!1315 for second order codes

Merge pull request !1315 from zongha/master

example/resnet50_imagenet2012_THOR/config_imagenet.py

@@ -0,0 +1,45 @@
+# Copyright 2020 Huawei Technologies Co., Ltd
+#
+# 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.
+# ============================================================================
+"""
+network config setting, will be used in train.py and eval.py
+"""
+from easydict import EasyDict as ed
+
+config = ed({
+    "class_num": 1000,
+    "batch_size": 32,
+    "loss_scale": 128,
+    "momentum": 0.9,
+    "weight_decay": 5e-4,
+    "epoch_size": 50,
+    "buffer_size": 1000,
+    "image_height": 224,
+    "image_width": 224,
+    "save_checkpoint": True,
+    "save_checkpoint_steps": 5004,
+    "keep_checkpoint_max": 20,
+    "save_checkpoint_path": "./",
+    "lr_init": 0.01,
+    "lr_end": 0.00001,
+    "lr_max": 0.1,
+    "warmup_epochs": 0,
+    "lr_decay_mode": "cosine",
+    "label_smooth": 1,
+    "label_smooth_factor": 0.1,
+    "lr": 0.1,
+    "T_max": 90,
+    "eta_min": 0,
+    "frequency": 278
+})

example/resnet50_imagenet2012_THOR/crossentropy.py

@@ -0,0 +1,41 @@
+# Copyright 2020 Huawei Technologies Co., Ltd
+#
+# 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.
+# ============================================================================
+"""CrossEntropy"""
+import mindspore.nn as nn
+from mindspore import Tensor
+from mindspore.common import dtype as mstype
+from mindspore.nn.loss.loss import _Loss
+from mindspore.ops import functional as F
+from mindspore.ops import operations as P
+
+
+class CrossEntropy(_Loss):
+    """CrossEntropy"""
+    def __init__(self, smooth_factor=0., num_classes=1000):
+        super(CrossEntropy, self).__init__()
+        self.onehot = P.OneHot()
+        self.on_value = Tensor(1.0 - smooth_factor, mstype.float32)
+        self.off_value = Tensor(1.0 * smooth_factor / (num_classes - 1), mstype.float32)
+        # self.cast = P.Cast()
+        self.ce = nn.SoftmaxCrossEntropyWithLogits()
+        self.mean = P.ReduceMean(False)
+
+    def construct(self, logit, label):
+        # one_hot_label = self.onehot(self.cast(label, mstype.int32),
+        #                F.shape(logit)[1], self.on_value, self.off_value)、
+        one_hot_label = self.onehot(label, F.shape(logit)[1], self.on_value, self.off_value)
+        loss = self.ce(logit, one_hot_label)
+        loss = self.mean(loss, 0)
+        return loss

example/resnet50_imagenet2012_THOR/dataset_imagenet.py

@@ -0,0 +1,80 @@
+# Copyright 2020 Huawei Technologies Co., Ltd
+#
+# 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.
+# ============================================================================
+"""
+create train or eval dataset.
+"""
+import os
+
+import mindspore.common.dtype as mstype
+import mindspore.dataset.engine as de
+import mindspore.dataset.transforms.c_transforms as C2
+import mindspore.dataset.transforms.vision.c_transforms as V_C
+
+
+def create_dataset(dataset_path, do_train, repeat_num=1, batch_size=32):
+    """
+    create a train or eval dataset
+    Args:
+        dataset_path(string): the path of dataset.
+        do_train(bool): whether dataset is used for train or eval.
+        repeat_num(int): the repeat times of dataset. Default: 1
+        batch_size(int): the batch size of dataset. Default: 32
+    Returns:
+        dataset
+    """
+
+    device_num = int(os.getenv("RANK_SIZE"))
+    rank_id = int(os.getenv("RANK_ID"))
+
+    if device_num == 1:
+        ds = de.ImageFolderDatasetV2(dataset_path, num_parallel_workers=8, shuffle=False)
+    else:
+        ds = de.ImageFolderDatasetV2(dataset_path, num_parallel_workers=8, shuffle=True,
+                                     num_shards=device_num, shard_id=rank_id)
+
+    image_size = 224
+    mean = [0.485 * 255, 0.456 * 255, 0.406 * 255]
+    std = [0.229 * 255, 0.224 * 255, 0.225 * 255]
+    if do_train:
+        transform_img = [
+            V_C.RandomCropDecodeResize(image_size, scale=(0.08, 1.0), ratio=(0.75, 1.333)),
+            V_C.RandomHorizontalFlip(prob=0.5),
+            V_C.Normalize(mean=mean, std=std),
+            V_C.HWC2CHW()
+        ]
+    else:
+        transform_img = [
+            V_C.Decode(),
+            V_C.Resize((256, 256)),
+            V_C.CenterCrop(image_size),
+            V_C.Normalize(mean=mean, std=std),
+            V_C.HWC2CHW()
+        ]
+    # type_cast_op = C2.TypeCast(mstype.float16)
+    type_cast_op = C2.TypeCast(mstype.int32)
+
+    ds = ds.map(input_columns="image", operations=transform_img, num_parallel_workers=8)
+    ds = ds.map(input_columns="label", operations=type_cast_op, num_parallel_workers=8)
+
+    # apply shuffle operations
+    # ds = ds.shuffle(buffer_size=config.buffer_size)
+
+    # apply batch operations
+    ds = ds.batch(batch_size, drop_remainder=True)
+
+    # apply dataset repeat operation
+    ds = ds.repeat(repeat_num)
+
+    return ds

