_block() param_attrs = self.desc.attr.map.data['param_attrs'] for key, param_attr in param_attrs.map.data.items(): param = paddle_helper.from_module_attr_to_param(param_attr) param['name'] = self.get_var_name_with_prefix(key) if (param['name'] not in global_block.vars): continue var = global_block.var(param['name']) global_block.create_parameter( shape=var.shape, dtype=var.dtype, type=var.type, lod_level=var.lod_level, error_clip=var.error_clip, stop_gradient=var.stop_gradient, is_data=var.is_data, **param) def _recover_variable_info(self, program): var_infos = self.desc.attr.map.data['var_infos'] for var_info in var_infos.map.data: idx = utils.from_module_attr_to_pyobj( var_infos.map.data[var_info].map.data['block_id']) stop_gradient = utils.from_module_attr_to_pyobj( var_infos.map.data[var_info].map.data['stop_gradient']) block = program.blocks[idx] var_name = self.get_var_name_with_prefix(var_info) if var_name in block.vars: var = block.vars[var_name] var.stop_gradient = stop_gradient def get_extra_info(self, key): return self.extra_info.get(key, None) def _generate_extra_info(self): for key in self.extra_info: self.__dict__["get_%s" % key] = functools.partial( self.get_extra_info, key=key) def _generate_module_info(self, module_info=None): if not module_info: self.module_info = {} else: if not utils.is_yaml_file(module_info): logger.critical("Module info file should be yaml format") exit(1) self.module_info = yaml_parser.parse(module_info) self.author = self.module_info.get('author', 'UNKNOWN') self.author_email = self.module_info.get('author_email', 'UNKNOWN') self.summary = self.module_info.get('summary', 'UNKNOWN') self.type = self.module_info.get('type', 'UNKNOWN') self.version = self.module_info.get('version', 'UNKNOWN') self.name = self.module_info.get('name', 'UNKNOWN') def _generate_sign_attr(self): self._check_signatures() for sign in self.signatures: self.__dict__[sign] = functools.partial( self.__call__, sign_name=sign) def get_vocab_path(self): for assets_file in self.assets: if "vocab.txt" in assets_file: return assets_file def _recover_from_desc(self): # recover signature for sign, module_var in self.desc.sign2var.items(): inputs = [] outputs = [] feed_names = [] fetch_names = [] for var in module_var.feed_desc: variable = self.program.global_block().vars[var.var_name] inputs.append(variable) feed_names.append(var.alias) for var in module_var.fetch_desc: variable = self.program.global_block().vars[var.var_name] outputs.append(variable) fetch_names.append(var.alias) self.signatures[sign] = create_signature( sign, inputs=inputs, outputs=outputs, feed_names=feed_names, fetch_names=fetch_names) # recover default signature default_signature_name = utils.from_module_attr_to_pyobj( self.desc.attr.map.data['default_signature']) self.default_signature = self.signatures[ default_signature_name] if default_signature_name else None # recover module info module_info = self.desc.attr.map.data['module_info'] self.name = utils.from_module_attr_to_pyobj( module_info.map.data['name']) self.author = utils.from_module_attr_to_pyobj( module_info.map.data['author']) self.author_email = utils.from_module_attr_to_pyobj( module_info.map.data['author_email']) self.version = utils.from_module_attr_to_pyobj( module_info.map.data['version']) self.type = utils.from_module_attr_to_pyobj( module_info.map.data['type']) self.summary = utils.from_module_attr_to_pyobj( module_info.map.data['summary']) # recover extra info extra_info = self.desc.attr.map.data['extra_info'] self.extra_info = {} for key, value in extra_info.map.data.items(): self.extra_info[key] = utils.from_module_attr_to_pyobj(value) # recover name prefix self.name_prefix = utils.from_module_attr_to_pyobj( self.desc.attr.map.data["name_prefix"]) def _generate_desc(self): # save fluid Parameter attr = self.desc.attr attr.type = module_desc_pb2.MAP param_attrs = attr.map.data['param_attrs'] param_attrs.type = module_desc_pb2.MAP for param in self.program.global_block().iter_parameters(): param_attr = param_attrs.map.data[param.name] paddle_helper.from_param_to_module_attr(param, param_attr) # save Variable Info var_infos = attr.map.data['var_infos'] var_infos.type = module_desc_pb2.MAP for block in self.program.blocks: for var in block.vars.values(): var_info = var_infos.map.data[var.name] var_info.type = module_desc_pb2.MAP utils.from_pyobj_to_module_attr( var.stop_gradient, var_info.map.data['stop_gradient']) utils.from_pyobj_to_module_attr(block.idx, var_info.map.data['block_id']) # save signarture info for key, sign in self.signatures.items(): var = self.desc.sign2var[sign.name] feed_desc = var.feed_desc fetch_desc = var.fetch_desc feed_names = sign.feed_names fetch_names = sign.fetch_names for index, input in