TransformerSentenceEncoder.Config

Component: TransformerSentenceEncoder

class TransformerSentenceEncoder.Config

Bases: TransformerSentenceEncoderBase.Config, ConfigBase

All Attributes (including base classes)

load_path: Optional[str] = None

save_path: Optional[str] = None

freeze: bool = False

shared_module_key: Optional[str] = None

output_dropout: float = 0.4

embedding_dim: int = 768

pooling: PoolingMethod = <PoolingMethod.CLS_TOKEN: 'cls_token'>

export: bool = False

projection_dim: int = 0

normalize_output_rep: bool = False

dropout: float = 0.1

attention_dropout: float = 0.1

activation_dropout: float = 0.1

ffn_embedding_dim: int = 3072

num_encoder_layers: int = 6

num_attention_heads: int = 8

num_segments: int = 2

use_position_embeddings: bool = True

offset_positions_by_padding: bool = True

apply_bert_init: bool = True

encoder_normalize_before: bool = True

activation_fn: str = 'relu'

max_seq_len: int = 128

multilingual: bool = False

freeze_embeddings: bool = False

n_trans_layers_to_freeze: int = 0

use_torchscript: bool = False

use_bias_finetuning: bool = False

Subclasses

• SparseTransformerSentenceEncoder.Config

Default JSON

{
    "load_path": null,
    "save_path": null,
    "freeze": false,
    "shared_module_key": null,
    "output_dropout": 0.4,
    "embedding_dim": 768,
    "pooling": "cls_token",
    "export": false,
    "projection_dim": 0,
    "normalize_output_rep": false,
    "dropout": 0.1,
    "attention_dropout": 0.1,
    "activation_dropout": 0.1,
    "ffn_embedding_dim": 3072,
    "num_encoder_layers": 6,
    "num_attention_heads": 8,
    "num_segments": 2,
    "use_position_embeddings": true,
    "offset_positions_by_padding": true,
    "apply_bert_init": true,
    "encoder_normalize_before": true,
    "activation_fn": "relu",
    "max_seq_len": 128,
    "multilingual": false,
    "freeze_embeddings": false,
    "n_trans_layers_to_freeze": 0,
    "use_torchscript": false,
    "use_bias_finetuning": false
}