File size: 17,261 Bytes
d7d2fb2 1b2e4da a050405 d7d2fb2 ad359f7 d7d2fb2 1b2e4da d7d2fb2 ad359f7 d7d2fb2 ad359f7 1b2e4da a050405 d7d2fb2 326b359 ad359f7 1b2e4da d7d2fb2 1b2e4da d7d2fb2 1b2e4da a050405 d7d2fb2 a050405 d7d2fb2 1b2e4da 6b1c605 d7d2fb2 ad359f7 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 |
import tiktoken
import torch
from torch.utils.data import Dataset, DataLoader
from typing import Tuple, Optional, Literal, List
from pathlib import Path
from tqdm import tqdm
import mmap
import numpy as np
import os
import json
from model import ModelArgs
# Turkish Tokenizer support
try:
from turkish_tokenizer import TurkishTokenizer as TurkishTokenizerBase
TURKISH_TOKENIZER_AVAILABLE = True
except ImportError:
TURKISH_TOKENIZER_AVAILABLE = False
TurkishTokenizerBase = None
#####################################
# TURKISH TOKENIZER WRAPPER
#####################################
class TurkishTokenizerWrapper:
def __init__(self):
if not TURKISH_TOKENIZER_AVAILABLE:
raise ImportError(
"turkish-tokenizer package is not installed. "
"Install it with: pip install turkish-tokenizer"
)
self.tokenizer = TurkishTokenizerBase()
self.name = "turkish-tokenizer"
def encode(self, text: str, allowed_special: Optional[set] = None) -> List[int]:
return self.tokenizer.encode(text)
def decode(self, tokens: List[int]) -> str:
return self.tokenizer.decode(tokens)
@property
def n_vocab(self) -> int:
"""Get vocabulary size"""
return self.tokenizer.vocab_size
@property
def max_token_value(self) -> int:
"""Get maximum token value"""
return self.n_vocab - 1
#####################################
# DATA
#####################################
class MemoryEfficientTextDataset(Dataset):
"""
Memory-efficient dataset that tokenizes on-the-fly from disk.
Instead of loading all data into RAM, it:
1. Memory-maps the text file
2. Pre-computes line offsets for fast random access
3. Tokenizes only the required chunks during __getitem__
"""
def __init__(self, file_path: str, tokenizer, args: ModelArgs, stride: Optional[int] = None, max_samples: Optional[int] = None):
self.file_path = Path(file_path)
self.tokenizer = tokenizer
self.max_seq_len = args.max_seq_len
self.stride = stride if stride is not None else self.max_seq_len // 2
if not self.file_path.exists():
raise FileNotFoundError(f"File not found: {self.file_path}")
print(f"📝 Creating memory-efficient dataset from {self.file_path.name}...")
# Get file size
file_size = self.file_path.stat().st_size
print(f" File size: {file_size / 1024**2:.1f} MB")
# Pre-tokenize to get sample count (lightweight - just count)
self._count_samples(max_samples)
print(f"✅ Dataset ready with {len(self.samples)} samples (using lazy loading)")
def _count_samples(self, max_samples: Optional[int]):
"""Count how many samples we can create without loading everything"""
# Read file in chunks to estimate token count
chunk_size = 1024 * 1024 # 1MB chunks
total_tokens = 0
with open(self.file_path, 'r', encoding='utf-8') as f:
while True:
chunk = f.read(chunk_size)
if not chunk:
break
# Quick estimate: ~4 chars per token for most languages
total_tokens += len(chunk) // 4
# Calculate number of samples
num_samples = max((total_tokens - self.max_seq_len - 1) // self.stride, 1)
if max_samples:
num_samples = min(num_samples, max_samples)
# Store sample metadata (start positions in tokens)
self.samples = list(range(num_samples))
self.estimated_total_tokens = total_tokens
def _get_text_chunk(self, token_start: int) -> str:
"""Get a chunk of text starting from approximate token position"""
