wehub-resource-sync
151 lines
5.3 KiB
Python
151 lines
5.3 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()
|
|
|
|
|
|
def get_program(mesh, placements, local_mesh_dim):
|
|
main_program = paddle.base.Program()
|
|
with paddle.base.program_guard(main_program):
|
|
x = paddle.static.data(name='x', shape=[64, 36, 24])
|
|
y = paddle.static.data(name='y', shape=[64, 36, 24])
|
|
x.stop_gradient = False
|
|
y.stop_gradient = False
|
|
dist_x = dist.shard_tensor(x, mesh, placements)
|
|
dist_y = dist.shard_tensor(y, mesh, placements)
|
|
local_tensors = dist.auto_parallel.api.moe_sub_mesh_tensors(
|
|
dist_x, mesh, local_mesh_dim, placements
|
|
)
|
|
out = dist.auto_parallel.api.moe_global_mesh_tensor(
|
|
local_tensors, mesh, placements, local_mesh_dim
|
|
)
|
|
loss = dist_y - out
|
|
|
|
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
|
|
|
|
|
|
class TestMoEApi(unittest.TestCase):
|
|
def test_1Dmesh_2experts(self):
|
|
mesh = dist.ProcessMesh([0, 1])
|
|
global_placements = [dist.Shard(0)]
|
|
local_mesh_dim = 0
|
|
dist_program = get_program(mesh, global_placements, local_mesh_dim)
|
|
ops = dist_program.global_block().ops
|
|
|
|
global_mesh = [0, 1]
|
|
global_dims_mapping = [0, -1, -1]
|
|
local_meshes = [[0], [1]]
|
|
local_dims_mapping = [-1, -1, -1]
|
|
self.check_results(
|
|
ops,
|
|
global_mesh,
|
|
global_dims_mapping,
|
|
local_meshes,
|
|
local_dims_mapping,
|
|
)
|
|
|
|
def test_2Dmesh_4experts(self):
|
|
mesh = dist.ProcessMesh([[0, 1], [2, 3], [4, 5], [6, 7]])
|
|
global_placements = [dist.Shard(0), dist.Shard(2)]
|
|
local_mesh_dim = -2
|
|
dist_program = get_program(mesh, global_placements, local_mesh_dim)
|
|
ops = dist_program.global_block().ops
|
|
|
|
global_mesh = [0, 1, 2, 3, 4, 5, 6, 7]
|
|
local_meshes = [[0, 1], [2, 3], [4, 5], [6, 7]]
|
|
global_dims_mapping = [0, -1, 1]
|
|
local_dims_mapping = [-1, -1, 1]
|
|
self.check_results(
|
|
ops,
|
|
global_mesh,
|
|
global_dims_mapping,
|
|
local_meshes,
|
|
local_dims_mapping,
|
|
)
|
|
|
|
def test_error(self):
|
|
mesh = dist.ProcessMesh([[0, 1], [2, 3], [4, 5], [6, 7]])
|
|
global_placements = [dist.Shard(0), dist.Shard(2)]
|
|
local_mesh_dim = -3
|
|
with self.assertRaises(ValueError):
|
|
dist_program = get_program(mesh, global_placements, local_mesh_dim)
|
|
|
|
with self.assertRaises(ValueError):
|
|
main_program = paddle.base.Program()
|
|
with paddle.base.program_guard(main_program):
|
|
x = paddle.static.data(name='x', shape=[64, 36, 24])
|
|
y = paddle.static.data(name='y', shape=[64, 36, 24])
|
|
x.stop_gradient = False
|
|
y.stop_gradient = False
|
|
dist_x = dist.shard_tensor(x, mesh, global_placements)
|
|
dist_y = dist.shard_tensor(y, mesh, global_placements)
|
|
local_tensors = dist.auto_parallel.api.moe_sub_mesh_tensors(
|
|
dist_x, None, local_mesh_dim, global_placements
|
|
)
|
|
|
|
def check_dist_attr(self, op, meshes, dims_mapping):
|
|
results = op.results()
|
|
self.assertEqual(len(results), len(meshes))
|
|
for i, result in enumerate(results):
|
|
dist_attr = result.dist_attr()
|
|
self.assertEqual(dist_attr.process_mesh.process_ids, meshes[i])
|
|
self.assertEqual(dist_attr.dims_mapping, dims_mapping)
|
|
|
|
def check_results(
|
|
self,
|
|
ops,
|
|
global_mesh,
|
|
global_dims_mapping,
|
|
local_meshes,
|
|
local_dims_mapping,
|
|
):
|
|
op_names = [
|
|
"dist_op.moe_sub_mesh_tensors",
|
|
"dist_op.moe_global_mesh_tensor",
|
|
]
|
|
ops_to_check = [op for op in ops if op.name() in op_names]
|
|
# moe_sub_mesh_tensors op
|
|
self.check_dist_attr(ops_to_check[0], local_meshes, local_dims_mapping)
|
|
# moe_global_mesh_tensor op
|
|
self.check_dist_attr(
|
|
ops_to_check[1], [global_mesh], global_dims_mapping
|
|
)
|
|
# grad op for moe_global_mesh_tensor
|
|
self.check_dist_attr(ops_to_check[2], local_meshes, local_dims_mapping)
|
|
# grad op for moe_sub_mesh_tensors op
|
|
self.check_dist_attr(
|
|
ops_to_check[3], [global_mesh], global_dims_mapping
|
|
)
|
|
|
|
|
|
if __name__ == "__main__":
|
|
unittest.main()
|