import warnings

from transformers.configuration_utils import PretrainedConfig


class MLAConfig(PretrainedConfig):

    model_type = 'mla'
    keys_to_ignore_at_inference = ['past_key_values']

    def __init__(
        self,
        hidden_size: int = 2048,
        num_hidden_layers: int = 24,
        num_heads: int = 16,
        q_lora_rank: int | None = 64,
        qk_rope_head_dim: int = 64,
        kv_lora_rank: int = 512,  # following the original Deepseek paper
        v_head_dim: int = 128,
        qk_nope_head_dim: int = 128,
        qk_head_dim: int | None = 192,  # qk_nope_head_dim + qk_rope_head_dim
        window_size: int | None = None,
        rope_theta: float | None = 10000.,
        max_position_embeddings: int = 2048,
        rope_scaling: dict | None = None,
        hidden_ratio: int | None = 4,
        intermediate_size: int | None = None,
        hidden_act: str = "swish",
        initializer_range: float = 0.02,
        elementwise_affine: bool | None = True,
        norm_eps: float = 1e-6,
        use_cache: bool = True,
        pad_token_id: int = None,
        bos_token_id: int = 1,
        eos_token_id: int = 2,
        tie_word_embeddings: bool = False,
        fuse_norm: bool = True,
        fuse_swiglu: bool = True,
        fuse_cross_entropy: bool = True,
        fuse_linear_cross_entropy: bool = False,
        use_l2warp: bool = False,
        vocab_size: int = 32000,
        **kwargs,
    ):
        self.hidden_size = hidden_size
        self.num_hidden_layers = num_hidden_layers
        self.num_heads = num_heads

        # MLA specific args
        self.q_lora_rank = q_lora_rank
        self.qk_rope_head_dim = qk_rope_head_dim
        self.kv_lora_rank = kv_lora_rank
        self.v_head_dim = v_head_dim
        self.qk_nope_head_dim = qk_nope_head_dim
        self.qk_head_dim = qk_head_dim
        self.rope_scaling = rope_scaling

        self.window_size = window_size
        self.rope_theta = rope_theta
        self.max_position_embeddings = max_position_embeddings

        self.hidden_ratio = hidden_ratio
        self.intermediate_size = intermediate_size
        self.hidden_act = hidden_act

        self.initializer_range = initializer_range
        self.elementwise_affine = elementwise_affine
        self.norm_eps = norm_eps
        self.use_cache = use_cache

        self.fuse_norm = fuse_norm
        self.fuse_swiglu = fuse_swiglu
        self.fuse_cross_entropy = fuse_cross_entropy
        self.fuse_linear_cross_entropy = fuse_linear_cross_entropy
        self.use_l2warp = use_l2warp
        self.vocab_size = vocab_size

        if fuse_cross_entropy and fuse_linear_cross_entropy:
            raise ValueError(
                "`fuse_cross_entropy` and `fuse_linear_cross_entropy` cannot be True at the same time.",
            )
        if fuse_linear_cross_entropy:
            warnings.warn(
                "`fuse_linear_cross_entropy` is enabled, which can improves memory efficiency "
                "at the potential cost of reduced precision. "
                "If you observe issues like loss divergence, consider disabling this setting.",
            )

        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,
        )