# Copyright (c) 2018 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 math import os import re import logging import numpy as np from .... import io from .... import core from .... import framework from ....executor import global_scope, Executor from ....framework import IrGraph from ....log_helper import get_logger from .quantization_pass import QuantizationTransformPass from .quantization_pass import QuantizationFreezePass from .quantization_pass import AddQuantDequantPass from .quantization_pass import _out_scale_op_list from .quantization_pass import _get_op_input_var_names from .quantization_pass import _get_op_output_var_names from .quantization_pass import _get_output_name_index from .quantization_pass import _channelwise_quant_axis1_ops __all__ = ['PostTrainingQuantization', 'WeightQuantization'] _logger = get_logger( __name__, logging.INFO, fmt='%(asctime)s-%(levelname)s: %(message)s') def _load_variable_data(scope, var_name): ''' Load variable value from scope ''' var_node = scope.find_var(var_name) assert var_node is not None, \ "Cannot find " + var_name + " in scope." return np.array(var_node.get_tensor()) def _set_variable_data(scope, place, var_name, np_value): ''' Set the value of var node by name, if the node exits, ''' assert isinstance(np_value, np.ndarray), \ 'The type of value should be numpy array.' var_node = scope.find_var(var_name) if var_node != None: tensor = var_node.get_tensor() tensor.set(np_value, place) def _all_persistable_var_names(program): persistable_var_names = [] for var in program.list_vars(): if var.persistable: persistable_var_names.append(var.name) return persistable_var_names def _remove_unused_var_nodes(graph): all_used_vars = set() ops = graph.all_op_nodes() for op_node in ops: for input_node in op_node.inputs: all_used_vars.add(input_node) for output_node in op_node.outputs: all_used_vars.add(output_node) all_used_vars = {n.node for n in all_used_vars} all_unused_vars = { n for n in filter(lambda node: node.node not in all_used_vars, graph.all_var_nodes()) } graph.safe_remove_nodes(all_unused_vars) return graph def _remove_ctrl_vars(graph): remove_ctr_vars = set() for node in graph.all_var_nodes(): if node.is_ctrl_var(): remove_ctr_vars.add(node) graph.safe_remove_nodes(remove_ctr_vars) return graph def _apply_pass(scope, graph, pass_name, attrs=None, attr_values=None, debug=False): ir_pass = core.get_pass(pass_name) cpp_graph = graph.graph if not cpp_graph.has('__param_scope__'): cpp_graph.set_not_owned('__param_scope__', scope) if attrs: assert attr_values and len(attrs) == len( attr_values), "Different number of pass attributes and their values." for attr, value in zip(attrs, attr_values): ir_pass.set(attr, value) ir_pass.apply(cpp_graph) if debug: graph.draw('.', 'qat_fp32_{}'.format(pass_name), graph.all_op_nodes()) _remove_unused_var_nodes(graph) return graph class PostTrainingQuantization(object): """ Utilizing post training quantization methon to quantize the FP32 model, and it uses calibrate data to get the quantization information for all quantized variables. """ def __init__(self, executor=None, scope=None, model_dir=None, model_filename=None, params_filename=None, batch_generator=None, sample_generator=None, batch_size=10, batch_nums=None, algo="KL", quantizable_op_type=["conv2d", "depthwise_conv2d", "mul"], is_full_quantize=False, activation_bits=8, weight_bits=8, activation_quantize_type='range_abs_max', weight_quantize_type='channel_wise_abs_max', optimize_model=False, is_use_cache_file=False, cache_dir="./temp_post_training"): ''' Constructor. Args: executor(fluid.Executor): The executor to load, run and save the quantized model. scope(fluid.Scope, optional): The scope of the program, use it to load and save variables. If scope=None, get scope by global_scope(). model_dir(str): The path of the fp32 model that will be quantized, and the model and params files are under the path. model_filename(str, optional): The name of file to load the inference program. If it is None, the default filename '__model__' will be used. Default is 'None'. params_filename(str, optional): The name of file to load all parameters. When all parameters were saved in a single binary file, set it as the real filename. If parameters were saved in separate files, set it as 'None'. Default is 'None'. batch_generator(Python Generator): The batch generator provides calibrate data for DataLoader, and it returns a batch every time. Note that, sample_generator and batch_generator, only one should be set. Beisdes, batch_generator supports lod tensor. sample_generator(Python Generator): The sample generator provides calibrate data for DataLoader, and it only returns a sample every time. Note that, sample_generator and batch_generator, only one should be set. Beisdes, sample_generator dose not support lod tensor. batch_size(int, optional): The batch size of DataLoader. Default is 10. batch_nums(int, optional): If