mindspore 2.7.0__cp310-cp310-win_amd64.whl → 2.7.0rc1__cp310-cp310-win_amd64.whl

mindspore/nn/layer/activation.py

@@ -1269,10 +1269,8 @@ class PReLUExt(Cell):
         :align: center
 
     .. note::
-        - Channel dim is the 2nd dim of input. When input has dims < 2, then there is
-          no channel dim and the number of channels = 1.
-        - In GE mode, the rank of the input tensor must be greater than 1;
-          otherwise, an error will be triggered.
+        Channel dim is the 2nd dim of input. When input has dims < 2, then there is
+        no channel dim and the number of channels = 1.
 
     Args:
         num_parameters (int, optional): number of `w` to learn. Although it takes an int as input,

mindspore/nn/layer/basic.py

@@ -209,7 +209,7 @@ class DropoutExt(Cell):
     Dropout is a means of regularization that reduces overfitting by preventing correlations between neuronal nodes.
     The operator randomly sets some neurons output to 0 according to `p`, which means the probability of discarding
     during training. And the return will be multiplied by :math:`\frac{1}{1-p}` during training.
-    During the reasoning, this layer returns the same Tensor as the `input`.
+    During the reasoning, this layer returns the same Tensor as the `x`.
 
     This technique is proposed in paper `Dropout: A Simple Way to Prevent Neural Networks from Overfitting
     <http://www.cs.toronto.edu/~rsalakhu/papers/srivastava14a.pdf>`_ and proved to be effective to reduce
@@ -228,13 +228,14 @@ class DropoutExt(Cell):
             If set to ``True`` , will do this operation in-place. Default: ``False`` .
 
     Inputs:
-        - **input** (Tensor) - The input of Dropout.
+        - **x** (Tensor) - The input of Dropout.
 
     Outputs:
-        Tensor, output tensor with the same shape as the `input`.
+        Tensor, output tensor with the same shape as the `x`.
 
     Raises:
-        TypeError: If the dtype of `inplace` is not bool.
+        TypeError: If the dtype of `p` is not float.
+        ValueError: If length of shape of `x` is less than 1.
 
     Supported Platforms:
         ``Ascend``
@@ -258,12 +259,17 @@ class DropoutExt(Cell):
         self.inplace = inplace
         self.generator_step = Tensor(12, mstype.int64)
 
-    def construct(self, input):
+    def construct(self, x):
         if not self.training or self.p == 0:
-            return input
+            return x
 
         seed, offset = default_generator._step(self.generator_step)  # pylint: disable=protected-access
-        return ops.auto_generate.func_dropout_ext_op(input, self.p, self.training, self.inplace, seed, offset)
+        out, _ = ops.auto_generate.dropout_ext_op(x, self.p, seed, offset)
+
+        if self.inplace:
+            x.copy_(out)
+            return x
+        return out
 
 
 class Dropout1d(Cell):

mindspore/nn/layer/image.py

@@ -111,7 +111,7 @@ def _convert_img_dtype_to_float32(img, max_val):
 @constexpr
 def _get_dtype_max(dtype):
     """get max of the dtype"""
-    np_type = mstype._dtype_to_nptype(dtype)  # pylint:disable=protected-access
+    np_type = mstype.dtype_to_nptype(dtype)
     if issubclass(np_type, numbers.Integral):
         dtype_max = np.float64(np.iinfo(np_type).max).item()
     else:

mindspore/nn/optim/adam.py

@@ -909,7 +909,9 @@ class AdamWeightDecay(Optimizer):
     Note:
         There is usually no connection between a optimizer and mixed precision. But when `FixedLossScaleManager` is used
         and `drop_overflow_update` in `FixedLossScaleManager` is set to False, optimizer needs to set the 'loss_scale'.
-        As this optimizer has no argument of `loss_scale`, so `loss_scale` needs to be processed by other means.
+        As this optimizer has no argument of `loss_scale`, so `loss_scale` needs to be processed by other means, refer
+        document `LossScale <https://www.mindspore.cn/tutorials/en/master/beginner/mixed_precision.html>`_ to
+        process `loss_scale` correctly.
 
