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
|
# Copyright (c) Meta Platforms, Inc. and affiliates
# Owner(s): ["oncall: distributed"]
import copy
from model_registry import MLPModule
import torch
from torch.distributed.pipelining._backward import (
stage_backward,
stage_backward_input,
stage_backward_weight,
)
from torch.testing._internal.common_utils import run_tests, TestCase
d_hid = 512
batch_size = 256
class StageBackwardTests(TestCase):
def test_stage_backward(self):
# MLP as a stage module
mod = MLPModule(d_hid)
x = torch.randn(batch_size, d_hid)
# As in a pipeline stage, the inputs to this stage requires gradients
x.requires_grad_(True)
target = torch.randn(batch_size, d_hid)
loss_fn = torch.nn.MSELoss(reduction="sum")
# Make a copy
ref_mod = copy.deepcopy(mod)
ref_x = x.detach().requires_grad_(x.requires_grad)
ref_target = target.detach()
# Forward and backward in stage manner
out = mod(x)
loss = loss_fn(out, target)
grad_inputs = stage_backward(
stage_output=loss,
output_grads=None,
input_values=(x,),
)
# Run reference
ref_out = ref_mod(ref_x)
ref_loss = loss_fn(ref_out, ref_target)
ref_loss.backward()
torch.testing.assert_close(grad_inputs[0], ref_x.grad)
# Every rank checks gradients
for name, p in mod.named_parameters():
ref_p = ref_mod.get_parameter(name)
try:
torch.testing.assert_close(p.grad, ref_p.grad)
except AssertionError:
print(f"Gradient test failed for {name}: {p.grad} vs {ref_p.grad}")
raise
def test_stage_backward_input(self):
# MLP as a stage module
mod = MLPModule(d_hid)
x = torch.randn(batch_size, d_hid)
# As in a pipeline stage, the inputs to this stage requires gradients
x.requires_grad_(True)
target = torch.randn(batch_size, d_hid)
loss_fn = torch.nn.MSELoss(reduction="sum")
# Make a copy
ref_mod = copy.deepcopy(mod)
ref_x = x.detach().requires_grad_(x.requires_grad)
ref_target = target.detach()
# Forward, then backward of loss with respect to inputs
out = mod(x)
loss = loss_fn(out, target)
dinputs, param_groups = stage_backward_input(
stage_outputs_or_loss=(loss,),
output_grads=None,
input_values=[x],
weights=mod.parameters(),
)
# Run reference
ref_out = ref_mod(ref_x)
ref_loss = loss_fn(ref_out, ref_target)
ref_loss.backward()
torch.testing.assert_close(x.grad, ref_x.grad)
torch.testing.assert_close(dinputs[0], ref_x.grad)
for name, p in mod.named_parameters():
# Check that the weight gradients were not updated
self.assertEqual(p.grad, None)
def test_stage_backward_weight(self):
# MLP as a stage module
mod = MLPModule(d_hid)
x = torch.randn(batch_size, d_hid)
# As in a pipeline stage, the inputs to this stage requires gradients
x.requires_grad_(True)
target = torch.randn(batch_size, d_hid)
loss_fn = torch.nn.MSELoss(reduction="sum")
# Make a copy
ref_mod = copy.deepcopy(mod)
ref_x = x.detach().requires_grad_(x.requires_grad)
ref_target = target.detach()
# Forward, then backward of loss with respect to inputs
out = mod(x)
loss = loss_fn(out, target)
dinputs, param_groups = stage_backward_input(
stage_outputs_or_loss=(loss,),
output_grads=None,
input_values=[x],
weights=mod.parameters(),
)
# backward of loss with respect to weights
stage_backward_weight(mod.parameters(), param_groups, retain_graph=True)
# Run reference
ref_out = ref_mod(ref_x)
ref_loss = loss_fn(ref_out, ref_target)
ref_loss.backward()
# Every rank checks gradients
for name, p in mod.named_parameters():
ref_p = ref_mod.get_parameter(name)
try:
torch.testing.assert_close(p.grad, ref_p.grad)
except AssertionError:
print(f"Gradient test failed for {name}: {p.grad} vs {ref_p.grad}")
raise
def test_stage_backward_weight_multiple_iters(self):
# MLP as a stage module
mod = MLPModule(d_hid)
inputs = []
for _ in range(10):
x = torch.randn(batch_size, d_hid)
inputs.append(x)
# As in a pipeline stage, the inputs to this stage requires gradients
x.requires_grad_(True)
target = torch.randn(batch_size, d_hid)
loss_fn = torch.nn.MSELoss(reduction="sum")
# Make a copy
ref_mod = copy.deepcopy(mod)
ref_inputs = []
for x in inputs:
ref_inputs.append(x.detach().requires_grad_(x.requires_grad))
ref_target = target.detach()
# Forward, then backward of loss with respect to inputs
for x in inputs:
out = mod(x)
loss = loss_fn(out, target)
dinputs, param_groups = stage_backward_input(
stage_outputs_or_loss=(loss,),
output_grads=None,
input_values=[x],
weights=mod.parameters(),
)
# backward of loss with respect to weights
stage_backward_weight(mod.parameters(), param_groups)
# Run reference
for ref_x in ref_inputs:
ref_out = ref_mod(ref_x)
ref_loss = loss_fn(ref_out, ref_target)
ref_loss.backward()
# Every rank checks gradients
for name, p in mod.named_parameters():
ref_p = ref_mod.get_parameter(name)
try:
torch.testing.assert_close(p.grad, ref_p.grad)
except AssertionError:
print(f"Gradient test failed for {name}: {p.grad} vs {ref_p.grad}")
raise
if __name__ == "__main__":
run_tests()
|