example/resnet50_imagenet2012_THOR/lr_generator.py

@@ -0,0 +1,126 @@
+# Copyright 2020 Huawei Technologies Co., Ltd
+#
+# 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.
+# ============================================================================
+"""learning rate generator"""
+import math
+
+import numpy as np
+
+
+def linear_warmup_lr(current_step, warmup_steps, base_lr, init_lr):
+    """linear_warmup_lr"""
+    lr_inc = (float(base_lr) - float(init_lr)) / float(warmup_steps)
+    lr = float(init_lr) + lr_inc * current_step
+    return lr
+
+
+def cosine_annealing_lr(lr, steps_per_epoch, warmup_epochs, max_epoch, T_max, eta_min=0, num_periods=0.5):
+    """linear_warmup_lr"""
+    base_lr = lr
+    warmup_init_lr = 0
+    total_steps = int(max_epoch * steps_per_epoch)
+    warmup_steps = int(warmup_epochs * steps_per_epoch)
+    decay_steps = total_steps - warmup_steps
+    lr_each_step = []
+    for i in range(total_steps):
+        if i < warmup_steps:
+            lr = linear_warmup_lr(i + 1, warmup_steps, base_lr, warmup_init_lr)
+        else:
+            # linear_decay = (total_steps - i) / decay_steps
+            cosine_decay = 0.5 * (1 + math.cos(math.pi * i / decay_steps))
+            decayed = cosine_decay
+            lr = base_lr * decayed
+        lr_each_step.append(lr)
+    return np.array(lr_each_step).astype(np.float32)
+
+
+def warmup_cosine_annealing_lr(lr, steps_per_epoch, warmup_epochs, max_epoch, T_max, eta_min=0, num_periods=0.5):
+    """warmup_cosine_annealing_lr"""
+    base_lr = lr
+    warmup_init_lr = 0
+    total_steps = int(max_epoch * steps_per_epoch * 0.99)
+    warmup_steps = int(warmup_epochs * steps_per_epoch)
+    decay_steps = total_steps - warmup_steps
+    lr_each_step = []
+    for i in range(total_steps):
+        if i < warmup_steps:
+            lr = linear_warmup_lr(i + 1, warmup_steps, base_lr, warmup_init_lr)
+        else:
+            linear_decay = (total_steps - i) / decay_steps
+            cosine_decay = 0.5 * (1 + math.cos(math.pi * 2 * num_periods * i / decay_steps))
+            decayed = linear_decay * cosine_decay
+            lr = base_lr * decayed + 0.000005
+        lr_each_step.append(lr)
+    return np.array(lr_each_step).astype(np.float32)
+
+
+def get_lr(global_step, lr_init, lr_end, lr_max, warmup_epochs, total_epochs, steps_per_epoch, lr_decay_mode):
+    """
+    generate learning rate array
+
+    Args:
+       global_step(int): total steps of the training
+       lr_init(float): init learning rate
+       lr_end(float): end learning rate
+       lr_max(float): max learning rate
+       warmup_epochs(int): number of warmup epochs
+       total_epochs(int): total epoch of training
+       steps_per_epoch(int): steps of one epoch
+       lr_decay_mode(string): learning rate decay mode, including steps, poly or default
+
+    Returns:
+       np.array, learning rate array
+    """
+    lr_each_step = []
+    total_steps = steps_per_epoch * total_epochs
+    warmup_steps = steps_per_epoch * warmup_epochs
+    if lr_decay_mode == 'steps':
+        decay_epoch_index = [0.3 * total_steps, 0.6 * total_steps, 0.8 * total_steps]
+        for i in range(total_steps):
+            if i < decay_epoch_index[0]:
+                lr = lr_max
+            elif i < decay_epoch_index[1]:
+                lr = lr_max * 0.1
+            elif i < decay_epoch_index[2]:
+                lr = lr_max * 0.01
+            else:
+                lr = lr_max * 0.001
+            lr_each_step.append(lr)
+    elif lr_decay_mode == 'poly':
+        if warmup_steps != 0:
+            inc_each_step = (float(lr_max) - float(lr_init)) / float(warmup_steps)
+        else:
+            inc_each_step = 0
+        for i in range(total_steps):
+            if i < warmup_steps:
+                lr = float(lr_init) + inc_each_step * float(i)
+            else:
+                base = (1.0 - (float(i) - float(warmup_steps)) / (float(total_steps) - float(warmup_steps)))
+                lr = float(lr_max) * base * base
+                if lr < 0.0:
+                    lr = 0.0
+            lr_each_step.append(lr)
+    else:
+        for i in range(total_steps):
+            if i < warmup_steps:
+                lr = lr_init + (lr_max - lr_init) * i / warmup_steps
+            else:
+                lr = lr_max - (lr_max - lr_end) * (i - warmup_steps) / (total_steps - warmup_steps)
+            lr_each_step.append(lr)
+
+    current_step = global_step
+    lr_each_step = np.array(lr_each_step).astype(np.float32)
+    learning_rate = lr_each_step[current_step:]
+
+    return learning_rate

