p47068915
/
mindspore
forked from mindspore-Ecosystem/mindspore
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

check arg is tensor with vm backend

This commit is contained in:
buxue 2020-06-10 15:14:28 +08:00
parent 444d9484d7
commit 0cd57ddc5d
5 changed files with 40 additions and 33 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
mindspore/ccsrc/pipeline/pipeline.cc
View File

@ -615,12 +615,16 @@ void ProcessVmArgInner(const py::tuple &args, const ResourcePtr &res, VectorRef
py::object arg = args[i];
auto ms_context = MsContext::GetInstance();
if (ms_context->backend_policy() == kMsConvert && py::isinstance<py::array>(arg)) {
MS_LOG(EXCEPTION) << "Args[" << i << "] is numpy array, not tensor";
MS_LOG(EXCEPTION) << "The " << i << "th arg is numpy array, not tensor.";
}
ValuePtr converted = nullptr;
bool succ = parse::ConvertData(arg, &converted);
if (!succ) {
MS_LOG(EXCEPTION) << "Args convert error";
MS_LOG(EXCEPTION) << "The " << i << "th arg convert failed.";
}
if (MsContext::GetInstance()->execution_mode() == 0 && !converted->isa<tensor::Tensor>()) {
MS_EXCEPTION(TypeError) << "For 'graph mode', the " << i << "th arg: " << converted->ToString()
<< " is not tensor.";
}
arg_list->push_back(converted);
}

4
mindspore/ccsrc/pipeline/pipeline_ge.cc
View File

@ -460,12 +460,12 @@ void ProcessGeArg(const std::map<std::string, ExecutorInfoPtr> &info, const py::
ValuePtr converted = nullptr;
bool succ = parse::ConvertData(args[i], &converted);
if (!succ) {
MS_LOG(EXCEPTION) << "Args convert error";
MS_LOG(EXCEPTION) << "The " << i << "th arg convert failed.";
}
if (converted->isa<tensor::Tensor>()) {
inputs->push_back(converted->cast<tensor::TensorPtr>());
} else {
MS_EXCEPTION(TypeError) << "Args " << converted->ToString() << " is not tensor";
MS_EXCEPTION(TypeError) << "The " << i << "th arg: " << converted->ToString() << " is not tensor.";
}
}
}

6
mindspore/ops/operations/math_ops.py
View File

@ -777,7 +777,7 @@ class Sub(_MathBinaryOp):
When the inputs are one tensor and one scalar,
the scalar only could be a constant.
Inputs:
Inputs:
- **input_x** (Union[Tensor, Number, bool]) - The first input is a number or
a bool or a tensor whose data type is number or bool.
- **input_y** (Union[Tensor, Number, bool]) - The second input is a number or
@ -1845,7 +1845,7 @@ class LogicalAnd(_LogicBinaryOp):
Inputs:
- **input_x** (Union[Tensor, bool]) - The first input is a bool or a tensor whose data type is bool.
- **input_y** (Union[Tensor, bool]) - The second input is a bool when the first input is a tensor or
a tensor whose data type is bool.
a tensor whose data type is bool.
Outputs:
Tensor, the shape is same as the shape after broadcasting, and the data type is bool.
@ -1875,7 +1875,7 @@ class LogicalOr(_LogicBinaryOp):
Inputs:
- **input_x** (Union[Tensor, bool]) - The first input is a bool or a tensor whose data type is bool.
- **input_y** (Union[Tensor, bool]) - The second input is a bool when the first input is a tensor or
a tensor whose data type is bool.
a tensor whose data type is bool.
Outputs:
Tensor, the shape is same as the shape after broadcasting,and the data type is bool.

18
tests/st/ops/gpu/test_cast_op.py
View File

@ -24,13 +24,15 @@ from mindspore.ops import operations as P
class Net(Cell):
def __init__(self):
def __init__(self, type0, type1):
super(Net, self).__init__()
self.Cast = P.Cast()
self.type0 = type0
self.type1 = type1
def construct(self, x0, type0, x1, type1):
output = (self.Cast(x0, type0),
self.Cast(x1, type1))
def construct(self, x0, x1):
output = (self.Cast(x0, self.type0),
self.Cast(x1, self.type1))
return output
@ -44,8 +46,8 @@ def test_cast():
t1 = mstype.float32
context.set_context(mode=context.GRAPH_MODE, device_target='GPU')
net = Net()
output = net(x0, t0, x1, t1)
net = Net(t0, t1)
output = net(x0, x1)
type0 = output[0].asnumpy().dtype
assert type0 == 'float16'
type1 = output[1].asnumpy().dtype
@ -62,8 +64,8 @@ def test_cast1():
t1 = mstype.float32
context.set_context(mode=context.GRAPH_MODE, device_target='GPU')
net = Net()
output = net(x0, t0, x1, t1)
net = Net(t0, t1)
output = net(x0, x1)
type0 = output[0].asnumpy().dtype
assert type0 == 'float32'
type1 = output[1].asnumpy().dtype

