Skip to content

unitorch.cli.losses¤

CELoss¤

Tip

core/loss/ce is the section for configuration of CELoss.

Bases: CELoss

Source code in src/unitorch/cli/losses/__init__.py 
26
27
28
29
30
31
32
def __init__(
    self,
    smoothing_alpha: Optional[float] = 0.0,
):
    super().__init__(
        smoothing_alpha,
    )

from_config classmethod ¤

from_config(config, **kwargs)
Source code in src/unitorch/cli/losses/__init__.py 
34
35
36
37
@classmethod
@config_defaults_init("core/loss/ce")
def from_config(cls, config, **kwargs):
    pass

forward ¤

forward(
    outputs: Union[
        ClassificationOutputs, SegmentationOutputs
    ],
    targets: Union[
        ClassificationTargets, SegmentationTargets
    ],
)
Source code in src/unitorch/cli/losses/__init__.py 
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
def forward(
    self,
    outputs: Union[ClassificationOutputs, SegmentationOutputs],
    targets: Union[ClassificationTargets, SegmentationTargets],
):
    if isinstance(outputs, ClassificationOutputs):
        outputs = outputs.outputs
    if isinstance(targets, ClassificationTargets):
        weights = targets.sample_weight
        targets = targets.targets

    if isinstance(outputs, SegmentationOutputs):
        outputs = torch.cat([m.view(-1, m.size(-1)) for m in outputs.masks], dim=0)
    if isinstance(targets, SegmentationTargets):
        weights = torch.empty(0)
        targets = torch.cat([m.view(-1) for m in targets.targets], dim=0)

    return super().forward(
        input=outputs,
        target=targets,
        sample_weight=weights if weights.numel() > 0 else None,
    )

BCELoss¤

Tip

core/loss/bce is the section for configuration of BCELoss.

Bases: BCELoss

Source code in src/unitorch/cli/losses/__init__.py 
65
66
67
68
def __init__(
    self,
):
    super().__init__()

from_config classmethod ¤

from_config(config, **kwargs)
Source code in src/unitorch/cli/losses/__init__.py 
70
71
72
73
@classmethod
@config_defaults_init("core/loss/bce")
def from_config(cls, config, **kwargs):
    pass

forward ¤

forward(
    outputs: Union[
        ClassificationOutputs, SegmentationOutputs
    ],
    targets: Union[
        ClassificationTargets, SegmentationTargets
    ],
)
Source code in src/unitorch/cli/losses/__init__.py 
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
def forward(
    self,
    outputs: Union[ClassificationOutputs, SegmentationOutputs],
    targets: Union[ClassificationTargets, SegmentationTargets],
):
    if isinstance(outputs, ClassificationOutputs):
        outputs = outputs.outputs
    if isinstance(targets, ClassificationTargets):
        weights = targets.sample_weight
        targets = targets.targets

    if isinstance(outputs, SegmentationOutputs):
        outputs = torch.cat([m.view(-1, m.size(-1)) for m in outputs.masks], dim=0)
    if isinstance(targets, SegmentationTargets):
        weights = torch.empty(0)
        targets = torch.cat([m.view(-1) for m in targets.targets], dim=0)

    return super().forward(
        input=outputs,
        target=targets,
        sample_weight=weights if weights.numel() > 0 else None,
    )

LMLoss¤

Tip

core/loss/lm is the section for configuration of LMLoss.

Bases: LMLoss

Source code in src/unitorch/cli/losses/__init__.py 
101
102
103
104
def __init__(
    self,
):
    super().__init__()

from_config classmethod ¤

from_config(config, **kwargs)
Source code in src/unitorch/cli/losses/__init__.py 
106
107
108
109
@classmethod
@config_defaults_init("core/loss/lm")
def from_config(cls, config, **kwargs):
    pass

forward ¤

forward(
    outputs: GenerationOutputs, targets: GenerationTargets
)
Source code in src/unitorch/cli/losses/__init__.py 
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
def forward(
    self,
    outputs: GenerationOutputs,
    targets: GenerationTargets,
):
    outputs = outputs.sequences
    masks = targets.masks
    weights = targets.sample_weight
    targets = targets.refs

    return super().forward(
        input=outputs,
        target=targets,
        masks=masks if masks.numel() > 0 else None,
        sample_weight=weights if weights.numel() > 0 else None,
    )

MSELoss¤

Tip

core/loss/mse is the section for configuration of MSELoss.

