xxq250
/
mindspore
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

!55448 Fix PyNative tasksink bug. 

Merge pull request !55448 from caifubi/master-pynative-tasksink-bugfix
This commit is contained in:
i-robot 2023-06-21 01:48:24 +00:00 committed by Gitee
parent c57deb1b67 01a33913b0
commit 8f13ed1829
No known key found for this signature in database
GPG Key ID: 173E9B9CA92EEF8F
5 changed files with 66 additions and 16 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
.jenkins/check/config/filter_pylint.txt
View File

@ -176,7 +176,7 @@
"mindspore/tests/st/networks/test_gpu_lstm.py" "redefined-outer-name"
"mindspore/tests/st/networks/test_gpu_lstm.py" "superfluous-parens"
"mindspore/tests/st/networks/test_gpu_alexnet.py" "unused-variable"
"mindspore/tests/st/networks/test_gpu_lenet.py" "unused-variable"
"mindspore/tests/st/networks/test_lenet.py" "unused-variable"
"mindspore/tests/st/ops/custom_ops_tbe/cus_add3.py" "unused-import"
"mindspore/tests/st/control/inner/test_002_single_for.py" "not-an-iterable"
"mindspore/tests/st/ops/ascend/test_aicpu_ops/test_strided_slice.py" "redefined-outer-name"

2
mindspore/ccsrc/pipeline/jit/action.cc
View File

@ -1205,7 +1205,7 @@ void SetRunMode(const FuncGraphPtr &func_graph, compile::Backend *backend_ptr) {
// GRAPH | normal network and if/for/switch scenario etc : MultiGraph path in MindRT.
MS_LOG(INFO) << "Run graph mode with multi graph sink.";
set_ctx(true, true, true);
set_ctx(true, true, !pynative_mode);
return;
}

2
tests/st/dataset/test_pynative_lenet_gpu_2p.py
View File

@ -25,5 +25,5 @@ def test_pynative_lenet_gpu_2p_mpi():
Description: test PyNative LeNet 2P with mpirun
Expectation: success, return_code==0
"""
return_code = os.system("mpirun -n 2 pytest -s test_gpu_lenet.py::test_train_net_fade_then_sink")
return_code = os.system("mpirun -n 2 pytest -s test_lenet.py::test_train_net_fade_then_sink")
assert return_code == 0

2
tests/st/networks/test_gpu_lenet_enable_rdr.py
View File

@ -5,7 +5,7 @@ import pytest
import mindspore.context as context
from tests.security_utils import security_off_wrap
from .test_gpu_lenet import test_train_lenet, test_train_and_eval_lenet
from .test_lenet import test_train_lenet, test_train_and_eval_lenet
# create config file for RDR
def create_config_file(path):

74
tests/st/networks/test_gpu_lenet.py → tests/st/networks/test_lenet.py
View File

