Config
- class dataset2vec.config.Dataset2VecConfig(*, activation_cls: ~typing.Type[~torch.nn.modules.module.Module] = <class 'torch.nn.modules.activation.ReLU'>, f_dense_hidden_size: int = 32, f_res_hidden_size: int = 32, f_res_n_layers: int = 3, f_block_repetitions: int = 7, f_out_size: int = 32, g_layers_sizes: list[int] = [32, 16, 8], h_dense_hidden_size: int = 16, h_res_hidden_size: int = 16, h_res_n_layers: int = 3, h_block_repetitions: int = 3, output_size: int = 16)
Bases:
BaseModelConfiguration of the Dataset2Vec encoder
- activation_cls: Type[Module]
Class of the activation function used in entire network.
- f_block_repetitions: int
Number of building blocks of the first stage.
Size of the hidden layers of the first stage.
- f_out_size: int
Dimensionality of the output of the first starge.
Size of the hidden layers of the residual blocks of the first stage.
- f_res_n_layers: int
Number of the layers of the residual block of the first stage.
- g_layers_sizes: list[int]
Sizes of the layers of the feed forward network in the second stage.
- h_block_repetitions: int
Number of building blocks of the third stage.
Size of the hidden layers of the feed forward net of the third stage.
Size of the hidden layers of the residual blocks of the third stage.
- h_res_n_layers: int
Number of layers of the residual block of the third stage.
- output_size: int
Output dimensionality of the encoder.
- class dataset2vec.config.OptimizerConfig(*, gamma: float = 1, optimizer_cls: ~typing.Type[~torch.optim.optimizer.Optimizer] = <class 'torch.optim.adam.Adam'>, learning_rate: float = 0.0001, weight_decay: float = 0.0001)
Bases:
BaseModelConfiguration of the Dataset2Vec training
- gamma: float
Scaling parameter for the calculation of the probability.
- learning_rate: float
Learning rate.
- optimizer_cls: Type[Optimizer]
Class of the optimizer.
- weight_decay: float
Weight decay.