def __init__(self): super(Net, self).__init__() self.bn = P.FusedBatchNorm() self.scale = Parameter(initializer('ones', [64]), name='scale') self.b = Parameter(initializer('zeros', [64]), name='b') self.mean = Parameter(initializer('ones', [64]), name='mean') self.variance = Parameter(initializer('zeros', [64]), name='variance')
def __init__(self, num_features, eps=1e-5, momentum=0.9, affine=True, gamma_init='ones', beta_init='zeros', moving_mean_init='zeros', moving_var_init='ones', use_batch_statistics=True): super(_BatchNorm, self).__init__() if num_features < 1: raise ValueError("num_features must be at least 1") if momentum < 0 or momentum > 1: raise ValueError( "momentum should be a number in range [0, 1], but got {}". format(momentum)) self.use_batch_statistics = use_batch_statistics self.num_features = num_features self.eps = eps self.moving_mean = Parameter(initializer(moving_mean_init, num_features), name="mean", requires_grad=False) self.moving_variance = Parameter(initializer(moving_var_init, num_features), name="variance", requires_grad=False) self.gamma = Parameter(initializer(gamma_init, num_features), name="gamma", requires_grad=affine) self.beta = Parameter(initializer(beta_init, num_features), name="beta", requires_grad=affine) if context.get_context("enable_ge"): self.is_ge_backend = True self.momentum = Tensor(1.0 - momentum, DT.float32) self.bn_train = P.BatchNorm(is_training=True, epsilon=self.eps) else: self.is_ge_backend = False self.momentum = 1.0 - momentum self.bn_train = P.FusedBatchNorm(mode=1, epsilon=self.eps, momentum=self.momentum) self.bn_infer = P.BatchNorm(is_training=False, epsilon=self.eps) data_parallel_strategy = ((1, ), (1, )) data_parallel_strategy_one = ((1, ), ()) self.sub_mean = P.Sub().set_strategy(data_parallel_strategy) self.sub_var = P.Sub().set_strategy(data_parallel_strategy) self.mul_mean = P.Mul().set_strategy(data_parallel_strategy_one) self.mul_var = P.Mul().set_strategy(data_parallel_strategy_one) self.assign_sub_mean = P.AssignSub().set_strategy( data_parallel_strategy) self.assign_sub_var = P.AssignSub().set_strategy( data_parallel_strategy)
def __init__(self, num_features, eps=1e-5, momentum=0.9, affine=True, gamma_init='ones', beta_init='zeros', moving_mean_init='zeros', moving_var_init='ones', use_batch_statistics=True, device_num_each_group=1): super(_BatchNorm, self).__init__() if num_features < 1: raise ValueError("num_features must be at least 1") if momentum < 0 or momentum > 1: raise ValueError( "momentum should be a number in range [0, 1], but got {}". format(momentum)) self.use_batch_statistics = use_batch_statistics self.num_features = num_features self.eps = eps self.moving_mean = Parameter(initializer(moving_mean_init, num_features), name="mean", requires_grad=False) self.moving_variance = Parameter(initializer(moving_var_init, num_features), name="variance", requires_grad=False) self.gamma = Parameter(initializer(gamma_init, num_features), name="gamma", requires_grad=affine) self.beta = Parameter(initializer(beta_init, num_features), name="beta", requires_grad=affine) self.group = check_int_positive(device_num_each_group) self.is_global = False if self.group != 1: self.rank_id = get_rank() self.rank_size = get_group_size() self.device_list = [i for i in range(0, self.rank_size)] self.rank_list = self.list_group(self.device_list, self.group) self.rank_list_idx = len(self.rank_list) for i in range(self.rank_list_idx): if self.rank_id in self.rank_list[i] and self.group != 1: self.is_global = True management.create_group('group' + str(i), self.rank_list[i]) self.all_reduce = P.AllReduce( P.ReduceOp.SUM, 'group' + str(i)).add_prim_attr('fusion', 1) self.shape = P.Shape() self.reduce_mean = P.ReduceMean(keep_dims=True) self.square = P.Square() self.sqrt = P.Sqrt() self.cast = P.Cast() self.dtype = P.DType() self.reshape = P.Reshape() self.is_ascend = context.get_context("device_target") == "Ascend" if