example/resnet50_imagenet2012_THOR/model/dataset_helper.py

@@ -0,0 +1,191 @@
+# Copyright 2020 Huawei Technologies Co., Ltd
+#
+# 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.
+# ============================================================================
+"""Dataset help for minddata dataset"""
+from mindspore import context
+from mindspore._checkparam import check_bool
+from mindspore.nn.wrap import GetNextSingleOp
+from mindspore.parallel._utils import _get_device_num, _get_global_rank, _get_parallel_mode
+from mindspore.train._utils import _exec_datagraph, _get_types_and_shapes, _to_tensor, \
+    _construct_tensor_list, _to_full_shapes, _to_full_tensor
+from mindspore.train.parallel_utils import ParallelMode
+
+
+class DatasetHelper:
+    """
+    Help function to use the Minddata dataset.
+
+    According to different context, change the iter of dataset, to use the same for loop in different context.
+
+    Note:
+        The iter of DatasetHelper will give one epoch data.
+
+    Args:
+        dataset (DataSet): The dataset.
+        dataset_sink_mode (bool): If true use GetNext to fetch the data, or else feed the data from host.
+            Default: True.
+
+    Examples:
+        >>> dataset_helper = DatasetHelper(dataset)
+        >>> for inputs in dataset_helper:
+        >>>     outputs = network(*inputs)
+    """
+
+    def __init__(self, dataset, first_order_iter=0, dataset_sink_mode=True):
+        check_bool(dataset_sink_mode)
+
+        iterclass = _DatasetIterGE
+        if not dataset_sink_mode:
+            iterclass = _DatasetIterFeed
+        elif not context.get_context("enable_ge"):
+            if context.get_context("enable_loop_sink"):
+                iterclass = _DatasetIterMSLoopSink
+            else:
+                iterclass = _DatasetIterMS
+
+        self.iter = iterclass(dataset, first_order_iter)
+
+    def __iter__(self):
+        return self.iter.__iter__()
+
+    # A temp solution for loop sink. Delete later
+    def types_shapes(self):
+        """Get the types and shapes from dataset on current config."""
+        return self.iter.types_shapes()
+
+    def loop_size(self):
+        """Get loop_size for every iteration."""
+        return self.iter.loop_size
+
+
+class _DatasetIter:
+    """Base iter for dataset help"""
+
+    def __init__(self, dataset):
+        self.loop_size = 1
+        if not hasattr(dataset, '__ME_INITED__'):
+            if not hasattr(dataset, '__loop_size__'):
+                self.loop_size = dataset.get_dataset_size()
+            else:
+                self.loop_size = dataset.__loop_size__
+            dataset.__ME_INITED__ = _exec_datagraph(dataset, self.loop_size).queue_name
+
+        self.ind = 0
+        self.dataset = dataset
+        dataset_types, dataset_shapes = _get_types_and_shapes(dataset)
+        self.dataset_types, self.dataset_shapes = dataset_types, dataset_shapes
+        # for self._parallel_mode equal to semi_auto_parallel or auto_parallel, use a complete tensor to
+        # compile, and slice tensor to run. The batch dimension of tensors for compile is device_number
+        # times the batch dimension of tensors for run
+        if _get_parallel_mode() in (ParallelMode.SEMI_AUTO_PARALLEL, ParallelMode.AUTO_PARALLEL):
+            device_num = _get_device_num()
+            self.dataset_shapes = _to_full_shapes(dataset_shapes, device_num)
+
+    def __iter__(self):
+        self.ind = 0
+        return self
+
+    def __next__(self):
+        if self.ind >= self.loop_count:
+            raise StopIteration()
+        self.ind += 1
+        return self.op()
+
+    def types_shapes(self):
+        return self.dataset_types, self.dataset_shapes
+
+    def get_loop_count(self, dataset):
+        loop_count = 1
+        if hasattr(dataset, '__loop_size__'):
+            loop_size = dataset.__loop_size__
+            loop_count = int(dataset.get_dataset_size() / loop_size)
+        return loop_count
+
+
+class _DatasetIterMSLoopSink(_DatasetIter):
+    """Iter for context (enable_loop_sink=True)"""
+
+    def __init__(self, dataset, first_order_iter):
+        super(_DatasetIterMSLoopSink, self).__init__(dataset)
+        # self.loop_count = self.get_loop_count(dataset)
+        loop_size = dataset.__loop_size__ + first_order_iter
+        self.loop_count = int(dataset.get_dataset_size() / loop_size) * 2
+
+        def op():
+            return tuple()
+
+        self.op = op
+
+
+class _DatasetIterMS(_DatasetIter):
+    """Iter for context (enable_loop_sink=False)"""
+
+    def __init__(self, dataset, first_order_order):
+        super(_DatasetIterMS, self).__init__(dataset)
+        self.loop_count = dataset.get_dataset_size()
+        self.loop_size = 1
+        queue_name = dataset.__ME_INITED__
+        self.op = GetNextSingleOp(self.dataset_types, self.dataset_shapes, queue_name)
+
+
+class _DatasetIterGE(_DatasetIter):
+    """Iter for ge"""
+
+    def __init__(self, dataset):
+        super(_DatasetIterGE, self).__init__(dataset)
+        self.loop_count = self.get_loop_count(dataset)
+        parallel_mode = _get_parallel_mode()
+        self.need_to_full = parallel_mode in (ParallelMode.SEMI_AUTO_PARALLEL, ParallelMode.AUTO_PARALLEL)
+        batch_expand_num = 1
+        if self.need_to_full:
+            batch_expand_num = _get_device_num()
+        tensor_list_run = _construct_tensor_list(self.dataset_types, self.dataset_shapes, batch_expand_num)
+
+        def op():
+            return tensor_list_run
+
+        self.op = op
+
+
+class _DatasetIterFeed:
+    """Iter for feed data"""
+
+    def __init__(self, dataset, first_order_order):
+        self.dataset = dataset
+        self.device_num = _get_device_num()
+        self.global_rank = _get_global_rank()
+        self.repeat_count = dataset.get_repeat_count()
+        self.repeat_ind = 0
+        self.loop_count = dataset.get_dataset_size()
+        self.ind = 0
+
+        parallel_mode = context.get_auto_parallel_context("parallel_mode")
+        self.need_to_full = parallel_mode in (ParallelMode.SEMI_AUTO_PARALLEL, ParallelMode.AUTO_PARALLEL)
+
+    def __iter__(self):
+        if self.repeat_ind % self.repeat_count == 0:
+            self.iter = self.dataset.__iter__()
+
+        self.repeat_ind += 1
+        self.ind = 0
+        return self
+
+    def __next__(self):
+        if self.ind >= self.loop_count:
+            raise StopIteration()
+        self.ind += 1
+        data = self.iter.__next__()
+        if self.need_to_full:
+            return _to_full_tensor(data, self.device_num, self.global_rank)
+        return _to_tensor(data)

