p19647583
/
mindspore
forked from mindspore-Ecosystem/mindspore
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

!44684 rm set_auto_parallel 

Merge pull request !44684 from yangzhenzhang/rm-set-auto-parallel
This commit is contained in:
i-robot 2022-10-28 06:56:30 +00:00 committed by Gitee
parent 85720e99b7 5d5388ad00
commit 384efc057f
No known key found for this signature in database
GPG Key ID: 173E9B9CA92EEF8F
165 changed files with 58 additions and 393 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

29
docs/api/api_python/nn/mindspore.nn.Cell.rst
View File

@ -37,13 +37,6 @@
返回：
Tuple类型，经过类型转换后的输入。
.. py:method:: auto_parallel_compile_and_run()
是否在‘AUTO_PARALLEL’或‘SEMI_AUTO_PARALLEL’模式下执行编译流程。
返回：
bool，`_auto_parallel_compile_and_run` 的值。
.. py:method:: bprop_debug
:property:
@ -245,15 +238,6 @@
- **KeyError** - 如果参数名称为空或包含"."。
- **TypeError** - 如果参数的类型不是Parameter。
.. py:method:: load_parameter_slice(params)
根据并行策略获取Tensor分片并替换原始参数。
请参考 `mindspore.common._Executor.compile` 源代码中的用法。
参数：
- **params** (dict) - 用于初始化数据图的参数字典。
.. py:method:: name_cells()
递归地获取一个Cell中所有子Cell的迭代器。
@ -414,12 +398,6 @@
返回：
Cell的输出。
.. py:method:: set_auto_parallel()
将Cell设置为自动并行模式。
.. note:: 如果一个Cell需要使用自动并行或半自动并行模式来进行训练、评估或预测，则该Cell需要调用此接口。
.. py:method:: set_boost(boost_type)
为了提升网络性能，可以配置boost内的算法让框架自动使能该算法来加速网络训练。
@ -497,13 +475,6 @@
- **task_sink** (bool) - 是否通过数据集方式传递数据。默认值：True。
.. py:method:: set_parallel_input_with_inputs(*inputs)
通过并行策略对输入张量进行切分。
参数：
- **inputs** (tuple) - construct方法的输入。
.. py:method:: set_param_fl(push_to_server=False, pull_from_server=False, requires_aggr=True)
设置参数与服务器交互的方式。

37
mindspore/python/mindspore/common/api.py
View File

@ -24,7 +24,6 @@ import time
import ast
import inspect
import importlib
from collections import OrderedDict
from functools import wraps
import numpy as np
import mindspore as ms
@ -40,7 +39,7 @@ from mindspore._c_expression import GraphExecutor_, Tensor, MetaTensor, CSRTenso
_ms_memory_recycle
from mindspore.parallel._ps_context import _is_role_pserver, _is_role_sched, _enable_distributed_mindrt
from mindspore.parallel._utils import _check_full_batch, _get_parameter_broadcast, _is_pynative_parallel, \
_get_pipeline_stages
_get_pipeline_stages, _is_in_auto_parallel_mode
from mindspore._checkparam import Validator
from mindspore.common._utils import is_shape_unknown
from mindspore.common.mutable import mutable
@ -893,16 +892,6 @@ def _function_forbid_reuse(func):
return func
def _get_auto_split_param_names(parameter_layout_dict):
auto_split_param_names = []
for key, value in parameter_layout_dict.items():
for dim in value[1]:
if dim != -1:
auto_split_param_names.append(key)
break
return auto_split_param_names
def _build_broadcast_graph(broadcast_params_dict, broadcast_phase):
"""Build broadcast graph."""
from mindspore.nn.wrap.cell_wrapper import _BroadCastCell
@ -918,18 +907,9 @@ def _build_broadcast_graph(broadcast_params_dict, broadcast_phase):
broadcast_params_dict[param_name].set_data(param)
def _parameter_broadcast(obj, auto_parallel_mode):
def _parameter_broadcast(obj):
"""Parameter broadcast."""
auto_split_param_names = []
if auto_parallel_mode:
auto_split_param_names = _get_auto_split_param_names(obj.parameter_layout_dict)
broadcast_params_dict = obj.parameters_broadcast_dict()
if auto_split_param_names and broadcast_params_dict:
broadcast_params_dict = OrderedDict()
for param_name, param in obj.parameters_broadcast_dict().items():
if param_name not in auto_split_param_names:
broadcast_params_dict[param_name] = param
broadcast_phase = "_broadcast_subgraph"
_build_broadcast_graph(broadcast_params_dict, broadcast_phase)
@ -974,20 +954,19 @@ class _PyNativeExecutor:
return self._executor(sens_param, obj, args)
@staticmethod
def parameter_broadcast(obj, phase, auto_parallel_mode):
def parameter_broadcast(obj, phase):
"""
Run broadcast for parameter.
Args:
obj (Cell): The cell instance.
phase (str): The phase of cell instance.
auto_parallel_mode (bool): The flag of running auto parallel.
Return:
None.
"""
if BROADCAST_PHASE not in phase and _get_parameter_broadcast():
_parameter_broadcast(obj, auto_parallel_mode)
_parameter_broadcast(obj)
def real_run_op(self, *args):
"""
@ -1342,7 +1321,7 @@ class _CellGraphExecutor:
if "train" in phase and (enable_compile_cache is True or enable_compile_cache == "1"):
self._graph_executor.set_compile_cache_dep_files(_get_compile_cache_dep_files())
def compile(self, obj, *args, phase='predict', do_convert=True, auto_parallel_mode=False, jit_config_dict=None):
def compile(self, obj, *args, phase='predict', do_convert=True, jit_config_dict=None):
"""
Compiles graph.
@ -1351,7 +1330,6 @@ class _CellGraphExecutor:
args (tuple): Function or cell input arguments.
phase (str): The name of compile phase. Default: 'predict'.
do_convert (bool): When set to True, convert ME graph to GE graph after compiling graph.
auto_parallel_mode: When set to True, use auto parallel mode to compile graph.
jit_config_dict (dict): Jit config for compile. Default: None.
Return:
@ -1395,10 +1373,11 @@ class _CellGraphExecutor:
if graph is None:
raise RuntimeError("Compile graph failed for phase {}.".format(phase))
auto_parallel_mode = _is_in_auto_parallel_mode()
if not auto_parallel_mode:
replace = obj.init_parameters_data(auto_parallel_mode=auto_parallel_mode)
self._update_param_node_default_input(phase, replace)
else:
elif 'skip_auto_parallel_compile' not in obj.get_flags().keys():
obj.parameter_layout_dict = self._graph_executor.get_parameter_layout(phase)
obj.parallel_parameter_name_list = self._graph_executor.get_parallel_parameter_name_list(phase)
if _get_pipeline_stages() > 1 and (not hasattr(obj, "is_first_iteration") or not obj.is_first_iteration):
@ -1413,7 +1392,7 @@ class _CellGraphExecutor:
elif "export" in phase:
self._build_data_graph(obj, phase)
elif BROADCAST_PHASE not in phase and _get_parameter_broadcast():
_parameter_broadcast(obj, auto_parallel_mode)
_parameter_broadcast(obj)
return phase, True

