1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
|
# Owner(s): ["oncall: distributed"]
from collections import defaultdict
from typing import Dict
import torch
from torch.distributed._tensor.experimental._tp_transform import (
tensor_parallel_transformation,
)
from torch.distributed.tensor.parallel.style import (
ColwiseParallel,
ParallelStyle,
RowwiseParallel,
)
from torch.testing._internal.common_utils import run_tests
from torch.testing._internal.distributed._tensor.common_dtensor import (
DTensorTestBase,
with_comms,
)
class MLPListModule(torch.nn.Module):
"""
A dummy model with list of MLPs.
"""
def __init__(self, num_mlps=3, bias=True):
super().__init__()
self.mlps = torch.nn.ModuleList()
for _ in range(num_mlps):
self.mlps.append(
torch.nn.Sequential(
torch.nn.Linear(6, 18),
torch.nn.ReLU(),
torch.nn.Linear(18, 6, bias=bias),
)
)
def forward(self, x: torch.Tensor) -> torch.Tensor:
x = torch.chunk(x, 2, dim=1)[0]
for mlp in self.mlps:
x = mlp(x)
return x + torch.ones_like(x)
class DummyModel(torch.nn.Module):
def __init__(self) -> None:
super().__init__()
self.fc = torch.nn.Linear(3, 5)
self.bn = torch.nn.BatchNorm1d(5)
def forward(self, x):
return self.bn(self.fc(x))
class TensorParallelTest(DTensorTestBase):
def setUp(self) -> None:
super().setUp()
def assert_has_c10d_ops(
self, gm: torch.fx.GraphModule, expected_ops_count: Dict[str, int]
) -> None:
actual_ops_count: Dict[str, int] = defaultdict(int)
for node in gm.graph.nodes:
if node.op == "call_function":
if "c10d_functional" in str(node.target):
actual_ops_count[str(node.target)] += 1
self.assertDictEqual(expected_ops_count, actual_ops_count)
@with_comms
def test_tp_transform_with_uncovered_op(self):
model = DummyModel().to(device=self.device_type)
inputs = (torch.randn(7, 3, requires_grad=False).to(device=self.device_type),)
with torch.no_grad():
res = model(*inputs)
exported_program = torch.export.export(
model,
inputs,
).run_decompositions()
tp_exported_program = tensor_parallel_transformation(
exported_program,
self.rank,
self.world_size,
self.device_type,
{"fc": ColwiseParallel},
)
tp_model = tp_exported_program.module()
with torch.no_grad():
tp_res = tp_model(*inputs)
self.assertEqual(res, tp_res)
# Expect all_gather to be inserted to distributed sharded fc resutls
self.assert_has_c10d_ops(
tp_exported_program.graph_module,
{
"_c10d_functional.all_gather_into_tensor.default": 1,
"_c10d_functional.wait_tensor.default": 1,
},
)
@with_comms
def test_tp_transform_e2e(self):
torch.manual_seed(0)
model = MLPListModule(2).to(device=self.device_type)
inputs = (torch.randn((10, 12)).to(device=self.device_type),)
parallel_strategies: Dict[str, ParallelStyle] = {
"mlps.0.0": ColwiseParallel,
"mlps.0.2": RowwiseParallel,
"mlps.1.0": ColwiseParallel,
"mlps.1.2": RowwiseParallel,
}
with torch.inference_mode():
res = model(*inputs)
exported_program = torch.export.export(
model,
inputs,
).run_decompositions()
tp_exported_program = tensor_parallel_transformation(
exported_program,
self.rank,
self.world_size,
self.device_type,
parallel_strategies,
)
tp_model = tp_exported_program.module()
with torch.inference_mode():
tp_res = tp_model(*inputs)
self.assertEqual(res, tp_res)
# Expect all_reduce to be inserted at the end of each MLP
self.assert_has_c10d_ops(
tp_exported_program.graph_module,
{
"_c10d_functional.all_reduce.default": 2,
"_c10d_functional.wait_tensor.default": 2,
},
)
@with_comms
def test_tp_transform_no_bias(self):
torch.manual_seed(0)
model = MLPListModule(1, bias=False).to(device=self.device_type)
inputs = (torch.randn((10, 12)).to(device=self.device_type),)
parallel_strategies: Dict[str, ParallelStyle] = {
"mlps.0.0": ColwiseParallel,
"mlps.0.2": RowwiseParallel,
}
with torch.inference_mode():
res = model(*inputs)
exported_program = torch.export.export(
model,
inputs,
).run_decompositions()
tp_exported_program = tensor_parallel_transformation(
exported_program,
self.rank,
self.world_size,
self.device_type,
parallel_strategies,
)
tp_model = tp_exported_program.module()
with torch.inference_mode():
tp_res = tp_model(*inputs)
self.assertEqual(res, tp_res)
self.assert_has_c10d_ops(
tp_exported_program.graph_module,
{
"_c10d_functional.all_reduce.default": 1,
"_c10d_functional.wait_tensor.default": 1,
},
)
if __name__ == "__main__":
run_tests()
|
