from transformers import PretrainedConfig

class HybridModelConfig(PretrainedConfig):
    model_type = "hybrid_model"
    keys_to_ignore_at_inference = ["past_key_values"]
    
    def __init__(
        self,
        vocab_size=151936,
        hidden_size=768,
        intermediate_size=2048,
        num_hidden_layers=12,
        num_attention_heads=12,
        # MLA compression dims (DeepSeek-style naming)
        kv_lora_rank=192,        # KV latent/compression dimension (d_c)
        q_lora_rank=384,         # Query latent/compression dimension (d_c1)
        qk_rope_head_dim=32,     # RoPE dimension per head (d_rotate)
        hidden_act="silu",
        max_position_embeddings=32768,
        initializer_range=0.02,
        rms_norm_eps=1e-6,
        use_cache=True,
        pad_token_id=0,
        bos_token_id=1,
        eos_token_id=2,
        tie_word_embeddings=False,
        rope_theta=10000.0,
        sliding_window=4096,
        attention_dropout=0.0,
        # MHC (Multi-Head Connections) settings
        mhc_num_streams=4,       # number of parallel streams (mhc_n)
        mhc_sinkhorn_iters=20,   # Sinkhorn-Knopp iterations (mhc_tmax)
        mhc_alpha_init=0.01,
        mhc_rmsnorm_eps=1e-6,
        mhc_stream_init="paper",
        mhc_readout_init="first",
        **kwargs,
    ):
        self.vocab_size = vocab_size
        self.max_position_embeddings = max_position_embeddings
        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.kv_lora_rank = kv_lora_rank
        self.q_lora_rank = q_lora_rank
        self.qk_rope_head_dim = qk_rope_head_dim

        self.sliding_window = sliding_window

        self.hidden_act = hidden_act
        self.initializer_range = initializer_range
        self.rms_norm_eps = rms_norm_eps
        self.use_cache = use_cache
        self.rope_theta = rope_theta
        self.attention_dropout = attention_dropout

        self.mhc_num_streams = mhc_num_streams
        self.mhc_sinkhorn_iters = mhc_sinkhorn_iters
        self.mhc_alpha_init = mhc_alpha_init
        self.mhc_rmsnorm_eps = mhc_rmsnorm_eps
        self.mhc_stream_init = mhc_stream_init
        self.mhc_readout_init = mhc_readout_init
        
        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,
        )