CoCalc -- test_scheduler

GitHub Repository: shivamshrirao/diffusers
Path: blob/main/tests/schedulers/test_scheduler_euler.py
¹⁴⁴⁸ views
1
import torch
2

3
from diffusers import EulerDiscreteScheduler
4
from diffusers.utils import torch_device
5

6
from .test_schedulers import SchedulerCommonTest
7

8

9
class EulerDiscreteSchedulerTest(SchedulerCommonTest):
10
    scheduler_classes = (EulerDiscreteScheduler,)
11
    num_inference_steps = 10
12

13
    def get_scheduler_config(self, **kwargs):
14
        config = {
15
            "num_train_timesteps": 1100,
16
            "beta_start": 0.0001,
17
            "beta_end": 0.02,
18
            "beta_schedule": "linear",
19
        }
20

21
        config.update(**kwargs)
22
        return config
23

24
    def test_timesteps(self):
25
        for timesteps in [10, 50, 100, 1000]:
26
            self.check_over_configs(num_train_timesteps=timesteps)
27

28
    def test_betas(self):
29
        for beta_start, beta_end in zip([0.00001, 0.0001, 0.001], [0.0002, 0.002, 0.02]):
30
            self.check_over_configs(beta_start=beta_start, beta_end=beta_end)
31

32
    def test_schedules(self):
33
        for schedule in ["linear", "scaled_linear"]:
34
            self.check_over_configs(beta_schedule=schedule)
35

36
    def test_prediction_type(self):
37
        for prediction_type in ["epsilon", "v_prediction"]:
38
            self.check_over_configs(prediction_type=prediction_type)
39

40
    def test_full_loop_no_noise(self):
41
        scheduler_class = self.scheduler_classes[0]
42
        scheduler_config = self.get_scheduler_config()
43
        scheduler = scheduler_class(**scheduler_config)
44

45
        scheduler.set_timesteps(self.num_inference_steps)
46

47
        generator = torch.manual_seed(0)
48

49
        model = self.dummy_model()
50
        sample = self.dummy_sample_deter * scheduler.init_noise_sigma
51
        sample = sample.to(torch_device)
52

53
        for i, t in enumerate(scheduler.timesteps):
54
            sample = scheduler.scale_model_input(sample, t)
55

56
            model_output = model(sample, t)
57

58
            output = scheduler.step(model_output, t, sample, generator=generator)
59
            sample = output.prev_sample
60

61
        result_sum = torch.sum(torch.abs(sample))
62
        result_mean = torch.mean(torch.abs(sample))
63

64
        assert abs(result_sum.item() - 10.0807) < 1e-2
65
        assert abs(result_mean.item() - 0.0131) < 1e-3
66

67
    def test_full_loop_with_v_prediction(self):
68
        scheduler_class = self.scheduler_classes[0]
69
        scheduler_config = self.get_scheduler_config(prediction_type="v_prediction")
70
        scheduler = scheduler_class(**scheduler_config)
71

72
        scheduler.set_timesteps(self.num_inference_steps)
73

74
        generator = torch.manual_seed(0)
75

76
        model = self.dummy_model()
77
        sample = self.dummy_sample_deter * scheduler.init_noise_sigma
78
        sample = sample.to(torch_device)
79

80
        for i, t in enumerate(scheduler.timesteps):
81
            sample = scheduler.scale_model_input(sample, t)
82

83
            model_output = model(sample, t)
84

85
            output = scheduler.step(model_output, t, sample, generator=generator)
86
            sample = output.prev_sample
87

88
        result_sum = torch.sum(torch.abs(sample))
89
        result_mean = torch.mean(torch.abs(sample))
90

91
        assert abs(result_sum.item() - 0.0002) < 1e-2
92
        assert abs(result_mean.item() - 2.2676e-06) < 1e-3
93

94
    def test_full_loop_device(self):
95
        scheduler_class = self.scheduler_classes[0]
96
        scheduler_config = self.get_scheduler_config()
97
        scheduler = scheduler_class(**scheduler_config)
98

99
        scheduler.set_timesteps(self.num_inference_steps, device=torch_device)
100

101
        generator = torch.manual_seed(0)
102

103
        model = self.dummy_model()
104
        sample = self.dummy_sample_deter * scheduler.init_noise_sigma
105
        sample = sample.to(torch_device)
106

107
        for t in scheduler.timesteps:
108
            sample = scheduler.scale_model_input(sample, t)
109

110
            model_output = model(sample, t)
111

112
            output = scheduler.step(model_output, t, sample, generator=generator)
113
            sample = output.prev_sample
114

115
        result_sum = torch.sum(torch.abs(sample))
116
        result_mean = torch.mean(torch.abs(sample))
117

118
        assert abs(result_sum.item() - 10.0807) < 1e-2
119
        assert abs(result_mean.item() - 0.0131) < 1e-3
120

121
Product

Resources

Company
