Book a Demo!
CoCalc Logo Icon
StoreFeaturesDocsShareSupportNewsAboutPoliciesSign UpSign In
shivamshrirao
GitHub Repository: shivamshrirao/diffusers
 Path: blob/main/tests/schedulers/test_scheduler_lms.py
1448 views
1
import torch

2


3
from diffusers import LMSDiscreteScheduler

4
from diffusers.utils import torch_device

5


6
from .test_schedulers import SchedulerCommonTest

7


8


9
class LMSDiscreteSchedulerTest(SchedulerCommonTest):

10
    scheduler_classes = (LMSDiscreteScheduler,)

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_time_indices(self):

41
        for t in [0, 500, 800]:

42
            self.check_over_forward(time_step=t)

43


44
    def test_full_loop_no_noise(self):

45
        scheduler_class = self.scheduler_classes[0]

46
        scheduler_config = self.get_scheduler_config()

47
        scheduler = scheduler_class(**scheduler_config)

48


49
        scheduler.set_timesteps(self.num_inference_steps)

50


51
        model = self.dummy_model()

52
        sample = self.dummy_sample_deter * scheduler.init_noise_sigma

53


54
        for i, t in enumerate(scheduler.timesteps):

55
            sample = scheduler.scale_model_input(sample, t)

56


57
            model_output = model(sample, t)

58


59
            output = scheduler.step(model_output, t, sample)

60
            sample = output.prev_sample

61


62
        result_sum = torch.sum(torch.abs(sample))

63
        result_mean = torch.mean(torch.abs(sample))

64


65
        assert abs(result_sum.item() - 1006.388) < 1e-2

66
        assert abs(result_mean.item() - 1.31) < 1e-3

67


68
    def test_full_loop_with_v_prediction(self):

69
        scheduler_class = self.scheduler_classes[0]

70
        scheduler_config = self.get_scheduler_config(prediction_type="v_prediction")

71
        scheduler = scheduler_class(**scheduler_config)

72


73
        scheduler.set_timesteps(self.num_inference_steps)

74


75
        model = self.dummy_model()

76
        sample = self.dummy_sample_deter * scheduler.init_noise_sigma

77


78
        for i, t in enumerate(scheduler.timesteps):

79
            sample = scheduler.scale_model_input(sample, t)

80


81
            model_output = model(sample, t)

82


83
            output = scheduler.step(model_output, t, sample)

84
            sample = output.prev_sample

85


86
        result_sum = torch.sum(torch.abs(sample))

87
        result_mean = torch.mean(torch.abs(sample))

88


89
        assert abs(result_sum.item() - 0.0017) < 1e-2

90
        assert abs(result_mean.item() - 2.2676e-06) < 1e-3

91


92
    def test_full_loop_device(self):

93
        scheduler_class = self.scheduler_classes[0]

94
        scheduler_config = self.get_scheduler_config()

95
        scheduler = scheduler_class(**scheduler_config)

96


97
        scheduler.set_timesteps(self.num_inference_steps, device=torch_device)

98


99
        model = self.dummy_model()

100
        sample = self.dummy_sample_deter * scheduler.init_noise_sigma

101
        sample = sample.to(torch_device)

102


103
        for i, t in enumerate(scheduler.timesteps):

104
            sample = scheduler.scale_model_input(sample, t)

105


106
            model_output = model(sample, t)

107


108
            output = scheduler.step(model_output, t, sample)

109
            sample = output.prev_sample

110


111
        result_sum = torch.sum(torch.abs(sample))

112
        result_mean = torch.mean(torch.abs(sample))

113


114
        assert abs(result_sum.item() - 1006.388) < 1e-2

115
        assert abs(result_mean.item() - 1.31) < 1e-3

116


117