         If parameters are not grouped, the `weight_decay` in optimizer will be applied on the network parameters without
         'beta' or 'gamma' in their names. Users can group parameters to change the strategy of decaying weight. When

mindspore/nn/optim/lamb.py

@@ -131,7 +131,9 @@ class Lamb(Optimizer):
     Note:
         There is usually no connection between a optimizer and mixed precision. But when `FixedLossScaleManager` is used
         and `drop_overflow_update` in `FixedLossScaleManager` is set to False, optimizer needs to set the 'loss_scale'.
-        As this optimizer has no argument of `loss_scale`, so `loss_scale` needs to be processed by other means.
+        As this optimizer has no argument of `loss_scale`, so `loss_scale` needs to be processed by other means. Refer
+        document `LossScale <https://www.mindspore.cn/tutorials/en/master/beginner/mixed_precision.html>`_ to
+        process `loss_scale` correctly.
 
         If parameters are not grouped, the `weight_decay` in optimizer will be applied on the network parameters without
         'beta' or 'gamma' in their names. Users can group parameters to change the strategy of decaying weight. When

mindspore/nn/optim/tft_wrapper.py

@@ -69,9 +69,10 @@ class OptTFTWrapper(Optimizer):
         tft_env = os.getenv("MS_ENABLE_TFT", "")
         if ("TTP:1" not in tft_env) and ("UCE:1" not in tft_env) and ("ARF:1" not in tft_env):
             raise ValueError("MindIO TFT regitster need custom switch on[MS_ENABLE_TFT='{TTP:1,UCE:1,ARF:1}']!")
+        mode = context.get_context("mode")
         device_target = context.get_context("device_target")
-        if device_target != "Ascend":
-            raise ValueError("MindIO adataper only support on Ascend device!")
+        if device_target != "Ascend" or mode != context.GRAPH_MODE:
+            raise ValueError("MindIO adataper only support on Ascend device with GRAPH Mode!")
         self.opt = opt
         self.report = TensorReport()
         self.report_end = TensorReport()

mindspore/nn/probability/distribution/_utils/utils.py

@@ -298,7 +298,7 @@ class CheckTuple(PrimitiveWithInfer):
         # The op is not used in a cell
         if isinstance(x, tuple):
             return x
-        if context.get_context("mode") == context.GRAPH_MODE:
+        if context.get_context("mode") == 0:
             return x["value"]
         raise TypeError(f"For {name}, input type must be a tuple.")
 
@@ -349,7 +349,7 @@ def set_param_type(args, hint_type):
     for name, arg in args.items():
         if hasattr(arg, 'dtype'):
             if isinstance(arg, np.ndarray):
-                cur_dtype = mstype._pytype_to_dtype(arg.dtype)  # pylint:disable=protected-access
+                cur_dtype = mstype.pytype_to_dtype(arg.dtype)
             else:
                 cur_dtype = arg.dtype
             if common_dtype is None:

mindspore/nn/wrap/cell_wrapper.py

@@ -23,7 +23,7 @@ from types import FunctionType, MethodType
 
 from mindspore import log as logger
 from mindspore.parallel._utils import _get_device_num, _get_gradients_mean,\
-    _get_parallel_mode, _get_enable_parallel_optimizer
+    _get_parallel_mode, _get_enable_parallel_optimizer, _is_pynative_parallel
 from mindspore.context import ParallelMode
 from mindspore import _checkparam as validator
 from mindspore import ops, nn
@@ -397,7 +397,8 @@ class TrainOneStepCell(Cell):
         self.reducer_flag = False
         self.grad_reducer = nn.Identity()
         self.parallel_mode = _get_parallel_mode()
-        self.reducer_flag = self.parallel_mode in (ParallelMode.DATA_PARALLEL, ParallelMode.HYBRID_PARALLEL)
+        self.reducer_flag = self.parallel_mode in (ParallelMode.DATA_PARALLEL, ParallelMode.HYBRID_PARALLEL) or \
+                            _is_pynative_parallel()
         if self.reducer_flag:
             self.mean = _get_gradients_mean()
             self.degree = _get_device_num()
@@ -859,7 +860,7 @@ class _BroadCastCell(Cell):
         from mindspore import context
         self.map_ = ops.Map()
         self.params = tuple(params)
-        if context.get_context("device_target") == "Ascend":
+        if context.get_context("device_target") == "Ascend" and context.get_context("mode") != context.PYNATIVE_MODE:
             rank_list = [id for id in range(0, get_group_size())]
             create_group("BroadcastWorldGroup", rank_list)
             self.broadcast = ops.Broadcast(0, group="BroadcastWorldGroup")
@@ -888,8 +889,6 @@ class PipelineCell(Cell):
         micro_size (int): MicroBatch size.
         stage_config (dict, optional): The stage configuration for each cell's execution in pipeline parallel.
             Default ``None``.
-        segment_config (dict, optional): The segment configuration for each cell's execution in pipeline parallel.
-            Default ``None``.
 
