model.py

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


class CausalSelfAttention(nn.Module):

    def __init__(self):
        super().__init__()
        # key, query, value projections for all heads, but in a batch
        self.c_attn = nn.Linear(n_embd, 3 * n_embd, bias=False)
        # output projection
        self.c_proj = nn.Linear(n_embd, n_embd, bias=False)
        # regularization
        self.attn_dropout = nn.Dropout(dropout)
        self.resid_dropout = nn.Dropout(dropout)
        # causal mask
        self.register_buffer(
            "tril",
            torch.tril(torch.ones(block_size, block_size)).view(
                1, 1, block_size, block_size
            ),
        )

    def forward(self, x):
        # batch size, sequence length, embedding dimensionality (n_embd)
        B, T, C = x.size()

        # calculate query, key, values for all heads in batch and move head forward to be the batch dim
        q, k, v = self.c_attn(x).split(n_embd, dim=2)
        k = k.view(B, T, n_head, C // n_head).transpose(1, 2)  # (B, nh, T, hs)
        q = q.view(B, T, n_head, C // n_head).transpose(1, 2)  # (B, nh, T, hs)
        v = v.view(B, T, n_head, C // n_head).transpose(1, 2)  # (B, nh, T, hs)

        # causal self-attention; Self-attend: (B, nh, T, hs) x (B, nh, hs, T) -> (B, nh, T, T)
        att = (q @ k.transpose(-2, -1)) * (1.0 / math.sqrt(k.size(-1)))
        att = att.masked_fill(self.tril[:, :, :T, :T] == 0, float("-inf"))
        att = F.softmax(att, dim=-1)
        att = self.attn_dropout(att)

        # (B, nh, T, T) x (B, nh, T, hs) -> (B, nh, T, hs)
        y = att @ v

        # re-assemble all head outputs side by side
        y = y.transpose(1, 2).contiguous().view(B, T, C)

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


class FeedForward(nn.Module):
    """a simple linear layer followed by a non-linearity"""

    def __init__(self, n_embd):
        super().__init__()
        self.net = nn.Sequential(
            nn.Linear(n_embd, 4 * n_embd),
            nn.ReLU(),
            nn.Linear(4 * n_embd, n_embd),
            nn.Dropout(dropout),
        )

    def forward(self, x):
        return self.net(x)


class Block(nn.Module):
    """Transformer block: communication followed by computation"""

    def __init__(self, n_embd):
        super().__init__()
        self.ln1 = nn.LayerNorm(n_embd, bias=False)
        self.attn = CausalSelfAttention()
        self.ln2 = nn.LayerNorm(n_embd, bias=False)
        self.ffwd = FeedForward(n_embd)

    def forward(self, x):
        x = x + self.attn(self.ln1(x))
        x = x + self.ffwd(self.ln2(x))
        return x


class GPT(nn.Module):

    def __init__(self):
        super().__init__()
        self.token_embedding_table = nn.Embedding(vocab_size, n_embd)
        self.position_embedding_table = nn.Embedding(block_size, n_embd)
        self.blocks = nn.Sequential(*[Block(n_embd) for _ in range(n_layer)])
        self.ln_f = nn.LayerNorm(n_embd)
        self.lm_head = nn.Linear(n_embd, vocab_size)

    def forward(self, context, targets=None):
        device = context.device
        B, T = context.shape

        tok_emb = self.token_embedding_table(context)  # (B, T, C)
        pos_emb = self.position_embedding_table(
            torch.arange(T, device=device)
        )  # (T, C)
        x = tok_emb + pos_emb  # (B, T, C)
        x = self.blocks(x)
        x = self.ln_f(x)
        logits = self.lm_head(x)  # (B, T, vocab_size)

        if targets is None:
            loss = None
        else:
            B, T, C = logits.shape
            logits = logits.view(B * T, C)
            targets = targets.view(B * T)
            loss = F.cross_entropy(logits, targets)

        return logits, loss

    @torch.no_grad()
    def generate(self, context, max_new_tokens, temperature=1.0):
        # context is a (B, T) array of indices in the current context
        for _ in range(max_new_tokens):
            # crop context to the last block_size tokens
            context_cropped = context[:, -block_size:]
            # get the predictions
            logits, _ = self(context_cropped)
            # pluck the logits at the last time step and scale by desired temperature
            logits = logits[:, -1, :] / temperature
            # apply softmax to get probabilites
            probs = F.softmax(logits, dim=-1)
            # sample from the distribution
            next_char = torch.multinomial(probs, num_samples=1)
            # append sampled index to the running sequence
            context = torch.cat((context, next_char), dim=1)

        return context