107
mindspore/python/mindspore/nn/cell.py
View File

@ -40,7 +40,6 @@ from mindspore.ops.operations import Cast
from mindspore.ops.primitive import Primitive
from mindspore.ops.operations import _inner_ops as inner
from mindspore.parallel.shard import Shard
from mindspore.parallel._tensor import _load_tensor_by_layout
class Cell(Cell_):
@ -84,12 +83,11 @@ class Cell(Cell_):
[Parameter (name=weight, shape=(240, 120, 4, 4), dtype=Float32, requires_grad=True)]
"""
IGNORE_LIST = ['_scope', '_cell_init_args', '_auto_prefix', '_cells', '_params', '_construct_inputs_names',
'_construct_inputs_num', '_create_time', '_func_graph_flags', '_parallel_inputs_run',
'_parameter_layout_dict', '_params_list', '_tensor_list', '_phase', '_auto_parallel_mode',
IGNORE_LIST = ['_scope', '_cell_init_args', '_auto_prefix', '_cells', '_params', '_create_time',
'_func_graph_flags', '_parameter_layout_dict', '_params_list', '_tensor_list', '_phase',
'_forward_pre_hook', '_forward_hook', '_enable_forward_pre_hook', '_enable_forward_hook',
'_bprop_debug', '_enable_backward_hook', '_cell_backward_hook', '_is_run', '_param_prefix',
'_attr_synced', 'pynative', 'requires_grad', '_auto_parallel_compile_and_run', 'cell_type']
'_attr_synced', 'pynative', 'requires_grad', 'cell_type']
def __init__(self, auto_prefix=True, flags=None):
Cell_.__init__(self, self._cell_tag)
@ -123,10 +121,6 @@ class Cell(Cell_):
if os.getenv('GC_COLLECT_IN_CELL') == '1':
gc.collect()
self._construct_inputs_num = 0
self._construct_inputs_names = []
self._auto_parallel_mode = False
self._parallel_inputs_run = None
if flags:
self.add_flags(**flags)
self._bprop_debug = False
@ -138,7 +132,6 @@ class Cell(Cell_):
self._cell_backward_hook = None
self._is_recursion_hook = False
self.cell_type = None
self._auto_parallel_compile_and_run = False
self.cast = Cast()
self._has_config_recompute = False
self._user_parameters = []
@ -385,7 +378,7 @@ class Cell(Cell_):
def _do_parameter_broadcast(self):
if context.get_auto_parallel_context("parallel_mode") == ParallelMode.DATA_PARALLEL:
if not self.parameter_broadcast_done:
_pynative_executor.parameter_broadcast(self, self.phase, self._auto_parallel_mode)
_pynative_executor.parameter_broadcast(self, self.phase)
self.parameter_broadcast_done = True
def run_construct(self, cast_inputs, kwargs):
@ -832,84 +825,20 @@ class Cell(Cell_):
"""
Replace parameters with sliced tensors by parallel strategies.
Please refer to the usage in source code of `mindspore.common._CellGraphExecutor.compile`.
Args:
params (dict): The parameters dictionary used for initializing the data graph.
Note:
This interface is deprecated.
"""
if params is None:
params = self.parameters_dict()
if isinstance(params, OrderedDict):
for key in params:
tensor = params[key].data
if key not in self.parameter_layout_dict:
logger.info("The layout dict does not contain the key %s.", key)
continue
if params[key].sliced:
logger.debug("The param %s is already sliced.", key)
continue
layout = self.parameter_layout_dict[key]
new_tensor = _load_tensor_by_layout(tensor, layout)
params[key].set_data(new_tensor, True)
else:
raise TypeError("For 'load_parameter_slice', the argument 'params' must be OrderedDict type, "
"but got {}.".format(type(params)))
logger.warning("'load_parameter_slice' function is deprecated.")
def _load_inputs(self, *inputs):
"""
Slice inputs tensors by parallel strategies.
Args:
inputs (Function or Cell): inputs of construct method.
"""
parallel_inputs_run = []
# judge if *args exists in input
if self.argspec[1] is not None:
prefix = self.argspec[1]
for i in range(len(inputs)):
key = prefix + str(i)
self._construct_inputs_names = self._construct_inputs_names + (key,)
self._construct_inputs_num = self._construct_inputs_num + 1
for i, tensor in enumerate(inputs):
key = self._construct_inputs_names[i]
# if input is not used, self.parameter_layout_dict may not contain the key
if key not in self.parameter_layout_dict:
logger.warning("Layout dict does not contain the key %s.", key)
parallel_inputs_run.append(tensor)
else:
layout = self.parameter_layout_dict[key]
new_tensor = _load_tensor_by_layout(tensor, layout)
parallel_inputs_run.append(new_tensor)
return tuple(parallel_inputs_run)
def set_parallel_input_with_inputs(self, *inputs):
"""
Slice inputs tensors by parallel strategies.
Args:
inputs (tuple): inputs of construct method.
Note:
This interface is deprecated.
"""
self._parallel_inputs_run = self._load_inputs(*inputs)
def _get_construct_inputs_number_and_name(self):
"""Compute self._construct_inputs_names and self._construct_inputs_num"""
from mindspore._extends.parse.parser import get_parse_method_of_class
fn = get_parse_method_of_class(self)
self.argspec = inspect.getfullargspec(fn)
self._construct_inputs_num = fn.__code__.co_argcount
self._construct_inputs_names = fn.__code__.co_varnames
if self._construct_inputs_num <= 0:
raise ValueError(f"For 'set_auto_parallel', the number of inputs must be greater than 0,"
f"but got {self._construct_inputs_num}.")
if self._construct_inputs_names[0] != 'self':
raise ValueError(f"First member of fn function must be self, but got {self._construct_inputs_names[0]}")
if self._construct_inputs_num - 1 > len(self._construct_inputs_names):
raise ValueError(f"Num of inputs must be greater than num of fn function members, num of inputs is \
{self._construct_inputs_names - 1}, num of fn function members is {len(self._construct_inputs_names)}")
self._construct_inputs_names = self._construct_inputs_names[1:self._construct_inputs_num]
self._construct_inputs_num = self._construct_inputs_num - 1
logger.warning("'set_parallel_input_with_inputs' function is deprecated.")
def set_inputs(self, *inputs):
"""
@ -975,7 +904,7 @@ class Cell(Cell_):
inputs (tuple): Inputs of the Cell object.
"""
if self._dynamic_shape_inputs is None or self._dynamic_shape_inputs[0] is None:
_cell_graph_executor.compile(self, *inputs, phase=self.phase, auto_parallel_mode=self._auto_parallel_mode,
_cell_graph_executor.compile(self, *inputs, phase=self.phase,
jit_config_dict=self._jit_config_dict)
else:
self._check_compile_dynamic_shape(*inputs)
@ -986,7 +915,6 @@ class Cell(Cell_):
self.saved_dynamic_shape = self._dynamic_shape_inputs
_cell_graph_executor.compile(self, *self._dynamic_shape_inputs, phase=self.phase,
auto_parallel_mode=self._auto_parallel_mode,
jit_config_dict=self._jit_config_dict)
logger.debug("Compiled Graph with dynamic shape")
@ -1003,7 +931,6 @@ class Cell(Cell_):
Returns:
Object, the result of executing.
"""
self._auto_parallel_compile_and_run = True
self.compile(*inputs)
new_inputs = _get_args_for_run(self, inputs)
@ -1013,10 +940,10 @@ class Cell(Cell_):
"""
Whether or not to execute compile and run in 'AUTO_PARALLEL' or 'SEMI_AUTO_PARALLEL' mode.
Returns:
bool, `_auto_parallel_compile_and_run` value.
Note:
This interface is deprecated.
"""
return self._auto_parallel_compile_and_run
logger.warning("'auto_parallel_compile_and_run' function is deprecated.")
def exec_checkpoint_graph(self):
"""Executes saving checkpoint graph operation."""
@ -1652,11 +1579,9 @@ class Cell(Cell_):
Set the cell to auto parallel mode.
Note:
If a cell needs to use the auto parallel or semi auto parallel mode for training, evaluation or prediction,
this interface needs to be called by the cell.
This interface is deprecated.
"""
self._auto_parallel_mode = True
self._get_construct_inputs_number_and_name()
logger.warning("'set_auto_parallel' function is deprecated.")
def set_jit_config(self, jit_config):
"""

7
mindspore/python/mindspore/parallel/_auto_parallel_context.py
View File

@ -429,13 +429,14 @@ class _AutoParallelContext:
Set the value of sharding strategy propagation in AUTO_PARALLEL mode. If True, the strategy-configured operators
will propagate the strategies to other operators with minimum redistribution cost; otherwise, the algorithm
will search the desired strategies. Default: False.
This attribute is replaced by context.set_auto_parallel(search_mode="sharding_propagation").
This attribute is replaced by context.set_auto_parallel_context(search_mode="sharding_propagation").
Args:
sharding_propagation (bool): Enable/disable strategy propagation.
"""
logger.warning("This attribute is replaced by context.set_auto_parallel(search_mode='sharding_propagation'), "
"and this attribute will be deleted in a future MindSpore version.")
logger.warning("This attribute is replaced by "
"context.set_auto_parallel_context(search_mode='sharding_propagation'), and this attribute will"
" be deleted in a future MindSpore version.")
self.check_context_handle()
if not isinstance(sharding_propagation, bool):
raise TypeError("For 'set_auto_parallel_context().set_sharding_propagation', "