     Supported Platforms:
         ``Ascend`` ``GPU``
@@ -901,7 +900,7 @@ class PipelineCell(Cell):
         >>> net = LeNet5()
         >>> net = nn.PipelineCell(net, 4)
     """
-    def __init__(self, network, micro_size, stage_config=None, segment_config=None):
+    def __init__(self, network, micro_size, stage_config=None):
         super(PipelineCell, self).__init__(auto_prefix=False)
         self.network = network
         self.micro_inputs = nn.CellList()
@@ -957,39 +956,6 @@ class PipelineCell(Cell):
                     print(cell_name)
                 raise KeyError("For 'PipelineCell', the argument 'stage_config' : {} is not "
                                "found in 'network' : {}".format(config_dict, network))
-        if segment_config is None:
-            return
-        self._config_segment(segment_config)
-
-
-    def _config_segment(self, segment_config=None):
-        """
-        Config segment num for cell.
-        """
-        config_dict = segment_config.copy()
-        for cell_name, cell in self.network.cells_and_names():
-            if cell_name in segment_config:
-                setattr(cell, "pipeline_segment", segment_config[cell_name])
-                del config_dict[cell_name]
-        if str(self.network) in segment_config:
-            setattr(self.network, "pipeline_segment", segment_config[str(self.network)])
-            del config_dict[str(self.network)]
-        # if there are any config elements left, print them
-        if config_dict:
-            for config_cell_name, config_segment_num in config_dict.items():
-                logger.error("pipeline_cell segment_config set pipeline_segment fail!")
-                logger.warning("config cell name:" + str(config_cell_name) +
-                               " config segment num:" + str(config_segment_num))
-            logger.warning("network:" + str(self.network))
-            logger.warning("cell name available:")
-            for cell_name, _ in self.network.cells_and_names():
-                logger.warning(cell_name)
-            raise KeyError("For 'PipelineCell', the argument 'segment_config' : {} is not "
-                           "found in 'network' : {}".format(config_dict, self.network))
-
-
-    def shard(self, in_strategy, out_strategy=None, parameter_plan=None, device="Ascend", level=0):
-        raise ValueError("For 'PipelineCell', no 'shard' on 'PipelineCell' is allowed.")
 
     def construct(self, *inputs):
         ret = None

mindspore/nn/wrap/grad_reducer.py

@@ -394,6 +394,7 @@ class DistributedGradReducer(Cell):
 
     def __init__(self, parameters, mean=None, degree=None, fusion_type=1, group=GlobalComm.WORLD_COMM_GROUP):
         super(DistributedGradReducer, self).__init__(auto_prefix=False)
+        self._check_parallel_mode()
         self.map_ = ops.Map()
         self.mean = mean
         if mean is None:
@@ -462,6 +463,13 @@ class DistributedGradReducer(Cell):
         new_grad = self.map_(ops.partial(_cast_datatype), datatypes, new_grad)
         return new_grad
 