example/resnet50_imagenet2012_THOR/model/grad_reducer_thor.py

@@ -0,0 +1,183 @@
+# Copyright 2020 Huawei Technologies Co., Ltd
+#
+# 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.
+# ============================================================================
+"""grad_reducer_thor"""
+import mindspore.common.dtype as mstype
+from mindspore.communication.management import GlobalComm, get_group_size
+from mindspore.nn.cell import Cell
+from mindspore.ops import functional as F, composite as C, operations as P
+from mindspore.ops.operations.comm_ops import AllReduce, ReduceOp
+
+reduce_opt = C.MultitypeFuncGraph("reduce_opt")
+
+_all_reduce_A = AllReduce()
+
+
+def _init_optimizer_allreduce(group):
+    global _all_reduce_A
+    _all_reduce_A = AllReduce(ReduceOp.SUM, GlobalComm.WORLD_COMM_GROUP)
+    _all_reduce_A.add_prim_attr('fusion', group)
+
+
+@reduce_opt.register("Function", "Number", "Tensor")
+def _tensors_allreduce_mean(mul, degree, grad):
+    degree = F.scalar_cast(degree, F.dtype(grad))
+    grad = _all_reduce_A(grad)
+    cast_op = P.Cast()
+    return mul(grad, cast_op(F.scalar_to_array(1.0 / degree), F.dtype(grad)))
+
+
+@reduce_opt.register("Bool", "Tensor")
+def _tensors_allreduce(allreduce_filter, grad):
+    if allreduce_filter:
+        return _all_reduce_A(grad)
+    return grad
+
+
+_get_datatype = C.MultitypeFuncGraph("_get_datatype")
+
+
+@_get_datatype.register("Tensor")
+def _tensors_get_datatype(grad):
+    """
+    Acquire gradient datatype.
+
+    Args:
+        grad (Tensor): The gradient tensor before operation.
+
+    Returns:
+        mstype, the datatype of gradient.
+    """
+    return F.dtype(grad)
+
+
+_cast_datatype = C.MultitypeFuncGraph("_cast_datatype")
+
+
+@_cast_datatype.register("TypeType", "Tensor")
+def _tensors_cast_datatype(datatype, grad):
+    """
+    Cast gradient to datatype.
+
+    Args:
+        datatype (mstype): the destination datatype of gradient.
+        grad (Tensor): The gradient tensor before operation.
+
+    Returns:
+        Tensor, the gradient tensor after operation.
+    """
+    return F.cast(grad, datatype)
+
+
+class DistributedGradReducerThor(Cell):
+    """
+    A distributed optimizer.
+
+    Constructs a gradient reducer Cell, which applies communication and average operations on
+    single-process gradient values.
+
+    Args:
+        parameters (list): the parameters to be updated.
+        mean (bool): When mean is true, the mean coefficient (degree) would apply on gradients. Default: False.
+        degree (int): The mean coefficient. Usually it equals to device number. Default: None.
+
+    Raises:
+        ValueError: If degree is not a int or less than 0.
+
+    Examples:
+        >>> from mindspore.communication import init, get_group_size
+        >>> from mindspore.ops import composite as C
+        >>> from mindspore.ops import operations as P
+        >>> from mindspore.ops import functional as F
+        >>> from mindspore import context
+        >>> from mindspore import nn
+        >>> from mindspore import ParallelMode, ParameterTuple
+        >>>
+        >>> device_id = int(os.environ["DEVICE_ID"])
+        >>> context.set_context(mode=context.GRAPH_MODE, device_target="Ascend", save_graphs=True,
+        >>>                     device_id=int(device_id), enable_hccl=True)
+        >>> init()
+        >>> context.reset_auto_parallel_context()
+        >>> context.set_auto_parallel_context(parallel_mode=ParallelMode.DATA_PARALLEL)
+        >>>
+        >>>
+        >>> class TrainingWrapper(nn.Cell):
+        >>>     def __init__(self, network, optimizer, sens=1.0):
+        >>>         super(TrainingWrapper, self).__init__(auto_prefix=False)
+        >>>         self.network = network
+        >>>         self.network.add_flags(defer_inline=True)
+        >>>         self.weights = ParameterTuple(network.trainable_params())
+        >>>         self.optimizer = optimizer
+        >>>         self.grad = C.GradOperation('grad', get_by_list=True, sens_param=True)
+        >>>         self.sens = sens
+        >>>         self.reducer_flag = False
+        >>>         self.grad_reducer = None
+        >>>         self.parallel_mode = context.get_auto_parallel_context("parallel_mode")
+        >>>         if self.parallel_mode in [ParallelMode.DATA_PARALLEL,
+        >>>                                            ParallelMode.HYBRID_PARALLEL]:
+        >>>             self.reducer_flag = True
+        >>>         if self.reducer_flag:
+        >>>             mean = context.get_auto_parallel_context("mirror_mean")
+        >>>             if mean.get_device_num_is_set():
+        >>>                 degree = context.get_auto_parallel_context("device_num")
+        >>>             else:
+        >>>                 degree = get_group_size()
+        >>>             self.grad_reducer = nn.DistributedGradReducer(optimizer.parameters, mean, degree)
+        >>>
+        >>>     def construct(self, *args):
+        >>>         weights = self.weights
+        >>>         loss = self.network(*args)
+        >>>         sens = P.Fill()(P.DType()(loss), P.Shape()(loss), self.sens)
+        >>>         grads = self.grad(self.network, weights)(*args, sens)
+        >>>         if self.reducer_flag:
+        >>>             # apply grad reducer on grads
+        >>>             grads = self.grad_reducer(grads)
+        >>>         return F.depend(loss, self.optimizer(grads))
+        >>>
+        >>> network = Net()
+        >>> optimizer = nn.Momentum(network.trainable_params(), learning_rate=0.1, momentum=0.9)
+        >>> train_cell = TrainingWrapper(network, optimizer)
+        >>> inputs = Tensor(np.ones([16, 16]).astype(np.float32))
+        >>> label = Tensor(np.zeros([16, 16]).astype(np.float32))
+        >>> grads = train_cell(inputs, label)
+    """
+
+    def __init__(self, parameters, group, mean=True, degree=None):
+        super(DistributedGradReducerThor, self).__init__(auto_prefix=False)
+        self.hyper_map = C.HyperMap()
+        self.mul = P.Mul()
+        if degree is None:
+            self.degree = get_group_size()
+        else:
+            if not isinstance(degree, int) or degree <= 0:
+                raise ValueError("Parameter 'degree' in DistributedGradReducer should large than 0 and be int")
+            self.degree = degree
+        self.mean = mean
+        self.allreduce_filter = tuple(x.layerwise_parallel is False for x in parameters)
+        _init_optimizer_allreduce(group)
+
+    def construct(self, grads):
+        # In some circumstances, the data precision of grads could be mixed with float16 and float32. Thus, the
+        # result of AllReduce is unreliable. To solve the problem, grads should be cast to float32 before AllReduce,
+        # and cast back after the operation.
+        datatypes = self.hyper_map(F.partial(_get_datatype), grads)
+        grads = self.hyper_map(F.partial(_cast_datatype, mstype.float32), grads)
+
+        if self.mean:
+            new_grad = self.hyper_map(F.partial(reduce_opt, self.mul, self.degree), grads)
+        else:
+            new_grad = self.hyper_map(F.partial(reduce_opt), self.allreduce_filter, grads)
+
+        new_grad = self.hyper_map(F.partial(_cast_datatype), datatypes, new_grad)
+        return new_grad

