Skip to content

GitLab

Switch branch/tag
Find file Normal viewHistoryPermalink
test_other_optimizer.py 4.67 KB
Newer Older
aiss's avatar
push dsv0.8.2 version  
aiss committed
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 
'''Copyright The Microsoft DeepSpeed Team'''

import deepspeed
from deepspeed.ops.op_builder import FusedLambBuilder

from unit.common import DistributedTest
from unit.simple_model import *

from unit.checkpoint.common import checkpoint_correctness_verification

import pytest


class TestOtherOptimizerCheckpoint(DistributedTest):
    world_size = 2

    @pytest.mark.skipif(not deepspeed.ops.__compatible_ops__[FusedLambBuilder.NAME],
                        reason="lamb is not compatible")
    def test_checkpoint_unfused_optimizer(self, tmpdir):
        config_dict = {
            "train_batch_size": 2,
            "steps_per_print": 1,
            "optimizer": {
                "type": "Lamb",
                "params": {
                    "lr": 0.00015
                }
            },
            "gradient_clipping": 1.0,
            "fp16": {
                "enabled": True
            },
            "scheduler": {
                "type": "OneCycle",
                "params": {
                    "cycle_first_step_size": 1000,
                    "cycle_first_stair_count": 500,
                    "cycle_second_step_size": 1000,
                    "cycle_second_stair_count": 500,
                    "decay_step_size": 1000,
                    "cycle_min_lr": 0.0001,
                    "cycle_max_lr": 0.0010,
                    "decay_lr_rate": 0.001,
                    "cycle_min_mom": 0.85,
                    "cycle_max_mom": 0.99,
                    "decay_mom_rate": 0.0
                }
            }
        }

        args = args_from_dict(tmpdir, config_dict)
        hidden_dim = 10
        models = [SimpleModel(hidden_dim, empty_grad=False) for _ in range(2)]

        # Load & verify optimizer states
        checkpoint_correctness_verification(config_dict,
                                            models=models,
                                            hidden_dim=hidden_dim,
                                            tmpdir=tmpdir,
                                            load_optimizer_states=True)

        # Ignore optimizer states
        checkpoint_correctness_verification(config_dict,
                                            models=models,
                                            hidden_dim=hidden_dim,
                                            tmpdir=tmpdir,
                                            load_optimizer_states=False)

    def test_checkpoint_fused_optimizer(self, tmpdir):
        config_dict = {
            "train_batch_size": 2,
            "steps_per_print": 1,
            "optimizer": {
                "type": "Adam",
                "params": {
                    "lr": 0.00015,
                    "betas": [0.8,
                              0.999],
                    "eps": 1e-8,
                    "weight_decay": 3e-7
                }
            },
            "fp16": {
                "enabled": True
            }
        }

        args = args_from_dict(tmpdir, config_dict)
        hidden_dim = 10
        models = [SimpleModel(hidden_dim, empty_grad=False) for _ in range(2)]

        # Load & verify optimizer states
        checkpoint_correctness_verification(config_dict,
                                            models=models,
                                            hidden_dim=hidden_dim,
                                            tmpdir=tmpdir,
                                            load_optimizer_states=True)

        # Ignore optimizer states
        checkpoint_correctness_verification(config_dict,
                                            models=models,
                                            hidden_dim=hidden_dim,
                                            tmpdir=tmpdir,
                                            load_optimizer_states=False)

    def test_checkpoint_fp32_optimizer(self, tmpdir):
        config_dict = {
            "train_batch_size": 2,
            "steps_per_print": 1,
            "optimizer": {
                "type": "Adam",
                "params": {
                    "lr": 0.00015,
                    "betas": [0.8,
                              0.999],
                    "eps": 1e-8,
                    "weight_decay": 3e-7
                }
            },
            "fp16": {
                "enabled": False
            }
        }

        args = args_from_dict(tmpdir, config_dict)
        hidden_dim = 10
        models = [SimpleModel(hidden_dim, empty_grad=False) for _ in range(2)]
        checkpoint_correctness_verification(config_dict,
                                            models=models,
                                            hidden_dim=hidden_dim,
                                            tmpdir=tmpdir,
                                            fp16=False)