batch_nums is not None, the number of calibrate data is batch_size*batch_nums. If batch_nums is None, use all data provided by sample_generator as calibrate data. algo(str, optional): If algo='KL', use KL-divergenc method to get the KL threshold for quantized activations and get the abs_max value for quantized weights. If algo='abs_max', get the abs max value for activations and weights. If algo= 'min_max', get the min and max value for quantized activations and weights. Default is KL. quantizable_op_type(list[str], optional): List the type of ops that will be quantized. Default is ["conv2d", "depthwise_conv2d", "mul"]. is_full_quantized(bool, optional): If set is_full_quantized as True, apply quantization to all supported quantizable op type. If set is_full_quantized as False, only apply quantization to the op type according to the input quantizable_op_type. activation_bits(int): quantization bit number for activation. weight_bits(int, optional): quantization bit number for weights. activation_quantize_type(str): quantization type for activation, now support 'range_abs_max', 'moving_average_abs_max' and 'abs_max'. This param only specifies the fake ops in saving quantized model. If it is 'range_abs_max' or 'moving_average_abs_max', we save the scale obtained by post training quantization in fake ops. Note that, if it is 'abs_max', the scale will not be saved in fake ops. weight_quantize_type(str): quantization type for weights, support 'abs_max' and 'channel_wise_abs_max'. This param only specifies the fake ops in saving quantized model, and we save the scale obtained by post training quantization in fake ops. Compared to 'abs_max', the model accuracy is usually higher when it is 'channel_wise_abs_max'. optimize_model(bool, optional): If set optimize_model as True, it applies some passes to the model before quantization, and it supports `conv2d/depthwise_conv2d + bn` pass so far. Some targets require the weights are quantized by tensor-wise method, which means the weights scale for all channel are the same. However, if fuse `conv2d/depthwise_conv2d + bn`, the weights scale for all channel will be different. In address this problem, fuse the pattern before quantization. Default False. is_use_cache_file(bool, optional): If set is_use_cache_file as False, all temp data will be saved in memory. If set is_use_cache_file as True, it will save temp data to disk. When the fp32 model is complex or the number of calibrate data is large, we should set is_use_cache_file as True. Defalut is False. cache_dir(str, optional): When is_use_cache_file is True, set cache_dir as the directory for saving temp data. Default is ./temp_post_training. Returns: None Examples: .. code-block:: python import paddle.fluid as fluid from paddle.fluid.contrib.slim.quantization import PostTrainingQuantization exe = fluid.Executor(fluid.CPUPlace()) model_dir = path/to/fp32_model_params # set model_filename as None when the filename is __model__, # otherwise set it as the real filename model_filename = None # set params_filename as None when all parameters were saved in # separate files, otherwise set it as the real filename params_filename = None save_model_path = path/to/save_model_path # prepare the sample generator according to the model, and the # sample generator must return a sample every time. The reference # document: https://www.paddlepaddle.org.cn/documentation/docs/zh # /user_guides/howto/prepare_data/use_py_reader.html sample_generator = your_sample_generator batch_size = 10 batch_nums = 10 algo = "KL" quantizable_op_type = ["conv2d", "depthwise_conv2d", "mul"] ptq = PostTrainingQuantization( executor=exe, sample_generator=sample_generator, model_dir=model_dir, model_filename=model_filename, params_filename=params_filename, batch_size=batch_size, batch_nums=batch_nums, algo=algo, quantizable_op_type=quantizable_op_type) ptq.quantize() ptq.save_quantized_model(save_model_path) ''' self._support_activation_quantize_type = [ 'range_abs_max', 'moving_average_abs_max', 'abs_max' ] self._support_weight_quantize_type = ['abs_max', 'channel_wise_abs_max'] self._support_algo_type = ['KL', 'abs_max', 'min_max'] self._support_quantize_op_type = \ list(set(QuantizationTransformPass._supported_quantizable_op_type + AddQuantDequantPass._supported_quantizable_op_type)) # Check inputs assert executor is not None, "The executor cannot be None." assert model_dir is not None, "The model_dir cannot be None." assert any([gen is not None] for gen in [sample_generator, batch_generator]), "The sample_generator and batch_generator " \ "cannot be None in the same time." assert batch_size > 0, "The batch_size should be greater than 0." assert algo in self._support_algo_type, \ "The algo should be KL, abs_max or min_max." assert activation_quantize_type in self._support_activation_quantize_type, \ "The activation_quantize_type ({}) should in ({}).".format( activation_quantize_type, self._support_activation_quantize_type) assert weight_quantize_type in self._support_weight_quantize_type, \ "The weight_quantize_type ({}) shoud in ({}).".format( weight_quantize_type, self._support_weight_quantize_type) # Save input params self._executor = executor self._scope = global_scope() if scope == None else scope self._model_dir = model_dir self._model_filename = model_filename self._params_filename = params_filename self._sample_generator = sample_generator self._batch_generator = batch_generator self._batch_size = batch_size self._batch_nums = batch_nums self._algo = algo self._activation_bits = activation_bits self._weight_bits = weight_bits self._activation_quantize_type = activation_quantize_type self._weight_quantize_type = weight_quantize_type self._is_full_quantize = is_full_quantize if is_full_quantize: self._quantizable_op_type = self._support_quantize_op_type else: self._quantizable_op_type = quantizable_op_type for op_type in self._quantizable_op_type: assert op_type in self._support_quantize_op_type, \ op_type + " is not supported for quantization." self._optimize_model = optimize_model self._is_use_cache_file = is_use_cache_file self._cache_dir = cache_dir if self._is_use_cache_file and not os.path.exists(self._cache_dir): os.mkdir(self._cache_dir) # Define variables self._place = self._executor.place self._program = None self._feed_list = None self._fetch_list = None self._data_loader = None self._out_scale_op_list = _out_scale_op_list self._quantized_weight_var_name = set() self._quantized_act_var_name = set() self.weight_op_pairs = {} self._sampling_data = {} self._quantized_var_kl_threshold = {} self._quantized_var_min = {} self._quantized_var_max = {} self._quantized_var_abs_max = {} def quantize(self): ''' Load the FP32 model, and use the calibrate data to calculate the forward-stage. Based on the sample data, we can get the quantization information, and obtain the final quantized model. Args: None Returns: the program of quantized model. ''' self._load_model_data() self._collect_target_varnames() self._set_activation_persistable() batch_id = 0 for data in self._data_loader(): self._executor.run(program=self._program, feed=data, fetch_list=self._fetch_list, return_numpy=False, scope=self._scope) if self._algo == "KL": self._sample_data(batch_id) else: self._sample_threshold() if batch_id % 5 == 0: _logger.info("Run batch: " + str(batch_id)) batch_id += 1 if self._batch_nums and batch_id >= self._batch_nums: break _logger.info("Finish all batch: " + str(batch_id)) self._reset_activation_persistable() if self._algo == "KL": self._calculate_kl_threshold() if self._algo in ["KL", "abs_max"]: self._update_program() else: self._save_input_threhold() self._save_output_threshold() return self._program def save_quantized_model(self, save_model_path, model_filename=None, params_filename=None): ''' Save the quantized model to the disk. Args: save_model_path(str): The path to save the quantized model. model_filename(str, optional): If the model_filename is None, save the model to '__model__'. Otherwise, save the model to the specified filename. Default: None. params_filename(str, optional): If the params_filename is None, save params to separted files. Otherwise, save all params to the specified filename. Returns: None ''' io.save_inference_model( dirname=save_model_path, model_filename=model_filename, params_filename=params_filename, feeded_var_names=self._feed_list, target_vars=self._fetch_list, executor=self._executor, main_program=self._program) def _load_model_data(self): ''' Load model and set data loader. ''' _logger.info("Load model and set data loader ...") [self._program, self._feed_list, self._fetch_list] = \ io.load_inference_model(dirname=self._model_dir, executor=self._executor, model_filename=self._model_filename, params_filename=self._params_filename) if self._program.num_blocks > 1: _logger.error("The post training quantization requires that the " "program only has one block.") if self._optimize_model: self._optimize_fp32_model() feed_vars = [framework._get_var(str(var_name), self._program) \ for var_name in self._feed_list] self._data_loader = io.DataLoader.from_generator( feed_list=feed_vars, capacity=3 * self._batch_size, iterable=True) if self._sample_generator is not None: self._data_loader.set_sample_generator( self._sample_generator, batch_size=self._batch_size, drop_last=True, places=self._place) elif self._batch_generator is not None: self._data_loader.set_batch_generator( self._batch_generator, places=self._place) def _optimize_fp32_model(self): ''' Fuse the `conv2d/depthwise_conv2d + bn` in FP32 model. ''' _logger.info("Optimize FP32 model ...") graph = IrGraph(core.Graph(self._program.desc), for_test=True) graph = _remove_ctrl_vars(graph) graph = _apply_pass(self._scope, graph, 'conv_bn_fuse_pass') graph = _apply_pass(self._scope, graph, 'depthwise_conv_bn_fuse_pass') graph = _apply_pass(self._scope, graph, 'conv_transpose_bn_fuse_pass') self._program = graph.to_program() def _collect_target_varnames(self): ''' Collect the variable names for sampling, and set activation variables to be persistable. ''' # TODO(juncaipeng), consider the name_scope of skip_quant _logger.info("Collect quantized variable names ...") def collect_var_name(var_name_list, persistable_var_names, op_type): for var_name in var_name_list: if var_name in persistable_var_names: self._quantized_weight_var_name.add(var_name) self.weight_op_pairs[var_name] = op_type else: self._quantized_act_var_name.add(var_name) persistable_var_names = _all_persistable_var_names(self._program) for op in self._program.global_block().ops: op_type = op.type if self._is_full_quantize and \ op_type not in self._quantizable_op_type: _logger.warning(op_type + " is not supported for quantization.") # For quantized ops, sample inputs and outputs if op_type in self._quantizable_op_type: collect_var_name( _get_op_input_var_names(op), persistable_var_names, op_type) collect_var_name( _get_op_output_var_names(op), persistable_var_names, op_type) # For other op, only sample output scale elif op_type in self._out_scale_op_list: collect_var_name( _get_op_output_var_names(op), persistable_var_names, op_type) def _set_activation_persistable(self): ''' Set activation variables to be persistable, so can obtain the tensor data in sample_data ''' for var in self._program.list_vars(): if var.name in self._quantized_act_var_name: var.persistable = True def _reset_activation_persistable(self): ''' Reset activations to be not persistable. ''' for var in self._program.list_vars(): if var.name in self._quantized_act_var_name: var.persistable = False def _sample_threshold(self): ''' Sample the input threshold(min, max, or abs_max) in every iterations. ''' assert self._algo in ["abs_max", "min_max"], \ "The algo should be abs_max or min_max for _sample_threshold." if self._algo == "abs_max": self._sample_threshold_abs_max() elif self._algo == "min_max": self._sample_threshold_min_max() def _sample_threshold_abs_max(self): assert self._algo == "abs_max", \ "The algo should be abs_max for _sample_threshold_abs_max." # Only calculate abs_max value for weight for once if self._quantized_var_abs_max == {}: for var_name in self._quantized_weight_var_name: var_tensor = _load_variable_data(self._scope, var_name) if self._weight_quantize_type == "abs_max": abs_max_value = float(np.max(np.abs(var_tensor))) elif self._weight_quantize_type == "channel_wise_abs_max": abs_max_value = [] if self.weight_op_pairs[ var_name] in _channelwise_quant_axis1_ops: for i in range(var_tensor.shape[1]): abs_max_value.append( float(np.max(np.abs(var_tensor[:, i])))) else: for i in range(var_tensor.shape[0]): abs_max_value.append( float(np.max(np.abs(var_tensor[i])))) self._quantized_var_abs_max[var_name] = abs_max_value for var_name in self._quantized_act_var_name: var_tensor = _load_variable_data(self._scope, var_name) abs_max_value = float(np.max(np.abs(var_tensor))) if (var_name not in self._quantized_var_abs_max) or \ (abs_max_value > self._quantized_var_abs_max[var_name]): self._quantized_var_abs_max[var_name] = abs_max_value def _sample_threshold_min_max(self): assert self._algo == "min_max", \ "The algo should be min_max for _sample_threshold_min_max." if self._quantized_var_min == {} and self._quantized_var_max == {}: for var_name in self._quantized_weight_var_name: var_tensor = _load_variable_data(self._scope, var_name) if self._weight_quantize_type == "abs_max": min_value = float(np.min(var_tensor)) max_value = float(np.max(var_tensor)) elif self._weight_quantize_type == "channel_wise_abs_max": min_value = [] max_value = [] if self.weight_op_pairs[ var_name] in _channelwise_quant_axis1_ops: for i in range(var_tensor.shape[1]): min_value.append(float(np.min(var_tensor[:, i]))) max_value.append(float(np.max(var_tensor[:, i]))) else: for i in range(var_tensor.shape[0]): min_value.append(float(np.min(var_tensor[i]))) max_value.append(float(np.max(var_tensor[i]))) self._quantized_var_min[var_name] = min_value self._quantized_var_max[var_name] = max_value for var_name in self._quantized_act_var_name: var_tensor = _load_variable_data(self._scope, var_name) min_value = float(np.min(var_tensor)) max_value = float(np.max(var_tensor)) if (var_name not in self._quantized_var_min) or \ (min_value < self._quantized_var_min[var_name]): self._quantized_var_min[var_name] = min_value if (var_name not in self._quantized_var_max) or \ (max_value > self._quantized_var_max[var_name]): self._quantized_var_max[var_name] = max_value def _save_input_threhold(self): ''' Save input threshold to the quantized op. ''' assert self._algo == "min_max", \ "The algo should be min_max to save input threshold." for op in self._program.global_block().ops: if op.type in self._quantizable_op_type: for var_name in _get_op_input_var_names(op): assert var_name