# Estimate byte position (rough: 4 chars per token, 1 byte per char for ASCII/UTF-8)
approx_byte_pos = token_start * 4
chunk_size = (self.max_seq_len + 1) * 8 # Read extra for safety
with open(self.file_path, 'r', encoding='utf-8', errors='ignore') as f:
f.seek(max(0, approx_byte_pos - 100)) # Seek with small buffer
return f.read(chunk_size)
def __len__(self):
return len(self.samples)
def __getitem__(self, idx) -> Tuple[torch.Tensor, torch.Tensor]:
"""Tokenize on-the-fly for this specific sample"""
# Calculate token position
token_start = idx * self.stride
# Get text chunk from file
text_chunk = self._get_text_chunk(token_start)
# Tokenize
try:
if hasattr(self.tokenizer, 'encode'):
tokens = self.tokenizer.encode(text_chunk, allowed_special={"<|endoftext|>"})
else:
tokens = self.tokenizer.encode(text_chunk)
except:
# Fallback for Turkish tokenizer
tokens = self.tokenizer.encode(text_chunk)
# Ensure we have enough tokens
if len(tokens) < self.max_seq_len + 1:
# Pad with zeros if needed
tokens = tokens + [0] * (self.max_seq_len + 1 - len(tokens))
# Create input/target pair
input_ids = torch.tensor(tokens[:self.max_seq_len], dtype=torch.long)
target_ids = torch.tensor(tokens[1:self.max_seq_len + 1], dtype=torch.long)
return input_ids, target_ids
class TextDataset(Dataset):
def __init__(self, txt: str, tokenizer, args: ModelArgs, stride: Optional[int] = None, max_samples: Optional[int] = None):
self.max_seq_len = args.max_seq_len
self.stride = stride if stride is not None else self.max_seq_len // 2
# Handle file paths efficiently with memory mapping
# Check if txt is a file path (avoid Path().exists() for long strings)
try:
path = Path(txt)
if len(txt) < 4096 and path.exists(): # Reasonable path length check
text_content = self._read_file_mmap(txt)
else:
text_content = txt
except (OSError, ValueError):
# If Path() fails or string is too long, treat as raw text
text_content = txt
# Validate input
if not text_content or len(text_content.strip()) < self.max_seq_len:
raise ValueError(f"Text too short. Need at least {self.max_seq_len} chars, got {len(text_content)}")
print(f"📝 Tokenizing {len(text_content):,} characters...")
# Tokenize with progress bar for large texts
token_ids = self._tokenize_with_progress(tokenizer, text_content)
# Create sliding windows with vectorized operations
self.samples = self._create_sliding_windows(token_ids, max_samples)
print(f"✅ Created {len(self.samples)} training samples")
def _read_file_mmap(self, file_path: str) -> str:
try:
with open(file_path, 'r', encoding='utf-8') as f:
with mmap.mmap(f.fileno(), 0, access=mmap.ACCESS_READ) as mm:
return mm.read().decode('utf-8', errors='ignore')
except Exception as e:
raise RuntimeError(f"Failed to read file {file_path}: {e}")
def _tokenize_with_progress(self, tokenizer, text: str) -> List[int]:
# Process in chunks for memory efficiency
chunk_size = 10_000_000 # 10MB chunks
tokens = []
if len(text) > chunk_size:
# Process large texts in chunks
pbar = tqdm(total=len(text), desc="Tokenizing", unit="char")
for i in range(0, len(text), chunk_size):
chunk = text[i:i + chunk_size]
chunk_tokens = tokenizer.encode(chunk, allowed_special={"<|endoftext|>"})
tokens.extend(chunk_tokens)
pbar.update(len(chunk))
pbar.close()
else:
# Single pass for smaller texts
tokens = tokenizer.encode(text, allowed_special={"<|endoftext|>"})
if not tokens:
raise ValueError("No tokens generated from input text")
return tokens
def _create_sliding_windows(self, token_ids: List[int], max_samples: Optional[int]) -> torch.Tensor:
if len(token_ids) < self.max_seq_len + 1:
raise ValueError(f"Not enough tokens. Need {self.max_seq_len + 1}, got {len(token_ids)}")
# Convert to numpy for faster slicing
tokens_array = np.array(token_ids, dtype=np.int64)
# Calculate number of windows
num_windows = (len(tokens_array) - self.max_seq_len - 1) // self.stride + 1
if max_samples:
num_windows = min(num_windows, max_samples)
# Pre-allocate tensors
inputs = torch.zeros(num_windows, self.max_seq_len, dtype=torch.long)
targets = torch.zeros(num_windows, self.max_seq_len, dtype=torch.long)
# Fill tensors efficiently
for i in range(num_windows):
start = i * self.stride
inputs[i] = torch.from_numpy(tokens_array[start:start + self.max_seq_len])
targets[i] = torch.from_numpy(tokens_array[start + 1:start + self.max_seq_len + 1])
# Stack into pairs (more memory efficient than separate lists)
self.samples = torch.stack([inputs, targets], dim=1)
return self.samples
def __len__(self):
return len(self.samples)
def __getitem__(self, idx) -> Tuple[torch.Tensor, torch.Tensor]:
"""Return (input_ids, target_ids) tuple"""
return self.samples[idx, 0], self.samples[idx, 1]
def create_dataloader(
txt: str,
args: ModelArgs,
stride: Optional[int] = None,
shuffle: bool = True,
drop_last: bool = True,
num_workers: int = 0,
pin_memory: bool = True,
persistent_workers: bool = False,
max_samples: Optional[int] = None,
use_turkish_tokenizer: bool = True,
use_memory_efficient: bool = True, # NEW: Use memory-efficient loading by default
is_val: bool = True
) -> DataLoader:
# Select tokenizer based on user preference
if use_turkish_tokenizer:
if not TURKISH_TOKENIZER_AVAILABLE:
raise ImportError(
"Turkish tokenizer requested but not available. "
"Install it with: pip install turkish-tokenizer"
)
tokenizer = TurkishTokenizerWrapper()
print(f"🇹🇷 Using Turkish Tokenizer (vocab size: {tokenizer.n_vocab:,})")
else:
# Use the best default tokenizer for your setup
# tiktoken's gpt2 is fast, well-tested, and has reasonable vocab size (~50k)
# For multilingual or code, consider "cl100k_base" or "o200k_base"
tokenizer_name = getattr(args, "tokenizer_name", "gpt2")
tokenizer = tiktoken.get_encoding(tokenizer_name)
print(f"📚 Using tiktoken tokenizer: {tokenizer_name} (vocab size: {tokenizer.n_vocab:,})")
# Create dataset with size validation
try:
# Check if txt is a file path
is_file_path = False
try:
path = Path(txt)
if len(txt) < 4096 and path.exists():
is_file_path = True
except (OSError, ValueError):
pass
# Use memory-efficient dataset for file paths
if use_memory_efficient and is_file_path:
print(f"💾 Using memory-efficient dataset (lazy loading from disk)")
dataset = MemoryEfficientTextDataset(
file_path=txt,
tokenizer=tokenizer,
args=args,
stride=stride,
max_samples=max_samples
)
else:
# Use original dataset (loads everything into RAM)
print(f"⚠️ Using in-memory dataset (loads all data into RAM)")
dataset = TextDataset(
txt=txt,
tokenizer=tokenizer,
args=args,
stride=stride,
max_samples=max_samples
)
except Exception as e:
raise RuntimeError(f"Failed to create dataset: {e}")
config_path = Path("config.json")
with open(config_path,"r") as f:
config = json.load(f)
val_batch_size = config["model"]["max_batch_size"] #* config["training"].get("val_batch_size_multiplier", 4)
if is_val:
batch_size = val_batch_size
else:
batch_size = args.max_batch_size
# Create DataLoader with optimized settings
dataloader = DataLoader(
dataset,
batch_size=batch_size,
shuffle=shuffle,
drop_last=drop_last,
num_workers=num_workers,
pin_memory=pin_memory,
persistent_workers=persistent_workers if num_workers > 0 else False,
prefetch_factor=2 if num_workers > 0 else None,
