from __future__ import annotations

from transformers import PretrainedConfig


class CustomTransformerConfig(PretrainedConfig):
    model_type = "custom_transformer"

    def __init__(
        self,
        vocab_size: int = 32000,
        hidden_size: int = 512,
        num_hidden_layers: int = 8,
        num_attention_heads: int = 8,
        num_key_value_heads: int = 2,
        ffn_hidden_size: int = 1365,
        max_position_embeddings: int = 2048,
        rope_theta: float = 10000.0,
        norm_eps: float = 1e-5,
        dropout: float = 0.0,
        qk_bias: bool = True,
        use_head_gating: bool = True,
        attn_res_mode: str = "full",
        attn_res_block_size: int = 4,
        tie_word_embeddings: bool = True,
        pad_token_id: int | None = None,
        bos_token_id: int | None = None,
        eos_token_id: int | None = None,
        **kwargs,
    ):
        super().__init__(
            pad_token_id=pad_token_id,
            bos_token_id=bos_token_id,
            eos_token_id=eos_token_id,
            tie_word_embeddings=tie_word_embeddings,
            **kwargs,
        )
        self.vocab_size = int(vocab_size)
        self.hidden_size = int(hidden_size)
        self.num_hidden_layers = int(num_hidden_layers)
        self.num_attention_heads = int(num_attention_heads)
        self.num_key_value_heads = int(num_key_value_heads)
        self.ffn_hidden_size = int(ffn_hidden_size)
        self.max_position_embeddings = int(max_position_embeddings)
        self.rope_theta = float(rope_theta)
        self.norm_eps = float(norm_eps)
        self.dropout = float(dropout)
        self.qk_bias = bool(qk_bias)
        self.use_head_gating = bool(use_head_gating)
        self.attn_res_mode = str(attn_res_mode)
        self.attn_res_block_size = int(attn_res_block_size)
        self.tie_word_embeddings = bool(tie_word_embeddings)
        self.use_cache = False

        if self.attn_res_mode not in ("full", "block", "none"):
            raise ValueError("attn_res_mode must be one of: 'full', 'block', 'none'")
        if self.num_attention_heads <= 0:
            raise ValueError("num_attention_heads must be positive")
        if self.num_key_value_heads <= 0:
            raise ValueError("num_key_value_heads must be positive")
        if self.hidden_size % self.num_attention_heads != 0:
            raise ValueError("hidden_size must be divisible by num_attention_heads")
        if self.num_attention_heads % self.num_key_value_heads != 0:
            raise ValueError("num_attention_heads must be divisible by num_key_value_heads")
        if self.ffn_hidden_size < 1:
            raise ValueError("ffn_hidden_size must be positive")
        if self.max_position_embeddings <= 0:
            raise ValueError("max_position_embeddings must be positive")