diff --git a/demo/dygraph/unstructured_pruning/evaluate.py b/demo/dygraph/unstructured_pruning/evaluate.py index 21e62b16d1e60c477903ff1022c7b5bce515a639..d2c6aa56c13cc2e0cae84f00a8f71e66b6bff05b 100644 --- a/demo/dygraph/unstructured_pruning/evaluate.py +++ b/demo/dygraph/unstructured_pruning/evaluate.py @@ -33,7 +33,7 @@ def compress(args): test_reader = None if args.data == "imagenet": import imagenet_reader as reader - val_dataset = reader.ImageNetDataset(data_dir='/data', mode='val') + val_dataset = reader.ImageNetDataset(mode='val') class_dim = 1000 elif args.data == "cifar10": normalize = T.Normalize( @@ -47,13 +47,12 @@ def compress(args): places = paddle.static.cuda_places( ) if args.use_gpu else paddle.static.cpu_places() - batch_size_per_card = int(args.batch_size / len(places)) valid_loader = paddle.io.DataLoader( val_dataset, places=places, drop_last=False, return_list=True, - batch_size=batch_size_per_card, + batch_size=args.batch_size, shuffle=False, use_shared_memory=True) @@ -70,15 +69,12 @@ def compress(args): y_data = paddle.to_tensor(data[1]) if args.data == 'cifar10': y_data = paddle.unsqueeze(y_data, 1) - end_time = time.time() logits = model(x_data) loss = F.cross_entropy(logits, y_data) acc_top1 = paddle.metric.accuracy(logits, y_data, k=1) acc_top5 = paddle.metric.accuracy(logits, y_data, k=5) - - acc_top1_ns.append(acc_top1.numpy()) - acc_top5_ns.append(acc_top5.numpy()) + end_time = time.time() if batch_id % args.log_period == 0: _logger.info( "Eval epoch[{}] batch[{}] - acc_top1: {}; acc_top5: {}; time: {}". diff --git a/demo/dygraph/unstructured_pruning/train.py b/demo/dygraph/unstructured_pruning/train.py index 3285cd167b9aa1124706d250e4e7a492e3ccfbca..39b6804bfca609db4506d4df17e96040623b3233 100644 --- a/demo/dygraph/unstructured_pruning/train.py +++ b/demo/dygraph/unstructured_pruning/train.py @@ -23,6 +23,7 @@ parser = argparse.ArgumentParser(description=__doc__) add_arg = functools.partial(add_arguments, argparser=parser) # yapf: disable add_arg('batch_size', int, 64 * 4, "Minibatch size.") +add_arg('batch_size_for_validation', int, 64, "Minibatch size for validation.") add_arg('use_gpu', bool, True, "Whether to use GPU or not.") 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.") @@ -121,7 +122,7 @@ def compress(args): places=place, drop_last=False, return_list=True, - batch_size=64, + batch_size=args.batch_size_for_validation, shuffle=False, use_shared_memory=True) step_per_epoch = int( @@ -146,15 +147,12 @@ def compress(args): y_data = paddle.to_tensor(data[1]) if args.data == 'cifar10': y_data = paddle.unsqueeze(y_data, 1) - end_time = time.time() logits = model(x_data) loss = F.cross_entropy(logits, y_data) acc_top1 = paddle.metric.accuracy(logits, y_data, k=1) acc_top5 = paddle.metric.accuracy(logits, y_data, k=5) - - acc_top1_ns.append(acc_top1.numpy()) - acc_top5_ns.append(acc_top5.numpy()) + end_time = time.time() if batch_id % args.log_period == 0: _logger.info( "Eval epoch[{}] batch[{}] - acc_top1: {}; acc_top5: {}; time: {}". diff --git a/demo/unstructured_prune/evaluate.py b/demo/unstructured_prune/evaluate.py index 62a08622b22072eb591809b576175975bfcb405e..e7f99b39cba6db3d0b8be22a79a065ba8626d2f9 100644 --- a/demo/unstructured_prune/evaluate.py +++ b/demo/unstructured_prune/evaluate.py @@ -20,7 +20,7 @@ _logger = get_logger(__name__, level=logging.INFO) parser = argparse.ArgumentParser(description=__doc__) add_arg = functools.partial(add_arguments, argparser=parser) # yapf: disable -add_arg('batch_size', int, 64*12, "Minibatch size.") +add_arg('batch_size', int, 64, "Minibatch size.") add_arg('use_gpu', bool, True, "Whether to use GPU or not.") add_arg('model', str, "MobileNet", "The target model.") add_arg('pruned_model', str, "models", "Whether to use pretrained model.") @@ -44,8 +44,8 @@ def compress(args): image_shape = "1,28,28" elif args.data == "imagenet": import imagenet_reader as reader - train_dataset = reader.ImageNetDataset(data_dir='/data', mode='train') - val_dataset = reader.ImageNetDataset(data_dir='/data', mode='val') + train_dataset = reader.ImageNetDataset(mode='train') + val_dataset = reader.ImageNetDataset(mode='val') class_dim = 1000 image_shape = "3,224,224" else: @@ -71,7 +71,6 @@ def compress(args): use_shared_memory=True, batch_size=batch_size_per_card, shuffle=False) - step_per_epoch = int(np.ceil(len(train_dataset) * 1. / args.batch_size)) # model definition model = models.