+    def _check_parallel_mode(self):
+        """check parallel mode"""
+        parallel_mode = context.get_auto_parallel_context('parallel_mode')
+        if context.get_context('mode') == context.GRAPH_MODE and parallel_mode in (
+                context.ParallelMode.SEMI_AUTO_PARALLEL, context.ParallelMode.AUTO_PARALLEL):
+            raise RuntimeError("{} can not use DistributedGradReducer in graph mode".format(parallel_mode))
+
 
 grad_scale = ops.MultitypeFuncGraph("grad_scale")
 shard_grad_scale = ops.MultitypeFuncGraph("shard_grad_scale")
@@ -579,6 +587,7 @@ class PipelineGradReducer(Cell):
     """
     def __init__(self, parameters, scale_sense=1.0, opt_shard=None):
         super(PipelineGradReducer, self).__init__(auto_prefix=False)
+        self._check_mode()
         self.accu_grads = parameters.clone(prefix="accu_grads", init="zeros")
         self.grad_reducer = Identity()
         self.degree = Tensor(1, mstype.float32)
@@ -600,3 +609,9 @@ class PipelineGradReducer(Cell):
             accu_grads = self.grad_reducer(self.accu_grads)
             new_grads = self.hyper_map(ops.partial(grad_scale, self.scale_sense * self.degree), grads, accu_grads)
         return new_grads
+
+    def _check_mode(self):
+        """check parallel mode"""
+        mode = context.get_context('mode')
+        if mode != context.GRAPH_MODE:
+            raise RuntimeError(f"PipelineGradReducer only support graph mode, but get {mode}")

mindspore/numpy/array_creations.py

@@ -127,7 +127,7 @@ def asarray_const(a, dtype=None):
         # If dtype is not specified, we keep consistent with numpy decision
         # only exceptions are: we use int/float32
         if dtype is None:
-            dtype = mstype._pytype_to_dtype(a.dtype)  # pylint:disable=protected-access
+            dtype = mstype.pytype_to_dtype(a.dtype)
             if dtype == mstype.float64:
                 dtype = mstype.float32
             elif dtype == mstype.int64:
@@ -138,7 +138,7 @@ def asarray_const(a, dtype=None):
     if isinstance(a, onp.ndarray) and dtype is None:
         if a.dtype is onp.dtype('object'):
             raise TypeError(f"For Tensor conversion, the input_data is {a} that contains unsupported element.")
-        dtype = mstype._pytype_to_dtype(a.dtype)  # pylint:disable=protected-access
+        dtype = mstype.pytype_to_dtype(a.dtype)
         a = Tensor.from_numpy(a)
 
     return Tensor(a, dtype=dtype)

mindspore/numpy/utils_const.py

@@ -70,7 +70,7 @@ def _check_dtype(dtype):
         elif dtype is float:
             dtype = mstype.float32
         else:
-            dtype = mstype._pytype_to_dtype(dtype)  # pylint:disable=protected-access
+            dtype = mstype.pytype_to_dtype(dtype)
     if dtype not in dtype_tuple:
         raise TypeError(f"only {all_types} are allowed for dtype, but got {type(dtype)}")
     return dtype

mindspore/ops/_grad_experimental/grad_inner_ops.py

@@ -23,6 +23,15 @@ from mindspore.ops.composite.multitype_ops.zeros_like_impl import zeros_like
 from mindspore.ops._grad_experimental.grad_base import bprop_getters
 
 
+@bprop_getters.register("raise")
+def get_bprop_raise(self):
+    """Grad definition for `raise` operation."""
+    def bprop(x, y, z, out, dout):
+        return x, y, z
+
+    return bprop
+
+
 @bprop_getters.register(inner.ParallelResizeBilinear)
 def get_bprop_parallel_resize_bilinear(self):
     """Grad definition for `ParallelResizeBilinear` operation."""

mindspore/ops/_op_impl/cpu/__init__.py

@@ -72,7 +72,6 @@ from .pyfunc import _pyfunc_cpu
 from .buffer_append import _buffer_append_cpu
 from .buffer_get import _buffer_get_cpu
 from .raise_op import _raise_cpu
-from .joinedstr_op import _joinedstr_cpu
 from .buffer_sample import _buffer_sample_cpu
 from .priority_replay_buffer import _prb_push_op_cpu
 from .priority_replay_buffer import _prb_sample_op_cpu

