"""
model.py
TurkishLLM Generative V1 Model Architecture (52.5M Parameters)
Decoder-Only Transformer with SwiGLU, RoPE, RMSNorm, Weight Tying and Causal Self-Attention.
Custom PyTorch Architecture.
"""

import math
from typing import Optional, Tuple
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.utils.checkpoint


class RMSNorm(nn.Module):
    def __init__(self, dim: int, eps: float = 1e-5):
        super().__init__()
        self.eps = eps
        self.weight = nn.Parameter(torch.ones(dim))

    def forward(self, x: torch.Tensor) -> torch.Tensor:
        norm = torch.rsqrt(x.pow(2).mean(-1, keepdim=True) + self.eps)
        return x * norm * self.weight


class RotaryEmbedding(nn.Module):
    def __init__(self, dim: int, max_seq_len: int = 1024, base: float = 10000.0):
        super().__init__()
        self.dim = dim
        self.max_seq_len = max_seq_len
        self.base = base
        inv_freq = 1.0 / (base ** (torch.arange(0, dim, 2).float() / dim))
        self.register_buffer("inv_freq", inv_freq, persistent=False)

    def forward(self, x: torch.Tensor, seq_len: int) -> Tuple[torch.Tensor, torch.Tensor]:
        t = torch.arange(seq_len, device=x.device, dtype=self.inv_freq.dtype)
        freqs = torch.outer(t, self.inv_freq)
        emb = torch.cat((freqs, freqs), dim=-1)
        return emb.cos(), emb.sin()


def rotate_half(x: torch.Tensor) -> torch.Tensor:
    x1 = x[..., :x.shape[-1] // 2]
    x2 = x[..., x.shape[-1] // 2:]
    return torch.cat((-x2, x1), dim=-1)


def apply_rotary_pos_emb(q: torch.Tensor, k: torch.Tensor, cos: torch.Tensor, sin: torch.Tensor) -> Tuple[torch.Tensor, torch.Tensor]:
    cos = cos.unsqueeze(0).unsqueeze(1)
    sin = sin.unsqueeze(0).unsqueeze(1)
    q_embed = (q * cos) + (rotate_half(q) * sin)
    k_embed = (k * cos) + (rotate_half(k) * sin)
    return q_embed, k_embed


class CausalSelfAttention(nn.Module):
    def __init__(self, d_model: int = 576, num_heads: int = 9):
        super().__init__()
        assert d_model % num_heads == 0, f"d_model {d_model} must be divisible by num_heads {num_heads}"
        self.d_model = d_model
        self.num_heads = num_heads
        self.head_dim = d_model // num_heads  # 64
        self.q_proj = nn.Linear(d_model, d_model, bias=False)
        self.k_proj = nn.Linear(d_model, d_model, bias=False)
        self.v_proj = nn.Linear(d_model, d_model, bias=False)
        self.out_proj = nn.Linear(d_model, d_model, bias=False)

    def forward(self, x: torch.Tensor, cos: torch.Tensor, sin: torch.Tensor, attention_mask: Optional[torch.Tensor] = None) -> torch.Tensor:
        B, S, D = x.shape
        q = self.q_proj(x).view(B, S, self.num_heads, self.head_dim).transpose(1, 2)
        k = self.k_proj(x).view(B, S, self.num_heads, self.head_dim).transpose(1, 2)
        v = self.v_proj(x).view(B, S, self.num_heads, self.head_dim).transpose(1, 2)

        q, k = apply_rotary_pos_emb(q, k, cos, sin)

        if attention_mask is not None:
            if attention_mask.dim() == 2:
                mask = attention_mask[:, None, None, :].bool()
                out = F.scaled_dot_product_attention(q, k, v, attn_mask=mask, is_causal=False)
            else:
                out = F.scaled_dot_product_attention(q, k, v, attn_mask=attention_mask, is_causal=False)
        else:
            out = F.scaled_dot_product_attention(q, k, v, is_causal=True)

        out = out.transpose(1, 2).contiguous().view(B, S, D)
        return self.out_proj(out)


class SwiGLUFFN(nn.Module):
    def __init__(self, d_model: int = 576, ffn_dim: int = 1536):
        super().__init__()
        self.gate_proj = nn.Linear(d_model, ffn_dim, bias=False)
        self.up_proj = nn.Linear(d_model, ffn_dim, bias=False)
        self.down_proj = nn.Linear(ffn_dim, d_model, bias=False)

    def forward(self, x: torch.Tensor) -> torch.Tensor:
        return self.down_proj(F.silu(self.gate_proj(x)) * self.up_proj(x))