example/resnet50_imagenet2012_THOR/model/thor.py

@@ -0,0 +1,202 @@
+# Copyright 2020 Huawei Technologies Co., Ltd
+#
+# 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.
+# ============================================================================
+"""momentum"""
+import mindspore.common.dtype as mstype
+from mindspore.common.initializer import initializer
+from mindspore.common.parameter import Parameter
+from mindspore.common.parameter import ParameterTuple
+from mindspore.common.tensor import Tensor
+from mindspore.nn.optim.optimizer import Optimizer
+from mindspore.ops import functional as F, composite as C, operations as P
+from mindspore.parallel._utils import _get_device_num, _get_mirror_mean
+
+from cus_ops.cus_matmul_cube_dense_right import CusMatMulCubeDenseRight
+from cus_ops.cus_matmul_cube_fracz_left_cast import CusMatMulCubeFraczLeftCast
+from cus_ops.cus_matmul_cube_dense_left import CusMatMulCubeDenseLeft
+from cus_ops.cus_matmul_cube_fracz_right_mul import CusMatMulCubeFraczRightMul
+from model.grad_reducer_thor import DistributedGradReducerThor
+
+momentum_opt = C.MultitypeFuncGraph("momentum_opt")
+
+
+@momentum_opt.register("Function", "Tensor", "Tensor", "Tensor", "Tensor", "Tensor")
+def _tensor_run_opt_ext(opt, learning_rate, momentum, gradient, weight, moment):
+    """Apply momentum optimizer to the weight parameter using Tensor."""
+    success = True
+    success = F.depend(success, opt(weight, moment, learning_rate, gradient, momentum))
+    return success
+
+
+op_add = P.AddN()
+apply_decay = C.MultitypeFuncGraph("apply_decay")
+
+
+@apply_decay.register("Number", "Bool", "Tensor", "Tensor")
+def _tensor_apply_decay(weight_decay, if_apply, weight, gradient):
+    """Get grad with weight_decay."""
+    if if_apply:
+        return op_add((weight * weight_decay, gradient))
+    return gradient
+
+
+class THOR(Optimizer):
+    """THOR"""
+    def __init__(self, params, learning_rate, momentum, matrix_A, matrix_G, A_inv_max, G_inv_max, weight_decay=0.0,
+                 loss_scale=1.0,
+                 decay_filter=lambda x: x.name not in []):
+        super(THOR, self).__init__(learning_rate, params, weight_decay, loss_scale)
+        if isinstance(momentum, float) and momentum < 0.0:
+            raise ValueError("momentum should be at least 0.0, but got momentum {}".format(momentum))
+        self.momentum = Parameter(Tensor(momentum, mstype.float32), name="momentum")
+        self.params = self.parameters
+        self.moments = self.params.clone(prefix="moments", init='zeros')
+        self.hyper_map = C.HyperMap()
+        self.opt = P.ApplyMomentum()
+        self.matrix_A = ParameterTuple(matrix_A)
+        self.matrix_G = ParameterTuple(matrix_G)
+        self.A_inv_max = ParameterTuple(A_inv_max)
+        self.G_inv_max = ParameterTuple(G_inv_max)
+        self.cube_matmul_left = CusMatMulCubeFraczLeftCast()
+        self.cube_matmul_left_fc = CusMatMulCubeDenseLeft()
+        self.cube_matmul_right_fc = CusMatMulCubeDenseRight()
+        self.cube_matmul_right_mul = CusMatMulCubeFraczRightMul()
+        self.transpose = P.Transpose()
+        self.shape = P.Shape()
+        self.reshape = P.Reshape()
+        self.mul = P.Mul()
+        self.weight_idx = []
+        for i in range(len(self.params)):
+            if "conv" in self.params[i].name or "end_point" in self.params[i].name:
+                self.weight_idx.append(i)
+        self.weight_idx.append(len(self.params))
+        self.feature_map = [1.0 / 12544, 1.0 / 3136, 1.0 / 3136, 1.0 / 3136, 1.0 / 3136, 1.0 / 3136, 1.0 / 3136,
+                            1.0 / 3136, 1.0 / 3136, 1.0 / 3136, 1.0 / 3136, 1.0 / 3136,
+                            1.0 / 784, 1.0 / 784, 1.0 / 784, 1.0 / 784, 1.0 / 784, 1.0 / 784, 1.0 / 784, 1.0 / 784,
+                            1.0 / 784, 1.0 / 784, 1.0 / 784, 1.0 / 784, 1.0 / 784,
+                            1.0 / 196, 1.0 / 196, 1.0 / 196, 1.0 / 196, 1.0 / 196, 1.0 / 196, 1.0 / 196, 1.0 / 196,
+                            1.0 / 196, 1.0 / 196, 1.0 / 196, 1.0 / 196, 1.0 / 196, 1.0 / 196, 1.0 / 196, 1.0 / 196,
+                            1.0 / 196, 1.0 / 196, 1.0 / 196,
+                            1.0 / 49, 1.0 / 49, 1.0 / 49, 1.0 / 49, 1.0 / 49, 1.0 / 49, 1.0 / 49, 1.0 / 49, 1.0 / 49,
+                            1.0]
+        mean = _get_mirror_mean()
+        degree = _get_device_num()
+        self.grad_reducer_Amax = DistributedGradReducerThor(self.parameters, 2, mean, degree)
+        self.grad_reducer_Gmax = DistributedGradReducerThor(self.parameters, 5, mean, degree)
+        self.grad_reducer_A = DistributedGradReducerThor(self.parameters, 3, mean, degree)
+        self.grad_reducer_G = DistributedGradReducerThor(self.parameters, 4, mean, degree)
+        self.matrix_A_inv = ()
+        self.matrix_G_inv = ()
+        self.matrix_max_inv = ()
+
+        for i in range(54):
+            self.matrix_max_inv = self.matrix_max_inv + (
+                Parameter(initializer(1, [1], mstype.float32), name="matrix_max" + str(i), requires_grad=False),)
+        self.log = P.Log()
+        self.exp = P.Exp()
+        self.sqrt = P.Sqrt()
+        self.matrix_max_inv = ParameterTuple(self.matrix_max_inv)
+        self.assign = P.Assign()
+        self.cast = P.Cast()
+        self.thor = True
+        self.weight_decay = weight_decay * loss_scale
+        self.decay_flags = tuple(decay_filter(x) for x in self.parameters)
+
+    def construct(self, gradients):
+        params = self.params
+        moments = self.moments
+        if self.thor:
+            matrix_A_allreduce = ()
+            matrix_G_allreduce = ()
+            matrix_A_max_allreduce = ()
+            matrix_G_max_allreduce = ()
+            for i in range(54):
+                g = gradients[i * 3]
+                matrix_A = self.matrix_A[i]
+                matrix_G = self.matrix_G[i]
+                A_max = self.A_inv_max[i]
+                G_max = self.G_inv_max[i]
+                matrix_A = F.depend(matrix_A, g)
+                matrix_G = F.depend(matrix_G, g)
+                A_max = F.depend(A_max, g)
+                G_max = F.depend(G_max, g)
+                matrix_A_allreduce = matrix_A_allreduce + (matrix_A,)
+                matrix_G_allreduce = matrix_G_allreduce + (matrix_G,)
+                matrix_A_max_allreduce = matrix_A_max_allreduce + (A_max,)
+                matrix_G_max_allreduce = matrix_G_max_allreduce + (G_max,)
+            matrix_A_allreduce = self.grad_reducer_A(matrix_A_allreduce)
+            matrix_G_allreduce = self.grad_reducer_G(matrix_G_allreduce)
+            matrix_A_max_allreduce = self.grad_reducer_Amax(matrix_A_max_allreduce)
+            matrix_G_max_allreduce = self.grad_reducer_Gmax(matrix_G_max_allreduce)
+            new_grads = ()
+            for i in range(54):
+                g = gradients[i * 3]
+                temp_a = matrix_A_allreduce[i]
+                temp_g = matrix_G_allreduce[i]
+                temp_a = self.cast(temp_a, mstype.float32)
+                temp_g = self.cast(temp_g, mstype.float32)
+                matrix_A_inv_max = self.log(matrix_A_max_allreduce[i])
+                matrix_A_inv_max = self.mul(matrix_A_inv_max, -1)
+                matrix_A_inv_max = self.exp(matrix_A_inv_max)
+                temp_a = self.mul(temp_a, matrix_A_inv_max)
+                matrix_G_inv_max = self.log(matrix_G_max_allreduce[i])
+                matrix_G_inv_max = self.mul(matrix_G_inv_max, -1)
+                matrix_G_inv_max = self.exp(matrix_G_inv_max)
+                temp_g = self.mul(temp_g, matrix_G_inv_max)
+                temp_max = self.mul(matrix_A_max_allreduce[i], matrix_G_max_allreduce[i])
+                temp_max = self.mul(temp_max, self.feature_map[i])
+                if i == 53:
+                    g = self.cube_matmul_left_fc(temp_g, g)
+                    g = self.cube_matmul_right_fc(g, temp_a, temp_max)
+                else:
+                    g = self.cube_matmul_left(temp_g, g)
+                    g = self.cube_matmul_right_mul(g, temp_a, temp_max)
+                fake_A = self.assign(self.matrix_A[i], temp_a)
+                fake_G = self.assign(self.matrix_G[i], temp_g)
+                fake_max = self.assign(self.matrix_max_inv[i], temp_max)
+                g = F.depend(g, fake_A)
+                g = F.depend(g, fake_G)
+                g = F.depend(g, fake_max)
+                if i == 53:
+                    new_grads = new_grads + (g,)
+                else:
+                    new_grads = new_grads + (g, gradients[i * 3 + 1], gradients[i * 3 + 2])
+            gradients = new_grads
+        else:
+            new_grads = ()
+            for i in range(54):
+                g = gradients[i * 3]
+                matrix_A = self.matrix_A[i]
+                matrix_G = self.matrix_G[i]
+                matrix_max = self.matrix_max_inv[i]
+                matrix_A = F.depend(matrix_A, g)
+                matrix_G = F.depend(matrix_G, g)
+                matrix_max = F.depend(matrix_max, g)
+                if i == 53:
+                    g = self.cube_matmul_left_fc(matrix_G, g)
+                    g = self.cube_matmul_right_fc(g, matrix_A, matrix_max)
+                    new_grads = new_grads + (g,)
+                else:
+                    g = self.cube_matmul_left(matrix_G, g)
+                    g = self.cube_matmul_right_mul(g, matrix_A, matrix_max)
+                    new_grads = new_grads + (g, gradients[i * 3 + 1], gradients[i * 3 + 2])
+            gradients = new_grads
+
+        if self.weight_decay > 0:
+            gradients = self.hyper_map(F.partial(apply_decay, self.weight_decay), self.decay_flags,
+                                       params, gradients)
+        gradients = self.scale_grad(gradients)
+        lr = self.get_lr()
+        success = self.hyper_map(F.partial(momentum_opt, self.opt, lr, self.momentum), gradients, params, moments)
+        return success

