Nanogpt

# NanoGPT Architecture

## Model Structure (~300 Lines)

NanoGPT implements a clean GPT-2 architecture in minimal code for educational purposes.

### Complete Model (model.py)

```python
import torch
import torch.nn as nn
from torch.nn import functional as F

class CausalSelfAttention(nn.Module):
    """Multi-head masked self-attention layer."""

    def __init__(self, config):
        super().__init__()
        assert config.n_embd % config.n_head == 0

        # Key, query, value projections for all heads (batched)
        self.c_attn = nn.Linear(config.n_embd, 3 * config.n_embd, bias=config.bias)
        # Output projection
        self.c_proj = nn.Linear(config.n_embd, config.n_embd, bias=config.bias)

        # Regularization
        self.attn_dropout = nn.Dropout(config.dropout)
        self.resid_dropout = nn.Dropout(config.dropout)

        self.n_head = config.n_head
        self.n_embd = config.n_embd
        self.dropout = config.dropout

        # Flash attention flag
        self.flash = hasattr(torch.nn.functional, 'scaled_dot_product_attention')

        if not self.flash:
            # Causal mask (lower triangular)
            self.register_buffer("bias", torch.tril(
                torch.ones(config.block_size, config.block_size)
            ).view(1, 1, config.block_size, config.block_size))

    def forward(self, x):
        B, T, C = x.size()  # batch, seq_len, embedding_dim

        # Calculate Q, K, V for all heads in batch
        q, k, v = self.c_attn(x).split(self.n_embd, dim=2)

        # Reshape for multi-head attention
        k = k.view(B, T, self.n_head, C // self.n_head).transpose(1, 2)  # (B, nh, T, hs)
        q = q.view(B, T, self.n_head, C // self.n_head).transpose(1, 2)  # (B, nh, T, hs)
        v = v.view(B, T, self.n_head, C // self.n_head).transpose(1, 2)  # (B, nh, T, hs)

        # Attention
        if self.flash:
            # Flash Attention (PyTorch 2.0+)
            y = torch.nn.functional.scaled_dot_product_attention(
                q, k, v,
                attn_mask=None,
                dropout_p=self.dropout if self.training else 0,
                is_causal=True
            )
        else:
            # Manual attention implementation
            att = (q @ k.transpose(-2, -1)) * (1.0 / math.sqrt(k.size(-1)))
            att = att.masked_fill(self.bias[:, :, :T, :T] == 0, float('-inf'))
            att = F.softmax(att, dim=-1)
            att = self.attn_dropout(att)
            y = att @ v  # (B, nh, T, hs)

        # Reassemble all head outputs
        y = y.transpose(1, 2).contiguous().view(B, T, C)

        # Output projection
        y = self.resid_dropout(self.c_proj(y))
        return y


class MLP(nn.Module):
    """Feedforward network (2-layer with GELU activation)."""

    def __init__(self, config):
        super().__init__()
        self.c_fc = nn.Linear(config.n_embd, 4 * config.n_embd, bias=config.bias)
        self.gelu = nn.GELU()
        self.c_proj = nn.Linear(4 * config.n_embd, config.n_embd, bias=config.bias)
        self.dropout = nn.Dropout(config.dropout)

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


class Block(nn.Module):
    """Transformer block (attention + MLP with residuals)."""

    def __init__(self, config):
        super().__init__()
        self.ln_1 = nn.LayerNorm(config.n_embd)
        self.attn = CausalSelfAttention(config)
        self.ln_2 = nn.LayerNorm(config.n_embd)
        self.mlp = MLP(config)

    def forward(self, x):
        x = x + self.attn(self.ln_1(x))  # Pre-norm + residual
        x = x + self.mlp(self.ln_2(x))   # Pre-norm + residual
        return x


@dataclass
class GPTConfig:
    """GPT model configuration."""
    block_size: int = 1024    # Max sequence length
    vocab_size: int = 50304   # GPT-2 vocab size (50257 rounded up for efficiency)
    n_layer: int = 12         # Number of layers
    n_head: int = 12