enumerate(sign.inputs): feed_var = feed_desc.add() feed_var.var_name = self.get_var_name_with_prefix(input.name) feed_var.alias = feed_names[index] for index, output in enumerate(sign.outputs): fetch_var = fetch_desc.add() fetch_var.var_name = self.get_var_name_with_prefix(output.name) fetch_var.alias = fetch_names[index] # save default signature utils.from_pyobj_to_module_attr( self.default_signature.name if self.default_signature else None, attr.map.data['default_signature']) # save name prefix utils.from_pyobj_to_module_attr(self.name_prefix, self.desc.attr.map.data["name_prefix"]) # save module info module_info = attr.map.data['module_info'] module_info.type = module_desc_pb2.MAP utils.from_pyobj_to_module_attr(self.name, module_info.map.data['name']) utils.from_pyobj_to_module_attr(self.version, module_info.map.data['version']) utils.from_pyobj_to_module_attr(self.author, module_info.map.data['author']) utils.from_pyobj_to_module_attr(self.author_email, module_info.map.data['author_email']) utils.from_pyobj_to_module_attr(self.type, module_info.map.data['type']) utils.from_pyobj_to_module_attr(self.summary, module_info.map.data['summary']) # save extra info extra_info = attr.map.data['extra_info'] extra_info.type = module_desc_pb2.MAP for key, value in self.extra_info.items(): utils.from_pyobj_to_module_attr(value, extra_info.map.data[key]) def __call__(self, sign_name, data, use_gpu=False, batch_size=1, **kwargs): self.check_processor() def _get_reader_and_feeder(data_format, data, place): def _reader(process_data): for item in zip(*process_data): yield item process_data = [] feed_name_list = [] for key in data_format: process_data.append([value['processed'] for value in data[key]]) feed_name_list.append(data_format[key]['feed_key']) feeder = fluid.DataFeeder(feed_list=feed_name_list, place=place) return functools.partial(_reader, process_data=process_data), feeder if self.last_call_name != sign_name: self.last_call_name = sign_name self.cache_feed_dict, self.cache_fetch_dict, self.cache_program = self.context( sign_name, for_test=True) feed_dict = self.cache_feed_dict fetch_dict = self.cache_fetch_dict program = self.cache_program fetch_list = list(set([value for key, value in fetch_dict.items()])) with fluid.program_guard(program): result = [] index = 0 try: _places = os.environ["CUDA_VISIBLE_DEVICES"] int(_places[0]) except: use_gpu = False place = fluid.CUDAPlace(0) if use_gpu else fluid.CPUPlace() exe = fluid.Executor(place=place) data = self.processor.preprocess( sign_name=sign_name, data_dict=data) data_format = self.processor.data_format(sign_name=sign_name) reader, feeder = _get_reader_and_feeder(data_format, data, place) reader = paddle.batch(reader, batch_size=batch_size) for batch in reader(): data_out = exe.run( feed=feeder.feed(batch), fetch_list=fetch_list, return_numpy=False) sub_data = { key: value[index:index + len(batch)] for key, value in data.items() } result += self.processor.postprocess(sign_name, data_out, sub_data, **kwargs) index += len(batch) return result def check_processor(self): if not self.processor: raise ValueError("This Module is not callable!") def context(self, sign_name=None, for_test=False, trainable=True, regularizer=None, max_seq_len=128, learning_rate=1e-3): """ Args: max_seq_len(int): maximum sequence length, this option is only available for BERT/ERNIE module """ if sign_name: if sign_name not in self.signatures: raise KeyError( "Module did not have a signature with name %s" % sign_name) signature = self.signatures[sign_name] else: inputs = [ input for signature in self.signatures.values() for input in signature.inputs ] outputs = [ output for signature in self.signatures.values() for output in signature.outputs ] feed_names = [ feed_name for signature in self.signatures.values() for feed_name in signature.feed_names ] fetch_names = [ fetch_name for signature in self.signatures.values() for fetch_name in signature.fetch_names ] signature = create_signature( name="hub_temp_signature", inputs=inputs, outputs=outputs, feed_names=feed_names, fetch_names=fetch_names, for_predict=False) program = self.program.clone(for_test=for_test) paddle_helper.remove_feed_fetch_op(program) if not for_test: paddle_helper.set_parameter_trainable(program, trainable) paddle_helper.set_parameter_learning_rate(program, learning_rate) paddle_helper.set_parameter_regularizer(program, regularizer) self._restore_parameter(program) self._recover_variable_info(program) paddle_helper.set_op_attr(program, is_test=for_test) feed_dict = {} fetch_dict = {} for index, var in enumerate(signature.inputs): feed_dict[index] = program.global_block().var(var.name) key = signature.feed_names[index] if key: feed_dict[key] = program.global_block().var(var.name) for index, var