File size: 8,210 Bytes
9ff262b |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 |
import os
import math
import pickle
import torch
import torch.nn as nn
import torch.optim as optim
import wandb
from dataset import get_dataloader
from model import GreyMatter
# prepare dataset
# data.txt -> train.txt, val.txt
def prepare_train_val_split(data_path="fineweb_subset.txt", train_percent=0.9):
# train.txt, val.txt exists skip
path = os.path.dirname(data_path)
train_path = os.path.join(path, "train.txt")
val_path = os.path.join(path, "val.txt")
if not os.path.exists(train_path) and not os.path.exists(val_path):
with open(data_path, "r", encoding="utf-8") as file:
text = file.read()
train_data = text[:int(len(text)*train_percent)]
val_data = text[int(len(text)*train_percent):]
with open(train_path, "w") as file:
file.write(train_data)
with open(val_path, "w") as file:
file.write(val_data)
print("Train Validation split complete")
else:
print("Train and Validation files exists, skipping split")
# return train_path, val_path
return train_path, val_path
def calculate_perplexity(loss):
"""Calculate perplexity from cross-entropy loss"""
return math.exp(loss)
def train(model, train_loader, val_loader, config, warmup_steps=1000):
device = config["device"]
model = model.to(device)
grad_accum_steps = config.get("grad_acc_step", 1)
optimizer = optim.AdamW(model.parameters(), lr=config["learning_rate"], weight_decay=config['weight_decay'])
criterion = nn.CrossEntropyLoss(ignore_index=0)
total_steps = len(train_loader) // grad_accum_steps * config["num_epochs"]
# Cosine LR scheduler with warmup
def lr_lambda(step):
if step < warmup_steps:
return step / warmup_steps
progress = (step - warmup_steps) / max(1, total_steps - warmup_steps)
return 0.5 * (1.0 + math.cos(math.pi * progress))
scheduler = torch.optim.lr_scheduler.LambdaLR(optimizer, lr_lambda)
train_losses, val_losses = [], []
train_perplexities, val_perplexities = [], []
global_step = 0
print(f"Starting training for {config['num_epochs']} epochs")
print(f"Total steps: {total_steps}, Warmup steps: {warmup_steps}")
print("-" * 60)
for epoch in range(config["num_epochs"]):
model.train()
epoch_train_loss = 0
scaled_loss = 0
num_batches = 0
optimizer.zero_grad()
for batch_idx, batch_tokens in enumerate(train_loader):
input_ids = batch_tokens[0].to(device)
targets = batch_tokens[1].to(device)
logits = model(input_ids)
loss = criterion(logits.view(-1, logits.size(-1)), targets.view(-1))
scaled_loss = loss / grad_accum_steps
scaled_loss.backward()
if (batch_idx + 1) % grad_accum_steps == 0:
torch.nn.utils.clip_grad_norm_(model.parameters(), 1.0)
optimizer.step()
scheduler.step() # update LR
optimizer.zero_grad()
global_step += 1
epoch_train_loss += loss.item()
num_batches += 1
# Print progress with perplexity
if batch_idx % 50 == 0:
current_perplexity = calculate_perplexity(loss.item())
current_lr = scheduler.get_last_lr()[0]
print(f"Epoch {epoch+1}/{config['num_epochs']}, Step {batch_idx}, "
f"Loss: {loss.item():.4f}, Perplexity: {current_perplexity:.2f}, "
f"LR: {current_lr:.6f}")
wandb.log({
"train/epoch": (epoch+1)/config['num_epochs'],
"train/step": batch_idx,
"train/loss": float(f"{loss.item():.4f}"),
"train/learning_rate": float(f"{current_lr:.6f}"),
"train/perplexity": float(f"{current_perplexity:.2f}")
})
average_train_loss = epoch_train_loss / num_batches
train_perplexity = calculate_perplexity(average_train_loss)
# validation
model.eval()
epoch_val_loss = 0
val_batches = 0
print("Running validation...")
with torch.no_grad():
for val_tokens in val_loader:
input_ids = val_tokens[0].to(device)
targets = val_tokens[1].to(device)
logits = model(input_ids)
loss = criterion(logits.view(-1, logits.size(-1)), targets.view(-1))
epoch_val_loss += loss.item()
val_batches += 1
average_val_loss = epoch_val_loss / val_batches
val_perplexity = calculate_perplexity(average_val_loss)
# Store metrics
train_losses.append(average_train_loss)
val_losses.append(average_val_loss)
train_perplexities.append(train_perplexity)
val_perplexities.append(val_perplexity)
# Print epoch summary
print("-" * 60)
print(f"EPOCH {epoch+1} SUMMARY:")
print(f"Train Loss: {average_train_loss:.4f} | Train Perplexity: {train_perplexity:.2f}")
print(f"Val Loss: {average_val_loss:.4f} | Val Perplexity: {val_perplexity:.2f}")
print(f"Global Step: {global_step}")
print("-" * 60)
wandb.log({
"val/epoch": epoch+1,
"val/loss": float(f"{average_val_loss:.4f}"),
"val/perplexity": float(f"{val_perplexity:.2f}"),
"val/global_step": global_step
})
if (epoch + 1) % 2 == 0:
checkpoint_path = f'checkpoint_epoch_{epoch+1}.pt'
torch.save({
'epoch': epoch + 1,
'model_state_dict': model.state_dict(),
'optimizer_state_dict': optimizer.state_dict(),
'scheduler_state_dict': scheduler.state_dict(),
'train_loss': average_train_loss,
'val_loss': average_val_loss,
'train_perplexity': train_perplexity,
'val_perplexity': val_perplexity,
'global_step': global_step,
'config': config
}, checkpoint_path)
print(f"Checkpoint saved: {checkpoint_path}")
return {
'train_losses': train_losses,
'val_losses': val_losses,
'train_perplexities': train_perplexities,
'val_perplexities': val_perplexities
}
def main():
wandb.login()
config = {
'vocab_size': 25000,
'seq_len': 1024,
'd_model': 768,
'n_heads': 8,
'n_layers': 12,
'd_ff': 4 * 768,
'max_seq_len': 1024,
'dropout': 0.1,
'batch_size': 8,
'grad_acc_step': 8, # effective_batch_size = batch_size * grad_acc_step
'learning_rate': 1e-4,
'weight_decay': 0.01,
'num_epochs': 3,
'train_split': 0.8,
'device': 'cuda' if torch.cuda.is_available() else 'cpu',
}
model = GreyMatter(
d_model=config["d_model"],
n_heads=config["n_heads"],
vocab_size=config["vocab_size"],
n_layers=config["n_layers"],
d_ff=config["d_ff"],
max_seq_len=config["max_seq_len"],
dropout=config["dropout"]
)
print(model.get_parameter_count())
n_params = model.get_parameter_count()[0]
wandb.init(project=f"greymatter-pretraining-{n_params}", config=config)
train_path, val_path = prepare_train_val_split(data_path="fineweb_subset.txt", train_percent=config ["train_split"]) # replace with your filename
# prepare data loader
train_dataloader = get_dataloader(file_path=train_path, seq_len=config["seq_len"], batch_size=config["batch_size"])
val_dataloader = get_dataloader(file_path=val_path, seq_len=config["seq_len"], batch_size=config["batch_size"])
metrics = train(model, train_dataloader, val_dataloader, config)
# Save the metrics
with open('training_metrics.pkl', 'wb') as f:
pickle.dump(metrics, f)
print("Training completed! Metrics saved to 'training_metrics.pkl'")
main()
|