@ -31,10 +31,11 @@ from mindspore.train import Accuracy
from mindspore.nn.optim import Momentum
from mindspore.ops import operations as P
from mindspore.ops import functional as F
from mindspore.train import Model, LossMonitor
from mindspore.train import Model, LossMonitor, Callback
from mindspore.common.initializer import TruncatedNormal
context.set_context(mode=context.GRAPH_MODE, device_target="GPU")
from mindspore.train.callback._callback import _handle_loss
from mindspore.common import JitConfig
from mindspore.common import set_seed
def conv(in_channels, out_channels, kernel_size, stride=1, padding=0):
@ -141,7 +142,7 @@ def test_train_lenet():
train_network = TrainOneStepCell(net_with_criterion, optimizer) # optimizer
train_network.set_train()
losses = []
for i in range(epoch):
for _ in range(epoch):
data = Tensor(np.ones([net.batch_size, 3, 32, 32]).astype(np.float32) * 0.01)
label = Tensor(np.ones([net.batch_size]).astype(np.int32))
loss = train_network(data, label).asnumpy()
@ -170,7 +171,7 @@ def test_train_lenet_memory_opt():
train_network = TrainOneStepCell(net_with_criterion, optimizer) # optimizer
train_network.set_train()
losses = []
for i in range(epoch):
for _ in range(epoch):
data = Tensor(np.ones([net.batch_size, 3, 32, 32]).astype(np.float32) * 0.01)
label = Tensor(np.ones([net.batch_size]).astype(np.int32))
loss = train_network(data, label).asnumpy()
@ -178,13 +179,16 @@ def test_train_lenet_memory_opt():
assert losses[-1] < 0.01
def create_dataset(data_path, batch_size=32, repeat_size=1,
num_parallel_workers=1):
def create_dataset(data_path, batch_size=32, repeat_size=1, num_parallel_workers=1, num_samples=None,
enable_shuffle=True):
"""
create dataset for train or test
"""
# define dataset
mnist_ds = ds.MnistDataset(data_path)
if num_samples is None:
mnist_ds = ds.MnistDataset(data_path)
else:
mnist_ds = ds.MnistDataset(data_path, num_samples=num_samples)
resize_height, resize_width = 32, 32
rescale = 1.0 / 255.0
@ -208,7 +212,8 @@ def create_dataset(data_path, batch_size=32, repeat_size=1,
# apply DatasetOps
buffer_size = 10000
mnist_ds = mnist_ds.shuffle(buffer_size=buffer_size) # 10000 as in LeNet train script
if enable_shuffle:
mnist_ds = mnist_ds.shuffle(buffer_size=buffer_size) # 10000 as in LeNet train script
mnist_ds = mnist_ds.batch(batch_size, drop_remainder=True)
mnist_ds = mnist_ds.repeat(repeat_size)
@ -250,7 +255,7 @@ def test_train_lenet_with_new_interface(num_classes=10, epoch=20, batch_size=32)
train_network = ForwardValueAndGrad(network=net_with_criterion, weights=weights, get_by_list=True, sens_param=True)
losses = []
for i in range(0, epoch):
for _ in range(0, epoch):
data = Tensor(np.ones([batch_size, 1, 32, 32]).astype(np.float32) * 0.01)
label = Tensor(np.ones([batch_size]).astype(np.int32))
sens = Tensor(np.ones([1]).astype(np.float32))
@ -282,7 +287,7 @@ def test_train_lenet_with_new_interface_tuple(num_classes=10, epoch=20, batch_si
train_network = ForwardValueAndGrad(network=net_with_criterion, weights=weights, get_by_list=True, sens_param=True)
losses = []
for i in range(0, epoch):
for _ in range(0, epoch):
data = Tensor(np.ones([batch_size, 1, 32, 32]).astype(np.float32) * 0.01)
label = Tensor(np.ones([batch_size]).astype(np.int32))
sens = Tensor(np.ones([1]).astype(np.float32))
@ -314,7 +319,7 @@ def test_train_lenet_with_new_interface_list(num_classes=10, epoch=20, batch_siz
train_network = ForwardValueAndGrad(network=net_with_criterion, weights=weights, get_by_list=True, sens_param=True)
losses = []
for i in range(0, epoch):
for _ in range(0, epoch):
data = Tensor(np.ones([batch_size, 1, 32, 32]).astype(np.float32) * 0.01)
label = Tensor(np.ones([batch_size]).astype(np.int32))
sens = Tensor(np.ones([1]).astype(np.float32))
@ -324,3 +329,48 @@ def test_train_lenet_with_new_interface_list(num_classes=10, epoch=20, batch_siz
losses.append(loss)
assert losses[-1].asnumpy() < 0.01
assert losses[-1].asnumpy() > 0.001
class CustomCallback(Callback):
"""
Assert the loss in every step_end.
"""
def __init__(self):
super(CustomCallback, self).__init__()
self.except_loss = [2.300181, 2.305321, 2.304834, 2.297870, 2.303635]
def step_end(self, run_context):
"""
Print training loss at the end of step.
Args:
run_context (RunContext): Include some information of the model. For more details,
please refer to :class:`mindspore.train.RunContext`.
"""
cb_params = run_context.original_args()
loss = _handle_loss(cb_params.net_outputs)
print("loss is ", loss, flush=True)
assert np.allclose(self.except_loss[cb_params.cur_step_num - 1], loss)
@pytest.mark.level0
@pytest.mark.platform_arm_ascend_training
@pytest.mark.env_onecard
def test_pynative_tasksink():
"""
Feature: PyNative Task Sink
Description: Enable PyNative Task Sink by setting JitConfig(jit_level="O2").
Expectation: No exception.
"""
set_seed(1)
np.random.seed(1)
context.set_context(mode=context.PYNATIVE_MODE, device_target="Ascend")
ds_train = create_dataset(os.path.join("/home/workspace/mindspore_dataset/mnist", "train"), 32, 1, 1, 160,
False)
network = LeNet5(10)
network.set_jit_config(JitConfig(jit_level="O2"))
net_loss = nn.SoftmaxCrossEntropyWithLogits(sparse=True, reduction="mean")
net_opt = nn.Momentum(network.trainable_params(), 0.01, 0.9)
model = Model(network, net_loss, net_opt, metrics={"Accuracy": Accuracy()})
model.train(1, ds_train, callbacks=[CustomCallback()], dataset_sink_mode=True)