wehub-resource-sync
128 lines
4.1 KiB
Python
128 lines
4.1 KiB
Python
# Copyright (c) 2024 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 unittest
|
|
|
|
import paddle
|
|
import paddle.distributed as dist
|
|
from paddle.distributed.auto_parallel.static.mix_to_dist_pass import (
|
|
apply_mix2dist_pass,
|
|
)
|
|
|
|
paddle.enable_static()
|
|
|
|
|
|
class TestDistReshape(unittest.TestCase):
|
|
def build_program(
|
|
self,
|
|
src_shape,
|
|
dst_shape,
|
|
src_mesh,
|
|
dst_mesh,
|
|
src_placements,
|
|
dst_placements,
|
|
):
|
|
main_program = paddle.base.Program()
|
|
with paddle.base.program_guard(main_program):
|
|
x = paddle.static.data(name='x', shape=src_shape)
|
|
x.stop_gradient = False
|
|
labels = paddle.static.data(name='labels', shape=dst_shape)
|
|
dist_x = dist.shard_tensor(x, src_mesh, src_placements)
|
|
dist_labels = dist.shard_tensor(labels, dst_mesh, dst_placements)
|
|
|
|
dist_y = dist.auto_parallel.moe_utils._dist_reshape(
|
|
dist_x, dst_shape, dst_mesh, dst_placements
|
|
)
|
|
loss = dist_y - dist_labels
|
|
|
|
dist_program = main_program.clone()
|
|
apply_mix2dist_pass(dist_program)
|
|
dist_loss_value = dist_program.global_block().ops[-1].result(0)
|
|
|
|
with paddle.static.program_guard(dist_program):
|
|
params_grads = paddle.autograd.ir_backward.append_backward(
|
|
dist_loss_value
|
|
)
|
|
|
|
return dist_program
|
|
|
|
def check_placements(self, fwd_op, bwd_op, x_placements, out_placements):
|
|
assert fwd_op.name() == "dist_op.dist_reshape"
|
|
assert bwd_op.name() == "dist_op.dist_reshape"
|
|
|
|
out = fwd_op.result(0)
|
|
assert out.dist_attr().placements == out_placements
|
|
x_grad = bwd_op.result(0)
|
|
assert x_grad.dist_attr().placements == x_placements
|
|
|
|
def test_case0(self):
|
|
src_shape = [64, 32]
|
|
dst_shape = [32, 64]
|
|
mesh = dist.ProcessMesh([[0, 1], [2, 3]])
|
|
x_placements = [dist.Shard(1), dist.Replicate()]
|
|
out_placements = [dist.Shard(1), dist.Replicate()]
|
|
dist_program = self.build_program(
|
|
src_shape, dst_shape, mesh, mesh, x_placements, out_placements
|
|
)
|
|
|
|
ops = dist_program.global_block().ops
|
|
fwd_op = ops[2]
|
|
bwd_op = ops[-1]
|
|
self.check_placements(fwd_op, bwd_op, x_placements, out_placements)
|
|
|
|
x = ops[0].result(0)
|
|
assert x.dist_attr().placements_attr == x_placements
|
|
|
|
out = fwd_op.result(0)
|
|
assert out.shape == dst_shape
|
|
assert out._local_shape == [32, 32]
|
|
|
|
x_grad = bwd_op.result(0)
|
|
assert x_grad.shape == src_shape
|
|
assert x_grad._local_shape == [64, 16]
|
|
|
|
def test_shard_on_multi_dim(self):
|
|
src_shape = [2, 64, 32]
|
|
dst_shape = [-1, 32]
|
|
src_mesh = dist.ProcessMesh([[0, 1], [2, 3]])
|
|
x_placements = [dist.Shard(0), dist.Shard(1)]
|
|
dst_mesh = dist.ProcessMesh([0, 1, 2, 3])
|
|
dst_placements = [dist.Shard(0)]
|
|
|
|
dist_program = self.build_program(
|
|
src_shape,
|
|
dst_shape,
|
|
src_mesh,
|
|
dst_mesh,
|
|
x_placements,
|
|
dst_placements,
|
|
)
|
|
|
|
ops = dist_program.global_block().ops
|
|
fwd_op = ops[2]
|
|
bwd_op = ops[-1]
|
|
self.check_placements(fwd_op, bwd_op, x_placements, dst_placements)
|
|
|
|
out = fwd_op.result(0)
|
|
assert out.shape == [128, 32]
|
|
assert out._local_shape == [32, 32]
|
|
|
|
x_grad = bwd_op.result(0)
|
|
assert x_grad.shape == src_shape
|
|
assert x_grad._local_shape == [1, 32, 32]
|
|
|
|
|
|
if __name__ == "__main__":
|
|
unittest.main()
|