Spaces:
Running
Running
File size: 12,564 Bytes
3096f4a |
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 |
"""Continuous training system for Veda Programming LLM"""
import os
import json
import shutil
from datetime import datetime
from typing import Optional
import threading
import time
import tensorflow as tf
from tensorflow import keras
from model import VedaProgrammingLLM
from tokenizer import VedaTokenizer
from database import db
from data_collector import collector
from config import (
MODEL_DIR, VERSIONS_DIR, VOCAB_SIZE, MAX_LENGTH,
D_MODEL, NUM_HEADS, NUM_LAYERS, FF_DIM, BATCH_SIZE,
MIN_SAMPLES_FOR_TRAINING, EPOCHS_PER_RETRAIN,
AUTO_TRAIN_INTERVAL_HOURS
)
class ContinuousTrainer:
"""Handles continuous learning and model updates"""
def __init__(self):
self.model: Optional[VedaProgrammingLLM] = None
self.tokenizer: Optional[VedaTokenizer] = None
self.is_training = False
self.training_progress = 0
self.last_training_time = None
self.model_version = self._get_current_version()
# Background training thread
self._training_thread = None
self._stop_background = False
def _get_current_version(self) -> str:
"""Get current model version"""
config_path = os.path.join(MODEL_DIR, "config.json")
if os.path.exists(config_path):
with open(config_path, 'r') as f:
config = json.load(f)
return config.get('version', 'v1.0')
return 'v1.0'
def _generate_version(self) -> str:
"""Generate new version string"""
return f"v{datetime.now().strftime('%Y%m%d_%H%M%S')}"
def load_model(self) -> bool:
"""Load the current model"""
config_path = os.path.join(MODEL_DIR, "config.json")
if not os.path.exists(config_path):
print("No existing model found.")
return False
try:
# Load config
with open(config_path, 'r') as f:
config = json.load(f)
# Load tokenizer
self.tokenizer = VedaTokenizer()
self.tokenizer.load(os.path.join(MODEL_DIR, "tokenizer.json"))
# Create model
self.model = VedaProgrammingLLM(
vocab_size=config['vocab_size'],
max_length=config['max_length'],
d_model=config['d_model'],
num_heads=config['num_heads'],
num_layers=config['num_layers'],
ff_dim=config['ff_dim']
)
# Build and load weights
dummy = tf.zeros((1, config['max_length']), dtype=tf.int32)
self.model(dummy)
self.model.load_weights(os.path.join(MODEL_DIR, "weights.h5"))
self.model_version = config.get('version', 'v1.0')
print(f"Model loaded: {self.model_version}")
return True
except Exception as e:
print(f"Error loading model: {e}")
return False
def save_model(self, version: str = None):
"""Save the current model"""
if self.model is None or self.tokenizer is None:
return
version = version or self._generate_version()
# Save to main directory
os.makedirs(MODEL_DIR, exist_ok=True)
self.model.save_weights(os.path.join(MODEL_DIR, "weights.h5"))
self.tokenizer.save(os.path.join(MODEL_DIR, "tokenizer.json"))
config = self.model.get_config()
config['version'] = version
config['last_trained'] = datetime.now().isoformat()
with open(os.path.join(MODEL_DIR, "config.json"), 'w') as f:
json.dump(config, f, indent=2)
# Save version backup
version_dir = os.path.join(VERSIONS_DIR, version)
os.makedirs(version_dir, exist_ok=True)
shutil.copy(
os.path.join(MODEL_DIR, "weights.h5"),
os.path.join(version_dir, "weights.h5")
)
shutil.copy(
os.path.join(MODEL_DIR, "tokenizer.json"),
os.path.join(version_dir, "tokenizer.json")
)
shutil.copy(
os.path.join(MODEL_DIR, "config.json"),
os.path.join(version_dir, "config.json")
)
self.model_version = version
print(f"Model saved: {version}")
def should_retrain(self) -> bool:
"""Check if retraining is needed"""
pending = collector.get_pending_count()
return pending >= MIN_SAMPLES_FOR_TRAINING
def prepare_training_data(self) -> tf.data.Dataset:
"""Prepare dataset for training"""
# Get all training samples
samples = collector.get_training_data(include_base=True)