37
tests/st/ops/gpu/test_rmsprop.py
View File

@ -25,24 +25,29 @@ context.set_context(mode=context.GRAPH_MODE, device_target="GPU")
class NetCenteredRMSProp(nn.Cell):
def __init__(self):
def __init__(self, lr, decay, momentum, epsilon):
super(NetCenteredRMSProp, self).__init__()
self.rms_opt = P.ApplyCenteredRMSProp()
self.lr = lr
self.decay = decay
self.momentum = momentum
self.epsilon = epsilon
def construct(self, var, g, mg, rms, mom, lr, decay, momentum, epsilon):
return self.rms_opt(var, mg, rms, mom, g, lr, decay, momentum, epsilon)
def construct(self, var, g, mg, rms, mom):
return self.rms_opt(var, mg, rms, mom, g, self.lr, self.decay, self.momentum, self.epsilon)
class NetRMSProp(nn.Cell):
def __init__(self, decay, momentum, epsilon):
def __init__(self, lr, decay, momentum, epsilon):
super(NetRMSProp, self).__init__()
self.lr = lr
self.decay = decay
self.momentum = momentum
self.epsilon = epsilon
self.rms_opt = P.ApplyRMSProp()
def construct(self, var, g, mg, rms, mom, lr):
return self.rms_opt(var, rms, mom, lr, g, self.decay, self.momentum, self.epsilon)
def construct(self, var, g, mg, rms, mom):
return self.rms_opt(var, rms, mom, self.lr, g, self.decay, self.momentum, self.epsilon)
def rmsprop_numpy(variable, gradients, mean_square, moment,
@ -82,16 +87,14 @@ def test_rmsprop():
if centered:
rmspropcented_numpy(variable_np, gradients_np, mean_gradients_np, mean_square_np, moment_np,
learning_rate, decay, momentum, epsilon)
net = NetCenteredRMSProp()
_ = net(variable_ms, gradients_ms, mean_gradients_ms, mean_square_ms,
moment_ms, learning_rate, decay, momentum, epsilon)
net = NetCenteredRMSProp(learning_rate, decay, momentum, epsilon)
_ = net(variable_ms, gradients_ms, mean_gradients_ms, mean_square_ms, moment_ms)
else:
rmsprop_numpy(variable_np, gradients_np, mean_square_np, moment_np,
learning_rate, decay, momentum, epsilon)
net = NetRMSProp(decay, momentum, epsilon)
_ = net(variable_ms, gradients_ms, mean_gradients_ms, mean_square_ms,
moment_ms, learning_rate)
net = NetRMSProp(learning_rate, decay, momentum, epsilon)
_ = net(variable_ms, gradients_ms, mean_gradients_ms, mean_square_ms, moment_ms)
error = np.ones(shape=variable_np.shape) * 10e-6
diff = variable_ms.asnumpy() - variable_np
@ -135,15 +138,13 @@ def test_rmspropcenter():
if centered:
rmspropcented_numpy(variable_np, gradients_np, mean_gradients_np, mean_square_np, moment_np,
learning_rate, decay, momentum, epsilon)
net = NetCenteredRMSProp()
_ = net(variable_ms, gradients_ms, mean_gradients_ms, mean_square_ms, moment_ms,
learning_rate, decay, momentum, epsilon)
net = NetCenteredRMSProp(learning_rate, decay, momentum, epsilon)
_ = net(variable_ms, gradients_ms, mean_gradients_ms, mean_square_ms, moment_ms)
else:
rmsprop_numpy(variable_np, gradients_np, mean_square_np, moment_np,
learning_rate, decay, momentum, epsilon)
net = NetRMSProp(decay, momentum, epsilon)
_ = net(variable_ms, gradients_ms, mean_gradients_ms, mean_square_ms, moment_ms,
learning_rate)
net = NetRMSProp(learning_rate, decay, momentum, epsilon)
_ = net(variable_ms, gradients_ms, mean_gradients_ms, mean_square_ms, moment_ms)
error = np.ones(shape=variable_np.shape) * 10e-6
diff = variable_ms.asnumpy() - variable_np