from transformers import PretrainedConfig


class ZgcmConfig(PretrainedConfig):
    model_type = "zgcm"

    def __init__(
        self,
        vocab_size=154880,
        hidden_size=4096,
        intermediate_size=11008,
        num_hidden_layers=32,
        num_attention_heads=32,
        num_key_value_heads=8,
        head_dim=128,
        hidden_act="silu",
        rms_norm_eps=1e-6,
        rope_theta=5000000.0,
        partial_rotary_factor=0.334,
        max_position_embeddings=16384,
        sliding_window=128,
        window_attn_skip_freq=6,
        layer_types=None,
        attention_gate_layers=None,
        attention_bias=False,
        tie_word_embeddings=False,
        pad_token_id=154831,
        eos_token_id=154820,
        bos_token_id=None,
        use_cache=True,
        **kwargs,
    ):
        super().__init__(
            pad_token_id=pad_token_id,
            eos_token_id=eos_token_id,
            bos_token_id=bos_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.head_dim = head_dim
        self.hidden_act = hidden_act
        self.rms_norm_eps = rms_norm_eps
        self.rope_theta = rope_theta
        self.partial_rotary_factor = partial_rotary_factor
        self.max_position_embeddings = max_position_embeddings
        self.sliding_window = sliding_window
        self.window_attn_skip_freq = window_attn_skip_freq
        self.layer_types = layer_types or [
            "full_attention" if (i + 1) % window_attn_skip_freq == 0 else "sliding_attention"
            for i in range(num_hidden_layers)
        ]
        self.attention_gate_layers = attention_gate_layers or [
            layer_type == "sliding_attention" for layer_type in self.layer_types
        ]
        self.attention_bias = attention_bias
        self.use_cache = use_cache
        self.auto_map = kwargs.get(
            "auto_map",
            {
                "AutoConfig": "configuration_zgcm.ZgcmConfig",
                "AutoModel": "modeling_zgcm.ZgcmModel",
                "AutoModelForCausalLM": "modeling_zgcm.ZgcmForCausalLM",
            },
        )

    def get_text_config(self, decoder=False):
        return self