import torch
import torch.nn as nn
import torch.nn.functional as F


class CausalSelfAttention(nn.Module):
    def __init__(self, n_embd, n_head):
        super().__init__()
        assert n_embd % n_head == 0
        self.n_head = n_head
        self.head_dim = n_embd // n_head
        self.c_attn = nn.Linear(n_embd, 3 * n_embd, bias=False)
        self.c_proj = nn.Linear(n_embd, n_embd, bias=False)

    def forward(self, x):
        B, T, C = x.size()
        q, k, v = self.c_attn(x).split(self.head_dim * self.n_head, dim=2)
        q = q.view(B, T, self.n_head, self.head_dim).transpose(1, 2)
        k = k.view(B, T, self.n_head, self.head_dim).transpose(1, 2)
        v = v.view(B, T, self.n_head, self.head_dim).transpose(1, 2)
        att = F.scaled_dot_product_attention(q, k, v, is_causal=True)
        att = att.transpose(1, 2).contiguous().view(B, T, C)
        return self.c_proj(att)


class MLP(nn.Module):
    def __init__(self, n_embd, n_inner):
        super().__init__()
        self.c_fc = nn.Linear(n_embd, n_inner, bias=False)
        self.c_proj = nn.Linear(n_inner, n_embd, bias=False)

    def forward(self, x):
        return self.c_proj(F.gelu(self.c_fc(x)))


class Block(nn.Module):
    def __init__(self, n_embd, n_head):
        super().__init__()
        self.ln_1 = nn.LayerNorm(n_embd)
        self.attn = CausalSelfAttention(n_embd, n_head)
        self.ln_2 = nn.LayerNorm(n_embd)
        self.mlp = MLP(n_embd, 4 * n_embd)

    def forward(self, x):
        x = x + self.attn(self.ln_1(x))
        x = x + self.mlp(self.ln_2(x))
        return x


class CharGPT(nn.Module):
    """Character-level causal transformer (nanoGPT-style). No bias in
    attention / FFN / lm_head; LayerNorm carries the affine bias."""

    def __init__(self, vocab_size, block_size, n_layer, n_head, n_embd):
        super().__init__()
        self.block_size = block_size
        self.transformer = nn.ModuleDict({
            "wte": nn.Embedding(vocab_size, n_embd),
            "wpe": nn.Embedding(block_size, n_embd),
            "drop": nn.Dropout(0.0),
            "h": nn.ModuleList([Block(n_embd, n_head) for _ in range(n_layer)]),
            "ln_f": nn.LayerNorm(n_embd),
        })
        self.lm_head = nn.Linear(n_embd, vocab_size, bias=False)

    def forward(self, idx, targets=None):
        B, T = idx.size()
        assert T <= self.block_size, f"block size {self.block_size} < {T}"
        pos = torch.arange(0, T, device=idx.device)
        x = self.transformer["drop"](
            self.transformer["wte"](idx) + self.transformer["wpe"](pos))
        for block in self.transformer["h"]:
            x = block(x)
        x = self.transformer["ln_f"](x)
        logits = self.lm_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=40):
        for _ in range(max_new_tokens):
            idx_cond = idx[:, -self.block_size:]
            logits, _ = self(idx_cond)
            logits = logits[:, -1, :] / temperature
            if top_k:
                v, _ = torch.topk(logits, min(top_k, logits.size(-1)))
                logits[logits < v[:, [-1]]] = float("-inf")
            probs = F.softmax(logits, dim=-1)
            idx = torch.cat((idx, torch.multinomial(probs, num_samples=1)), dim=1)
        return idx


def from_config(config):
    return CharGPT(
        vocab_size=config["vocab_size"],
        block_size=config["block_size"],
        n_layer=config["n_layer"],
        n_head=config["n_head"],
        n_embd=config["n_embd"],
    )


if __name__ == "__main__":
    import json
    cfg = json.load(open("config.json"))
    m = from_config(cfg)
    print("params:", sum(p.numel() for p in m.parameters()))
    # tiny smoke test
    x = torch.randint(0, cfg["vocab_size"], (1, 32))
    logits, loss = m(x, x)
    print("smoke loss:", loss.item())