diff --git a/demo/auto_prune/README.md b/demo/auto_prune/README.md index 9ada4cf4c848ef24e296aa3ebf4724c707354199..0f6b5bb80ce8182e3d10aad29e5c0ae12659c5de 100644 --- a/demo/auto_prune/README.md +++ b/demo/auto_prune/README.md @@ -1,5 +1,5 @@ 该示例介绍如何使用自动裁剪。 -该示例使用默认会自动下载并使用MNIST数据。支持以下模型： +该示例需要使用IMAGENET数据，以及预训练模型。支持以下模型： - MobileNetV1 - MobileNetV2 @@ -15,7 +15,7 @@ ## 2. 运行示例 -提供两种自动裁剪模式，直接以裁剪目标进行一次自动裁剪，和多次迭代的方式进行裁剪。 +提供两种自动裁剪模式，直接以裁剪目标进行一次自动裁剪，和多次迭代的方式进行裁剪。 ###2.1一次裁剪 diff --git a/demo/auto_prune/train.py b/demo/auto_prune/train.py index ff55a54597ce7e923764f9e6f8dff232d2d28658..f0b2a2b0b4cd7cd235af55268d1f2c44b11a8339 100644 --- a/demo/auto_prune/train.py +++ b/demo/auto_prune/train.py @@ -8,10 +8,10 @@ import math import time import numpy as np import paddle.fluid as fluid +sys.path[0] = os.path.join(os.path.dirname("__file__"), os.path.pardir) from paddleslim.prune import AutoPruner from paddleslim.common import get_logger from paddleslim.analysis import flops -sys.path.append(sys.path[0] + "/../") import models from utility import add_arguments, print_arguments @@ -23,7 +23,7 @@ add_arg = functools.partial(add_arguments, argparser=parser) add_arg('batch_size', int, 64 * 4, "Minibatch size.") add_arg('use_gpu', bool, True, "Whether to use GPU or not.") add_arg('model', str, "MobileNet", "The target model.") -add_arg('pretrained_model', str, "../pretrained_model/MobileNetV1_pretained", "Whether to use pretrained model.") +add_arg('pretrained_model', str, "../pretrained_model/MobileNetV1_pretrained", "Whether to use pretrained model.") add_arg('lr', float, 0.1, "The learning rate used to fine-tune pruned model.") add_arg('lr_strategy', str, "piecewise_decay", "The learning rate decay strategy.") add_arg('l2_decay', float, 3e-5, "The l2_decay parameter.") @@ -32,7 +32,7 @@ add_arg('num_epochs', int, 120, "The number of total epochs add_arg('total_images', int, 1281167, "The number of total training images.") parser.add_argument('--step_epochs', nargs='+', type=int, default=[30, 60, 90], help="piecewise decay step") add_arg('config_file', str, None, "The config file for compression with yaml format.") -add_arg('data', str, "mnist", "Which data to use. 'mnist' or 'imagenet'") +add_arg('data', str, "imagenet", "Which data to use. 'mnist' or 'imagenet'") add_arg('log_period', int, 10, "Log period in batches.") add_arg('test_period', int, 10, "Test period in epoches.") # yapf: enable @@ -90,8 +90,8 @@ def compress(args): raise ValueError("{} is not supported.".format(args.data)) image_shape = [int(m) for m in image_shape.split(",")] - assert args.model in model_list, "{} is not in lists: {}".format( - args.model, model_list) + assert args.model in model_list, "{} is not in lists: {}".format(args.model, + model_list) image = fluid.layers.data(name='image', shape=image_shape, dtype='float32') label = fluid.layers.data(name='label', shape=[1], dtype='int64') # model definition @@ -111,8 +111,7 @@ def compress(args): if args.pretrained_model: def if_exist(var): - return os.path.exists( - os.path.join(args.pretrained_model, var.name)) + return os.path.exists(os.path.join(args.pretrained_model, var.name)) fluid.io.load_vars(exe, args.pretrained_model, predicate=if_exist) @@ -145,10 +144,9 @@ def compress(args): acc_top5_ns.append(np.mean(acc_top5_n)) batch_id += 1 - _logger.info("Final eval epoch[{}] - acc_top1: {}; acc_top5: {}". - format(epoch, - np.mean(np.array(acc_top1_ns)), - np.mean(np.array(acc_top5_ns)))) + _logger.info("Final eval epoch[{}] - acc_top1: {}; acc_top5: {}".format( + epoch, + np.mean(np.array(acc_top1_ns)), np.mean(np.array(acc_top5_ns)))) return np.mean(np.array(acc_top1_ns)) def train(epoch, program): diff --git a/demo/auto_prune/train_finetune.py b/demo/auto_prune/train_finetune.py index 0dcb56bfe7988475cc7c63ebde257897456bfb22..bc56ffa0568520634aab5c294a03c05813f13a54 100644 --- a/demo/auto_prune/train_finetune.py +++ b/demo/auto_prune/train_finetune.py @@ -21,7 +21,7 @@ add_arg('batch_size', int, 64 * 4, "Minibatch size.") add_arg('use_gpu', bool, True, "Whether to use GPU or not.") add_arg('model', str, "MobileNet", "The target model.") add_arg('model_save_dir', str, "./", "checkpoint model.") -add_arg('pretrained_model', str, "../pretrained_model/MobileNetV1_pretained", "Whether to use pretrained model.") +add_arg('pretrained_model', str, "../pretrained_model/MobileNetV1_pretrained", "Whether to use pretrained model.") add_arg('lr', float, 0.01, "The learning rate used to fine-tune pruned model.") add_arg('lr_strategy', str, "piecewise_decay", "The learning rate decay strategy.") add_arg('l2_decay', float, 3e-5, "The l2_decay parameter.") @@ -81,8 +81,8 @@ def compress(args): class_dim = 1000 image_shape = "3,224,224" image_shape = [int(m) for m in image_shape.split(",")] - assert args.model in model_list, "{} is not in lists: {}".format( - args.model, model_list) + assert args.model in model_list, "{} is not in lists: {}".format(args.model, + model_list) image = fluid.layers.data(name='image', shape=image_shape, dtype='float32') label = fluid.layers.data(name='label', shape=[1], dtype='int64') # model definition @@ -158,8 +158,7 @@ def compress(args): train_program, feed=train_feeder.feed(data), fetch_list=[ - avg_cost.name, acc_top1.name, acc_top5.name, - "learning_rate" + avg_cost.name, acc_top1.name, acc_top5.name, "learning_rate" ]) end_time = time.time() loss_n = np.mean(loss_n) @@ -177,8 +176,7 @@ def compress(args): #if "_weights" in param.name and "conv1_weights" not in param.name: if "_sep_weights" in param.name: params.append(param.name) - print("fops before pruning: {}".format( - flops(fluid.default_main_program()))) + print("fops before pruning: {}".format(flops(fluid.default_main_program()))) pruned_program_iter = fluid.default_main_program() pruned_val_program_iter = val_program for ratios in ratiolist: diff --git a/demo/auto_prune/train_iterator.py b/demo/auto_prune/train_iterator.py index 8b7ca07f768e661476a5943dbc623291a37f939a..be80d892e0911e1ac592a7bfb8ab1f995b4f87f2 100644 --- a/demo/auto_prune/train_iterator.py +++ b/demo/auto_prune/train_iterator.py @@ -24,7 +24,7 @@ add_arg = functools.partial(add_arguments, argparser=parser) add_arg('batch_size', int, 64 * 4, "Minibatch size.") add_arg('use_gpu', bool, True, "Whether to use GPU or not.") add_arg('model', str, "MobileNet", "The target model.") -add_arg('pretrained_model', str, "../pretrained_model/MobileNetV1_pretained", "Whether to use pretrained model.") +add_arg('pretrained_model', str, "../pretrained_model/MobileNetV1_pretrained", "Whether to use pretrained model.") add_arg('model_save_dir', str, "./", "checkpoint model.") add_arg('lr', float, 0.1, "The learning rate used to fine-tune pruned model.") add_arg('lr_strategy', str, "piecewise_decay", "The learning rate decay strategy.") @@ -34,7 +34,7 @@ add_arg('num_epochs', int, 120, "The number of total epochs add_arg('total_images', int, 1281167, "The number of total training images.") parser.add_argument('--step_epochs', nargs='+', type=int, default=[30, 60, 90], help="piecewise decay step") add_arg('config_file', str, None, "The config file for compression with yaml format.") -add_arg('data', str, "mnist", "Which data to use. 