Skip to content

Data types

EvaluationType

Bases: str, Enum

EvaluationType

Passed to Trainer.evaluate(..), and controls which phase callbacks should be triggered (if at all).

Attributes:
    TEST
    VALIDATION
Source code in common/data_types/enum/evaluation_type.py 
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
class EvaluationType(str, Enum):
    """
    EvaluationType

    Passed to Trainer.evaluate(..), and controls which phase callbacks should be triggered (if at all).

        Attributes:
            TEST
            VALIDATION

    """

    TEST = "TEST"
    VALIDATION = "VALIDATION"

MultiGPUMode

Bases: str, Enum

MultiGPUMode

Attributes:
    OFF                       - Single GPU Mode / CPU Mode
    DATA_PARALLEL             - Multiple GPUs, Synchronous
    DISTRIBUTED_DATA_PARALLEL - Multiple GPUs, Asynchronous
Source code in common/data_types/enum/multi_gpu_mode.py 
 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
class MultiGPUMode(str, Enum):
    """
    MultiGPUMode

        Attributes:
            OFF                       - Single GPU Mode / CPU Mode
            DATA_PARALLEL             - Multiple GPUs, Synchronous
            DISTRIBUTED_DATA_PARALLEL - Multiple GPUs, Asynchronous
    """

    OFF = "Off"
    DATA_PARALLEL = "DP"
    DISTRIBUTED_DATA_PARALLEL = "DDP"
    AUTO = "AUTO"

    @classmethod
    def dict(cls):
        """
        return dictionary mapping from the mode name (in call string cases) to the enum value
        """
        out_dict = dict()
        for mode in MultiGPUMode:
            out_dict[mode.value] = mode
            out_dict[mode.name] = mode
            out_dict[stringcase.capitalcase(mode.name)] = mode
            out_dict[stringcase.camelcase(mode.name)] = mode
            out_dict[stringcase.lowercase(mode.name)] = mode
        out_dict[False] = MultiGPUMode.OFF
        return out_dict

dict() classmethod

return dictionary mapping from the mode name (in call string cases) to the enum value

Source code in common/data_types/enum/multi_gpu_mode.py 
20
21
22
23
24
25
26
27
28
29
30
31
32
33
@classmethod
def dict(cls):
    """
    return dictionary mapping from the mode name (in call string cases) to the enum value
    """
    out_dict = dict()
    for mode in MultiGPUMode:
        out_dict[mode.value] = mode
        out_dict[mode.name] = mode
        out_dict[stringcase.capitalcase(mode.name)] = mode
        out_dict[stringcase.camelcase(mode.name)] = mode
        out_dict[stringcase.lowercase(mode.name)] = mode
    out_dict[False] = MultiGPUMode.OFF
    return out_dict

StrictLoad

Bases: Enum

Wrapper for adding more functionality to torch's strict_load parameter in load_state_dict(). Attributes: OFF - Native torch "strict_load = off" behaviour. See nn.Module.load_state_dict() documentation for more details. ON - Native torch "strict_load = on" behaviour. See nn.Module.load_state_dict() documentation for more details. NO_KEY_MATCHING - Allows the usage of SuperGradient's adapt_checkpoint function, which loads a checkpoint by matching each layer's shapes (and bypasses the strict matching of the names of each layer (ie: disregards the state_dict key matching)). This implementation assumes order of layers in the state_dict and model are the same since it goes layer by layer and as name suggest does not use key matching, relying only on index of each weight. KEY_MATCHING - Loose load strategy that loads the state dict from checkpoint into model only for common keys and also handling the case when shapes of the tensors in the state dict and model are different for the same key (Such layers will be skipped).

Source code in common/data_types/enum/strict_load.py 
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
class StrictLoad(Enum):
    """
    Wrapper for adding more functionality to torch's strict_load parameter in load_state_dict().
    Attributes:
        OFF              - Native torch "strict_load = off" behaviour. See nn.Module.load_state_dict() documentation for more details.
        ON               - Native torch "strict_load = on" behaviour. See nn.Module.load_state_dict() documentation for more details.
        NO_KEY_MATCHING  - Allows the usage of SuperGradient's adapt_checkpoint function, which loads a checkpoint by matching each
                           layer's shapes (and bypasses the strict matching of the names of each layer (ie: disregards the state_dict key matching)).
                           This implementation assumes order of layers in the state_dict and model are the same since it goes layer by layer and as name
                           suggest does not use key matching, relying only on index of each weight.
        KEY_MATCHING     - Loose load strategy that loads the state dict from checkpoint into model only for common keys and also handling the
                           case when shapes of the tensors in the state dict and model are different for the same key (Such layers will be skipped).
    """

    OFF = False
    ON = True
    NO_KEY_MATCHING = "no_key_matching"
    KEY_MATCHING = "key_matching"