|
import torch |
|
import torch.nn as nn |
|
import torch.nn.functional as F |
|
from torch.utils.data import Dataset, DataLoader |
|
from torch.optim.lr_scheduler import CosineAnnealingLR |
|
import math |
|
from tqdm import tqdm |
|
import json |
|
from tokenizers import Tokenizer |
|
from datetime import datetime |
|
import gc |
|
|
|
class GPTConfig: |
|
def __init__( |
|
self, |
|
vocab_size=22588, |
|
n_embd=768, |
|
n_head=12, |
|
n_layer=8, |
|
dropout=0.1, |
|
block_size=256, |
|
learning_rate=3e-4, |
|
max_epochs=50, |
|
batch_size=8, |
|
grad_clip=1.0, |
|
): |
|
self.vocab_size = vocab_size |
|
self.n_embd = n_embd |
|
self.n_head = n_head |
|
self.n_layer = n_layer |
|
self.dropout = dropout |
|
self.block_size = block_size |
|
self.learning_rate = learning_rate |
|
self.max_epochs = max_epochs |
|
self.batch_size = batch_size |
|
self.grad_clip = grad_clip |
|
|
|
|
|
class SelfAttention(nn.Module): |
|
def __init__(self, config): |
|
super().__init__() |
|
assert config.n_embd % config.n_head == 0 |
|
self.w_k = nn.Linear(config.n_embd, config.n_embd) |
|
self.w_q = nn.Linear(config.n_embd, config.n_embd) |
|
self.w_v = nn.Linear(config.n_embd, config.n_embd) |
|
self.attn_drop = nn.Dropout(config.dropout) |
|
self.resid_drop = nn.Dropout(config.dropout) |
|
self.proj = nn.Linear(config.n_embd, config.n_embd) |
|
self.n_head = config.n_head |
|
self.n_embd = config.n_embd |
|
|
|
def forward(self, x): |
|
B, T, C = x.size() |
|
k = self.w_k(x).view(B, T, self.n_head, C // self.n_head).transpose(1, 2) |
|
q = self.w_q(x).view(B, T, self.n_head, C // self.n_head).transpose(1, 2) |
|
v = self.w_v(x).view(B, T, self.n_head, C // self.n_head).transpose(1, 2) |
|
|
|
att = (q @ k.transpose(-2, -1)) * (1.0 / math.sqrt(k.size(-1))) |
|
att = F.softmax(att, dim=-1) |
|
att = self.attn_drop(att) |
|
y = att @ v |
|
y = y.transpose(1, 2).contiguous().view(B, T, C) |
|
y = self.resid_drop(self.proj(y)) |
|
return y |
|
|
|
class Block(nn.Module): |
|
def __init__(self, config): |
|
super().__init__() |
|
self.ln1 = nn.LayerNorm(config.n_embd) |
|
self.attn = SelfAttention(config) |
|
self.ln2 = nn.LayerNorm(config.n_embd) |
|
self.mlp = nn.Sequential( |
|
nn.Linear(config.n_embd, 4 * config.n_embd), |
|
nn.GELU(), |
|
nn.Linear(4 * config.n_embd, config.n_embd), |
|
nn.Dropout(config.dropout), |
|
) |
|
|
|
def forward(self, x): |
|
x = x + self.attn(self.ln1(x)) |
|
x = x + self.mlp(self.ln2(x)) |
|
return x |
|
|
|
class GPT(nn.Module): |
|
def __init__(self, config): |
|
super().__init__() |
|
self.tok_emb = nn.Embedding(config.vocab_size, config.n_embd) |
|
self.pos_emb = nn.Parameter(torch.zeros(1, config.block_size, config.n_embd)) |
|
self.drop = nn.Dropout(config.dropout) |
|
self.blocks = nn.ModuleList([Block(config) for _ in range(config.n_layer)]) |
|
self.ln_f = nn.LayerNorm(config.n_embd) |
|
self.head = nn.Linear(config.n_embd, config.vocab_size, bias=False) |
|
|
|
self.block_size = config.block_size |
|
self.apply(self._init_weights) |
|
|
|
def _init_weights(self, module): |
|
if isinstance(module, (nn.Linear, nn.Embedding)): |
|
module.weight.data.normal_(mean=0.0, std=0.02) |
|
if isinstance(module, nn.Linear) and module.bias is not None: |
|
module.bias.data.zero_() |
|
elif isinstance(module, nn.LayerNorm): |
|
module.bias.data.zero_() |
|
module.weight.data.fill_(1.0) |
|
|
|
def forward(self, idx, targets=None): |
|
b, t = idx.size() |
|
assert t <= self.block_size, f"Cannot forward sequence of length {t}, block size is only {self.block_size}" |
|
|
|
token_embeddings = self.tok_emb(idx) |
|
position_embeddings = self.pos_emb[:, :t, :] |
|
x = self.drop(token_embeddings + position_embeddings) |
|
for block in self.blocks: |
|
x = block(x) |
|
x = self.ln_f(x) |
|
logits = self.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 |
|
|
|
|
|
class WikiTextDataset(Dataset): |
|
def __init__(self, texts, tokenizer, max_length=256): |
|
self.tokenizer = tokenizer |
|
self.max_length = max_length |
|
|
|
print("Tokenizing texts...") |
|
self.examples = [] |
