wehub-resource-sync
223 lines
7.7 KiB
Python
223 lines
7.7 KiB
Python
# Copyright (c) 2023 PaddlePaddle Authors. All Rights Reserved.
|
|
#
|
|
# Licensed under the Apache License, Version 2.0 (the "License");
|
|
# you may not use this file except in compliance with the License.
|
|
# You may obtain a copy of the License at
|
|
#
|
|
# http://www.apache.org/licenses/LICENSE-2.0
|
|
#
|
|
# Unless required by applicable law or agreed to in writing, software
|
|
# distributed under the License is distributed on an "AS IS" BASIS,
|
|
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
# See the License for the specific language governing permissions and
|
|
# limitations under the License.
|
|
|
|
import os
|
|
|
|
import numpy as np
|
|
|
|
import paddle
|
|
import paddle.distributed as dist
|
|
from paddle.framework import core
|
|
|
|
|
|
class TestReshardNdMesh:
|
|
def __init__(self):
|
|
self._shape = eval(os.getenv("shape"))
|
|
self._dtype = os.getenv("dtype")
|
|
self._seeds = eval(os.getenv("seeds"))
|
|
self._backend = os.getenv("backend")
|
|
self._mesh = dist.ProcessMesh([[0], [1]], dim_names=["x", "y"])
|
|
self._other_mesh = dist.ProcessMesh([[1], [0]], dim_names=["x", "y"])
|
|
|
|
def test_shard_partial_to_shard_replicated(self, dev_ctx):
|
|
paddle.seed(self._seeds)
|
|
value = paddle.uniform(self._shape, self._dtype)
|
|
|
|
input_tensor = dist.shard_tensor(
|
|
value, self._mesh, [dist.Partial(), dist.Shard(0)]
|
|
)
|
|
|
|
# check the shape of input tensor
|
|
in_expected_shape = list(self._shape)
|
|
in_expected_shape[0] = in_expected_shape[0] // self._mesh.shape[1]
|
|
assert np.equal(input_tensor._local_shape, in_expected_shape).all()
|
|
|
|
# check the value of input tensor
|
|
in_expected_local_tensor_list = paddle.split(
|
|
value, num_or_sections=self._mesh.shape[1], axis=0
|
|
)
|
|
index = dist.get_rank() % self._mesh.shape[1]
|
|
if dist.get_rank() // self._mesh.shape[1] == 0:
|
|
np.testing.assert_equal(
|
|
input_tensor._local_value().numpy(),
|
|
in_expected_local_tensor_list[index].numpy(),
|
|
)
|
|
else:
|
|
zeros = paddle.zeros(in_expected_shape)
|
|
np.testing.assert_equal(
|
|
input_tensor._local_value().numpy(), zeros.numpy()
|
|
)
|
|
|
|
out = dist.reshard(
|
|
input_tensor, self._mesh, [dist.Replicate(), dist.Shard(0)]
|
|
)
|
|
|
|
np.testing.assert_equal(
|
|
out._local_value().numpy(),
|
|
in_expected_local_tensor_list[index].numpy(),
|
|
)
|
|
|
|
def test_shard_partial_to_replicated(self, dev_ctx):
|
|
paddle.seed(self._seeds)
|
|
value = paddle.uniform(self._shape, self._dtype)
|
|
|
|
input_tensor = dist.shard_tensor(
|
|
value, self._mesh, [dist.Partial(), dist.Shard(0)]
|
|
)
|
|
|
|
# check the shape of input tensor
|
|
in_expected_shape = list(self._shape)
|
|
in_expected_shape[0] = in_expected_shape[0] // self._mesh.shape[1]
|
|
assert np.equal(input_tensor._local_shape, in_expected_shape).all()
|
|
|
|
# check the value of input tensor
|
|
in_expected_local_tensor_list = paddle.split(
|
|
value, num_or_sections=self._mesh.shape[1], axis=0
|
|
)
|
|
index = dist.get_rank() % self._mesh.shape[1]
|
|
if dist.get_rank() // self._mesh.shape[1] == 0:
|
|
np.testing.assert_equal(
|
|
input_tensor._local_value().numpy(),
|
|
in_expected_local_tensor_list[index].numpy(),
|
|
)
|
|
else:
|
|
zeros = paddle.zeros(in_expected_shape)
|
|
np.testing.assert_equal(
|
|
input_tensor._local_value().numpy(), zeros.numpy()
|
|
)
|
|
|
|
out = dist.reshard(
|
|
input_tensor, self._mesh, [dist.Replicate(), dist.Replicate()]
|
|
)
|
|
|
|
np.testing.assert_equal(out._local_value().numpy(), value.numpy())
|
|
|
|