in enumerate(signature.outputs): fetch_dict[index] = program.global_block().var(var.name) key = signature.fetch_names[index] if key: fetch_dict[key] = program.global_block().var(var.name) # update BERT/ERNIE's input tensor's sequence length to max_seq_len if "bert" in self.name or self.name.startswith("ernie"): MAX_SEQ_LENGTH = 512 if max_seq_len > MAX_SEQ_LENGTH or max_seq_len <= 0: raise ValueError( "max_seq_len({}) should be in the range of [1, {}]".format( MAX_SEQ_LENGTH)) logger.info( "Set maximum sequence length of input tensor to {}".format( max_seq_len)) if self.name.startswith("ernie_v2"): feed_list = [ "input_ids", "position_ids", "segment_ids", "input_mask", "task_ids" ] logger.warning( "%s will exploite task_id, the arguement use_taskid of Reader class must be True." % self.name) else: feed_list = [ "input_ids", "position_ids", "segment_ids", "input_mask" ] logger.warning( "%s has no task_id, the arguement use_taskid of Reader class must be False." % self.name) for tensor_name in feed_list: seq_tensor_shape = [-1, max_seq_len, 1] logger.info("The shape of input tensor[{}] set to {}".format( tensor_name, seq_tensor_shape)) program.global_block().var( feed_dict[tensor_name].name).desc.set_shape( seq_tensor_shape) # record num parameters loaded by paddlehub num_param_loaded = 0 for param in program.global_block().iter_parameters(): num_param_loaded += 1 logger.info( "%d pretrained paramaters loaded by PaddleHub" % num_param_loaded) return feed_dict, fetch_dict, program def get_name_prefix(self): return self.name_prefix def get_var_name_with_prefix(self, var_name): return self.get_name_prefix() + var_name def _check_signatures(self): if not self.signatures: raise ValueError("Signatures should not be None") for key, sign in self.signatures.items(): if not isinstance(sign, Signature): raise TypeError( "Item in Signatures shoule be an instance of paddlehub.Signature" ) for input in sign.inputs: _tmp_program = input.block.program if not self.program == _tmp_program: raise ValueError( "All input and outputs variables in signature should come from the same Program" ) for output in sign.outputs: _tmp_program = output.block.program if not self.program == _tmp_program: raise ValueError( "All input and outputs variables in signature should come from the same Program" ) def serialize_to_path(self, path=None, exe=None): self._check_signatures() self._generate_desc() # create module path for saving if path is None: path = os.path.join(".", self.name) self.helper = ModuleHelper(path) utils.mkdir(self.helper.module_dir) # create module pb module_desc = module_desc_pb2.ModuleDesc() logger.info("PaddleHub version = %s" % version.hub_version) logger.info("PaddleHub Module proto version = %s" % version.module_proto_version) logger.info("Paddle version = %s" % paddle.__version__) feeded_var_names = [ input.name for key, sign in self.signatures.items() for input in sign.inputs ] target_vars = [ output for key, sign in self.signatures.items() for output in sign.outputs ] feeded_var_names = list(set(feeded_var_names)) target_vars = list(set(target_vars)) # save inference program program = self.program.clone() for block in program.blocks: for op in block.ops: if "op_callstack" in op.all_attrs(): op._set_attr("op_callstack", [""]) if not exe: place = fluid.CPUPlace() exe = fluid.Executor(place=place) utils.mkdir(self.helper.model_path()) fluid.io.save_inference_model( self.helper.model_path(), feeded_var_names=list(feeded_var_names), target_vars=list(target_vars), main_program=program, executor=exe) with open(os.path.join(self.helper.model_path(), "__model__"), "rb") as file: program_desc_str = file.read() rename_program = fluid.framework.Program.parse_from_string( program_desc_str) varlist = { var: block for block in rename_program.blocks for var in block.vars if self.get_name_prefix() not in var } for var, block in varlist.items(): old_name = var new_name = self.get_var_name_with_prefix(old_name) block._rename_var(old_name, new_name) utils.mkdir(self.helper.model_path()) with open( os.path.join(self.helper.model_path(), "__model__"), "wb") as f: f.write(rename_program.desc.serialize_to_string()) for file in os.listdir(self.helper.model_path()): if (file == "__model__" or self.get_name_prefix() in file): continue os.rename( os.path.join(self.helper.model_path(), file), os.path.join(self.helper.model_path(), self.get_var_name_with_prefix(file))) # create processor file if self.processor: self._dump_processor() # create assets self._dump_assets() # create check info checker = ModuleChecker(self.helper.module_dir) checker.generate_check_info() # Serialize module_desc pb module_pb = self.desc.SerializeToString() with open(self.helper.module_desc_path(), "wb") as f: f.write(module_pb)