mindspore/ops/auto_generate/cpp_create_prim_instance_helper.py

@@ -96,8 +96,6 @@ op_args_default_value = {
     "ConvTranspose2D": {"bias": None, "stride": 1, "padding": 0, "output_padding": 0, "groups": 1, "dilation": 1},
     "Correlate": {"pad_mode": 'valid'},
     "CountNonZero": {"dim": None},
-    "CrossEntropyLossGrad": {"weight": None, "grad_zloss": None, "lse_for_zloss": None, "reduction": 'mean', "ignore_index": -100, "label_smoothing": 0.0, "lse_square_scale_for_zloss": 0.0},
-    "CrossEntropyLoss": {"weight": None, "reduction": 'mean', "ignore_index": -100, "label_smoothing": 0.0, "lse_square_scale_for_zloss": 0.0, "return_zloss": False},
     "Cross": {"dim": -65530},
     "CumsumExt": {"dtype": None},
     "CumProd": {"exclusive": False, "reverse": False},
@@ -184,7 +182,6 @@ op_args_default_value = {
     "InplaceAddExt": {"alpha": 1},
     "InplaceClampScalar": {"min": None, "max": None},
     "InplaceClampTensor": {"min": None, "max": None},
-    "InplaceCopy": {"non_blocking": False},
     "InplaceDivMods": {"rounding_mode": None},
     "InplaceDivMod": {"rounding_mode": None},
     "InplaceElu": {"alpha": 1.0},
@@ -242,8 +239,6 @@ op_args_default_value = {
     "Meshgrid": {"indexing": 'xy'},
     "MinimumGrad": {"grad_x": True, "grad_y": True},
     "MinDim": {"keepdim": False},
-    "Mla": {"attn_mask": None, "deq_scale_qk": None, "deq_scale_pv": None, "q_seq_lens": None, "context_lens": None, "head_num": 32, "scale_value": 0.0, "kv_head_num": 1, "mask_mode": 'MASK_NONE', "is_ring": 0},
-    "MlaPreprocess": {"param_cache_mode": 0},
     "MoeDistributeCombine": {"tp_send_counts": None, "x_active_mask": None, "activate_scale": None, "weight_scale": None, "group_list": None, "expand_scales": None, "group_ep": None, "group_tp": None, "tp_world_size": 0, "tp_rank_id": 0, "expert_shard_type": 0, "shared_expert_num": 0, "shared_export_rank_num": 0, "global_bs": 0, "out_dtype": 0, "common_quant_mode": 0, "group_list_type": 0},
     "MoeDistributeDispatch": {"expert_scales": None, "scales": None, "x_active_mask": None, "group_ep": None, "group_tp": None, "tp_world_size": 0, "tp_rank_id": 0, "expert_shard_type": 0, "shared_expert_num": 0, "shared_expert_rank_num": 0, "quant_mode": 0, "global_bs": 0, "expert_token_nums_type": 0},
     "MoeTokenPermuteGrad": {"num_topk": 1, "padded_mode": False},
@@ -305,7 +300,6 @@ op_args_default_value = {
     "RFFTFreq": {"d": 1.0, "dtype": None},
     "RFFTN": {"s": None, "dim": None, "norm": None},
     "RFFT": {"n": None, "dim": -1, "norm": None},
-    "RingAttentionUpdate": {"actual_seq_qlen": None, "layout": 'SBH'},
     "RmsNorm": {"epsilon": 1e-6},
     "Roll": {"dims": None},
     "RotaryPositionEmbeddingGrad": {"dx": None, "mode": 0},
@@ -421,6 +415,7 @@ op_args_default_value = {
     "DeprecatedBaddbmm": {"beta": 1, "alpha": 1},
     "DeprecatedBincount": {"weights": None, "minlength": 0},
     "DeprecatedChunk": {"axis": 0},
+    "DeprecatedInplaceCopy": {"non_blocking": False},
     "DeprecatedCountNonzero": {"axis": (), "keep_dims": False, "dtype": None},
     "DeprecatedCumsum": {"axis": None, "dtype": None},
     "DeprecatedDiv": {"rounding_mode": None},
@@ -483,8 +478,6 @@ op_labels = {
     "InplaceAddmm": {"side_effect_mem": True},
     "InplaceAddsExt": {"side_effect_mem": True},
     "InplaceAddExt": {"side_effect_mem": True},
-    "InplaceBernoulliScalar": {"side_effect_mem": True},
-    "InplaceBernoulliTensor": {"side_effect_mem": True},
     "InplaceCopy": {"side_effect_mem": True},
     "InplaceDivMods": {"side_effect_mem": True},
     "InplaceDivMod": {"side_effect_mem": True},
@@ -513,8 +506,6 @@ op_labels = {
     "InplacePut": {"side_effect_mem": True},
     "InplaceRandom": {"side_effect_mem": True},
     "InplaceReLU": {"side_effect_mem": True},
-    "InplaceRemainderTensorScalar": {"side_effect_mem": True},
-    "InplaceRemainderTensorTensor": {"side_effect_mem": True},
     "InplaceScatterAdd": {"side_effect_mem": True},
     "InplaceScatterSrc": {"side_effect_mem": True},
     "InplaceScatterSrcReduce": {"side_effect_mem": True},
@@ -526,9 +517,7 @@ op_labels = {
     "InplaceTanh": {"side_effect_mem": True},
     "InplaceThreshold": {"side_effect_mem": True},
     "InplaceUniform": {"side_effect_mem": True},
-    "KvScaleCache": {"side_effect_mem": True},
     "Log": {"cust_aicpu": 'Log', "base": -1.0, "scale": 1.0, "shift": 0.0},
-    "MlaPreprocess": {"side_effect_mem": True},
     "NarrowView": {"side_effect_mem": True},
     "ReshapeAndCache": {"side_effect_mem": True},
     "ResizeD": {"mode": 'linear'},
@@ -543,4 +532,5 @@ op_labels = {
     "UnstackExtView": {"side_effect_mem": True},
     "KVCacheScatterUpdate": {"side_effect_mem": True},
     "InplaceExponential": {"side_effect_mem": True},
+    "DeprecatedInplaceCopy": {"side_effect_mem": True},
 }