def test_partial_to_partial(self, dev_ctx):
|
|
a = paddle.ones(self._shape)
|
|
|
|
input_tensor = dist.shard_tensor(
|
|
a, self._mesh, [dist.Partial(), dist.Replicate()]
|
|
)
|
|
|
|
if dist.get_rank() // self._mesh.shape[1] == 0:
|
|
np.testing.assert_equal(
|
|
input_tensor._local_value().numpy(), a.numpy()
|
|
)
|
|
else:
|
|
zeros = paddle.zeros(self._shape)
|
|
np.testing.assert_equal(
|
|
input_tensor._local_value().numpy(), zeros.numpy()
|
|
)
|
|
|
|
out = dist.reshard(
|
|
input_tensor, self._mesh, [dist.Replicate(), dist.Partial()]
|
|
)
|
|
|
|
if dist.get_rank() % self._mesh.shape[1] == 0:
|
|
np.testing.assert_equal(out._local_value().numpy(), a.numpy())
|
|
else:
|
|
zeros = paddle.zeros(self._shape)
|
|
np.testing.assert_equal(out._local_value().numpy(), zeros.numpy())
|
|
|
|
assert np.equal(out.shape, input_tensor.shape).all()
|
|
assert np.equal(out._local_shape, input_tensor._local_shape).all()
|
|
|
|
def test_shard_to_shard(self, dev_ctx):
|
|
a = paddle.ones(self._shape)
|
|
|
|
in_shard_specs = [None for i in range(len(self._shape))]
|
|
in_shard_specs[1] = "y"
|
|
|
|
out_shard_specs = [None for i in range(len(self._shape))]
|
|
out_shard_specs[0] = "x"
|
|
|
|
input_tensor = dist.shard_tensor(
|
|
a, self._mesh, [dist.Replicate(), dist.Shard(1)]
|
|
)
|
|
|
|
in_expected_shape = list(self._shape)
|
|
in_expected_shape[1] = in_expected_shape[1] // self._mesh.shape[1]
|
|
assert np.equal(input_tensor._local_shape, in_expected_shape).all()
|
|
|
|
out = dist.reshard(
|
|
input_tensor, self._mesh, [dist.Shard(0), dist.Replicate()]
|
|
)
|
|
|
|
out_expected_shape = list(self._shape)
|
|
out_expected_shape[0] = out_expected_shape[0] // self._mesh.shape[0]
|
|
assert np.equal(input_tensor._local_shape, in_expected_shape).all()
|
|
|
|
assert np.equal(out.shape, input_tensor.shape).all()
|
|
|
|
def test_partial_replicate_to_shard_replicated(self, dev_ctx):
|
|
paddle.seed(self._seeds)
|
|
a = paddle.randn(self._shape).astype(self._dtype)
|
|
|
|
input_tensor = dist.shard_tensor(
|
|
a, self._mesh, [dist.Partial(), dist.Replicate()]
|
|
)
|
|
out = dist.reshard(
|
|
input_tensor, self._mesh, [dist.Shard(0), dist.Replicate()]
|
|
)
|
|
|
|
# check the value of input tensor
|
|
out_expected_local_tensor_list = paddle.split(
|
|
a, num_or_sections=self._mesh.shape[0], axis=0
|
|
)
|
|
index = dist.get_rank() % self._mesh.shape[0]
|
|
np.testing.assert_equal(
|
|
out._local_value().numpy(),
|
|
out_expected_local_tensor_list[index].numpy(),
|
|
)
|
|
assert np.equal(out.shape, input_tensor.shape).all()
|
|
|
|
def same_mesh_reshard(self):
|
|
if self._backend == "cpu":
|
|
paddle.set_device("cpu")
|
|
place = paddle.CPUPlace()
|
|
elif self._backend == "gpu":
|
|
place = paddle.CUDAPlace(dist.get_rank())
|
|
|
|
dev_ctx = core.DeviceContext.create(place)
|
|
|
|
self.test_partial_to_partial(dev_ctx)
|
|
self.test_shard_to_shard(dev_ctx)
|
|
self.test_shard_partial_to_shard_replicated(dev_ctx)
|
|
self.test_shard_partial_to_replicated(dev_ctx)
|
|
|
|
if self._backend == "gpu":
|
|
# reduce_scatter is not supported on CPU
|
|
self.test_partial_replicate_to_shard_replicated(dev_ctx)
|
|
|
|
def cross_mesh_reshard(self):
|
|
a = paddle.zeros([20, 20])
|
|
a = dist.shard_tensor(
|
|
a,
|
|
self._mesh,
|
|
[
|
|
dist.Partial(dist.ReduceType.kRedSum),
|
|
dist.Partial(dist.ReduceType.kRedSum),
|
|
],
|
|
)
|
|
dist.reshard(a, self._other_mesh, [dist.Shard(0), dist.Shard(1)])
|
|
|
|
def run_test_case(self):
|
|
self.same_mesh_reshard()
|
|
if self._backend == "gpu":
|
|
self.cross_mesh_reshard()
|
|
|
|
|
|
if __name__ == '__main__':
|
|
TestReshardNdMesh().run_test_case()
|