in self._quantized_var_min assert var_name in self._quantized_var_max op._set_attr(var_name + ".min", self._quantized_var_min[var_name]) op._set_attr(var_name + ".max", self._quantized_var_max[var_name]) def _sample_data(self, iter): ''' Sample the tensor data of quantized variables, applied in every iteration. ''' assert self._algo == "KL", "The algo should be KL to sample data." if self._is_use_cache_file: for var_name in self._quantized_act_var_name: var_tensor = _load_variable_data(self._scope, var_name) var_tensor = var_tensor.ravel() save_path = os.path.join( self._cache_dir, var_name.replace("/", ".") + "_" + str(iter) + ".npy") np.save(save_path, var_tensor) else: for var_name in self._quantized_act_var_name: if var_name not in self._sampling_data: self._sampling_data[var_name] = [] var_tensor = _load_variable_data(self._scope, var_name) var_tensor = var_tensor.ravel() self._sampling_data[var_name].append(var_tensor) def _calculate_kl_threshold(self): ''' Calculate the KL threshold of quantized variables. ''' _logger.info("Calculate KL threshold ...") assert self._algo == "KL", "The algo should be KL to calculate kl threshold." # Abs_max threshold for weights for var_name in self._quantized_weight_var_name: weight_data = _load_variable_data(self._scope, var_name) if self._weight_quantize_type == "abs_max": weight_threshold = float(np.max(np.abs(weight_data))) elif self._weight_quantize_type == "channel_wise_abs_max": weight_threshold = [] if self.weight_op_pairs[ var_name] in _channelwise_quant_axis1_ops: for i in range(weight_data.shape[1]): weight_threshold.append( float(np.max(np.abs(weight_data[:, i])))) else: for i in range(weight_data.shape[0]): weight_threshold.append( float(np.max(np.abs(weight_data[i])))) self._quantized_var_kl_threshold[var_name] = weight_threshold # KL threshold for activations if self._is_use_cache_file: for var_name in self._quantized_act_var_name: sampling_data = [] filenames = [f for f in os.listdir(self._cache_dir) \ if re.match(var_name.replace("/", ".") + '_[0-9]+.npy', f)] for filename in filenames: file_path = os.path.join(self._cache_dir, filename) sampling_data.append(np.load(file_path)) os.remove(file_path) sampling_data = np.concatenate(sampling_data) self._quantized_var_kl_threshold[var_name] = \ self._get_kl_scaling_factor(np.abs(sampling_data)) else: for var_name in self._quantized_act_var_name: self._sampling_data[var_name] = np.concatenate( self._sampling_data[var_name]) self._quantized_var_kl_threshold[var_name] = \ self._get_kl_scaling_factor(np.abs(self._sampling_data[var_name])) def _update_program(self): ''' Use QuantizationTransformPass and AddQuantDequantPass to insert fake_quantize, fake_dequantize and fake_quant_dequant op. Besides, save all kl threshold to the scale var node. ''' _logger.info("Update the program ...") graph = IrGraph(core.Graph(self._program.desc), for_test=True) # use QuantizationTransformPass to insert fake_quant/fake_dequantize op major_quantizable_op_types = [] for op_type in QuantizationTransformPass._supported_quantizable_op_type: if op_type in self._quantizable_op_type: major_quantizable_op_types.append(op_type) transform_pass = QuantizationTransformPass( scope=self._scope, place=self._place, weight_bits=self._weight_bits, activation_bits=self._activation_bits, activation_quantize_type=self._activation_quantize_type, weight_quantize_type=self._weight_quantize_type, quantizable_op_type=major_quantizable_op_types) transform_pass.apply(graph) # use AddQuantDequantPass to insert fake_quant_dequant op minor_quantizable_op_types = [] for op_type in AddQuantDequantPass._supported_quantizable_op_type: if op_type in self._quantizable_op_type: minor_quantizable_op_types.append(op_type) add_quant_dequant_pass = AddQuantDequantPass( scope=self._scope, place=self._place, quantizable_op_type=minor_quantizable_op_types) add_quant_dequant_pass.apply(graph) # save abs_max or KL threshold to scale var node if self._algo == "KL": scale_dict = self._quantized_var_kl_threshold else: scale_dict = self._quantized_var_abs_max for key, val in scale_dict.items(): _set_variable_data( self._scope, self._place, key + ".scale", np.array( [val], dtype=np.float32)) _set_variable_data( self._scope, self._place, key + ".quant_dequant.scale", np.array( [val], dtype=np.float32)) # apply QuantizationFreezePass, and obtain the final quant model freeze_pass = QuantizationFreezePass( scope=self._scope, place=self._place, weight_bits=self._weight_bits, activation_bits=self._activation_bits, weight_quantize_type=self._weight_quantize_type, quantizable_op_type=major_quantizable_op_types) freeze_pass.apply(graph) self._program = graph.to_program() def _save_output_threshold(self): ''' Save output threshold to the quantized op. ''' def save_info(op_node, out_var_name, threshold_map, out_info_name, quantized_type): assert out_var_name in threshold_map, \ "The output ({}) of {} node does not