from transformers import PretrainedConfig
from typing import List, Literal, Optional


_SAMPLING_TYPE = Literal['fp32_gumbel', 'fp16_gumbel', 'multinomial']


class CharmenElectraConfig(PretrainedConfig):
    # model_type = "SzegedAI/charmen-electra"
    # _name_or_path = "SzegedAI/charmen-electra"
    architectures = [
        "CharmenElectraModel"
    ]

    def __init__(
        self,
        downsampling_factor: int = 4,
        max_block_size: int = 4,
        score_consensus_attn: bool = True,
        upsample_output: bool = True,
        sampling: _SAMPLING_TYPE = 'fp32_gumbel',
        attention_probs_dropout_prob: float = 0.1,
        embedding_size: int = 768,
        hidden_act: str = "gelu",
        hidden_dropout_prob: float = 0.1,
        hidden_size: int = 512,
        initializer_range: float = 0.02,
        intermediate_size: int = 2048,
        layer_norm_eps: float = 1e-12,
        max_position_embeddings: int = 1024,
        model_type: str = "electra",
        num_attention_heads: int = 8,
        num_hidden_layers: int = 12,
        pad_token_id: int = 0,
        position_embedding_type: str = "absolute",
        summary_activation: str = "gelu",
        summary_last_dropout: float = 0.1,
        summary_type: str = "first",
        summary_use_proj: bool = True,
        type_vocab_size: int = 2,
        vocab_size: int = 261,
        classifier_dropout: Optional[float] = None,
        **kwargs,
    ):
        self.downsampling_factor = downsampling_factor
        self.max_block_size = max_block_size
        self.score_consensus_attn = score_consensus_attn
        self.upsample_output = upsample_output
        self.attention_probs_dropout_prob = attention_probs_dropout_prob
        self.embedding_size = embedding_size
        self.hidden_act = hidden_act
        self.hidden_dropout_prob = hidden_dropout_prob
        self.hidden_size = hidden_size
        self.initializer_range = initializer_range
        self.intermediate_size = intermediate_size
        self.layer_norm_eps = layer_norm_eps
        self.max_position_embeddings = max_position_embeddings
        self.model_type = model_type
        self.num_attention_heads = num_attention_heads
        self.num_hidden_layers = num_hidden_layers
        self.pad_token_id = pad_token_id
        self.position_embedding_type = position_embedding_type
        self.summary_activation = summary_activation
        self.summary_last_dropout = summary_last_dropout
        self.summary_type = summary_type
        self.summary_use_proj = summary_use_proj
        self.type_vocab_size = type_vocab_size
        self.vocab_size = vocab_size
        self.sampling = sampling
        self.classifier_dropout = classifier_dropout
        super().__init__(**kwargs)