mindspore/ops/auto_generate/gen_extend_func.py

@@ -118,11 +118,11 @@ def add(input, other, alpha=1):
         input (Union[Tensor, number.Number, bool]): The first input is a number.Number or
             a bool or a tensor whose data type is
             `number <https://www.mindspore.cn/docs/en/master/api_python/mindspore/mindspore.dtype.html>`_ or
-            `bool <https://www.mindspore.cn/docs/en/master/api_python/mindspore/mindspore.dtype.html>`_.
+            `bool_ <https://www.mindspore.cn/docs/en/master/api_python/mindspore/mindspore.dtype.html>`_.
         other (Union[Tensor, number.Number, bool]): The second input, is a number.Number or
             a bool or a tensor whose data type is
             `number <https://www.mindspore.cn/docs/en/master/api_python/mindspore/mindspore.dtype.html>`_ or
-            `bool <https://www.mindspore.cn/docs/en/master/api_python/mindspore/mindspore.dtype.html>`_.
+            `bool_ <https://www.mindspore.cn/docs/en/master/api_python/mindspore/mindspore.dtype.html>`_.
         alpha (number.Number): A scaling factor applied to `other`, default 1.
     
     Returns:
@@ -2003,11 +2003,11 @@ def sub(input, other, alpha=1):
         input (Union[Tensor, number.Number, bool]): The first input is a number.Number or
             a bool or a tensor whose data type is
             `number <https://www.mindspore.cn/docs/en/master/api_python/mindspore/mindspore.dtype.html>`_ or
-            `bool <https://www.mindspore.cn/docs/en/master/api_python/mindspore/mindspore.dtype.html>`_.
+            `bool_ <https://www.mindspore.cn/docs/en/master/api_python/mindspore/mindspore.dtype.html>`_.
         other (Union[Tensor, number.Number, bool]): The second input, is a number.Number or
             a bool or a tensor whose data type is
             `number <https://www.mindspore.cn/docs/en/master/api_python/mindspore/mindspore.dtype.html>`_ or
-            `bool <https://www.mindspore.cn/docs/en/master/api_python/mindspore/mindspore.dtype.html>`_.
+            `bool_ <https://www.mindspore.cn/docs/en/master/api_python/mindspore/mindspore.dtype.html>`_.
         alpha (number.Number): A scaling factor applied to `other`, default 1.
     
     Returns: