import math
from typing import Any, Optional

import mlx.core as mx
import mlx.nn as nn
from mlx_lm.models.activations import swiglu
from mlx_lm.models.mla import MultiLinear
from mlx_lm.models.rope_utils import initialize_rope
from mlx_lm.models.switch_layers import SwitchGLU

from ..base import (
    LanguageModelOutput,
    create_attention_mask,
    scaled_dot_product_attention,
)
from .config import ModelConfig, TextConfig


class YoutuAttention(nn.Module):
    """Multi-Latent Attention (MLA) from DeepSeek-V2/V3 architecture."""

    def __init__(self, config: TextConfig):
        super().__init__()
        self.config = config
        self.hidden_size = config.hidden_size
        self.num_heads = config.num_attention_heads
        self.q_lora_rank = config.q_lora_rank
        self.qk_rope_head_dim = config.qk_rope_head_dim
        self.kv_lora_rank = config.kv_lora_rank
        self.v_head_dim = config.v_head_dim
        self.qk_nope_head_dim = config.qk_nope_head_dim
        self.q_head_dim = config.qk_nope_head_dim + config.qk_rope_head_dim

        self.scale = self.q_head_dim**-0.5

        if config.rope_scaling is not None:
            mscale_all_dim = config.rope_scaling.get("mscale_all_dim", 0)
            if mscale_all_dim:
                scaling_factor = config.rope_scaling.get("factor", 1.0)
                if scaling_factor > 1:
                    s = 0.1 * mscale_all_dim * math.log(scaling_factor) + 1.0
                    self.scale = self.scale * s * s

        if self.q_lora_rank is None:
            self.q_proj = nn.Linear(
                self.hidden_size, self.num_heads * self.q_head_dim, bias=False
            )
        else:
            self.q_a_proj = nn.Linear(
                self.hidden_size, self.q_lora_rank, bias=config.attention_bias
            )
            self.q_a_layernorm = nn.RMSNorm(self.q_lora_rank, eps=config.rms_norm_eps)
            self.q_b_proj = nn.Linear(
                self.q_lora_rank, self.num_heads * self.q_head_dim, bias=False
            )

        self.kv_a_proj_with_mqa = nn.Linear(
            self.hidden_size,
            self.kv_lora_rank + self.qk_rope_head_dim,
            bias=config.attention_bias,
        )
        self.kv_a_layernorm = nn.RMSNorm(self.kv_lora_rank, eps=config.rms_norm_eps)
        self.embed_q = MultiLinear(
            self.qk_nope_head_dim, self.kv_lora_rank, self.num_heads
        )
        self.unembed_out = MultiLinear(
            self.kv_lora_rank, self.v_head_dim, self.num_heads
        )

        self.o_proj = nn.Linear(
            self.num_heads * self.v_head_dim,
            self.hidden_size,
            bias=config.attention_bias,
        )

        self.rope = initialize_rope(
            self.qk_rope_head_dim,
            base=config.rope_theta,
            traditional=config.rope_traditional,
            scaling_config=config.rope_scaling,
            max_position_embeddings=config.max_position_embeddings,
        )

    def __call__(
        self,
        x: mx.array,
        mask: Optional[mx.array] = None,
        cache: Optional[Any] = None,
    ) -> mx.array:
        B, L, _ = x.shape

        if self.q_lora_rank is None:
            q = self.q_proj(x)
        else:
            q = self.q_b_proj(self.q_a_layernorm(self.q_a_proj(x)))

        q = q.reshape(B, L, self.num_heads, self.q_head_dim).transpose(0, 2, 1, 3)
        q_nope, q_pe = mx.split(q, [self.qk_nope_head_dim], axis=-1)

        compressed_kv = self.kv_a_proj_with_mqa(x)
        compressed_kv, k_pe = mx.split(compressed_kv, [self.kv_lora_rank], axis=-1)
        k_pe = k_pe.reshape(B, L, 1, self.qk_rope_head_dim).transpose(0, 2, 1, 3)
        kv_latent = self.kv_a_layernorm(compressed_kv)

        offset = cache.offset if cache is not None else 0
        q_pe = self.rope(q_pe, offset)
        k_pe = self.rope(k_pe, offset)

        kv_latent = mx.expand_dims(kv_latent, axis=1)

        if cache is not None:
            kv_latent, k_pe = cache.update_and_fetch(kv_latent, k_pe)

        pe_scores = (q_pe * self.scale) @ k_pe.swapaxes(-1, -2)
        if mask is not None:
            pe_scores = mx.where(
                mask,
                pe_scores,
                mx.array(mx.finfo(pe_scores.dtype).min, pe_scores.dtype),
            )

        if L == 1:
            q_nope = self.embed_q(q_nope)
            k = v = kv_latent
        else:
            k = self.embed_q(kv_latent, transpose=False)
            v = self.unembed_out(kv_latent)

        output = scaled_dot_product_attention(
            q_nope, k, v, cache=cache, scale=self.scale, mask=pe_scores
        )
        if L == 1:
            output = self.unembed_out(output)

        output = output.transpose(0, 2, 1, 3).reshape(B, L, -1)
        return self.o_proj(output)


class YoutuMLP(nn.Module):
    def __init__(self, config: TextConfig, hidden_size=None, intermediate_size=None):
        super().__init__()
        self.hidden_size = config.hidden_size if hidden_size is None else hidden_size
        self.intermediate_size = (
            config.intermediate_size if intermediate_size is None else intermediate_size
        )
        self.gate_proj = nn.Linear(
            self.hidden_size, self.intermediate_size, bias=config.mlp_bias
        )
        self.up_proj = nn.Linear(
            self.hidden_size, self.intermediate_size, bias=config.mlp_bias
        )
        self.down_proj = nn.Linear(
            self.intermediate_size, self.hidden_size, bias=config.mlp_bias
        )

    def __call__(self, x) -> mx.array:
        return self.down_proj(swiglu(self.gate_proj(x), self.up_proj(x)))


@mx.compile
def group_expert_select(
    gates,
    e_score_correction_bias,
    top_k,
    n_group,
    topk_group,
    routed_scaling_factor,
    norm_topk_prob,
):
    scores = mx.sigmoid(gates.astype(mx.float32))
    orig_scores = scores
    scores = scores + e_score_correction_bias
    if n_group > 1:
        scores = mx.unflatten(scores, axis=-1, shape=(n_group, -1))
        group_scores = mx.topk(scores, 2, axis=-1).sum(axis=-1, keepdims=True)
        k = n_group - topk_group
        group_idx = mx.argpartition(group_scores, kth=k - 1, axis=-2)[..., :k, :]
        scores = mx.put_along_axis(
            scores, mx.stop_gradient(group_idx), mx.array(0.0), axis=-2
        )
        scores = mx.flatten(scores, -2, -1)

    k = top_k
    inds = mx.argpartition(-scores, kth=k - 1, axis=-1)[..., :k]
    scores = mx.take_along_axis(orig_scores, inds, axis=-1)
    if top_k > 1 and norm_topk_prob:
        denominator = scores.sum(axis=-1, keepdims=True)
        scores = scores / denominator
    scores = scores * routed_scaling_factor

    return inds, scores


class MoEGate(nn.Module):
    def __init__(self, config: TextConfig):
        super().__init__()
        self.config = config
        self.top_k = config.num_experts_per_tok
        self.norm_topk_prob = config.norm_topk_prob
        self.n_routed_experts = config.n_routed_experts
        self.routed_scaling_factor = config.routed_scaling_factor
        self.n_group = config.n_group
        self.topk_group = config.topk_group
        self.weight = mx.zeros((self.n_routed_experts, config.hidden_size))
        self.e_score_correction_bias = mx.zeros((self.n_routed_experts,))

    def __call__(self, x):
        return group_expert_select(
            x @ self.weight.T,
            self.e_score_correction_bias,
            self.top_k,
            self.n_group,
            self.topk_group,
            self.routed_scaling_factor,
            self.norm_topk_prob,
        )


class YoutuMoE(nn.Module):
    def __init__(self, config: TextConfig):
        super().__init__()
        self.config = config
        self.num_experts_per_tok = config.num_experts_per_tok

        if SwitchGLU is not None:
            self.switch_mlp = SwitchGLU(
                config.hidden_size,
                config.moe_intermediate_size,
                config.n_routed_experts,
            )
        else:
            raise ImportError("SwitchGLU not available. Cannot use MoE.")

        self.gate = MoEGate(config)
        if config.n_shared_experts is not None:
            intermediate_size = config.moe_intermediate_size * config.n_shared_experts
            self.shared_experts = YoutuMLP(
                config=config, intermediate_size=intermediate_size
            )

    def __call__(self, x):
        inds, scores = self.gate(x)
        y = self.switch_mlp(x, inds)
        y = (y * scores[..., None]).sum(axis=-2).astype(y.dtype)
        if self.config.n_shared_experts is not None:
            y = y + self.shared_experts(x)
        return y


class YoutuDecoderLayer(nn.Module):
    def __init__(self, config: TextConfig, layer_idx: int):
        super().__init__()
        self.hidden_size = config.hidden_size
        self.self_attn = YoutuAttention(config)

        # Use MoE if configured
        if (
            config.n_routed_experts is not None
            and layer_idx >= config.first_k_dense_replace
            and layer_idx % config.moe_layer_freq == 0
        ):
            self.mlp = YoutuMoE(config)
        else:
            self.mlp = YoutuMLP(config)

        self.input_layernorm = nn.RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
        self.post_attention_layernorm = nn.RMSNorm(
            config.hidden_size, eps=config.rms_norm_eps
        )

    def __call__(
        self,
        x: mx.array,
        mask: Optional[mx.array] = None,
        cache: Optional[Any] = None,
    ) -> mx.array:
        r = self.self_attn(self.input_layernorm(x), mask, cache)
        h = x + r
        r = self.mlp(self.post_attention_layernorm(h))
        out = h + r
        return out


class YoutuModel(nn.Module):
    def __init__(self, config: TextConfig):
        super().__init__()
        self.config = config
        self.vocab_size = config.vocab_size
        self.num_hidden_layers = config.num_hidden_layers
        self.embed_tokens = nn.Embedding(config.vocab_size, config.hidden_size)
        self.layers = [
            YoutuDecoderLayer(config=config, layer_idx=i)
            for i in range(config.num_hidden_layers)
        ]
        self.norm = nn.RMSNorm(config.hidden_size, eps=config.rms_norm_eps)

    def __call__(
        self,
        inputs: mx.array,
        inputs_embeds: Optional[mx.array] = None,
        mask: Optional[mx.array] = None,
        cache=None,
    ):
        if inputs_embeds is None:
            h = self.embed_tokens(inputs)
        else:
            h = inputs_embeds

        if cache is None:
            cache = [None] * len(self.layers)

        if mask is None:
            mask = create_attention_mask(h, cache[0], return_array=True)

        for layer, c in zip(self.layers, cache):
            h = layer(h, mask, c)

        return self.norm(h)


class LanguageModel(nn.Module):
    def __init__(self, config: TextConfig, model_config: ModelConfig = None):
        super().__init__()
        self.config = config
        self.model_config = model_config
        self.model_type = config.model_type
        self.model = YoutuModel(config)

        if not config.tie_word_embeddings:
            self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False)

    def __call__(
        self,
        inputs: mx.array,
        inputs_embeds: Optional[mx.array] = None,
        mask: Optional[mx.array] = None,
        cache=None,
        **kwargs,
    ):
        out = self.model(inputs, cache=cache, inputs_embeds=inputs_embeds, mask=mask)
        if self.config.tie_word_embeddings:
            out = self.model.embed_tokens.as_linear(out)
        else:
            out = self.lm_head(out)
        return LanguageModelOutput(logits=out)

    @property
    def layers(self):
        return self.model.layers

    @property
    def head_dim(self):
        return self.config.qk_nope_head_dim + self.config.qk_rope_head_dim

    @property
    def n_kv_heads(self):
        return self.config.num_key_value_heads