have threshold.".format( out_var_name, op_node.type) op_node._set_attr(out_info_name, threshold_map[var_name]) if op_node.type in self._quantizable_op_type: op._set_attr("quantization_type", quantized_type) def analysis_and_save_info(op_node, out_var_name): argname_index = _get_output_name_index(op_node, out_var_name) assert argname_index is not None, \ out_var_name + " is not the output of the op" if self._algo == "KL": # For compatibility, we save output threshold by two methods. save_info(op_node, out_var_name, self._quantized_var_kl_threshold, "out_threshold", "post_kl") save_info( op_node, out_var_name, self._quantized_var_kl_threshold, argname_index[0] + str(argname_index[1]) + "_threshold", "post_kl") elif self._algo == "abs_max": save_info(op_node, out_var_name, self._quantized_var_abs_max, "out_threshold", "post_abs_max") save_info( op_node, out_var_name, self._quantized_var_abs_max, argname_index[0] + str(argname_index[1]) + "_threshold", "post_kl") elif self._algo == "min_max": save_info(op_node, out_var_name, self._quantized_var_min, "out_min", "post_min_max") save_info(op_node, out_var_name, self._quantized_var_max, "out_max", "post_min_max") for op in self._program.global_block().ops: if op.type in (self._quantizable_op_type + self._out_scale_op_list): out_var_names = _get_op_output_var_names(op) assert len(out_var_names) == 1, "Post training " + \ "quantization only support one output for " + op.type for var_name in out_var_names: analysis_and_save_info(op, var_name) def _get_kl_scaling_factor(self, activation_blob, num_quantized_bins=255): ''' Using the KL-divergenc method to get the more precise scaling factor. ''' max_val = np.max(activation_blob) min_val = np.min(activation_blob) if min_val >= 0: hist, hist_edeges = np.histogram( activation_blob, bins=2048, range=(min_val, max_val)) ending_iter = 2047 starting_iter = int(ending_iter * 0.7) else: _logger.error("Please first apply abs to activation_blob.") bin_width = hist_edeges[1] - hist_edeges[0] P_sum = len(np.array(activation_blob).ravel()) min_kl_divergence = 0 min_kl_index = 0 kl_inited = False for i in range(starting_iter, ending_iter + 1): reference_distr_P = hist[0:i].tolist() outliers_count = sum(hist[i:2048]) if reference_distr_P[i - 1] == 0: continue reference_distr_P[i - 1] += outliers_count reference_distr_bins = reference_distr_P[:] candidate_distr_Q = hist[0:i].tolist() num_merged_bins = int(i / num_quantized_bins) candidate_distr_Q_quantized = [0] * num_quantized_bins j_start = 0 j_end = num_merged_bins for idx in range(num_quantized_bins): candidate_distr_Q_quantized[idx] = sum(candidate_distr_Q[ j_start:j_end]) j_start += num_merged_bins j_end += num_merged_bins if (idx + 1) == num_quantized_bins - 1: j_end = i candidate_distr_Q = self._expand_quantized_bins( candidate_distr_Q_quantized, reference_distr_bins) Q_sum = sum(candidate_distr_Q) kl_divergence = self._safe_entropy(reference_distr_P, P_sum, candidate_distr_Q, Q_sum) if not kl_inited: min_kl_divergence = kl_divergence min_kl_index = i kl_inited = True elif kl_divergence < min_kl_divergence: min_kl_divergence = kl_divergence min_kl_index = i else: pass if min_kl_index == 0: while starting_iter > 0: if hist[starting_iter] == 0: starting_iter -= 1 continue else: break min_kl_index = starting_iter return (min_kl_index + 0.5) * bin_width def _expand_quantized_bins(self, quantized_bins, reference_bins): ''' ''' expanded_quantized_bins = [0] * len(reference_bins) num_merged_bins = int(len(reference_bins) / len(quantized_bins)) j_start = 0 j_end = num_merged_bins for idx in range(len(quantized_bins)): zero_count = reference_bins[j_start:j_end].count(0) num_merged_bins = j_end - j_start if zero_count == num_merged_bins: avg_bin_ele = 0 else: avg_bin_ele = quantized_bins[idx] / ( num_merged_bins - zero_count + 0.0) for idx1 in range(j_start, j_end): expanded_quantized_bins[idx1] = (0 if reference_bins[idx1] == 0 else avg_bin_ele) j_start += num_merged_bins j_end += num_merged_bins if (idx + 1) == len(quantized_bins) - 1: j_end = len(reference_bins) return expanded_quantized_bins def _safe_entropy(self, reference_distr_P, P_sum, candidate_distr_Q, Q_sum): ''' Calculate the entropy. ''' assert len(reference_distr_P) == len(candidate_distr_Q) tmp_sum1 = 0 tmp_sum2 = 0 for idx in range(len(reference_distr_P)): p_idx = reference_distr_P[idx] q_idx = candidate_distr_Q[idx] if p_idx == 0: tmp_sum1 += 0 tmp_sum2 += 0 else: if q_idx == 0: _logger.error("Fatal error!, idx = " + str(idx) + " qindex = 0! p_idx = " + str(p_idx)) tmp_sum1 += p_idx * (math.log(Q_sum * p_idx)) tmp_sum2 += p_idx * (math.log(P_sum * q_idx)) return (tmp_sum1 - tmp_sum2) / P_sum class WeightQuantization(object): _supported_quantizable_op_type = ['conv2d', 'depthwise_conv2d', 'mul'] _supported_weight_quantize_type = ['channel_wise_abs_max', 'abs_max'] def __init__(self, model_dir, model_filename=None, params_filename=None): ''' This class