"""Hugging Face configuration for REX."""

from __future__ import annotations

from transformers import PretrainedConfig


class RexConfig(PretrainedConfig):
    model_type = "rex"

    def __init__(
        self,
        vocab_size: int = 50_257,
        max_seq_len: int = 2048,
        d_model: int = 1536,
        n_heads: int = 16,
        n_layers: int = 8,
        recurrence_steps: int = 2,
        ffn_dim: int = 3968,
        dropout: float = 0.0,
        norm_eps: float = 1e-5,
        tie_embeddings: bool = True,
        use_step_embeddings: bool = True,
        initializer_range: float = 0.02,
        tokenizer_name: str = "gpt2",
        **kwargs,
    ):
        super().__init__(tie_word_embeddings=tie_embeddings, **kwargs)
        self.vocab_size = vocab_size
        self.max_seq_len = max_seq_len
        self.d_model = d_model
        self.n_heads = n_heads
        self.n_layers = n_layers
        self.recurrence_steps = recurrence_steps
        self.ffn_dim = ffn_dim
        self.dropout = dropout
        self.norm_eps = norm_eps
        self.tie_embeddings = tie_embeddings
        self.use_step_embeddings = use_step_embeddings
        self.initializer_range = initializer_range
        self.tokenizer_name = tokenizer_name

    def to_core_dict(self) -> dict[str, object]:
        return {
            "vocab_size": self.vocab_size,
            "max_seq_len": self.max_seq_len,
            "d_model": self.d_model,
            "n_heads": self.n_heads,
            "n_layers": self.n_layers,
            "recurrence_steps": self.recurrence_steps,
            "ffn_dim": self.ffn_dim,
            "dropout": self.dropout,
            "norm_eps": self.norm_eps,
            "tie_embeddings": self.tie_embeddings,
            "use_step_embeddings": self.use_step_embeddings,
            "initializer_range": self.initializer_range,
        }