from transformers.configuration_utils import PretrainedConfig


class LowOnMindConfig(PretrainedConfig):
    """Config do LowOnMind.

    Derivada do DynamicMindConfig (DedeProGames/DynamicMind-Mini), com
    `use_qk_norm` e `initializer_range` adicionais.
    """

    model_type = "lowonmind"

    def __init__(
        self,
        vocab_size=1024,
        hidden_size=64,
        intermediate_size=136,
        num_hidden_layers=6,
        num_attention_heads=4,
        num_key_value_heads=2,
        max_position_embeddings=512,
        rms_norm_eps=1e-5,
        rope_theta=10000.0,
        attention_dropout=0.0,
        use_qk_norm=True,
        initializer_range=0.02,
        tie_word_embeddings=True,
        use_cache=False,
        bos_token_id=0,
        eos_token_id=0,
        pad_token_id=1,
        **kwargs,
    ):
        super().__init__(
            bos_token_id=bos_token_id,
            eos_token_id=eos_token_id,
            pad_token_id=pad_token_id,
            tie_word_embeddings=tie_word_embeddings,
            **kwargs,
        )

        self.vocab_size = vocab_size
        self.hidden_size = hidden_size
        self.intermediate_size = intermediate_size
        self.num_hidden_layers = num_hidden_layers
        self.num_attention_heads = num_attention_heads
        self.num_key_value_heads = num_key_value_heads
        self.max_position_embeddings = max_position_embeddings
        self.rms_norm_eps = rms_norm_eps
        self.rope_theta = rope_theta
        self.attention_dropout = attention_dropout
        self.use_qk_norm = use_qk_norm
        self.initializer_range = initializer_range
        self.use_cache = use_cache