minigpt.py

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

# hyperparameters
batch_size = 32  # how many independent sequences will we process in parallel?
block_size = 32  # what is the maximum context length for predictions?
max_iters = 5000
eval_interval = 100
learning_rate = 1e-3
device = "cuda" if torch.cuda.is_available() else "cpu"
eval_iters = 200
n_embd = 64
n_head = 4
n_layer = 4
dropout = 0.20
# ------------

torch.manual_seed(1337)

with open("./data/shakespeare.txt", "r", encoding="utf-8") as f:
    text = f.read()

# here are all the unique characters that occur in this text
chars = sorted(list(set(text)))
vocab_size = len(chars)
# create a mapping from characters to integers
stoi = {ch: i for i, ch in enumerate(chars)}
itos = {i: ch for i, ch in enumerate(chars)}
encode = lambda s: [
    stoi[c] for c in s
]  # encoder: take a string, output a list of integers
decode = lambda l: "".join(
    [itos[i] for i in l]
)  # decoder: take a list of integers, output a string

# Train and test splits
data = torch.tensor(encode(text), dtype=torch.long)
n = int(0.9 * len(data))  # first 90% will be train, rest val
train_data = data[:n]
val_data = data[n:]


# data loading
def get_batch(split):
    # generate a small batch of data of inputs x and targets y
    data = train_data if split == "train" else val_data
    ix = torch.randint(len(data) - block_size, (batch_size,))
    x = torch.stack([data[i : i + block_size] for i in ix])
    y = torch.stack([data[i + 1 : i + block_size + 1] for i in ix])
    x, y = x.to(device), y.to(device)
    return x, y


class DummyModel(nn.Module):
    def __init__(self, vocab_sizes):
        super().__init__()
        self.embedding_layer = nn.Embedding(vocab_size, vocab_size)

    def forward(self, x, targets=None):
        logits = self.embedding_layer(x)

        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


# Estimate loss function
# For train and eval split (output a dict)
# Set model to eval and then back to train
# Evaluate loss over multiple batch attempts
# model is global
@torch.no_grad()
def estimate_loss():
    loss = {}
    splits = ["train", "val"]
    model.eval()
    for split in splits:
        losses = []
        for _ in range(eval_iters):
            x, y = get_batch(split)
            _, l = model(x, y)
            losses.append(l)
        losses = torch.tensor(losses)
        loss[split] = torch.mean(losses).item()
        model.train()
    return loss


# Build a head of self-attention model
# Needs to have the key, query and value matrix
# Register a triangular 1 matrix as a buffer (to preserve sequential info)
# Standardize with softmax and use dropout layer on the wei matrix


class Head(nn.Module):
    def __init__(self, head_size):
        super().__init__()
        self.query = nn.Linear(n_embd, head_size)
        self.key = nn.Linear(n_embd, head_size)
        self.value = nn.Linear(n_embd, head_size)
        self.register_buffer("tril", torch.tril(torch.ones((block_size, block_size))))

        self.dropout = nn.Dropout(dropout)

    def forward(self, x):
        B, T, C = x.shape
        query = self.query(x)
        key = self.key(x)
        wei = query @ key.transpose(-2, -1) * C**0.5
        wei = wei.masked_fill(self.tril[:T, :T] == 0, float("-inf"))
        wei = F.softmax(wei, -1)
        wei = self.dropout(wei)
        value = self.value(x)
        out = wei @ value
        return out


# Create multi-head module
# Has a list of heads layer
# Forward concatenate the heads and then dropout on (projection of x)
class MultiHead(nn.Module):
    def __init__(self, n_head, head_size):
        super().__init__()
        self.heads = nn.ModuleList([Head(head_size) for _ in range(n_head)])
        self.proj = nn.Linear(n_embd, n_embd)
        self.dropout = nn.Dropout(dropout)

    def forward(self, x):
        x = torch.cat([head(x) for head in self.heads], dim=-1)
        x = self.dropout(self.proj(x))
        return x


# Implement two layers feedfoward with non linearity (relu in middle)
# + dropout at the hand (Sequential net)
class FeedForward(nn.Module):
    def __init__(self, n_embd):
        super().__init__()
        self.net = nn.Sequential(
            nn.Linear(n_embd, n_embd * 4),
            nn.ReLU(),
            nn.Linear(n_embd * 4, n_embd),
            nn.Dropout(),
        )

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

        return x


# Assemble the war beast
# The block
# head_size: head_size * n_head = n_embd
# The Multi-Head
# The ff net
# Two normalization layer
# The original embedding data is passed to along the
# multi-head/ffnet output
class Block(nn.Module):
    def __init__(self, n_head, n_embd):
        super().__init__()
        head_size = n_embd // n_head
        self.mh = MultiHead(n_head, head_size)
        self.ffwd = FeedForward(n_embd)
        self.ln1 = nn.LayerNorm(n_embd)
        self.ln2 = nn.LayerNorm(n_embd)

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

        return x


# BabyPt model
# one embedding table
# one positional table
# n_layer Blocks
# layer norm again
# Linear output layer

# For generation
# Logit of last time step
# Crop provided idx to block size
# Last step concat the predicted idx recursively
# So forward and generate methods


class BabyPT(nn.Module):
    def __init__(self, n_embd):
        super().__init__()
        self.token_embedding = nn.Embedding(vocab_size, n_embd)
        self.positional_embedding = nn.Embedding(block_size, n_embd)
        self.blocks = nn.Sequential(*[Block(n_head, n_embd) for _ in range(n_layer)])
        self.ln = nn.LayerNorm(n_embd)
        self.linear_head = nn.Linear(n_embd, vocab_size)

    def forward(self, x, targets=None):
        x = self.token_embedding(x)
        x = x + self.positional_embedding(
            torch.tensor([T for T in range(x.shape[1])]).to(device)
        )
        x = self.blocks(x)
        x = self.ln(x)
        x = self.linear_head(x)

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

        return x, loss

    def generate(self, idx, max_token):
        for _ in range(max_token):
            idx_cond = idx[:, -block_size:]
            x, _ = self.forward(idx_cond)
            # Last available time-step
            x = x[:, -1, :]
            probs = torch.softmax(x, dim=-1)
            idx_new = torch.multinomial(probs, num_samples=1)
            idx = torch.cat((idx, idx_new), dim=-1)

        return decode(idx.cpu().numpy().ravel())


# Gotta add a training routine now
if __name__ == "__main__":
    x, y = get_batch("train")
    x, y = x.to(device), y.to(device)

    model = BabyPT(n_embd).to(device)
    optimizer = torch.optim.AdamW(model.parameters(), lr=2e-4)

    n_epochs = 50000

    for epoch in range(n_epochs):
        logit, loss = model(x, y)

        optimizer.zero_grad()
        loss.backward()
        optimizer.step()

        if epoch % 100 == 0:
            loss, current = loss.item(), epoch
            print(f"loss: {loss:>7f},  {epoch}")

            y_hat = model.eval().generate(
                torch.tensor([15]).to(device).reshape(1, 1), 500
            )
            model.train()

            print(y_hat)