class TransformerBlock(nn.Module):
    def __init__(self, d_model: int = 576, num_heads: int = 9, ffn_dim: int = 1536):
        super().__init__()
        self.norm1 = RMSNorm(d_model)
        self.attn = CausalSelfAttention(d_model, num_heads)
        self.norm2 = RMSNorm(d_model)
        self.ffn = SwiGLUFFN(d_model, ffn_dim)

    def forward(self, x: torch.Tensor, cos: torch.Tensor, sin: torch.Tensor, attention_mask: Optional[torch.Tensor] = None) -> torch.Tensor:
        x = x + self.attn(self.norm1(x), cos, sin, attention_mask=attention_mask)
        x = x + self.ffn(self.norm2(x))
        return x


class TurkishLLMGenerativeV1(nn.Module):
    """
    TurkishLLM Generative V1 Model Architecture (52,508,736 Parameters)
    - Layers: 12
    - Hidden Dim (d_model): 576
    - Heads: 9 (Head Dim: 64)
    - FFN: SwiGLU (dim: 1536)
    - Max Context: 1024
    - Vocab Size (Embedding Capacity): 8192
    - Active Tokenizer Vocab Size: 4096
    - Normalization: RMSNorm (eps=1e-5)
    - Positional Embedding: RoPE (base=10000.0)
    - Weight Tying: lm_head.weight == token_embedding.weight
    """
    def __init__(
        self,
        vocab_size: int = 8192,
        d_model: int = 576,
        num_layers: int = 12,
        num_heads: int = 9,
        ffn_dim: int = 1536,
        max_seq_len: int = 1024
    ):
        super().__init__()
        self.vocab_size = vocab_size
        self.d_model = d_model
        self.num_layers = num_layers
        self.num_heads = num_heads
        self.ffn_dim = ffn_dim
        self.max_seq_len = max_seq_len

        self.token_embedding = nn.Embedding(vocab_size, d_model)
        self.rotary_emb = RotaryEmbedding(d_model // num_heads, max_seq_len=max_seq_len)

        self.blocks = nn.ModuleList([
            TransformerBlock(d_model, num_heads, ffn_dim)
            for _ in range(num_layers)
        ])

        self.norm = RMSNorm(d_model)
        self.lm_head = nn.Linear(d_model, vocab_size, bias=False)
        self.lm_head.weight = self.token_embedding.weight

        self.gradient_checkpointing = False
        self.apply(self._init_weights)

    def _init_weights(self, module):
        if isinstance(module, nn.Linear):
            torch.nn.init.normal_(module.weight, mean=0.0, std=0.02)
            if module.bias is not None:
                torch.nn.init.zeros_(module.bias)
        elif isinstance(module, nn.Embedding):
            torch.nn.init.normal_(module.weight, mean=0.0, std=0.02)

    def enable_gradient_checkpointing(self, enabled: bool = True):
        self.gradient_checkpointing = enabled

    def get_num_params(self) -> int:
        return sum(p.numel() for p in self.parameters())

    def forward(
        self,
        input_ids: torch.Tensor,
        labels: Optional[torch.Tensor] = None,
        attention_mask: Optional[torch.Tensor] = None
    ) -> Tuple[torch.Tensor, Optional[torch.Tensor]]:
        B, S = input_ids.shape
        assert S <= self.max_seq_len, f"Sequence length {S} exceeds maximum context {self.max_seq_len}"

        x = self.token_embedding(input_ids)
        cos, sin = self.rotary_emb(x, S)

        for block in self.blocks:
            if self.gradient_checkpointing and self.training:
                x = torch.utils.checkpoint.checkpoint(
                    block, x, cos, sin, attention_mask, use_reentrant=False
                )
            else:
                x = block(x, cos, sin, attention_mask=attention_mask)

        x = self.norm(x)
        logits = self.lm_head(x)

        loss = None
        if labels is not None:
            shift_logits = logits[..., :-1, :].contiguous()
            shift_labels = labels[..., 1:].contiguous()
            loss = F.cross_entropy(
                shift_logits.view(-1, self.vocab_size).float(),
                shift_labels.view(-1),
                ignore_index=-100
            )

        return logits, loss


def create_model() -> TurkishLLMGenerativeV1:
    """Instantiate and verify the exact 52.5M Generative V1 architecture."""
    model = TurkishLLMGenerativeV1()
    param_count = sum(p.numel() for p in model.parameters())
    assert param_count == 52_508_736, f"PARAM COUNT MISMATCH! Expected 52,508,736, got {param_count:,}"
    return model