context.get_context("enable_ge"): self.is_ge_backend = True self.momentum = Tensor(1.0 - momentum, mstype.float32) else: self.is_ge_backend = False self.momentum = 1.0 - momentum if self.is_ge_backend or self.is_ascend: self.bn_train = P.BatchNorm(is_training=True, epsilon=self.eps) else: self.bn_train = P.FusedBatchNorm(mode=1, epsilon=self.eps, momentum=self.momentum) self.bn_infer = P.BatchNorm(is_training=False, epsilon=self.eps) data_parallel_strategy = ((1, ), (1, )) data_parallel_strategy_one = ((1, ), ()) self.sub_mean = P.Sub().set_strategy(data_parallel_strategy) self.sub_var = P.Sub().set_strategy(data_parallel_strategy) self.mul_mean = P.Mul().set_strategy(data_parallel_strategy_one) self.mul_var = P.Mul().set_strategy(data_parallel_strategy_one) self.assign_sub_mean = P.AssignSub().set_strategy( data_parallel_strategy) self.assign_sub_var = P.AssignSub().set_strategy( data_parallel_strategy)
'desc_inputs': [0], }), ('MaxPoolWithArgmax_ValueError_2', { 'block': (lambda _: P.MaxPoolWithArgmax(ksize='1'), {'exception': TypeError}), 'desc_inputs': [0], }), ('MaxPoolWithArgmax_ValueError_3', { 'block': (lambda _: P.MaxPoolWithArgmax(ksize=-2), {'exception': ValueError}), 'desc_inputs': [0], }), ('MaxPoolWithArgmax_ValueError_4', { 'block': (lambda _: P.MaxPoolWithArgmax(strides=-1), {'exception': ValueError}), 'desc_inputs': [0], }), ('FusedBatchNorm_ValueError_1', { 'block': (lambda _: P.FusedBatchNorm(mode="1", epsilon=1e-5, momentum=0.1), {'exception': TypeError}), 'desc_inputs': [0], }), ('FusedBatchNorm_ValueError_2', { 'block': (lambda _: P.FusedBatchNorm(mode=2, epsilon=1e-5, momentum=0.1), {'exception': ValueError}), 'desc_inputs': [0], }), ('FusedBatchNorm_ValueError_3', { 'block': (lambda _: P.FusedBatchNorm(mode=0, epsilon=-1e-5, momentum=0.1), {'exception': ValueError}), 'desc_inputs': [0], }), ('FusedBatchNorm_ValueError_4', { 'block': (lambda _: P.FusedBatchNorm(mode=0, epsilon=1e-5, momentum=-0.1), {'exception': ValueError}), 'desc_inputs': [0], }), ('FusedBatchNorm_ValueError_5', {
def __init__(self, num_features, eps=1e-5, momentum=0.9, affine=True, gamma_init='ones', beta_init='zeros', moving_mean_init='zeros', moving_var_init='ones', use_batch_statistics=None, device_num_each_group=1, input_dims='2d', data_format='NCHW'): super(_BatchNorm, self).__init__() if num_features < 1: raise ValueError("num_features must be at least 1") if momentum < 0 or momentum > 1: raise ValueError("momentum should be a number in range [0, 1], but got {}".format(momentum)) self.format = validator.check_string(data_format, ['NCHW', 'NHWC'], 'format', self.cls_name) if context.get_context("device_target") != "GPU" and self.format == "NHWC": raise ValueError("NHWC format only support in GPU target.") self.use_batch_statistics = use_batch_statistics self.num_features = num_features self.eps = eps self.input_dims = input_dims self.moving_mean = Parameter(initializer( moving_mean_init, num_features), name="mean", requires_grad=False) self.moving_variance = Parameter(initializer( moving_var_init, num_features), name="variance", requires_grad=False) self.gamma = Parameter(initializer( gamma_init, num_features), name="gamma", requires_grad=affine) self.beta = Parameter(initializer( beta_init, num_features), name="beta", requires_grad=affine) self.group = validator.check_positive_int(device_num_each_group) self.is_global = False if self.group != 1: self.rank_id = get_rank() self.rank_size = get_group_size() self.device_list = [i for i in range(0, self.rank_size)] self.rank_list = self.list_group(self.device_list, self.group) self.rank_list_idx = len(self.rank_list) for i in range(self.rank_list_idx): if self.rank_id in self.rank_list[i] and self.group != 1: self.is_global = True management.create_group('group' + str(i), self.rank_list[i]) self.all_reduce = P.AllReduce(P.ReduceOp.SUM, 'group' + str(i)).add_prim_attr('fusion', 1) self.shape = P.Shape() self.reduce_mean = P.ReduceMean(keep_dims=True) self.square = P.Square() self.sqrt = P.Sqrt() self.cast = P.Cast() self.dtype = P.DType() self.reshape = P.Reshape() self.is_ascend = context.get_context("device_target") == "Ascend" self.is_gpu = context.get_context("device_target") == "GPU" self.is_graph_mode = context.get_context("mode") == context.GRAPH_MODE self.momentum = 1.0 - momentum if context.get_context("enable_ge"): self.is_ge_backend = True else: self.is_ge_backend = False if self.is_graph_mode and (self.is_ge_backend or self.is_ascend): self.bn_train = P.BatchNorm(is_training=True, epsilon=self.eps) elif self.is_gpu: self.bn_train = P.FusedBatchNormEx(mode=1, epsilon=self.eps, momentum=self.momentum, data_format=self.format) else: self.bn_train = P.FusedBatchNorm(mode=1, epsilon=self.eps, momentum=self.momentum) self.bn_infer = P.BatchNorm(is_training=False, epsilon=self.eps, data_format=self.format) self.enable_global_sync = self.is_global and (self.is_ge_backend or (self.is_graph_mode and self.is_ascend)) self.enable_default_train = self.is_graph_mode and not self.is_global and \ (self.is_ge_backend or self.is_ascend) data_parallel_strategy = ((1,), (1,)) data_parallel_strategy_one = ((1,), ()) self.sub_mean = P.Sub().shard(data_parallel_strategy) self.sub_var = P.Sub().shard(data_parallel_strategy) self.mul_mean = P.Mul().shard(data_parallel_strategy_one) self.mul_var = P.Mul().shard(data_parallel_strategy_one) self.assign_sub_mean = P.AssignSub().shard(data_parallel_strategy) self.assign_sub_var = P.AssignSub().shard(data_parallel_strategy)
('LogSoftmaxGrad', { 'block': G.LogSoftmaxGrad(), 'desc_inputs': [[16, 1234], [16, 1234]], 'desc_bprop': [[64, 2]], 'skip': ['backward']}), ('LayerNorm', { 'block': P.LayerNorm(), 'desc_inputs': [[2, 16], [16], [16]], 'desc_bprop': [[2, 16], [2, 16], [2, 16]]}), ('LayerNormGrad', { 'block': G.LayerNormGrad(), 'desc_inputs': [[2, 16], [2, 16], [2, 16], [2, 16], [16]], 'desc_bprop': [[2, 16], [16], [16]], 'skip': ['backward']}), ('FusedBatchNorm', { 'block': P.FusedBatchNorm(), 'desc_inputs': [[128, 64, 32, 64], [64], [64], [64], [64]], 'desc_bprop': [[128, 64, 32, 64], [64], [64], [64], [64]], 'skip': []}), ('FusedBatchNormGrad', { 'block': G.FusedBatchNormGrad(), 'desc_inputs': [[128, 64, 32, 64], [128, 64, 32, 64], [64], [64], [64]], 'desc_bprop': [[128, 64, 32, 64], [64], [64], [64], [64]], 'skip': ['backward']}), ('BatchNorm', { 'block': P.BatchNorm(), 'desc_inputs': [[128, 64, 32, 32], [64], [64], [64], [64]], 'desc_bprop': [[128, 64, 32, 32], [64], [64], [64], [64]], 'skip': []}), ('BatchNormGrad', { 'block': G.BatchNormGrad(),
}), ('MaxPoolWithArgmax_ValueError_3', { 'block': (lambda _: P.MaxPoolWithArgmax(ksize=-2), { 'exception': ValueError }), 'desc_inputs': [0], }), ('MaxPoolWithArgmax_ValueError_4', { 'block': (lambda _: P.MaxPoolWithArgmax(strides=-1), { 'exception': ValueError }), 'desc_inputs': [0], }), ('FusedBatchNorm_ValueError_1', { 'block': (lambda _: P.FusedBatchNorm(mode="1", epsilon=1e-5, momentum=0.1), { 'exception': TypeError }), 'desc_inputs': [0], }), ('FusedBatchNorm_ValueError_2', { 'block': (lambda _: P.FusedBatchNorm(mode=2, epsilon=1e-5, momentum=0.1), { 'exception': ValueError }), 'desc_inputs': [0], }), ('FusedBatchNorm_ValueError_3', { 'block': (lambda _: P.FusedBatchNorm(mode=0, epsilon=-1e-5, momentum=0.1), { 'exception': ValueError
# ============================================================================ from mindspore.ops import operations as P from mindspore.ops import Primitive make_tuple = Primitive('make_tuple') tuple_getitem = Primitive('tuple_getitem') conv = P.Conv2D(out_channel=64, kernel_size=7, mode=1, pad_mode="valid", pad=0, stride=1, dilation=1, group=1) bn = P.FusedBatchNorm() relu = P.ReLU() conv_bn1 = Primitive('ConvBN1') bn2_relu = Primitive('BN2Relu') class FnDict: def __init__(self): self.fnDict = {} def __call__(self, fn): self.fnDict[fn.__name__] = fn def __getitem__(self, name): return self.fnDict[name]