__dict__[args.model]() @@ -103,12 +102,7 @@ def compress(args): for batch_id, data in enumerate(valid_loader): start_time = time.time() acc_top1_n, acc_top5_n = exe.run( - program, - feed={ - "image": data[0].get('image'), - "label": data[0].get('label'), - }, - fetch_list=[acc_top1.name, acc_top5.name]) + program, feed=data, fetch_list=[acc_top1.name, acc_top5.name]) end_time = time.time() if batch_id % args.log_period == 0: _logger.info( diff --git a/demo/unstructured_prune/train.py b/demo/unstructured_prune/train.py index 2698753d7f8537afdaae118ae0d36175ac5af549..bcab1f73a3cc641ca6f510538a15dea3d7191268 100644 --- a/demo/unstructured_prune/train.py +++ b/demo/unstructured_prune/train.py @@ -20,6 +20,7 @@ parser = argparse.ArgumentParser(description=__doc__) add_arg = functools.partial(add_arguments, argparser=parser) # yapf: disable add_arg('batch_size', int, 64 * 4, "Minibatch size.") +add_arg('batch_size_for_validation', int, 64, "Minibatch size for validation.") 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_pretrained", "Whether to use pretrained model.") @@ -123,7 +124,7 @@ def compress(args): drop_last=False, return_list=False, use_shared_memory=True, - batch_size=batch_size_per_card, + batch_size=args.batch_size_for_validation, shuffle=False) step_per_epoch = int(np.ceil(len(train_dataset) * 1. / args.batch_size)) diff --git a/docs/zh_cn/api_cn/dygraph/pruners/unstructured_pruner.rst b/docs/zh_cn/api_cn/dygraph/pruners/unstructured_pruner.rst index 63111975148e4140ccecfa60971f3c1fe89394d9..e990a54c75988f44934d42db0410c1a7337ae8b7 100644 --- a/docs/zh_cn/api_cn/dygraph/pruners/unstructured_pruner.rst +++ b/docs/zh_cn/api_cn/dygraph/pruners/unstructured_pruner.rst @@ -7,7 +7,7 @@ UnstructuredPruner .. py:class:: paddleslim.UnstructuredPruner(model, mode, threshold=0.01, ratio=0.3, skip_params_func=None) -`源代码 `_ +`源代码 `_ 对于神经网络中的参数进行非结构化稀疏。非结构化稀疏是指，根据某些衡量指标，将不重要的参数置0。其不按照固定结构剪裁（例如一个通道等），这是和结构化剪枝的主要区别。 @@ -23,11 +23,16 @@ UnstructuredPruner **示例代码：** -此示例不能直接运行，因为需要定义和加载模型，详细用法请参考 `这里 `_ - .. code-block:: python + import paddle from paddleslim import UnstructuredPruner + from paddle.vision.models import LeNet as net + import numpy as np + + place = paddle.set_device('cpu') + model = net(num_classes=10) + pruner = UnstructuredPruner(model, mode='ratio', ratio=0.5) .. @@ -38,13 +43,19 @@ UnstructuredPruner **示例代码：** - 此示例不能直接运行，因为需要定义和加载模型，详细用法请参考 `这里 `_ - .. code-block:: python from paddleslim import UnstructuredPruner + from paddle.vision.models import LeNet as net + import numpy as np + + place = paddle.set_device('cpu') + model = net(num_classes=10) pruner = UnstructuredPruner(model, mode='ratio', ratio=0.5) + + print(pruner.threshold) pruner.step() + print(pruner.threshold) # 可以看出，这里的threshold和上面打印的不同，这是因为step函数根据设定的ratio更新了threshold数值，便于剪裁操作。 .. @@ -54,13 +65,23 @@ UnstructuredPruner **示例代码：** - 此示例不能直接运行，因为需要定义和加载模型，详细用法请参考 `这里 `_ - .. code-block:: python from paddleslim import UnstructuredPruner - pruner = UnstructuredPruner(model, mode='ratio', ratio=0.5) + from paddle.vision.models import LeNet as net + import numpy as np + + place = paddle.set_device('cpu') + model = net(num_classes=10) + pruner = UnstructuredPruner(model, mode='threshold', threshold=0.5) + + density = UnstructuredPruner.total_sparse(model) + print(density) + model(paddle.to_tensor( + np.random.uniform(0, 1, [16, 1, 28, 28]), dtype='float32')) pruner.update_params() + density = UnstructuredPruner.total_sparse(model) + print(density) # 可以看出，这里打印的模型稠密度与上述不同，这是因为update_params()函数置零了所有绝对值小于0.5的权重。 .. @@ -78,13 +99,17 @@ UnstructuredPruner **示例代码：** - 此示例不能直接运行，因为需要定义和加载模型，详细用法请参考 `这里 `_ - .. code-block:: python from paddleslim import UnstructuredPruner - density = UnstructuredPruner.total_sparse(model) + from paddle.vision.models import LeNet as net + import numpy as np + place = paddle.set_device('cpu') + model = net(num_classes=10) + density = UnstructuredPruner.total_sparse(model) + print(density) + .. .. py:method:: paddleslim.UnstructuredPruner.summarize_weights(model, ratio=0.1) @@ -102,12 +127,17 @@ UnstructuredPruner **示例代码：** - 此示例不能直接运行，因为需要定义和加载模型，详细用法请参考 `这里 `_ - .. code-block:: python from paddleslim import UnstructuredPruner + from paddle.vision.models import LeNet as net + import numpy as np + + place = paddle.set_device('cpu') + model = net(num_classes=10) pruner = UnstructuredPruner(model, mode='ratio', ratio=0.5) - threshold = pruner.summarize_weights(model, ratio=0.1) + + threshold = pruner.summarize_weights(model, 0.5) + print(threshold) .. diff --git a/docs/zh_cn/api_cn/static/prune/unstructured_prune_api.rst b/docs/zh_cn/api_cn/static/prune/unstructured_prune_api.rst index 2622fe48055e378b05787098c02dbdd6f5dba86a..1d2d080d577cac5a2b257dafb3980d9414887176 100644 --- a/docs/zh_cn/api_cn/static/prune/unstructured_prune_api.rst +++ b/docs/zh_cn/api_cn/static/prune/unstructured_prune_api.rst @@ -24,13 +24,30 @@ UnstrucuturedPruner **示例代码：** -此示例不能直接运行，因为需要加载数据和模型，详细demo请参照 `这里 `_ - .. code-block:: python - from paddleslim.prune import UnstructuredPruner - pruner = UnstructuredPruner() + import paddle as paddle + import paddle.fluid as fluid + from paddleslim.prune import UnstructuredPruner + + paddle.enable_static() + + train_program = paddle.static.default_main_program() + startup_program = paddle.static.default_startup_program() + + with fluid.program_guard(train_program, startup_program): + image = fluid.data(name='x', shape=[None, 1, 28, 28]) + label = fluid.data(name='label', shape=[None, 1], dtype='int64') + conv = fluid.layers.conv2d(image, 32, 1) + feature = fluid.layers.fc(conv, 10, act='softmax') + cost = fluid.layers.cross_entropy(input=feature, label=label) + avg_cost = fluid.layers.mean(x=cost) + place = paddle.static.cpu_places()[0] + exe = paddle.static.Executor(place) + exe.run(startup_program) + + pruner = UnstructuredPruner(paddle.static.default_main_program(), 'ratio', ratio=0.5, place=place) .. .. py:method:: paddleslim.prune.unstructured_pruner.UnstructuredPruner.step() @@ -39,33 +56,71 @@ UnstrucuturedPruner **示例代码：** - 此示例不能直接运行，因为需要加载数据和模型，详细demo请参照 `这里 `_ - .. code-block:: python + import paddle as paddle + import paddle.fluid as fluid from paddleslim.prune import UnstructuredPruner - - pruner = UnstructuredPruner( - paddle.static.default_main_program(), 'ratio', scope=paddle.static.global_scope(), place=paddle.static.cpu_places()[0]) - pruner.step() + paddle.enable_static() + + train_program = paddle.static.default_main_program() + startup_program = paddle.static.default_startup_program() + + with fluid.program_guard(train_program, startup_program): + image = fluid.data(name='x', shape=[None, 1, 28, 28]) + label = fluid.data(name='label', shape=[None, 1], dtype='int64') + conv = fluid.layers.conv2d(image, 32, 1) + feature = fluid.layers.fc(conv, 10, act='softmax') + cost = fluid.layers.cross_entropy(input=feature, label=label) + avg_cost = fluid.layers.mean(x=cost) + + place = paddle.static.cpu_places()[0] + exe = paddle.static.Executor(place) + exe.run(startup_program) + + pruner = UnstructuredPruner(paddle.static.default_main_program(), 'ratio', ratio=0.5, place=place) + print(pruner.threshold) + pruner.step() + print(pruner.threshold) # 可以看出，这里的threshold和上面打印的不同，这是因为step函数根据设定的ratio更新了threshold数值，便于剪裁操作。 .. .. py:method:: paddleslim.prune.unstructured_pruner.UnstructuredPruner.update_params() - 每一步优化后，重制模型中本来是0的权重。这一步通常用于模型evaluation和save之前，确保模型的稀疏率。 + 每一步优化后，重制模型中本来是0的权重。这一步通常用于模型evaluation和save之前，确保模型的稀疏率。但是，在训练过程中，由于step()函数会调用该方法，故不需要开发者在训练过程中额外调用了。 **示例代码：** - 此示例不能直接运行，因为需要加载数据和模型，详细demo请参照 `这里 `_ - .. code-block:: python + import paddle as paddle + import paddle.fluid as fluid from paddleslim.prune import UnstructuredPruner - pruner = UnstructuredPruner( - paddle.static.default_main_program(), 'ratio', scope=paddle.static.global_scope(), place=paddle.static.cpu_places()[0]) + paddle.enable_static() + + train_program = paddle.static.default_main_program() + startup_program = paddle.static.default_startup_program() + + with fluid.program_guard(train_program, startup_program): + image = fluid.data(name='x', shape=[None, 1, 28, 28]) + label = fluid.data(name='label', shape=[None, 1], dtype='int64') + conv = fluid.layers.conv2d(image, 32, 1) + feature = fluid.layers.fc(conv, 10, act='softmax') + cost = fluid.layers.cross_entropy(input=feature, label=label) + avg_cost = fluid.layers.mean(x=cost) + + place = paddle.static.cpu_places()[0] + exe = paddle.static.Executor(place) + exe.run(startup_program) + + pruner = UnstructuredPruner(paddle.static.default_main_program(), 'threshold', threshold=0.5, place=place) + density = UnstructuredPruner.total_sparse(paddle.static.default_main_program()) + print(density) + pruner.step() pruner.update_params() + density = UnstructuredPruner.total_sparse(paddle.static.default_main_program()) + print(density) # 可以看出，这里打印的模型稠密度与上述不同，这是因为update_params()函数置零了所有绝对值小于0.5的权重。 .. @@ -83,13 +138,31 @@ UnstrucuturedPruner **示例代码：** - 此示例不能直接运行，因为需要加载数据和模型，详细demo请参照 `这里 `_ - .. code-block:: python + import paddle as paddle + import paddle.fluid as fluid from paddleslim.prune import UnstructuredPruner + paddle.enable_static() + + train_program = paddle.static.default_main_program() + startup_program = paddle.static.default_startup_program() + + with fluid.program_guard(train_program, startup_program): + image = fluid.data(name='x', shape=[None, 1, 28, 28]) + label = fluid.data(name='label', shape=[None, 1], dtype='int64') + conv = fluid.layers.conv2d(image, 32, 1) + feature = fluid.layers.fc(conv, 10, act='softmax') + cost = fluid.layers.cross_entropy(input=feature, label=label) + avg_cost = fluid.layers.mean(x=cost) + + place = paddle.static.cpu_places()[0] + exe = paddle.static.Executor(place) + exe.run(startup_program) + density = UnstructuredPruner.total_sparse(paddle.static.default_main_program()) + print(density) .. @@ -108,14 +181,31 @@ UnstrucuturedPruner **示例代码：** - 此示例不能直接运行，因为需要加载数据和模型，详细demo请参照 `这里 `_ - .. code-block:: python + import paddle as paddle + import paddle.fluid as fluid from paddleslim.prune import UnstructuredPruner - pruner = UnstructuredPruner( - paddle.static.default_main_program(), 'ratio', scope=paddle.static.global_scope(), place=paddle.static.cpu_places()[0]) - threshold = pruner.summarize_weights(paddle.static.default_main_program(), 1.0) + paddle.enable_static() + + train_program = paddle.static.default_main_program() + startup_program = paddle.static.default_startup_program() + + with fluid.program_guard(train_program, startup_program): + image = fluid.data(name='x', shape=[None, 1, 28, 28]) + label = fluid.data(name='label', shape=[None, 1], dtype='int64') + conv = fluid.layers.conv2d(image, 32, 1) + feature = fluid.layers.fc(conv, 10, act='softmax') + cost = fluid.layers.cross_entropy(input=feature, label=label) + avg_cost = fluid.layers.mean(x=cost) + + place = paddle.static.cpu_places()[0] + exe = paddle.static.Executor(place) + exe.run(startup_program) + + threshold = pruner.summarize_weights(paddle.static.default_main_program(), ratio=0.5) + print(threshold) + ..