)
return dataloader, tokenizer
# Convenience function for downloading sample data
def get_sample_data(url: str = "https://raw.githubusercontent.com/karpathy/char-rnn/master/data/tinyshakespeare/input.txt") -> str:
"""Download sample text data for testing"""
try:
import requests
response = requests.get(url)
response.raise_for_status()
return response.text
except Exception as e:
print(f"⚠️ Could not download sample data: {e}")
return ""
if __name__ == "__main__":
print("=" * 60)
print("TOKENIZER TESTING")
print("=" * 60)
# Choose which tokenizer to test
USE_TURKISH = True # Set to False to test tiktoken instead
if USE_TURKISH and TURKISH_TOKENIZER_AVAILABLE:
print("\n🇹🇷 Testing Turkish Tokenizer")
tokenizer = TurkishTokenizerWrapper()
print(f"📚 Tokenizer: {tokenizer.name}")
print(f"📊 Vocabulary Size: {tokenizer.n_vocab:,}")
print(f"📝 Max Token Value: {tokenizer.max_token_value:,}")
else:
# Test different tokenizers
tokenizer_name = "gpt2" # Change to "cl100k_base" or "o200k_base" to test others
tokenizer = tiktoken.get_encoding(tokenizer_name)
print(f"\n📚 Tokenizer: {tokenizer_name}")
print(f"📊 Vocabulary Size: {tokenizer.n_vocab:,}")
print(f"📝 Max Token Value: {tokenizer.max_token_value:,}")
print(f"🔤 Name: {tokenizer.name}")
# Test encoding/decoding
if USE_TURKISH and TURKISH_TOKENIZER_AVAILABLE:
test_samples = [
"Merhaba Dünya!",
"İstanbul'da yaşıyorum ve Türkçe dilini öğreniyorum.",
"Kitap okumak çok güzeldir ve bilgi verir.",
"Türkiye Cumhuriyeti'nin başkenti Ankara'dır.",
"Yapay zeka ve makine öğrenmesi teknolojileri gelişiyor.",
]
else:
test_samples = [
"Hello, world!",
"The quick brown fox jumps over the lazy dog.",
"Machine learning is fascinating.",
"print('Hello, World!')", # Code sample
"日本語のテキスト", # Non-English
]
print("\n" + "=" * 60)
print("ENCODING EXAMPLES")
print("=" * 60)
for text in test_samples:
tokens = tokenizer.encode(text)
decoded = tokenizer.decode(tokens)
print(f"\nText: {text}")
print(f"Tokens ({len(tokens)}): {tokens}")
print(f"Token range: [{min(tokens)}, {max(tokens)}]")
print(f"Decoded: {decoded}")
# Test with actual data
print("\n" + "=" * 60)
print("DATALOADER TESTING")
print("=" * 60)
sample_text = get_sample_data()
if sample_text:
print(f"\n📄 Sample text length: {len(sample_text):,} characters")
# Tokenize sample
if USE_TURKISH and TURKISH_TOKENIZER_AVAILABLE:
full_tokens = tokenizer.encode(sample_text)
else:
full_tokens = tokenizer.encode(sample_text, allowed_special={"<|endoftext|>"})
print(f"🔢 Total tokens: {len(full_tokens):,}")
print(f"📈 Unique tokens used: {len(set(full_tokens)):,}")
print(f"📊 Vocabulary coverage: {len(set(full_tokens)) / tokenizer.n_vocab * 100:.2f}%")
# Create dataloader
args = ModelArgs(max_seq_len=128, max_batch_size=16)
dataloader = create_dataloader(
sample_text,
args,
num_workers=0,
max_samples=100,
use_turkish_tokenizer=USE_TURKISH and TURKISH_TOKENIZER_AVAILABLE
)
print(f"\n⚙️ DataLoader Config:")
print(f" Sequence length: {args.max_seq_len}")
print(f" Batch size: {args.max_batch_size}")
print(f" Total batches: {len(dataloader)}")
# Test first batch
for batch_idx, (input_ids, target_ids) in enumerate(dataloader):
print(f"\n🎯 Batch {batch_idx}:")
print(f" input_ids shape: {input_ids.shape}")
print(f" target_ids shape: {target_ids.shape}")
print(f" input_ids range: [{input_ids.min().item()}, {input_ids.max().item()}]")
print(f" Sample input (first 10 tokens): {input_ids[0, :10].tolist()}")
print(f" Decoded: {tokenizer.decode(input_ids[0, :10].tolist())}")
break
print("\n" + "=" * 60)
print("✅ Testing complete!")
print("=" * 60) |