15
mindspore/python/mindspore/train/model.py
View File

@ -32,8 +32,8 @@ from mindspore._checkparam import check_input_data, check_output_data, Validator
from mindspore.train.callback import _InternalCallbackParam, RunContext, _CallbackManager, Callback, TimeMonitor
from mindspore.train.callback import __all__ as internal_cb_names
from mindspore import context
from mindspore.parallel._utils import _get_parallel_mode, _get_device_num, _get_global_rank, \
_get_parameter_broadcast, _device_number_check, _parameter_broadcast_check, _parallel_predict_check, \
from mindspore.parallel._utils import _get_parallel_mode, _get_device_num, _get_parameter_broadcast, \
_device_number_check, _parameter_broadcast_check, _parallel_predict_check, \
_reset_op_id_with_offset
from mindspore.parallel._ps_context import _is_role_worker, _is_role_pserver, _is_role_sched, _is_ps_mode, \
_cache_enable, _enable_distributed_mindrt
@ -212,7 +212,6 @@ class Model:
self._process_amp_args(kwargs)
self._parallel_mode = _get_parallel_mode()
self._device_number = _get_device_num()
self._global_rank = _get_global_rank()
self._parameter_broadcast = _get_parameter_broadcast()
self._metrics = metrics
@ -323,7 +322,6 @@ class Model:
# If need to check if loss_fn is not None, but optimizer is None
if self._parallel_mode in (ParallelMode.SEMI_AUTO_PARALLEL, ParallelMode.AUTO_PARALLEL):
network.set_auto_parallel()
if self._optimizer is None:
# In this case, multiple optimizer(s) is supposed to be included in 'self._network'
_set_multi_subgraphs()
@ -371,14 +369,11 @@ class Model:
if self._optimizer is None:
# In this case, multiple optimizer(s) is supposed to be included in 'self._network'
_set_multi_subgraphs()
self._eval_network.set_auto_parallel()
def _build_predict_network(self):
"""Build the network for prediction."""
self._predict_network = self._network
if self._parallel_mode in (ParallelMode.SEMI_AUTO_PARALLEL, ParallelMode.AUTO_PARALLEL):
# Unlike the cases in build_train_network() and build_eval_network(), 'multi_subgraphs' is not set
self._predict_network.set_auto_parallel()
# Unlike the cases in build_train_network() and build_eval_network(), 'multi_subgraphs' is not set
def _clear_metrics(self):
"""Clear metrics local values."""
@ -451,9 +446,6 @@ class Model:
network.phase = phase
self._backbone_is_train = is_train
if self._parallel_mode in (ParallelMode.SEMI_AUTO_PARALLEL, ParallelMode.AUTO_PARALLEL):
network.set_auto_parallel()
return dataset_helper, network
def _check_network_mode(self, network, is_train):
@ -1638,7 +1630,6 @@ class Model:
predict_net = self._predict_network
# Unlike the cases in build_train_network() and build_eval_network(), 'multi_subgraphs' is not set
predict_net.set_auto_parallel()
predict_net = self._check_network_mode(predict_net, False)
predict_net.compile(*predict_data)
return predict_net.parameter_layout_dict

7
mindspore/python/mindspore/train/serialization.py
View File

@ -54,7 +54,7 @@ from mindspore.compression.export import quant_export
from mindspore.parallel._cell_wrapper import get_allgather_cell
from mindspore.parallel._tensor import _load_tensor, _get_tensor_strategy, _get_tensor_slice_index
from mindspore.parallel._tensor import _reshape_param_data, _reshape_param_data_with_weight
from mindspore.parallel._utils import _infer_rank_list, _remove_repeated_slices
from mindspore.parallel._utils import _infer_rank_list, _remove_repeated_slices, _is_in_auto_parallel_mode
from mindspore.parallel._parallel_serialization import _convert_to_list, _convert_to_layout, _build_searched_strategy, \
_restore_group_info_list
from mindspore.train._utils import read_proto
@ -1344,9 +1344,8 @@ def _msfunc_info(net, *inputs):
def _cell_info(net, *inputs):
"""Get mindir stream and net dict of cell"""
phase_name = "predict" if net._auto_parallel_mode else "export.mindir"
graph_id, _ = _executor.compile(net, *inputs, phase=phase_name,
do_convert=False, auto_parallel_mode=net._auto_parallel_mode)
phase_name = "predict" if _is_in_auto_parallel_mode() else "export.mindir"
graph_id, _ = _executor.compile(net, *inputs, phase=phase_name, do_convert=False)
# pylint: disable=protected-access
mindir_stream = _executor._get_func_graph_proto(net, graph_id, 'mind_ir')
# clean obfuscation config to prevent the next call

4
mindspore/python/mindspore/train/train_thor/model_thor.py
View File

@ -26,7 +26,6 @@ import math
from mindspore.train.callback import RunContext
from mindspore import context
from mindspore import nn
from mindspore.context import ParallelMode
from mindspore.train.model import Model
from mindspore.train.dataset_helper import connect_network_with_dataset
from mindspore.parallel._utils import _need_to_full, _to_full_tensor
@ -140,9 +139,6 @@ class ModelThor(Model):
network.set_train(is_train)
network.phase = phase
if self._parallel_mode in (ParallelMode.SEMI_AUTO_PARALLEL, ParallelMode.AUTO_PARALLEL):
network.set_auto_parallel()
return dataset_helper, network
def _train_gpu_sink_step(self, cb_params, inputs, list_callback, iter_first_order, run_context):

2
tests/st/model_zoo_tests/resnet50/train.py
View File

@ -236,8 +236,6 @@ def train_net():
loss_scale = FixedLossScaleManager(config.loss_scale, drop_overflow_update=False)
train_network = build_train_network(net, opt, loss, level="O2", boost_level=config.boost_mode,
loss_scale_manager=loss_scale, keep_batchnorm_fp32=False)
if config.run_distribute:
train_network.set_auto_parallel()
for _ in range(500):
image = Tensor(np.random.rand(32, 3, 224, 224), dtype=mindspore.float32)
label = Tensor(np.random.randint(0, 10, [32]), dtype=mindspore.int32)

1
tests/ut/python/parallel/test_adafactor.py
View File

@ -61,7 +61,6 @@ def compile_net(net):
scale_parameter=scale_parameter, relative_step=relative_step,
warmup_init=warmup_init, compression=compression)
train_net = TrainOneStepCell(net, optimizer)
train_net.set_auto_parallel()
train_net.set_train()
_cell_graph_executor.compile(train_net, _x, _b)
context.reset_auto_parallel_context()

1
tests/ut/python/parallel/test_add_relu_redistribution.py
View File

@ -62,7 +62,6 @@ class Grad(nn.Cell):
def compile_net(net, x, y):
net.set_auto_parallel()
net.set_train()
_cell_graph_executor.compile(net, x, y)

4
tests/ut/python/parallel/test_arithmetic.py
View File

@ -51,7 +51,6 @@ class GradWrap(nn.Cell):
def compile_net(net, x, y, b):
net.set_auto_parallel()
net.set_train()
_cell_graph_executor.compile(net, x, y, b)
@ -799,7 +798,6 @@ def test_assign_sub():
return grad_all(self.network)(x)
def compile_sub_net(net, x):
net.set_auto_parallel()
net.set_train()
_cell_graph_executor.compile(net, x)
@ -853,7 +851,6 @@ def test_assign_add():
return grad_all(self.network)(x)
def compile_sub_net(net, x):
net.set_auto_parallel()
net.set_train()
_cell_graph_executor.compile(net, x)
@ -907,7 +904,6 @@ def test_assign():
return grad_all(self.network)(x)
def compile_sub_net(net, x):
net.set_auto_parallel()
net.set_train()
_cell_graph_executor.compile(net, x)

1
tests/ut/python/parallel/test_auto_parallel_BN_PReLU.py
View File

@ -76,6 +76,5 @@ def test_auto_parallel_bn_with_prelu():
x = Tensor(np.random.rand(16, 16, 32, 64), dtype=ms.float32)
net = GradWrap(NetWithLoss(Net()))
net.set_auto_parallel()
net.set_train()
_cell_graph_executor.compile(net, x)

1
tests/ut/python/parallel/test_auto_parallel_activation.py
View File

@ -46,7 +46,6 @@ _b = Tensor(np.ones([64, 32]), dtype=ms.float32)
def compile_net(net):
optimizer = Momentum(net.trainable_params(), learning_rate=0.1, momentum=0.9)
train_net = TrainOneStepCell(net, optimizer)
train_net.set_auto_parallel()
train_net.set_train()
_cell_graph_executor.compile(train_net, _x, _b)
context.reset_auto_parallel_context()

1
tests/ut/python/parallel/test_auto_parallel_adasum.py
View File

@ -53,7 +53,6 @@ def compile_net(net, by_grad=True):
else:
optimizer = AdaSumByDeltaWeightWrapCell(Momentum(net.trainable_params(), learning_rate=0.1, momentum=0.9))
train_net = TrainOneStepCell(net, optimizer)
train_net.set_auto_parallel()
train_net.set_train()
_cell_graph_executor.compile(train_net, _x, _b)
context.reset_auto_parallel_context()

1
tests/ut/python/parallel/test_auto_parallel_arithmetic.py
View File

@ -56,7 +56,6 @@ class GradWrap(nn.Cell):
def compile_net(net, x, y, b, phase):
net.set_auto_parallel()
net.set_train()
_cell_graph_executor.compile(net, x, y, b, phase=phase)

1
tests/ut/python/parallel/test_auto_parallel_assign_sub_with_ref_key.py
View File

@ -63,7 +63,6 @@ def test_auto_parallel_assign_sub_with_ref_key():
net = NetWithLoss(nn.PReLU(4))
context.set_auto_parallel_context(parallel_mode="auto_parallel")
net.set_auto_parallel()
reset_op_id()
net.set_train()

1
tests/ut/python/parallel/test_auto_parallel_cast.py
View File

@ -82,7 +82,6 @@ def test_double_star_graph():
net = NetWithLoss(Net())
context.set_auto_parallel_context(parallel_mode="auto_parallel")
net.set_auto_parallel()
reset_op_id()
net.set_train()

1
tests/ut/python/parallel/test_auto_parallel_common_parameter.py
View File

@ -75,6 +75,5 @@ def test_common_parameter():
net = GradWrap(NetWithLoss(Net()))
context.set_auto_parallel_context(parallel_mode="auto_parallel")
net.set_auto_parallel()
net.set_train()
_cell_graph_executor.compile(net, x, y, z)

1
tests/ut/python/parallel/test_auto_parallel_data_parallel_mode.py
View File

@ -53,7 +53,6 @@ class NetRecursive(nn.Cell):
return self.mul_net(out1, out2)
def compile_net(net, x, y):
net.set_auto_parallel()
net.set_train()
_cell_graph_executor.compile(net, x, y)

2
tests/ut/python/parallel/test_auto_parallel_double_sources.py
View File

@ -82,7 +82,6 @@ def test_double_source_graph():
net = GradWrap(NetWithLoss(Net()))
context.set_auto_parallel_context(parallel_mode="auto_parallel")
net.set_auto_parallel()
net.set_train()
_cell_graph_executor.compile(net, x, y, z, w, a)
@ -118,6 +117,5 @@ def test_double_source_complex_graph():
net = GradWrap(NetWithLoss(Net()))
context.set_auto_parallel_context(parallel_mode="auto_parallel")
net.set_auto_parallel()
net.set_train()
_cell_graph_executor.compile(net, x, y, z, w, a)

1
tests/ut/python/parallel/test_auto_parallel_double_star.py
View File

@ -86,6 +86,5 @@ def test_double_star_graph():
net = GradWrap(NetWithLoss(Net()))
context.set_auto_parallel_context(parallel_mode="auto_parallel")
net.set_auto_parallel()
net.set_train()
_cell_graph_executor.compile(net, x, y, z, w, a, b, c)

1
tests/ut/python/parallel/test_auto_parallel_double_subgraphs.py
View File

@ -113,7 +113,6 @@ def test_double_subgraphs():
context.set_auto_parallel_context(device_num=8, global_rank=0)
context.set_auto_parallel_context(parallel_mode="auto_parallel")
net = TrainStepWarp(NetWithLoss(Net()))
net.set_auto_parallel()
x = Tensor(np.ones([8, 8, 8, 8]), dtype=ms.float32)
reset_op_id()

1
tests/ut/python/parallel/test_auto_parallel_fc_nobias.py
View File

@ -73,6 +73,5 @@ def test_two_matmul():
net = GradWrap(NetWithLoss(Net()))
context.set_auto_parallel_context(parallel_mode="auto_parallel")
net.set_auto_parallel()
net.set_train()
_cell_graph_executor.compile(net, x, y, b)

1
tests/ut/python/parallel/test_auto_parallel_for_loop.py
View File

@ -130,7 +130,6 @@ def test_auto_parallel():
context.set_auto_parallel_context(parallel_mode="auto_parallel", device_num=16, global_rank=0)
_set_algo_single_loop(True)
net = Full(_w1, 3)
net.set_auto_parallel()
net.set_train()
_cell_graph_executor.compile(net, _x, phase='train')
num_ops = _cell_graph_executor._get_num_parallel_ops(net)

1
tests/ut/python/parallel/test_auto_parallel_for_loop_multi_subgraph.py
View File

@ -131,7 +131,6 @@ def test_double_subgraphs():
_set_algo_single_loop(True)
net = TrainStepWarp(NetWithLoss(Net()))
_set_multi_subgraphs()
net.set_auto_parallel()
x = Tensor(np.ones([8, 8, 8, 8]), dtype=ms.float32)
reset_op_id()

1
tests/ut/python/parallel/test_auto_parallel_for_loop_reshape.py
View File

@ -134,7 +134,6 @@ def test_auto_parallel():
context.set_auto_parallel_context(parallel_mode="auto_parallel", device_num=16, global_rank=0)
_set_algo_single_loop(True)
net = Full(_w1, 3)
net.set_auto_parallel()
net.set_train()
_cell_graph_executor.compile(net, _x, phase='train')
num_ops = _cell_graph_executor._get_num_parallel_ops(net)

1
tests/ut/python/parallel/test_auto_parallel_four_matmul.py
View File

@ -52,7 +52,6 @@ class GradWrap(nn.Cell):
def compile_net(net, x, y, z, w, b):
net.set_auto_parallel()
net.set_train()
_cell_graph_executor.compile(net, x, y, z, w, b)

1
tests/ut/python/parallel/test_auto_parallel_inference.py
View File

@ -52,6 +52,5 @@ def test_inference_phase():
net_with_loss = WithLossCell(net, loss)
train_network = TrainOneStepCell(net_with_loss, optimizer)
train_network.set_train()
train_network.set_auto_parallel()
_ = train_network(predict, label)

1
tests/ut/python/parallel/test_auto_parallel_l2normalize.py
View File

@ -71,7 +71,6 @@ def test_auto_parallel_l2normalize():
context.set_auto_parallel_context(device_num=8, global_rank=0)
net = NetWithLoss(Net())
context.set_auto_parallel_context(parallel_mode="auto_parallel")
net.set_auto_parallel()
reset_op_id()
x = Tensor(np.ones([128, 64, 64]), dtype=ms.float32)

1
tests/ut/python/parallel/test_auto_parallel_matmul_drop.py
View File

@ -69,7 +69,6 @@ def test_two_matmul_dropout():
context.set_auto_parallel_context(device_num=8, global_rank=0)
net = GradWrap(NetWithLoss(Net()))
context.set_auto_parallel_context(parallel_mode="auto_parallel")
net.set_auto_parallel()
x = Tensor(np.ones([128, 32]), dtype=ms.float32)
y = Tensor(np.ones([32, 64]), dtype=ms.float32)

1
tests/ut/python/parallel/test_auto_parallel_matmul_prelu.py
View File

@ -76,7 +76,6 @@ def test_matmul_prelu():
net = NetWithLoss(Net())
context.set_auto_parallel_context(parallel_mode="auto_parallel")
net.set_auto_parallel()
reset_op_id()
net.set_train()

1
tests/ut/python/parallel/test_auto_parallel_multi_graph.py
View File

@ -61,7 +61,6 @@ label_ = Tensor(np.random.randn(128, 96).astype(np.float32), dtype=ms.float32)
def compile_net(net):
optimizer = Momentum(net.trainable_params(), learning_rate=0.1, momentum=0.9)
train_net = TrainOneStepCell(net, optimizer)
train_net.set_auto_parallel()
train_net.set_train()
_cell_graph_executor.compile(train_net, inputs_, label_)
context.reset_auto_parallel_context()

4
tests/ut/python/parallel/test_auto_parallel_onehot.py
View File

@ -98,7 +98,6 @@ def test_auto_parallel_arithmetic():
context.set_auto_parallel_context(device_num=8, global_rank=0)
net = GradWrap(NetWithLoss(Net()))
context.set_auto_parallel_context(parallel_mode="auto_parallel")
net.set_auto_parallel()
x = Tensor(np.ones([64, 32]), dtype=ms.float32)
y = Tensor(np.ones([32, 64]), dtype=ms.float32)
@ -125,7 +124,8 @@ def test_auto_parallel_arithmetic_model():
return out2
context.reset_auto_parallel_context()
context.set_auto_parallel_context(device_num=8, global_rank=0, parallel_mode=ParallelMode.AUTO_PARALLEL, dataset_strategy="data_parallel")
context.set_auto_parallel_context(device_num=8, global_rank=0, parallel_mode=ParallelMode.AUTO_PARALLEL,
dataset_strategy="data_parallel")
net = NetOneHot()
x = Tensor(np.ones([8, 32]), dtype=ms.float32)

1
tests/ut/python/parallel/test_auto_parallel_parameter_cast.py
View File

@ -74,7 +74,6 @@ def test_common_parameter():
net = NetWithLoss(Net())
context.set_auto_parallel_context(parallel_mode="auto_parallel")
net.set_auto_parallel()
reset_op_id()
net.set_train()

1
tests/ut/python/parallel/test_auto_parallel_partial_strategy.py
View File

@ -80,6 +80,5 @@ def test_four_matmul_linear():
net = GradWrap(NetWithLoss(Net(strategy1)))
context.set_auto_parallel_context(parallel_mode="auto_parallel")
net.set_auto_parallel()
net.set_train()
_cell_graph_executor.compile(net, x, y, z, w, b)

1
tests/ut/python/parallel/test_auto_parallel_reduce_method.py
View File

@ -52,7 +52,6 @@ class GradWrap(nn.Cell):
def compile_net(net, x, y, b):
net.set_auto_parallel()
net.set_train()
_cell_graph_executor.compile(net, x, y, b)

2
tests/ut/python/parallel/test_auto_parallel_reshape.py
View File

@ -82,13 +82,11 @@ class GradWrapTwoInput(nn.Cell):
def compile_graph(net, parallel_mode, device_num, x):
context.set_auto_parallel_context(device_num=device_num, global_rank=0, parallel_mode=parallel_mode)
net.set_auto_parallel()
net.set_train()
_cell_graph_executor.compile(net, x)
def compile_graph_two_input(net, parallel_mode, device_num, x, y):
context.set_auto_parallel_context(device_num=device_num, global_rank=0, parallel_mode=parallel_mode)
net.set_auto_parallel()
net.set_train()
_cell_graph_executor.compile(net, x, y)

1
tests/ut/python/parallel/test_auto_parallel_rhombus.py
View File

@ -52,7 +52,6 @@ class GradWrap(nn.Cell):
def compile_net(net, x, y, b):
net.set_auto_parallel()
net.set_train()
_cell_graph_executor.compile(net, x, y, b)

1
tests/ut/python/parallel/test_auto_parallel_segment_min.py
View File

@ -70,6 +70,5 @@ def test_auto_parallel_unsortedsegmentmin():
indices = Tensor(np.random.randint(16, size=(16,)), ms.int32)
net = GradWrap(NetWithLoss(Net(16)))
net.set_auto_parallel()
net.set_train()
_cell_graph_executor.compile(net, x, indices)

1
tests/ut/python/parallel/test_auto_parallel_segment_sum.py
View File

@ -70,6 +70,5 @@ def test_auto_parallel_unsortedsegmentsum():
indices = Tensor(np.random.randint(16, size=(16, 16)))
net = GradWrap(NetWithLoss(Net(16)))
net.set_auto_parallel()
net.set_train()
_cell_graph_executor.compile(net, x, indices)

1
tests/ut/python/parallel/test_auto_parallel_shard_propagation.py
View File

@ -47,7 +47,6 @@ _b = Tensor(np.ones([64, 32]), dtype=ms.float32)
def compile_net(net):
optimizer = Momentum(net.trainable_params(), learning_rate=0.1, momentum=0.9)
train_net = TrainOneStepCell(net, optimizer)
train_net.set_auto_parallel()
train_net.set_train()
_cell_graph_executor.compile(train_net, _x, _b)
context.reset_auto_parallel_context()

1
tests/ut/python/parallel/test_auto_parallel_shard_propagation2.py
View File

@ -48,7 +48,6 @@ _b = Tensor(np.ones([64, 32]), dtype=ms.float32)
def compile_net(net):
optimizer = Momentum(net.trainable_params(), learning_rate=0.1, momentum=0.9)
train_net = TrainOneStepCell(net, optimizer)
train_net.set_auto_parallel()
train_net.set_train()
_cell_graph_executor.compile(train_net, _x, _b)
context.reset_auto_parallel_context()

1
tests/ut/python/parallel/test_auto_parallel_shard_propagation3.py
View File

@ -49,7 +49,6 @@ _b = Tensor(np.ones([64, 32000]), dtype=ms.float32)
def compile_net(net):
optimizer = Momentum(net.trainable_params(), learning_rate=0.1, momentum=0.9)
train_net = TrainOneStepCell(net, optimizer)
train_net.set_auto_parallel()
train_net.set_train()
_cell_graph_executor.compile(train_net, _x, _b)
context.reset_auto_parallel_context()

1
tests/ut/python/parallel/test_auto_parallel_softmax_loss.py
View File

@ -65,7 +65,6 @@ def test_softmax_cross_entropy_loss_auto_parallel():
context.set_auto_parallel_context(device_num=8, global_rank=0)
net = GradWrap(NetWithLoss(Net()))
context.set_auto_parallel_context(parallel_mode="auto_parallel")
net.set_auto_parallel()
x = Tensor(np.ones([64, 32]), dtype=ms.float32)
y = Tensor(np.ones([64, 32]), dtype=ms.float32)

4
tests/ut/python/parallel/test_auto_parallel_star_partial_strategy.py
View File

@ -90,7 +90,6 @@ def test_star_strategy_consistency1():
"relu2": ((2, 2),), "add": ((1, 8), (1, 8))}
net = NetWithLoss(Net(strategy_dict))
context.set_auto_parallel_context(parallel_mode="auto_parallel")
net.set_auto_parallel()
reset_op_id()
net.set_train()
_cell_graph_executor.compile(net, x, phase='train')
@ -105,7 +104,6 @@ def test_star_strategy_consistency2():
"relu2": ((2, 2),), "add": ((8, 1), (8, 1))}
net = NetWithLoss(Net(strategy_dict))
context.set_auto_parallel_context(parallel_mode="auto_parallel")
net.set_auto_parallel()
reset_op_id()
net.set_train()
_cell_graph_executor.compile(net, x, phase='train')
@ -120,7 +118,6 @@ def test_star_strategy_consistency3():
"relu2": ((4, 1),), "add": ((2, 2), (2, 2))}
net = NetWithLoss(Net(strategy_dict))
context.set_auto_parallel_context(parallel_mode="auto_parallel")
net.set_auto_parallel()
reset_op_id()
net.set_train()
_cell_graph_executor.compile(net, x, phase='train')
@ -135,7 +132,6 @@ def test_star_strategy_consistency4():
"relu2": None, "add": None}
net = NetWithLoss(Net(strategy_dict))
context.set_auto_parallel_context(parallel_mode="auto_parallel")
net.set_auto_parallel()
reset_op_id()
with pytest.raises(RuntimeError):
net.set_train()

1
tests/ut/python/parallel/test_auto_parallel_transformer.py
View File

@ -115,6 +115,5 @@ def test_dmnet_train_step():
input_ = Tensor(np.ones([4096, 4096]).astype(np.float32) * 0.01)
net = GradWrap(NetWithLoss(MultiTransformer()))
context.set_auto_parallel_context(parallel_mode="auto_parallel")
net.set_auto_parallel()
net.set_train()
_cell_graph_executor.compile(net, input_)

1
tests/ut/python/parallel/test_auto_parallel_transpose.py
View File

@ -77,7 +77,6 @@ def test_two_matmul_transpose():
net = NetWithLoss(Net())
context.set_auto_parallel_context(parallel_mode="auto_parallel")
net.set_auto_parallel()
reset_op_id()
net.set_train()

1
tests/ut/python/parallel/test_auto_parallel_triangle_overwrite.py
View File

@ -71,7 +71,6 @@ def test_triangle_strategy_consistency():
x = Tensor(np.ones([128, 1000]), dtype=ms.float32)
net = NetWithLoss(Net())
context.set_auto_parallel_context(parallel_mode="auto_parallel")
net.set_auto_parallel()
reset_op_id()
net.set_train()

1
tests/ut/python/parallel/test_auto_parallel_tuple_depend.py
View File

@ -78,7 +78,6 @@ def test_virtual_dataset_3_input():
context.set_auto_parallel_context(parallel_mode="auto_parallel")
context.set_auto_parallel_context(device_num=8, global_rank=0)
net = GradWrap(NetWithLoss(Net()))
net.set_auto_parallel()
x = Tensor(np.ones([128, 32]), dtype=ms.float32)
y = Tensor(np.ones([32, 64]), dtype=ms.float32)
b = Tensor(np.ones([64, 2048]), dtype=ms.float32)

1
tests/ut/python/parallel/test_auto_parallel_two_bn.py
View File

@ -76,7 +76,6 @@ def test_two_bn():
context.set_auto_parallel_context(parallel_mode="auto_parallel")
net = NetWithLoss(Net())
x = Tensor(np.ones([64, 64]), dtype=ms.float32)
net.set_auto_parallel()
net.set_train()
set_algo_parameters(elementwise_op_strategy_follow=True)
reset_op_id()

1
tests/ut/python/parallel/test_auto_parallel_two_matmul.py
View File

@ -155,7 +155,6 @@ def test_two_matmul():
net = NetWithLoss(Net())
context.set_auto_parallel_context(parallel_mode="auto_parallel")
net.set_auto_parallel()
reset_op_id()
net.set_train()

1
tests/ut/python/parallel/test_auto_parallel_two_partial_matmul.py
View File

@ -74,6 +74,5 @@ def test_four_matmul_linear():
net = GradWrap(NetWithLoss(Net(strategy1)))
context.set_auto_parallel_context(parallel_mode="auto_parallel")
net.set_auto_parallel()
net.set_train()
_cell_graph_executor.compile(net, x, y)

1
tests/ut/python/parallel/test_auto_parallel_zig_zag.py
View File

@ -80,6 +80,5 @@ def test_zig_zag_graph():
net = GradWrap(NetWithLoss(Net()))
context.set_auto_parallel_context(parallel_mode="auto_parallel")
net.set_auto_parallel()
net.set_train()
_cell_graph_executor.compile(net, x, y, z, w, a)

1
tests/ut/python/parallel/test_auto_star_elimination.py
View File

@ -92,6 +92,5 @@ def test_marin_loss():
net = GradWrap(NetWithLoss(MarginCE()))
context.set_auto_parallel_context(parallel_mode="auto_parallel")
net.set_auto_parallel()
net.set_train()
_cell_graph_executor.compile(net, x, y)

1
tests/ut/python/parallel/test_batch_matmul.py
View File

@ -48,7 +48,6 @@ _b = Tensor(np.ones([128, 64, 16]), dtype=ms.float32)
def compile_net(net):
optimizer = Momentum(net.trainable_params(), learning_rate=0.1, momentum=0.9)
train_net = TrainOneStepCell(net, optimizer)
train_net.set_auto_parallel()
train_net.set_train()
_cell_graph_executor.compile(train_net, _x, _b)
context.reset_auto_parallel_context()

2
tests/ut/python/parallel/test_batch_parallel.py
View File

@ -112,7 +112,6 @@ def test_batch():
strategy3 = ((4, 1, 1, 2), (4, 1, 1, 2))
net = GradWrap(NetWithLoss(Net(strategy1, strategy2, strategy3)))
net.set_auto_parallel()
x = Tensor(np.ones([128, 16, 34, 34]), dtype=ms.float32)
w1 = Tensor(np.ones([128, 8, 32, 32]), dtype=ms.float32)
@ -134,7 +133,6 @@ def test_batch_shape_less_than_devices():
strategy3 = None
net = GradWrap(NetWithLoss(Net(strategy1, strategy2, strategy3)))
net.set_auto_parallel()
x = Tensor(np.ones([128, 16, 34, 34]), dtype=ms.float32)
w1 = Tensor(np.ones([128, 8, 32, 32]), dtype=ms.float32)

1
tests/ut/python/parallel/test_batch_parallel_dropout.py
View File

@ -69,7 +69,6 @@ def test_batch_parallel_dropout():
context.set_auto_parallel_context(device_num=8, global_rank=0)
net = GradWrap(NetWithLoss(Net()))
context.set_auto_parallel_context(parallel_mode="semi_auto_parallel")
net.set_auto_parallel()
x = Tensor(np.ones([128, 32]), dtype=ms.float32)
y = Tensor(np.ones([32, 64]), dtype=ms.float32)

1
tests/ut/python/parallel/test_batch_parallel_tensoradd.py
View File

@ -67,7 +67,6 @@ def test_matmul_add():
net = GradWrap(NetWithLoss(Net()))
context.set_auto_parallel_context(parallel_mode="semi_auto_parallel")
net.set_auto_parallel()
x = Tensor(np.ones([64, 32]), dtype=ms.float32)
y = Tensor(np.ones([32, 64]), dtype=ms.float32)

1
tests/ut/python/parallel/test_batchmm.py
View File

@ -61,7 +61,6 @@ def compile_net(net):
context.set_context(mode=context.GRAPH_MODE)
optimizer = Momentum(net.trainable_params(), learning_rate=0.1, momentum=0.9)
train_net = TrainOneStepCell(net, optimizer)
train_net.set_auto_parallel()
train_net.set_train()
_cell_graph_executor.compile(train_net, _x)
context.reset_auto_parallel_context()

2
tests/ut/python/parallel/test_batchnorm.py
View File

@ -50,7 +50,6 @@ _b = Tensor(np.ones([32, 16, 8, 8]), dtype=ms.float32)
def compile_net(net):
optimizer = Momentum(net.trainable_params(), learning_rate=0.1, momentum=0.9)
train_net = TrainOneStepCell(net, optimizer)
train_net.set_auto_parallel()
train_net.set_train()
_cell_graph_executor.compile(train_net, _x, _b)
context.reset_auto_parallel_context()
@ -117,7 +116,6 @@ _b1 = Tensor(np.ones([32, 8]), dtype=ms.float32)
def compile_net2(net):
optimizer = Momentum(net.trainable_params(), learning_rate=0.1, momentum=0.9)
train_net = TrainOneStepCell(net, optimizer)
train_net.set_auto_parallel()
train_net.set_train()
_cell_graph_executor.compile(train_net, _x1, _b1)
context.reset_auto_parallel_context()

1
tests/ut/python/parallel/test_batchparallel_replace_shape.py
View File

@ -49,7 +49,6 @@ _b = Tensor(np.ones([128, 64, 32]), dtype=ms.float32)
def compile_net(net):
optimizer = Momentum(net.trainable_params(), learning_rate=0.1, momentum=0.9)
train_net = TrainOneStepCell(net, optimizer)
train_net.set_auto_parallel()
train_net.set_train()
_cell_graph_executor.compile(train_net, _x, _b)
context.reset_auto_parallel_context()

1
tests/ut/python/parallel/test_bounding_box_encode.py
View File

@ -44,7 +44,6 @@ class Net(Cell):
def compile_net(net: Cell, *inputs):
net.set_auto_parallel()
net.set_train()
_cell_graph_executor.compile(net, *inputs)
context.reset_auto_parallel_context()

2
tests/ut/python/parallel/test_broadcast_to.py
View File

@ -71,7 +71,6 @@ def compile_net(net):
context.set_context(mode=context.GRAPH_MODE)
optimizer = Momentum(net.trainable_params(), learning_rate=0.1, momentum=0.9)
train_net = TrainOneStepCell(net, optimizer)
train_net.set_auto_parallel()
train_net.set_train()
_cell_graph_executor.compile(train_net, _x1)
context.reset_auto_parallel_context()
@ -81,7 +80,6 @@ def compile_net2(net):
context.set_context(mode=context.GRAPH_MODE)
optimizer = Momentum(net.trainable_params(), learning_rate=0.1, momentum=0.9)
train_net = TrainOneStepCell(net, optimizer)
train_net.set_auto_parallel()
train_net.set_train()
_cell_graph_executor.compile(train_net, _x1, _x2)
context.reset_auto_parallel_context()

1
tests/ut/python/parallel/test_comm_not_recompute.py
View File

@ -81,7 +81,6 @@ def compile_net(mp_comm_recompute, recompute_slice_activation):
label = Tensor(np.zeros([32, 768]).astype(np.float32))
net = train_step_with_loss_warp(DenseMutMulNet(mp_comm_recompute=mp_comm_recompute,
recompute_slice_activation=recompute_slice_activation))
net.set_auto_parallel()
net.set_train()
_cell_graph_executor.compile(net, input_, label)
_Context().set_backend_policy("ge")

1
tests/ut/python/parallel/test_comparison_function_info.py
View File

@ -52,7 +52,6 @@ class GradWrap(nn.Cell):
def compile_net(net, x, y, b):
net.set_auto_parallel()
net.set_train()
_cell_graph_executor.compile(net, x, y, b)

1
tests/ut/python/parallel/test_concat.py
View File

@ -86,7 +86,6 @@ w3 = Tensor(np.ones([64, 64, 32]), dtype=ms.float32)
def compile_net(net):
optimizer = Momentum(net.trainable_params(), learning_rate=0.1, momentum=0.9)
train_net = TrainOneStepCell(net, optimizer)
train_net.set_auto_parallel()
train_net.set_train()
_cell_graph_executor.compile(train_net, _x, _b)
context.reset_auto_parallel_context()

1
tests/ut/python/parallel/test_conv2d.py
View File

@ -56,7 +56,6 @@ _b = Tensor(np.ones([32, 16, 8, 8]), dtype=ms.float32)
def compile_net(net, input_x=_x):
optimizer = Momentum(net.trainable_params(), learning_rate=0.1, momentum=0.9)
train_net = TrainOneStepCell(net, optimizer)
train_net.set_auto_parallel()
train_net.set_train()
_cell_graph_executor.compile(train_net, input_x, _b)
context.reset_auto_parallel_context()

1
tests/ut/python/parallel/test_conv2d_transpose.py
View File

@ -75,7 +75,6 @@ _b = Tensor(np.ones([32, 16, 8, 8]), dtype=ms.float32)
def compile_net(net):
optimizer = Momentum(net.trainable_params(), learning_rate=0.1, momentum=0.9)
train_net = TrainOneStepCell(net, optimizer)
train_net.set_auto_parallel()
train_net.set_train()
_cell_graph_executor.compile(train_net, _x, _b)
context.reset_auto_parallel_context()

1
tests/ut/python/parallel/test_crop_and_resize.py
View File

@ -47,7 +47,6 @@ class Net(Cell):
def compile_net(net: Cell, *inputs):
net.set_auto_parallel()
net.set_train()
_cell_graph_executor.compile(net, *inputs)
context.reset_auto_parallel_context()

1
tests/ut/python/parallel/test_cus_matmul_dds.py
View File

@ -110,7 +110,6 @@ def compile_graph(batch_size, num_heads, dp, mp, auto=False, shard=True):
context.set_auto_parallel_context(parallel_mode="semi_auto_parallel")
x = Tensor(np.ones((batch_size * 1024, num_heads * 128)), ms.float32)
net = GradWrap(NetWithLoss(Net(batch_size, num_heads, dp, mp, shard=shard)))
net.set_auto_parallel()
net.set_train()
_cell_graph_executor.compile(net, x)

1
tests/ut/python/parallel/test_device_num_not_power_of_2.py
View File

@ -75,7 +75,6 @@ _b1 = Tensor(np.ones([32 * 3]), dtype=ms.float32)
def compile_net(net, change_input=False):
optimizer = Momentum(net.trainable_params(), learning_rate=0.1, momentum=0.9)
train_net = TrainOneStepCell(net, optimizer)
train_net.set_auto_parallel()
train_net.set_train()
if change_input:
_cell_graph_executor.compile(train_net, _x1, _b1)

1
tests/ut/python/parallel/test_different_type_for_div_op.py
View File

@ -40,7 +40,6 @@ class GradWrap(nn.Cell):
def compile_net(net, x, y):
net.set_auto_parallel()
net.set_train()
_cell_graph_executor.compile(net, x, y)

1
tests/ut/python/parallel/test_dropout.py
View File

@ -52,7 +52,6 @@ _b = Tensor(np.ones([128, 64]), dtype=ms.float32)
def compile_net(net):
optimizer = Momentum(net.trainable_params(), learning_rate=0.1, momentum=0.9)
train_net = TrainOneStepCell(net, optimizer)
train_net.set_auto_parallel()
train_net.set_train()
_cell_graph_executor.compile(train_net, _x, _b)
context.reset_auto_parallel_context()

1
tests/ut/python/parallel/test_dropout_do_mask.py
View File

@ -54,7 +54,6 @@ _b = Tensor(np.ones([128, 64]), dtype=ms.float32)
def compile_net(net):
optimizer = Momentum(net.trainable_params(), learning_rate=0.1, momentum=0.9)
train_net = TrainOneStepCell(net, optimizer)
train_net.set_auto_parallel()
train_net.set_train()
_cell_graph_executor.compile(train_net, _x, _b)
context.reset_auto_parallel_context()

1
tests/ut/python/parallel/test_dsd_matmul.py
View File

@ -126,7 +126,6 @@ def compile_graph(batch_size, num_heads, dp, mp, auto=False, shard=True):
context.set_auto_parallel_context(parallel_mode="semi_auto_parallel")
x = Tensor(np.ones((batch_size * 1024, num_heads * 128)), ms.float32)
net = GradWrap(NetWithLoss(Net(batch_size, num_heads, dp, mp, shard=shard)))
net.set_auto_parallel()
net.set_train()
_cell_graph_executor.compile(net, x)

2
tests/ut/python/parallel/test_dynamic_shape.py
View File

@ -83,7 +83,6 @@ def test_unique_column_split():
net = Net()
optimizer = Momentum(net.trainable_params(), learning_rate=0.1, momentum=0.9)
train_net = TrainOneStepCell(net, optimizer)
train_net.set_auto_parallel()
train_net.set_train()
_cell_graph_executor.compile(train_net, x)
@ -117,6 +116,5 @@ def test_unique_row_split():
net = Net()
optimizer = Momentum(net.trainable_params(), learning_rate=0.1, momentum=0.9)
train_net = TrainOneStepCell(net, optimizer)
train_net.set_auto_parallel()
train_net.set_train()
_cell_graph_executor.compile(train_net, x)

1
tests/ut/python/parallel/test_element_wise_function.py
View File

@ -52,7 +52,6 @@ class GradWrap(nn.Cell):
def compile_net(net, x, y, b):
net.set_auto_parallel()
net.set_train()
_cell_graph_executor.compile(net, x, y, b)

5
tests/ut/python/parallel/test_embeddinglookup.py
View File

@ -61,7 +61,6 @@ def test_embeddinglookup_reducescatter_false():
shape = [8, 8]
offset = 8
net = NetWithLoss(Net(shape, offset))
net.set_auto_parallel()
x = Tensor(np.ones([64, 32]), dtype=ms.float32)
y = Tensor(np.ones([8, 32, 8]), dtype=ms.float32)
@ -73,7 +72,6 @@ def test_embeddinglookup_reducescatter_true():
shape = [8, 8]
offset = 8
net = NetWithLoss(Net(shape, offset))
net.set_auto_parallel()
x = Tensor(np.ones([64, 32]), dtype=ms.float32)
y = Tensor(np.ones([8, 32, 8]), dtype=ms.float32)
@ -85,7 +83,6 @@ def test_embeddinglookup_reducescatter_false_grad():
shape = [8, 8]
offset = 8
net = GradWrap(NetWithLoss(Net(shape, offset)))
net.set_auto_parallel()
x = Tensor(np.ones([64, 32]), dtype=ms.float32)
y = Tensor(np.ones([8, 32, 8]), dtype=ms.float32)
@ -97,7 +94,6 @@ def test_embeddinglookup_reducescatter_true_grad():
shape = [8, 8]
offset = 8
net = GradWrap(NetWithLoss(Net(shape, offset)))
net.set_auto_parallel()
x = Tensor(np.ones([64, 32]), dtype=ms.float32)
y = Tensor(np.ones([8, 32, 8]), dtype=ms.float32)
@ -114,7 +110,6 @@ def test_embeddinglookup_semi_auto1():
strategy2 = ((4, 1, 2), (4, 2, 1))
net = GradWrap(NetWithLoss(Net(shape, offset, strategy1, strategy2, "CPU")))
net.set_auto_parallel()
x = Tensor(np.ones([64 // 8, 64]), dtype=ms.float32)
y = Tensor(np.ones([64 // 8, 64, 64]), dtype=ms.float32)
net.set_train()

3
tests/ut/python/parallel/test_eval.py
View File

@ -55,10 +55,9 @@ _w1 = Tensor(np.ones([64, 64]), dtype=ms.float32)
_b = Tensor(np.ones([64, 64]), dtype=ms.float32)
def compile_net(net, input_data, label, is_train=True):
net.set_auto_parallel()
net.set_train(mode=is_train)
phase = "train" if is_train else "eval"
_cell_graph_executor.compile(net, input_data, label, phase=phase, auto_parallel_mode=True)
_cell_graph_executor.compile(net, input_data, label, phase=phase)
def test_train_and_eval():
"""

1
tests/ut/python/parallel/test_expand_dims.py
View File

@ -61,7 +61,6 @@ _b = Tensor(np.ones([128, 64, 32, 1]), dtype=ms.float32)
def compile_net(net):
optimizer = Momentum(net.trainable_params(), learning_rate=0.1, momentum=0.9)
train_net = TrainOneStepCell(net, optimizer)
train_net.set_auto_parallel()
train_net.set_train()
_cell_graph_executor.compile(train_net, _x, _b)
context.reset_auto_parallel_context()

1
tests/ut/python/parallel/test_forward_graph.py
View File

@ -44,7 +44,6 @@ _b = Tensor(np.ones([128, 64, 32]), dtype=ms.float32)
def compile_net(net):
net.set_auto_parallel()
net.set_train()
_cell_graph_executor.compile(net, _x, _b)
context.reset_auto_parallel_context()

3
tests/ut/python/parallel/test_gather_field_split.py
View File

@ -38,10 +38,9 @@ class Net(Cell):
def compile_net(net, x):
net.set_auto_parallel()
net.set_train()
b = Tensor(np.ones([64, 8]), dtype=ms.float32)
phase, _ = _cell_graph_executor.compile(net, x, b, auto_parallel_mode=True)
phase, _ = _cell_graph_executor.compile(net, x, b)
context.reset_auto_parallel_context()
return phase

1
tests/ut/python/parallel/test_gather_v2.py
View File

@ -66,7 +66,6 @@ class Net(nn.Cell):
def compile_graph(net, device_num, parallel_mode, x, y):
context.set_auto_parallel_context(device_num=device_num, global_rank=0, parallel_mode=parallel_mode)
net.set_auto_parallel()
net.set_train()
_cell_graph_executor.compile(net, x, y)

1
tests/ut/python/parallel/test_gatherd.py
View File

@ -47,7 +47,6 @@ _b = Tensor(np.ones([16, 32, 64]), dtype=ms.float32)
def compile_net(net):
optimizer = Momentum(net.trainable_params(), learning_rate=0.1, momentum=0.9)
train_net = TrainOneStepCell(net, optimizer)
train_net.set_auto_parallel()
train_net.set_train()
_cell_graph_executor.compile(train_net, _x, _b)
context.reset_auto_parallel_context()

1
tests/ut/python/parallel/test_get_next.py
View File

@ -64,7 +64,6 @@ class GradWrap(nn.Cell):
def compile_net(net):
net.set_auto_parallel()
_cell_graph_executor.compile(net)
def test_get_next_single():

3
tests/ut/python/parallel/test_get_parameter_layout.py
View File

@ -50,10 +50,9 @@ def test_get_parameter_layout():
weight = Tensor(np.ones([64, 32]), dtype=ms.float32)
net = Net(strategy1, strategy2, weight)
net.set_auto_parallel()
net.set_train()
exe = me._cell_graph_executor
exe.compile(net, x, phase='train', auto_parallel_mode=True)
exe.compile(net, x, phase='train')
x_layout = ([8], [0, -1], [32, 32], 0, True, '') # device_arrangement = [2, 4], tensor_map = [1, -1]
weight_layout = ([2, 4], [0, -1], [16, 32], 0, True, '') # device_arrangement = [2, 4], tensor_map = [0, -1]
expect_dict = {'x': x_layout, 'w1': weight_layout}

1
tests/ut/python/parallel/test_gpu_dropout.py
View File

@ -63,7 +63,6 @@ class Net(nn.Cell):
def compile_graph(net, device_num, parallel_mode, x, y):
context.set_auto_parallel_context(device_num=device_num, global_rank=0, parallel_mode=parallel_mode)
net.set_auto_parallel()
net.set_train()
_cell_graph_executor.compile(net, x, y)

1
tests/ut/python/parallel/test_hybird_parallel_activation.py
View File

@ -52,7 +52,6 @@ class GradWrap(nn.Cell):
def compile_net(net, x, y, b):
net.set_auto_parallel()
net.set_train()
_cell_graph_executor.compile(net, x, y, b)

12
tests/ut/python/parallel/test_initializer_weight_slice.py
View File

@ -64,9 +64,8 @@ def check_initializer_weight_slice(init_name="Uniform", using_seed=False):
weight1 = initializer(init_name, [32, 32], ms.float32)
weight2 = initializer(init_name, [32, 32], ms.float32)
net = Net(strategy1, strategy2, weight1, weight2)
net.set_auto_parallel()
net.set_train()
exe.compile(net, x, auto_parallel_mode=True, phase='train')
exe.compile(net, x, phase='train')
hccl.rank_id = rank_save
return net.parameters_dict()['w1'].data.asnumpy(), net.parameters_dict()['w2'].data.asnumpy()
@ -123,9 +122,8 @@ def test_wrong_order_set_parallel_mode_with_initializer():
exe = me._cell_graph_executor
x = Tensor(np.ones([32, 32]), dtype=ms.float32)
context.set_auto_parallel_context(parallel_mode="semi_auto_parallel", device_num=8, global_rank=0)
net.set_auto_parallel()
with pytest.raises(RuntimeError):
exe.compile(net, x, auto_parallel_mode=True, phase='train')
exe.compile(net, x, phase='train')
def test_wrong_order_set_same_parallel_mode_with_initializer():
@ -143,8 +141,7 @@ def test_wrong_order_set_same_parallel_mode_with_initializer():
exe = me._cell_graph_executor
x = Tensor(np.ones([32, 32]), dtype=ms.float32)
context.set_auto_parallel_context(parallel_mode="auto_parallel", device_num=8, global_rank=0)
net.set_auto_parallel()
exe.compile(net, x, auto_parallel_mode=True, phase='train')
exe.compile(net, x, phase='train')
def test_wrong_order_set_parallel_mode_without_initializer():
@ -161,5 +158,4 @@ def test_wrong_order_set_parallel_mode_without_initializer():
exe = me._cell_graph_executor
x = Tensor(np.ones([32, 32]), dtype=ms.float32)
context.set_auto_parallel_context(parallel_mode="semi_auto_parallel", device_num=8, global_rank=0)
net.set_auto_parallel()
exe.compile(net, x, auto_parallel_mode=True, phase='train')
exe.compile(net, x, phase='train')

1
tests/ut/python/parallel/test_iou.py
View File

@ -42,7 +42,6 @@ class Net(Cell):
def compile_net(net: Cell):
net.set_train()
net.set_auto_parallel()
_cell_graph_executor.compile(net, _anchor_boxes, _gt_boxes)
context.reset_auto_parallel_context()

1
tests/ut/python/parallel/test_l2normalize.py
View File

@ -74,7 +74,6 @@ def test_l2normalize_matmul():
strategy3 = ((1, 1, 8), (1, 1, 8))
net = GradWrap(NetWithLoss(Net(strategy1, strategy2, strategy3)))
context.set_auto_parallel_context(parallel_mode="semi_auto_parallel")
net.set_auto_parallel()
x = Tensor(np.ones([128, 32, 64]), dtype=ms.float32)
y = Tensor(np.ones([128, 32, 64]), dtype=ms.float32)

1
tests/ut/python/parallel/test_layer_norm.py
View File

@ -55,7 +55,6 @@ _b = Tensor(np.ones([16, 64, 32, 16]), dtype=ms.float32)
def compile_net(net):
optimizer = Momentum(net.trainable_params(), learning_rate=0.1, momentum=0.9)
train_net = TrainOneStepCell(net, optimizer)
train_net.set_auto_parallel()
train_net.set_train()
_cell_graph_executor.compile(train_net, _x, _b)
context.reset_auto_parallel_context()

1
tests/ut/python/parallel/test_layer_norm_further.py
View File

@ -153,7 +153,6 @@ def compile_net(net):
optimizer = Momentum(net.trainable_params(),
learning_rate=0.1, momentum=0.9)
train_net = TrainOneStepCell(net, optimizer)
train_net.set_auto_parallel()
train_net.set_train()
_cell_graph_executor.compile(train_net, _x, _b)
context.reset_auto_parallel_context()

1
tests/ut/python/parallel/test_linear.py
View File

@ -71,7 +71,6 @@ def test_linear():
strategy2 = ((2, 8),)
strategy3 = ((16, 1), (16, 1))
net = GradWrap(NetWithLoss(Net(strategy0, strategy1, strategy2), strategy3))
net.set_auto_parallel()
x = Tensor(np.ones([64, 32]), dtype=ms.float32)
y = Tensor(np.ones([64, 32]), dtype=ms.float32)

1
tests/ut/python/parallel/test_loop_two_matmul.py
View File

@ -98,7 +98,6 @@ def test_two_matmul():
print(strategy1, strategy2)
net = GradWrap(NetWithLoss(Net(strategy1, strategy2)))
context.set_auto_parallel_context(parallel_mode="semi_auto_parallel")
net.set_auto_parallel()
net.set_train()
_cell_graph_executor.compile(net, x, y, b)
count = count + 1

1
tests/ut/python/parallel/test_loss_and_optimizer.py
View File

@ -40,7 +40,6 @@ class NetWithLoss(nn.Cell):
def compile_net(net, x, b):
net.set_auto_parallel()
net.set_train()
_cell_graph_executor.compile(net, x, b)

3
tests/ut/python/parallel/test_manual_embedding_lookup.py
View File

@ -70,9 +70,8 @@ def compile_net(net):
optimizer = LazyAdam(net.trainable_params(), learning_rate=0.1)
optimizer.sparse_opt.set_device("CPU")
train_net = TrainOneStepCell(net, optimizer)
train_net.set_auto_parallel()
train_net.set_train()
_cell_graph_executor.compile(train_net, _x, _b, auto_parallel_mode=True)
_cell_graph_executor.compile(train_net, _x, _b)
context.reset_auto_parallel_context()

4
tests/ut/python/parallel/test_manual_gatherv2.py
View File

@ -82,16 +82,14 @@ _b = Tensor(np.ones([64, 8]), dtype=ms.float32)
def compile_net(net):
optimizer = Momentum(net.trainable_params(), learning_rate=0.1, momentum=0.9)
train_net = TrainOneStepCell(net, optimizer)
train_net.set_auto_parallel()
train_net.set_train()
_cell_graph_executor.compile(train_net, _x, _b, auto_parallel_mode=True)
_cell_graph_executor.compile(train_net, _x, _b)
context.reset_auto_parallel_context()
def compile_net_and_return_strategy(net: Cell, *inputs):
optimizer = Momentum(net.trainable_params(), learning_rate=0.1, momentum=0.9)
train_net = TrainOneStepCell(net, optimizer)
net.set_auto_parallel()
net.set_train()
_cell_graph_executor.compile(train_net, *inputs, phase='train')
strategies = _cell_graph_executor._get_shard_strategy(train_net)

1
tests/ut/python/parallel/test_matmul_dropout.py
View File

@ -126,7 +126,6 @@ def test_two_matmul_dropout():
strategy3 = ((1, 8), (8, 1))
net = GradWrap(NetWithLoss(Net(strategy1, strategy2, strategy3)))
context.set_auto_parallel_context(parallel_mode="semi_auto_parallel")
net.set_auto_parallel()
x = Tensor(np.ones([128, 32]), dtype=ms.float32)
y = Tensor(np.ones([32, 64]), dtype=ms.float32)

1
tests/ut/python/parallel/test_matmul_tensor.py
View File

@ -54,7 +54,6 @@ class GradWrap(nn.Cell):
def compile_net(net, x, y):
net.set_auto_parallel()
net.set_train()
_cell_graph_executor.compile(net, x, y)

Some files were not shown because too many files have changed in this diff Show More