example/resnet50_imagenet2012_THOR/run_distribute_train_new.sh

@@ -0,0 +1,56 @@
+#!/bin/bash
+# Copyright 2020 Huawei Technologies Co., Ltd
+#
+# 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.
+# ============================================================================
+
+if [ $# != 3 ]
+then
+    echo "Usage: sh run_distribute_train.sh [MINDSPORE_HCCL_CONFIG_PATH] [DATASET_PATH] [DEVICE_NUM]"
+exit 1
+fi
+
+if [ ! -f $1 ]
+then
+    echo "error: DMINDSPORE_HCCL_CONFIG_PATH=$1 is not a file"
+exit 1
+fi
+
+if [ ! -d $2 ]
+then
+    echo "error: DATASET_PATH=$2 is not a directory"
+exit 1
+fi
+
+ulimit -u unlimited
+export DEVICE_NUM=$3
+export RANK_SIZE=$3
+export MINDSPORE_HCCL_CONFIG_PATH=$1
+
+for((i=0; i<${DEVICE_NUM}; i++))
+do
+    export DEVICE_ID=$i
+    export RANK_ID=$i
+    rm -rf ./train_parallel$i
+    mkdir ./train_parallel$i
+    cp *.py ./train_parallel$i
+    cp *.sh ./train_parallel$i
+    cp -r second_order ./train_parallel$i/second_order
+    cp -r test_ops ./train_parallel$i/test_ops
+    cd ./train_parallel$i || exit
+    echo "start training for rank $RANK_ID, device $DEVICE_ID"
+
+    env > env.log
+    python train_0517_1.py --do_train=True --run_distribute=True --device_num=$DEVICE_NUM --dataset_path=$2 > log 2>&1 &
+    cd ..
+done

example/resnet50_imagenet2012_THOR/train.py

@@ -0,0 +1,143 @@
+# Copyright 2020 Huawei Technologies Co., Ltd
+#
+# 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.
+# ============================================================================
+"""train_imagenet."""
+import argparse
+import os
+import random
+
+import mindspore.dataset.engine as de
+from mindspore import Tensor
+from mindspore import context
+from mindspore.communication.management import init
+from mindspore.parallel._auto_parallel_context import auto_parallel_context
+from mindspore.train.callback import ModelCheckpoint, CheckpointConfig, LossMonitor, TimeMonitor
+from mindspore.train.loss_scale_manager import FixedLossScaleManager
+from mindspore.train.model import ParallelMode
+from second_order.model_second_order import Model
+from second_order.resnet import resnet50
+from second_order.thor import THOR
+
+import numpy as np
+from config_imagenet import config
+from crossentropy import CrossEntropy
+from dataset_imagenet import create_dataset
+from lr_generator import  warmup_cosine_annealing_lr
+
+random.seed(1)
+np.random.seed(1)
+de.config.set_seed(1)
+
+parser = argparse.ArgumentParser(description='Image classification')
+parser.add_argument('--run_distribute', type=bool, default=False, help='Run distribute')
+parser.add_argument('--device_num', type=int, default=1, help='Device num.')
+parser.add_argument('--do_train', type=bool, default=True, help='Do train or not.')
+parser.add_argument('--do_eval', type=bool, default=False, help='Do eval or not.')
+parser.add_argument('--dataset_path', type=str, default=None, help='Dataset path')
+
+args_opt = parser.parse_args()
+device_id = int(os.getenv('DEVICE_ID'))
+
+context.set_context(mode=context.GRAPH_MODE, device_target="Ascend", save_graphs=True, device_id=device_id)
+context.set_context(enable_task_sink=True)
+context.set_context(enable_loop_sink=True)
+context.set_context(enable_mem_reuse=True)
+
+
+def get_second_order_lr(global_step, lr_init, decay, total_epochs, steps_per_epoch):
+    """get_second_order_lr"""
+    lr_each_step = []
+    total_steps = steps_per_epoch * total_epochs
+    for i in range(total_steps):
+        epoch = (i + 1) / steps_per_epoch
+        base = (1.0 - float(epoch) / total_epochs) ** decay
+        lr_local = lr_init * base
+        lr_each_step.append(lr_local)
+    current_step = global_step
+    lr_each_step = np.array(lr_each_step).astype(np.float32)
+    print("learning_rate_is=====", lr_each_step)
+    learning_rate = lr_each_step[current_step:]
+    return learning_rate
+
+
+def get_second_order_damping(global_step, damping_init, decay_rate, total_epochs, steps_per_epoch):
+    """get_second_order_damping"""
+    damping_each_step = []
+    total_steps = steps_per_epoch * total_epochs
+    for step in range(total_steps):
+        epoch = (step + 1) / steps_per_epoch
+        damping_here = damping_init * (decay_rate ** (epoch / 10))
+        damping_each_step.append(damping_here)
+
+    current_step = global_step
+    damping_each_step = np.array(damping_each_step).astype(np.float32)
+    damping_now = damping_each_step[current_step:]
+    print("damping_is=========", damping_now)
+    return damping_now
+
+
+if __name__ == '__main__':
+    if args_opt.do_eval:
+        print("eval")
+    else:
+        if args_opt.run_distribute:
+            context.set_auto_parallel_context(device_num=args_opt.device_num, parallel_mode=ParallelMode.DATA_PARALLEL,
+                                              mirror_mean=True, parameter_broadcast=True)
+            auto_parallel_context().set_all_reduce_fusion_split_indices([80], "hccl_world_groupsum1")
+            auto_parallel_context().set_all_reduce_fusion_split_indices([27], "hccl_world_groupsum3")
+            auto_parallel_context().set_all_reduce_fusion_split_indices([27], "hccl_world_groupsum4")
+            init()
+        else:
+            print(" ")
+
+    epoch_size = config.epoch_size
+    damping = get_second_order_damping(0, 0.03, 0.87, 50, 5004)
+    net = resnet50(class_num=config.class_num, damping=damping, loss_scale=config.loss_scale,
+                   frequency=config.frequency)
+
+    if not config.label_smooth:
+        config.label_smooth_factor = 0.0
+    loss = CrossEntropy(smooth_factor=config.label_smooth_factor, num_classes=config.class_num)
+    if args_opt.do_train:
+        dataset = create_dataset(dataset_path=args_opt.dataset_path, do_train=True,
+                                 repeat_num=epoch_size, batch_size=config.batch_size)
+        step_size = dataset.get_dataset_size()
+
+        loss_scale = FixedLossScaleManager(config.loss_scale, drop_overflow_update=False)
+        lr = Tensor(warmup_cosine_annealing_lr(0.035,
+                                               step_size,
+                                               config.warmup_epochs,
+                                               50,
+                                               config.T_max,
+                                               config.eta_min))
+        opt = THOR(filter(lambda x: x.requires_grad, net.get_parameters()), lr,
+                   config.momentum, damping, config.frequency,
+                   filter(lambda x: 'matrix_A' in x.name, net.get_parameters()),
+                   filter(lambda x: 'matrix_G' in x.name, net.get_parameters()),
+                   filter(lambda x: 'spatial_norm' in x.name, net.get_parameters()),
+                   config.weight_decay, config.loss_scale)
+
+        model = Model(net, loss_fn=loss, optimizer=opt, amp_level='O2', loss_scale_manager=loss_scale,
+                      keep_batchnorm_fp32=False, metrics={'acc'}, frequency=config.frequency)
+
+        time_cb = TimeMonitor(data_size=step_size)
+        loss_cb = LossMonitor()
+        cb = [time_cb, loss_cb]
+        if config.save_checkpoint:
+            config_ck = CheckpointConfig(save_checkpoint_steps=config.save_checkpoint_steps,
+                                         keep_checkpoint_max=config.keep_checkpoint_max)
+            ckpt_cb = ModelCheckpoint(prefix="resnet", directory=config.save_checkpoint_path, config=config_ck)
+            cb += [ckpt_cb]
+
+        model.train(epoch_size, dataset, callbacks=cb)