Bases: MSELoss

Source code in src/unitorch/cli/losses/__init__.py 
131
132
133
134
def __init__(
    self,
):
    super().__init__()

from_config classmethod ¤

from_config(config, **kwargs)
Source code in src/unitorch/cli/losses/__init__.py 
136
137
138
139
@classmethod
@config_defaults_init("core/loss/mse")
def from_config(cls, config, **kwargs):
    pass

forward ¤

forward(
    outputs: Union[
        ClassificationOutputs, SegmentationOutputs
    ],
    targets: Union[
        ClassificationTargets, SegmentationTargets
    ],
)
Source code in src/unitorch/cli/losses/__init__.py 
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
def forward(
    self,
    outputs: Union[ClassificationOutputs, SegmentationOutputs],
    targets: Union[ClassificationTargets, SegmentationTargets],
):
    if isinstance(outputs, ClassificationOutputs):
        outputs = outputs.outputs
    if isinstance(targets, ClassificationTargets):
        weights = targets.sample_weight
        targets = targets.targets

    if isinstance(outputs, SegmentationOutputs):
        outputs = torch.cat([m.view(-1) for m in outputs.masks], dim=0)
    if isinstance(targets, SegmentationTargets):
        weights = torch.empty(0)
        targets = torch.cat([m.view(-1) for m in targets.targets], dim=0)

    return super().forward(
        input=outputs,
        target=targets,
        sample_weight=weights if weights.numel() > 0 else None,
    )

ListMLELoss¤

Tip

core/loss/ranking/listmle is the section for configuration of ListMLELoss.

Bases: ListMLELoss

Source code in src/unitorch/cli/losses/ranking.py 
19
20
21
22
def __init__(
    self,
):
    super().__init__()

from_config classmethod ¤

from_config(config, **kwargs)
Source code in src/unitorch/cli/losses/ranking.py 
24
25
26
27
@classmethod
@config_defaults_init("core/loss/ranking/listmle")
def from_config(cls, config, **kwargs):
    pass

forward ¤

forward(outputs: RankingOutputs, targets: RankingTargets)
Source code in src/unitorch/cli/losses/ranking.py 
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
def forward(
    self,
    outputs: RankingOutputs,
    targets: RankingTargets,
):
    outputs = outputs.outputs
    masks = targets.masks
    weights = targets.sample_weight
    targets = targets.targets

    return super().forward(
        input=outputs,
        target=targets,
        masks=masks if masks.numel() > 0 else None,
        sample_weight=weights if weights.numel() > 0 else None,
    )

ApproxNDCGLoss¤

Tip

core/loss/ranking/approxndcg is the section for configuration of ApproxNDCGLoss.

Bases: ApproxNDCGLoss

Source code in src/unitorch/cli/losses/ranking.py 
49
50
51
52
53
54
55
def __init__(
    self,
    alpha: Optional[float] = 10.0,
):
    super().__init__(
        alpha=alpha,
    )

from_config classmethod ¤

from_config(config, **kwargs)
Source code in src/unitorch/cli/losses/ranking.py 
57
58
59
60
@classmethod
@config_defaults_init("core/loss/ranking/approxndcg")
def from_config(cls, config, **kwargs):
    pass

forward ¤

forward(outputs: RankingOutputs, targets: RankingTargets)
Source code in src/unitorch/cli/losses/ranking.py 
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
def forward(
    self,
    outputs: RankingOutputs,
    targets: RankingTargets,
):
    outputs = outputs.outputs
    masks = targets.masks
    weights = targets.sample_weight
    targets = targets.targets

    return super().forward(
        input=outputs,
        target=targets,
        masks=masks if masks.numel() > 0 else None,
        sample_weight=weights if weights.numel() > 0 else None,
    )

ApproxMRRLoss¤

Tip

core/loss/ranking/approxmrr is the section for configuration of ApproxMRRLoss.

Bases: ApproxMRRLoss

Source code in src/unitorch/cli/losses/ranking.py 
82
83
84
85
86
87
88
def __init__(
    self,
    alpha: Optional[float] = 0.0,
):
    super().__init__(
        alpha=alpha,
    )

from_config classmethod ¤

from_config(config, **kwargs)
Source code in src/unitorch/cli/losses/ranking.py 
90
91
92
93
@classmethod
@config_defaults_init("core/loss/ranking/approxmrr")
def from_config(cls, config, **kwargs):
    pass

forward ¤

forward(outputs: RankingOutputs, targets: RankingTargets)
Source code in src/unitorch/cli/losses/ranking.py 
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
def forward(
    self,
    outputs: RankingOutputs,
    targets: RankingTargets,
):
    outputs = outputs.outputs
    masks = targets.masks
    weights = targets.sample_weight
    targets = targets.targets

    return super().forward(
        input=outputs,
        target=targets,
        masks=masks if masks.numel() > 0 else None,
        sample_weight=weights if weights.numel() > 0 else None,
    )