"""
model.py — A tiny GPT (Transformer) written FROM SCRATCH.

Nothing here is downloaded or pretrained. Every layer below is built
by hand from basic math operations. The only import is `torch`, which
is a math library (matrix multiply, softmax, gradients) — not a model.

This is exactly the architecture from the diagram:
  embedding + position  ->  N blocks of (self-attention + feed-forward)  ->  output head
"""

import math
import torch
import torch.nn as nn
from torch.nn import functional as F


class CausalSelfAttention(nn.Module):
    """Self-attention: each token looks at earlier tokens and mixes their meaning."""

    def __init__(self, n_embd, n_head, block_size, dropout):
        super().__init__()
        assert n_embd % n_head == 0
        self.n_head = n_head
        self.n_embd = n_embd
        # One linear layer produces Query, Key and Value together.
        self.qkv = nn.Linear(n_embd, 3 * n_embd, bias=False)
        self.proj = nn.Linear(n_embd, n_embd, bias=False)
        self.attn_drop = nn.Dropout(dropout)
        self.resid_drop = nn.Dropout(dropout)
        # A mask so a token can only attend to itself and tokens BEFORE it.
        self.register_buffer(
            "mask",
            torch.tril(torch.ones(block_size, block_size)).view(1, 1, block_size, block_size),
        )

    def forward(self, x):
        B, T, C = x.shape
        q, k, v = self.qkv(x).split(self.n_embd, dim=2)
        hs = C // self.n_head
        # split into multiple heads
        q = q.view(B, T, self.n_head, hs).transpose(1, 2)
        k = k.view(B, T, self.n_head, hs).transpose(1, 2)
        v = v.view(B, T, self.n_head, hs).transpose(1, 2)
        # attention scores = Query . Key, scaled
        att = (q @ k.transpose(-2, -1)) / math.sqrt(hs)
        att = att.masked_fill(self.mask[:, :, :T, :T] == 0, float("-inf"))
        att = F.softmax(att, dim=-1)          # turn scores into weights that sum to 1
        att = self.attn_drop(att)
        y = att @ v                            # mix the Values by those weights
        y = y.transpose(1, 2).contiguous().view(B, T, C)
        return self.resid_drop(self.proj(y))


class MLP(nn.Module):
    """The feed-forward network: think about each token on its own."""

    def __init__(self, n_embd, dropout):
        super().__init__()
        self.fc = nn.Linear(n_embd, 4 * n_embd)
        self.proj = nn.Linear(4 * n_embd, n_embd)
        self.drop = nn.Dropout(dropout)

    def forward(self, x):
        return self.drop(self.proj(F.gelu(self.fc(x))))


class Block(nn.Module):
    """One Transformer block = attention + feed-forward (this gets repeated N times)."""

    def __init__(self, n_embd, n_head, block_size, dropout):
        super().__init__()
        self.ln1 = nn.LayerNorm(n_embd)
        self.attn = CausalSelfAttention(n_embd, n_head, block_size, dropout)
        self.ln2 = nn.LayerNorm(n_embd)
        self.mlp = MLP(n_embd, dropout)

    def forward(self, x):
        x = x + self.attn(self.ln1(x))   # the "+" is a residual connection
        x = x + self.mlp(self.ln2(x))
        return x


class GPT(nn.Module):
    def __init__(self, vocab_size, block_size, n_embd=192, n_head=6, n_layer=6, dropout=0.1):
        super().__init__()
        self.block_size = block_size
        self.tok_emb = nn.Embedding(vocab_size, n_embd)   # token -> vector
        self.pos_emb = nn.Embedding(block_size, n_embd)   # position -> vector
        self.drop = nn.Dropout(dropout)
        self.blocks = nn.ModuleList(
            [Block(n_embd, n_head, block_size, dropout) for _ in range(n_layer)]
        )
        self.ln_f = nn.LayerNorm(n_embd)
        self.head = nn.Linear(n_embd, vocab_size, bias=False)  # vector -> score per character
        self.apply(self._init)

    def _init(self, m):
        if isinstance(m, (nn.Linear, nn.Embedding)):
            nn.init.normal_(m.weight, mean=0.0, std=0.02)
            if isinstance(m, nn.Linear) and m.bias is not None:
                nn.init.zeros_(m.bias)

    def forward(self, idx, targets=None):
        B, T = idx.shape
        pos = torch.arange(T, device=idx.device)
        x = self.drop(self.tok_emb(idx) + self.pos_emb(pos))
        for b in self.blocks:
            x = b(x)
        x = self.ln_f(x)
        logits = self.head(x)
        loss = None
        if targets is not None:
            loss = F.cross_entropy(logits.view(-1, logits.size(-1)), targets.view(-1))
        return logits, loss

    @torch.no_grad()
    def generate(self, idx, max_new_tokens, temperature=0.8, top_k=None):
        """Predict one character at a time, feeding each back in (autoregressive)."""
        for _ in range(max_new_tokens):
            idx_cond = idx[:, -self.block_size:]
            logits, _ = self(idx_cond)
            logits = logits[:, -1, :] / temperature
            if top_k is not None:
                v, _ = torch.topk(logits, min(top_k, logits.size(-1)))
                logits[logits < v[:, [-1]]] = float("-inf")
            probs = F.softmax(logits, dim=-1)
            nxt = torch.multinomial(probs, num_samples=1)
            idx = torch.cat([idx, nxt], dim=1)
        return idx