import torch
import torch.nn as nn
import math

class PositionalEncoding(nn.Module):
    def __init__(self, d_model, max_len=8192):
        super().__init__()
        pe = torch.zeros(max_len, d_model)
        position = torch.arange(0, max_len).unsqueeze(1).float()
        div_term = torch.exp(torch.arange(0, d_model, 2).float() * -(math.log(10000.0) / d_model))
        pe[:, 0::2] = torch.sin(position * div_term)
        pe[:, 1::2] = torch.cos(position * div_term)
        self.register_buffer("pe", pe)

    def forward(self, x):
        t = x.size(1)
        return x + self.pe[:t].unsqueeze(0)

class SmallCodeTransformer(nn.Module):
    def __init__(self, vocab_size, d_model=512, nhead=8, nlayers=6, dim_feed=2048, max_len=8192):
        super().__init__()
        self.token_emb = nn.Embedding(vocab_size, d_model)
        self.pos = PositionalEncoding(d_model, max_len)
        encoder_layer = nn.TransformerEncoderLayer(d_model, nhead, dim_feed, dropout=0.1, activation="gelu")
        self.encoder = nn.TransformerEncoder(encoder_layer, nlayers)
        self.ln = nn.LayerNorm(d_model)
        self.head = nn.Linear(d_model, vocab_size, bias=False)
        self._init_weights()

    def _init_weights(self):
        nn.init.normal_(self.token_emb.weight, mean=0.0, std=0.02)
        nn.init.normal_(self.head.weight, mean=0.0, std=0.02)

    def forward(self, input_ids, attention_mask=None):
        x = self.token_emb(input_ids)
        x = self.pos(x)
        x = x.permute(1,0,2)
        x = self.encoder(x, src_key_padding_mask=(attention_mask==0) if attention_mask is not None else None)
        x = x.permute(1,0,2)
        x = self.ln(x)
        return self.head(x)