from transformers.configuration_utils import PretrainedConfig

class TinyImageGenConfig(PretrainedConfig):
    model_type = "tinyimagegen"

    def __init__(
        self,
        image_size: int = 32,
        in_channels: int = 3,
        patch_size: int = 4,
        hidden_size: int = 32,
        num_hidden_layers: int = 6,
        num_attention_heads: int = 4,
        num_key_value_heads: int = 2,
        intermediate_size: int = 48,
        swiglu_interval: int = 3,
        num_lanes: int = 4,
        use_xsa: bool = False,
        use_per_head_gating: bool = False,
        rope_theta: float = 2500.0,
        rms_norm_eps: float = 1e-5,
        initializer_range: float = 0.02,
        **kwargs,
    ):
        self.image_size = image_size
        self.in_channels = in_channels
        self.patch_size = patch_size
        self.hidden_size = hidden_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.intermediate_size = intermediate_size
        self.swiglu_interval = swiglu_interval
        self.num_lanes = num_lanes
        self.use_xsa = use_xsa
        self.use_per_head_gating = use_per_head_gating
        self.rope_theta = rope_theta
        self.rms_norm_eps = rms_norm_eps
        self.initializer_range = initializer_range
        self.head_dim = hidden_size // num_attention_heads
        self.num_patches_side = image_size // patch_size
        self.num_patches = self.num_patches_side ** 2
        self.patch_dim = in_channels * (patch_size ** 2)

        self.auto_map = {
            "AutoConfig": "configuration_tinyimagegen.TinyImageGenConfig",
            "AutoModel": "modeling_tinyimagegen.TinyImageGenModelForImageDiffusion",
            "AutoModelForImageDiffusion": "modeling_tinyimagegen.TinyImageGenModelForImageDiffusion",
        }

        super().__init__(**kwargs)