quantizes the weight of some ops to reduce the size of model or improve the perforemace. Args: model_dir(str): The path of the fp32 model that will be quantized, and the model and params files are under the path. model_filename(str, optional): The name of file to load the inference program. If it is None, the default filename '__model__' will be used. Default is 'None'. params_filename(str, optional): The name of file to load all parameters. When all parameters were saved in a single binary file, set it as the real filename. If parameters were saved in separate files, set it as 'None'. Default is 'None'. ''' self._model_dir = model_dir self._model_filename = model_filename self._params_filename = params_filename def quantize_weight_to_int(self, save_model_dir, save_model_filename=None, save_params_filename=None, quantizable_op_type=["conv2d", "mul"], weight_bits=8, weight_quantize_type="channel_wise_abs_max", generate_test_model=False, threshold_rate=0.0): ''' In order to reduce the size of model, this api quantizes the weight of some ops from float32 to int8/16. In the inference stage, the quantized weight will be dequantized to float32 again. Args: save_model_dir(str): The path to save the quantized model. save_model_filename(str, optional): The name of file to save the inference program. If it is None, the default filename '__model__' will be used. Default is 'None'. save_params_filename(str, optional): The name of file to save all parameters. If it is None, parameters were saved in separate files. If it is not None, all parameters were saved in a single binary file. quantizable_op_type(list[str], optional): The list of ops that will be quantized, and the quantized ops should be contained in ["conv2d", "depthwise_conv2d", "mul"]. Default is ["conv2d","mul"]. weight_bits(int, optional): The bits for the quantized weight, and it should be 8 or 16. Default is 8. weight_quantize_type(str, optional): quantization type for weights, support 'channel_wise_abs_max' and 'abs_max'. Set it as 'channel_wise_abs_max', the accuracy performs better. generate_test_model(bool, optional): If set generate_test_model as True, it saves a fake quantized model, in which the weights are quantized and dequantized. We can use PaddlePaddle to load the fake quantized model and test the accuracy on GPU or CPU. threshold_rate(float, optional): This api uses abs_max methd to quantize the weight from float32 to int8/16, and the abs max value is important for quantization diff. When the abs_max value is far away from the center of the numerical distribution, we can set threshold_rate between 1e-6 and 1e-8, so the abs max value will be optimized. Default is 0.0. ''' for op_type in quantizable_op_type: assert op_type in self._supported_quantizable_op_type, \ "Input error:" + op_type + \ " is not supported for weight quantization." assert weight_bits in [8, 16], \ "Input error: weight_bits should be 8 or 16." assert weight_quantize_type in self._supported_weight_quantize_type, \ "Input error: weight_quantize_type should in {}".format( self._supported_weight_quantize_type) quantized_model_dir = os.path.join(save_model_dir, "quantized_model") self._quantize_weight_to_int(quantized_model_dir, save_model_filename, save_params_filename, quantizable_op_type, weight_bits, weight_quantize_type, False, threshold_rate) if generate_test_model: test_model_dir = os.path.join(save_model_dir, "test_model") self._quantize_weight_to_int( test_model_dir, save_model_filename, save_params_filename, quantizable_op_type, weight_bits, weight_quantize_type, True, threshold_rate) def _quantize_weight_to_int(self, save_model_dir, save_model_filename, save_params_filename, quantizable_op_type, weight_bits, weight_quantize_type, for_test, threshold_rate): """ Generate quantized model or fake quantized model. """ # Load model place = core.CPUPlace() exe = Executor(place) scope = global_scope() [program, feed_list, fetch_list] = \ io.load_inference_model(dirname=self._model_dir, executor=exe, model_filename=self._model_filename, params_filename=self._params_filename) quantized_ops = [] for index in range(program.num_blocks): block = program.block(index) for op in block.ops: if op.type in quantizable_op_type: quantized_ops.append(op) # Quantize weights persistable_var_names = _all_persistable_var_names(program) for op in quantized_ops: for var_name in op.input_arg_names: if var_name in persistable_var_names: if weight_quantize_type == "abs_max": self._weight_abs_max_quantization( scope, place, weight_bits, threshold_rate, op, var_name, for_test) elif weight_quantize_type == "channel_wise_abs_max": self._weight_channel_wise_abs_max_quantization( scope, place, weight_bits, op, var_name, for_test) io.save_inference_model( dirname=save_model_dir, feeded_var_names=feed_list, target_vars=fetch_list, executor=exe, main_program=program, model_filename=save_model_filename, params_filename=save_params_filename) def _weight_abs_max_quantization(self, scope, place, weight_bits, threshold_rate, op, var_name, for_test): ''' Use abs_max method to quantize weight. ''' quantize_range = (1 << (weight_bits - 1)) - 1 save_weight_dtype = np.int8 if weight_bits == 8 else np.int16 # Get quantized scale and weight data weight_data = _load_variable_data(scope, var_name) if abs(threshold_rate) < 1e-10: threshold_value = np.max(np.abs(weight_data)) else: threshold_value = self._calculate_threshold(\ weight_data, threshold_rate) weight_data[weight_data > threshold_value] = threshold_value weight_data[weight_data < -threshold_value] = -threshold_value scale = threshold_value / quantize_range quantized_weight_data = \ np.around(weight_data / scale).astype(save_weight_dtype) # Set weight data if not for_test: _set_variable_data(scope, place, var_name, quantized_weight_data) else: dequantized_weight_data = \ (quantized_weight_data * scale).astype(np.float32) _set_variable_data(scope, place, var_name, dequantized_weight_data) # Save info op._set_attr('quantization_type', 'post_weight_abs_max') op._set_attr('quantize_weight_bits', weight_bits) op._set_attr(var_name + "_quant_scale", [scale]) # Save as list def _weight_channel_wise_abs_max_quantization( self, scope, place, weight_bits, op, var_name, for_test): ''' Use channel_wise_abs_max method to quantize weight. ''' quantize_range = (1 << (weight_bits - 1)) - 1 save_weight_dtype = np.int8 if weight_bits == 8 else np.int16 # Get quantized scale and weight data weight_data = _load_variable_data(scope, var_name) if op.type == "mul": scales, quantized_weight_data = \ self._mul_channel_wise_quantization(weight_data, quantize_range, save_weight_dtype) elif op.type in ["conv2d", "depthwise_conv2d"]: scales, quantized_weight_data = \ self._conv_channel_wise_quantization(weight_data, quantize_range, save_weight_dtype) else: _logger.error(op.type + " is not supported by weight quantization") # Set weight data if not for_test: _set_variable_data(scope, place, var_name, quantized_weight_data) else: if op.type == "mul": dequantized_weight_data = \ self._mul_channel_wise_dequantization(quantized_weight_data, scales) elif op.type in ["conv2d", "depthwise_conv2d"]: dequantized_weight_data = \ self._conv_channel_wise_dequantization(quantized_weight_data, scales) else: _logger.error(op.type + " is not supported by weight quantization") _set_variable_data(scope, place, var_name, dequantized_weight_data) # Save info op._set_attr('quantization_type', 'post_weight_channel_wise_abs_max') op._set_attr('quantize_weight_bits', weight_bits) op._set_attr(var_name + "_quant_scale", scales) def _conv_channel_wise_quantization(self, weight_data, quantize_range, save_weight_dtype): ''' Get channel wise scale for the weights of conv2d and depthwise_conv2d, and quantize the weights. ''' scales = [] quantized_weight_data = np.zeros_like( weight_data, dtype=save_weight_dtype) channel_num = weight_data.shape[0] for i in range(channel_num): scale = np.max(np.abs(weight_data[i])) / quantize_range scales.append(scale) quantized_weight_data[i] = \ np.around(weight_data[i] / scale).astype(save_weight_dtype) return scales, quantized_weight_data def _conv_channel_wise_dequantization(self, quantized_weight_data, scales): ''' For conv2d and depthwise_conv2d, dequantize the weights to fp32. ''' dequantized_weight_data = np.zeros_like( quantized_weight_data, dtype=np.float32) for i in range(len(scales)): dequantized_weight_data[i] = \ (quantized_weight_data[i] * scales[i]).astype(np.float32) return dequantized_weight_data def _mul_channel_wise_quantization(self, weight_data, quantize_range, save_weight_dtype): ''' Get channel wise scale for the weights of conv2d and depthwise_conv2d, and quantize the weights. ''' scales = [] quantized_weight_data = np.zeros_like( weight_data, dtype=save_weight_dtype) channel_num = weight_data.shape[-1] for i in range(channel_num): scale = np.max(np.abs(weight_data[:, i])) / quantize_range scales.append(scale) quantized_weight_data[:, i] = \ np.around(weight_data[:, i] / scale).astype(save_weight_dtype) return scales, quantized_weight_data def _mul_channel_wise_dequantization(self, quantized_weight_data, scales): ''' For mul, dequantize the weights to fp32. ''' dequantized_weight_data = np.zeros_like( quantized_weight_data, dtype=np.float32) for i in range(len(scales)): dequantized_weight_data[:, i] = \ (quantized_weight_data[:, i] * scales[i]).astype(np.float32) return dequantized_weight_data def _calculate_threshold(self, input, threshold_rate, histogram_bins=5000): input_abs = np.abs(input) hist, hist_edeges = np.histogram( input_abs, bins=histogram_bins, range=(0, np.max(input_abs))) hist = hist / float(sum(hist)) hist_sum = 0 hist_index = 0 for i in range(len(hist)): hist_sum += hist[i] if hist_sum >= 1.0 - threshold_rate: hist_index = i + 1 break bin_width = hist_edeges[1] - hist_edeges[0] return hist_index * bin_width