def load(cls, fn: str, conf: dict):
'''
load prediction model from filename
'''
device = conf.get('DEVICE')
self = torch.load(fn, map_location=device)
Config.__init__(self, conf)
return self
def __init__(self, query_model: nn.Module, candidate_model: nn.Module, hidden_size, conf=None):
nn.Module.__init__(self)
Config.__init__(self, conf)
# save model.
self.query_model = query_model
self.candidate_model = candidate_model
# projection layer.
self.proj = nn.Sequential(
nn.Linear(3 * hidden_size, hidden_size, bias=self.USE_BIAS),
activation[self.ACT_NAME](),
nn.Linear(hidden_size, 2, bias=self.USE_BIAS)
)
def __init__(self, config=None):
nn.Module.__init__(self)
Config.__init__(self, config)
self._unfreeze(self.FINETUNE_LAYER_RANGE)
def __init__(self, config=None):
nn.Module.__init__(self)
Config.__init__(self, config)
self.proto = Protonet.Prototype()