|
|
|
for text in tqdm(texts): |
|
tokens = self.tokenizer.encode(text).ids |
|
for i in range(0, len(tokens) - max_length, max_length // 2): |
|
chunk = tokens[i:i + max_length] |
|
if len(chunk) < max_length: |
|
chunk = chunk + [0] * (max_length - len(chunk)) |
|
self.examples.append(chunk) |
|
|
|
def __len__(self): |
|
return len(self.examples) |
|
|
|
def __getitem__(self, idx): |
|
tokens = self.examples[idx] |
|
return torch.tensor(tokens[:-1]), torch.tensor(tokens[1:]) |
|
|
|
def train(): |
|
|
|
torch.cuda.empty_cache() |
|
gc.collect() |
|
|
|
print("Loading Wikipedia data...") |
|
with open('az_wiki_data.json', 'r', encoding='utf-8') as f: |
|
wiki_data = json.load(f) |
|
|
|
texts = [page['text'] for page in wiki_data.values()] |
|
tokenizer = Tokenizer.from_file("az_tokenizer.json") |
|
|
|
dataset = WikiTextDataset(texts, tokenizer) |
|
train_size = int(0.9 * len(dataset)) |
|
val_size = len(dataset) - train_size |
|
train_dataset, val_dataset = torch.utils.data.random_split(dataset, [train_size, val_size]) |
|
|
|
config = GPTConfig() |
|
|
|
train_loader = DataLoader( |
|
train_dataset, |
|
batch_size=config.batch_size, |
|
shuffle=True, |
|
num_workers=2, |
|
pin_memory=True |
|
) |
|
|
|
val_loader = DataLoader( |
|
val_dataset, |
|
batch_size=config.batch_size, |
|
shuffle=False, |
|
num_workers=2, |
|
pin_memory=True |
|
) |
|
|
|
model = GPT(config) |
|
model = model.to('cuda') |
|
print(f"Number of parameters: {sum(p.numel() for p in model.parameters())/1e6:.2f}M") |
|
|
|
optimizer = torch.optim.AdamW(model.parameters(), lr=config.learning_rate) |
|
scheduler = CosineAnnealingLR(optimizer, T_max=config.max_epochs) |
|
scaler = torch.amp.GradScaler() |
|
|
|
def run_epoch(split, epoch_num=0): |
|
is_train = split == 'train' |
|
model.train(is_train) |
|
if not is_train: |
|
model.eval() |
|
|
|
loader = train_loader if is_train else val_loader |
|
losses = [] |
|
|
|
pbar = tqdm(enumerate(loader), total=len(loader)) if is_train else enumerate(loader) |
|
|
|
for it, (x, y) in pbar: |
|
|
|
torch.cuda.empty_cache() |
|
|
|
x = x.to('cuda', non_blocking=True) |
|
y = y.to('cuda', non_blocking=True) |
|
|
|
with torch.amp.autocast(device_type='cuda'): |
|
logits, loss = model(x, y) |
|
|
|
losses.append(loss.item()) |
|
|
|
if is_train: |
|
scaler.scale(loss).backward() |
|
scaler.unscale_(optimizer) |
|
torch.nn.utils.clip_grad_norm_(model.parameters(), config.grad_clip) |
|
scaler.step(optimizer) |
|
scaler.update() |
|
optimizer.zero_grad(set_to_none=True) |
|
|
|
pbar.set_description(f"epoch {epoch_num+1} iter {it}: train loss {loss.item():.5f}") |
|
|
|
|
|
del x, y, logits |
|
if is_train: |
|
del loss |
|
|
|
mean_loss = torch.tensor(losses).mean().item() |
|
return mean_loss |
|
|
|
best_val_loss = float('inf') |
|
|
|
try: |
|
for epoch in range(config.max_epochs): |
|
print(f"\nEpoch {epoch+1}/{config.max_epochs}") |
|
|
|
train_loss = run_epoch('train', epoch_num=epoch) |
|
|
|
with torch.no_grad(): |
|
val_loss = run_epoch('val') |
|
|
|
scheduler.step() |
|
|
|
if val_loss < best_val_loss: |
|
best_val_loss = val_loss |
|
print(f"Saving best model with val_loss: {val_loss:.4f}") |
|
torch.save(model.state_dict(), 'best_model.pt') |
|
|
|
print(f"Epoch {epoch+1}: train_loss: {train_loss:.4f}, val_loss: {val_loss:.4f}") |
|
|
|
if (epoch + 1) % 5 == 0: |
|
torch.save({ |
|
'epoch': epoch, |
|
'model_state_dict': model.state_dict(), |
|
'optimizer_state_dict': optimizer.state_dict(), |
|
'scheduler_state_dict': scheduler.state_dict(), |
|
'train_loss': train_loss, |
|
'val_loss': val_loss, |
|
}, f'checkpoint_epoch_{epoch+1}.pt') |
|
|
|
except KeyboardInterrupt: |
|
print('Training interrupted, saving checkpoint...') |
|
torch.save({ |
|
'epoch': epoch, |
|
'model_state_dict': model.state_dict(), |
|
'optimizer_state_dict': optimizer.state_dict(), |
|
'scheduler_state_dict': scheduler.state_dict(), |
|
'train_loss': train_loss, |
|
'val_loss': val_loss, |
|
}, 'interrupt_checkpoint.pt') |
|
|
|
if __name__ == '__main__': |
|
train() |