'mnist' or 'imagenet'") +add_arg('data', str, "imagenet", "Which data to use. 'mnist' or 'imagenet'") add_arg('log_period', int, 10, "Log period in batches.") add_arg('test_period', int, 10, "Test period in epoches.") # yapf: enable @@ -96,8 +96,8 @@ def compress(args): raise ValueError("{} is not supported.".format(args.data)) image_shape = [int(m) for m in image_shape.split(",")] - assert args.model in model_list, "{} is not in lists: {}".format( - args.model, model_list) + assert args.model in model_list, "{} is not in lists: {}".format(args.model, + model_list) image = fluid.layers.data(name='image', shape=image_shape, dtype='float32') label = fluid.layers.data(name='label', shape=[1], dtype='int64') # model definition @@ -117,8 +117,7 @@ def compress(args): if args.pretrained_model: def if_exist(var): - return os.path.exists( - os.path.join(args.pretrained_model, var.name)) + return os.path.exists(os.path.join(args.pretrained_model, var.name)) # fluid.io.load_vars(exe, args.pretrained_model, predicate=if_exist) @@ -151,10 +150,9 @@ def compress(args): acc_top5_ns.append(np.mean(acc_top5_n)) batch_id += 1 - _logger.info("Final eval epoch[{}] - acc_top1: {}; acc_top5: {}". - format(epoch, - np.mean(np.array(acc_top1_ns)), - np.mean(np.array(acc_top5_ns)))) + _logger.info("Final eval epoch[{}] - acc_top1: {}; acc_top5: {}".format( + epoch, + np.mean(np.array(acc_top1_ns)), np.mean(np.array(acc_top5_ns)))) return np.mean(np.array(acc_top1_ns)) def train(epoch, program): diff --git a/paddleslim/prune/auto_pruner.py b/paddleslim/prune/auto_pruner.py index 54ae8f3aab6c6047677661a66e0ddd7fd0d3d3e9..6ace8546567fe7cce7ac8ec971d6a81d59546109 100644 --- a/paddleslim/prune/auto_pruner.py +++ b/paddleslim/prune/auto_pruner.py @@ -41,12 +41,12 @@ class AutoPruner(object): params(list): The names of parameters to be pruned. init_ratios(list|float): Init ratios used to pruned parameters in `params`. List means ratios used for pruning each parameter in `params`. - The length of `init_ratios` should be equal to length of params when `init_ratios` is a list. + The length of `init_ratios` should be equal to length of params when `init_ratios` is a list. If it is a scalar, all the parameters in `params` will be pruned by uniform ratio. None means get a group of init ratios by `pruned_flops` of `pruned_latency`. Default: None. pruned_flops(float): The percent of FLOPS to be pruned. Default: None. pruned_latency(float): The percent of latency to be pruned. Default: None. - server_addr(tuple): A tuple of server ip and server port for controller server. + server_addr(tuple): A tuple of server ip and server port for controller server. init_temperature(float): The init temperature used in simulated annealing search strategy. reduce_rate(float): The decay rate used in simulated annealing search strategy. max_try_times(int): The max number of trying to generate legal tokens. @@ -191,14 +191,14 @@ class AutoPruner(object): paddle.fluid.Program: The pruned program. """ self._current_ratios = self._next_ratios() - pruned_program, _, _ = self._pruner.prune( + pruned_program, self._param_backup, _ = self._pruner.prune( program, self._scope, self._params, self._current_ratios, place=self._place, only_graph=False, - param_backup=self._param_backup) + param_backup=True) pruned_val_program = None if eval_program is not None: pruned_val_program, _, _ = self._pruner.prune(