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
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
|
# Owner(s): ["oncall: distributed"]
import os
import sys
import torch
import torch.distributed as dist
import torch.distributed.checkpoint as dist_cp
import torch.nn as nn
from torch.distributed._tensor import init_device_mesh
from torch.distributed.checkpoint.state_dict import (
get_model_state_dict,
get_state_dict,
set_model_state_dict,
set_state_dict,
StateDictOptions,
)
from torch.distributed.fsdp import FullyShardedDataParallel as FSDP
from torch.testing._internal.common_distributed import skip_if_lt_x_gpu
from torch.testing._internal.common_utils import run_tests, TEST_WITH_DEV_DBG_ASAN
from torch.testing._internal.distributed._tensor.common_dtensor import (
DTensorTestBase,
with_comms,
)
from torch.testing._internal.distributed.checkpoint_utils import with_temp_dir
if not dist.is_available():
print("Distributed not available, skipping tests", file=sys.stderr)
sys.exit(0)
if TEST_WITH_DEV_DBG_ASAN:
print(
"Skip dev-asan as torch + multiprocessing spawn have known issues",
file=sys.stderr,
)
sys.exit(0)
DIM = 500
class PreTrainedModel(nn.Module):
def __init__(self) -> None:
super().__init__()
self.layer1 = nn.Linear(DIM, DIM)
self.layer2 = nn.Linear(DIM, DIM)
self.layer3 = nn.Linear(DIM, DIM)
self.sequential = nn.Sequential(nn.Linear(DIM, DIM), nn.ReLU())
self.module_list = nn.ModuleList([nn.Linear(DIM, DIM), nn.ReLU()])
self.relu = nn.ReLU()
def forward(self, batch):
x = self.relu(self.layer1(batch))
x = self.relu(self.layer2(x))
x = self.relu(self.layer3(x))
x = self.sequential(x)
x = self.module_list[1](self.module_list[0](x))
return x
class FineTuningModel(nn.Module):
def __init__(self) -> None:
super().__init__()
self.pretrain = PreTrainedModel()
for p in self.pretrain.parameters():
p.requires_grad = False
self.layer1 = nn.Linear(DIM, DIM)
self.layer2 = nn.Linear(DIM, DIM)
self.layer3 = nn.Linear(DIM, DIM)
self.relu = nn.ReLU()
def forward(self, batch):
x = self.relu(self.pretrain(batch))
x = self.relu(self.layer1(x))
x = self.relu(self.layer2(x))
x = self.relu(self.layer3(x))
return x
class TestFineTuning(DTensorTestBase):
@property
def world_size(self) -> int:
return min(4, torch.cuda.device_count())
@property
def backend(self):
return "cpu:gloo,cuda:nccl"
def pretrain(self, pretrain_dir: str) -> None:
device_mesh = init_device_mesh(self.device_type, (self.world_size,))
model = PreTrainedModel().cuda()
model = FSDP(model, device_mesh=device_mesh)
optim = torch.optim.Adam(model.parameters(), lr=1e-3)
# Training
for i in range(3):
batch = torch.rand(32, DIM, device="cuda")
loss = model(batch).sum()
loss.backward()
optim.step()
optim.zero_grad()
# Save state_dict
model_state_dict, optim_state_dict = get_state_dict(model, optimizers=optim)
saved_state_dict = {"model": model_state_dict, "optim": optim_state_dict}
dist_cp.save(
state_dict=saved_state_dict,
storage_writer=dist_cp.FileSystemWriter(pretrain_dir),
)
def finetune(self, pretrain_dir: str, finetune_dir: str) -> None:
device_mesh = init_device_mesh(self.device_type, (self.world_size,))
model = FineTuningModel().cuda()
# TODO: make the parallelism more complicated, e.g., using 2D + DDP.
model = FSDP(model, use_orig_params=True, device_mesh=device_mesh)
optim = torch.optim.Adam(model.parameters(), lr=1e-3)
# Simulate that the fine tuning restart after 3 iterations
for i in range(2):
# Load pretrain submodules checkpoint
pretrain_state_dict = get_model_state_dict(
model,
submodules={model.pretrain},
options=StateDictOptions(keep_submodule_prefixes=False),
)
dist_cp.load(
{"model": pretrain_state_dict},
storage_reader=dist_cp.FileSystemReader(pretrain_dir),
)
set_model_state_dict(
model,
model_state_dict={model.pretrain: pretrain_state_dict},
options=StateDictOptions(strict=False),
)
try:
# Load training submodules checkpoint
model_state_dict, optim_state_dict = get_state_dict(
model,
optimizers=optim,
options=StateDictOptions(ignore_frozen_params=True),
)
dist_cp.load_state_dict(
{"model": model_state_dict, "optim": optim_state_dict},
storage_reader=dist_cp.FileSystemReader(pretrain_dir),
)
set_state_dict(
model,
optimizers=optim,
model_state_dict=model_state_dict,
optim_state_dict=optim_state_dict,
options=StateDictOptions(strict=False),
)
except KeyError:
# If this is the first round of the fine tuning, then nothing is saved.
# If this is the restart of the fine tuning, then checkpoint should exit.
self.assertEqual(i, 0)
# Training
for j in range(3):
batch = torch.rand(32, DIM, device="cuda")
loss = model(batch).sum()
loss.backward()
optim.step()
optim.zero_grad()
# Save state_dict
model_state_dict, optim_state_dict = get_state_dict(
model,
optimizers=optim,
options=StateDictOptions(ignore_frozen_params=True),
)
saved_state_dict = {"model": model_state_dict, "optim": optim_state_dict}
dist_cp.save(
state_dict=saved_state_dict,
storage_writer=dist_cp.FileSystemWriter(finetune_dir),
)
@skip_if_lt_x_gpu(4)
@with_comms
@with_temp_dir
def test_fine_tuning(self) -> None:
self.assertTrue(os.path.exists(self.temp_dir))
pretrain_dir = os.path.join(self.temp_dir, "pretrain")
finetune_dir = os.path.join(self.temp_dir, "finetune")
print(pretrain_dir, finetune_dir)
if self.rank == 0:
os.mkdir(pretrain_dir)
os.mkdir(finetune_dir)
dist.barrier()
os.sync()
self.assertTrue(os.path.exists(pretrain_dir))
self.assertTrue(os.path.exists(finetune_dir))
self.pretrain(pretrain_dir)
self.finetune(pretrain_dir, finetune_dir)
if __name__ == "__main__":
run_tests()
|