if not samples:
return None
# Combine all samples
all_text = '\n\n'.join(samples)
# Fit or update tokenizer
if self.tokenizer is None:
self.tokenizer = VedaTokenizer(vocab_size=VOCAB_SIZE)
self.tokenizer.fit([all_text])
# Encode
all_tokens = self.tokenizer.encode(all_text)
# Create sequences
sequences = []
stride = MAX_LENGTH // 2
for i in range(0, len(all_tokens) - MAX_LENGTH - 1, stride):
seq = all_tokens[i:i + MAX_LENGTH + 1]
if len(seq) == MAX_LENGTH + 1:
sequences.append(seq)
if len(sequences) < 5:
stride = max(1, MAX_LENGTH // 8)
sequences = []
for i in range(0, len(all_tokens) - MAX_LENGTH - 1, stride):
seq = all_tokens[i:i + MAX_LENGTH + 1]
if len(seq) == MAX_LENGTH + 1:
sequences.append(seq)
import numpy as np
sequences = np.array(sequences)
X = sequences[:, :-1]
y = sequences[:, 1:]
dataset = tf.data.Dataset.from_tensor_slices((X, y))
dataset = dataset.shuffle(buffer_size=min(1000, len(sequences)))
dataset = dataset.batch(BATCH_SIZE)
dataset = dataset.prefetch(tf.data.AUTOTUNE)
print(f"Prepared {len(sequences)} training sequences")
return dataset
def train(
self,
epochs: int = EPOCHS_PER_RETRAIN,
callback=None
) -> dict:
"""Train/retrain the model"""
if self.is_training:
return {'status': 'error', 'message': 'Training already in progress'}
self.is_training = True
self.training_progress = 0
try:
# Prepare data
dataset = self.prepare_training_data()
if dataset is None:
self.is_training = False
return {'status': 'error', 'message': 'No training data available'}
# Create/update model
if self.model is None:
self.model = VedaProgrammingLLM(
vocab_size=self.tokenizer.vocabulary_size,
max_length=MAX_LENGTH,
d_model=D_MODEL,
num_heads=NUM_HEADS,
num_layers=NUM_LAYERS,
ff_dim=FF_DIM
)
# Compile
self.model.compile(
optimizer=keras.optimizers.Adam(learning_rate=1e-4),
loss=keras.losses.SparseCategoricalCrossentropy(from_logits=True),
metrics=['accuracy']
)
# Build
dummy = tf.zeros((1, MAX_LENGTH), dtype=tf.int32)
self.model(dummy)
# Custom callback for progress
class ProgressCallback(keras.callbacks.Callback):
def __init__(self, trainer, total_epochs):
self.trainer = trainer
self.total_epochs = total_epochs
def on_epoch_end(self, epoch, logs=None):
self.trainer.training_progress = (epoch + 1) / self.total_epochs * 100
callbacks = [ProgressCallback(self, epochs)]
if callback:
callbacks.append(callback)
# Train
history = self.model.fit(
dataset,
epochs=epochs,
callbacks=callbacks,
verbose=1
)
# Save model
new_version = self._generate_version()
self.save_model(new_version)
# Mark samples as used
new_samples = collector.get_new_training_data()
if new_samples:
sample_ids = [s['id'] for s in new_samples]
db.mark_as_used_for_training(sample_ids)
# Record training run
final_loss = history.history['loss'][-1]
final_acc = history.history.get('accuracy', [0])[-1]
samples_count = len(new_samples) if new_samples else 0
db.save_training_run(
samples_used=samples_count,
epochs=epochs,
final_loss=final_loss,
final_accuracy=final_acc,
model_version=new_version
)
self.last_training_time = datetime.now()
self.is_training = False
self.training_progress = 100
return {
'status': 'success',
'version': new_version,
'loss': final_loss,
'accuracy': final_acc,
'samples_used': samples_count
}
except Exception as e:
self.is_training = False
import traceback
traceback.print_exc()
return {'status': 'error', 'message': str(e)}
def train_async(self, epochs: int = EPOCHS_PER_RETRAIN):
"""Start training in background thread"""
if self.is_training:
return False
def train_thread():
result = self.train(epochs=epochs)
print(f"Background training completed: {result}")
self._training_thread = threading.Thread(target=train_thread)
self._training_thread.start()
return True
def start_auto_training(self):
"""Start automatic retraining scheduler"""
def auto_train_loop():
while not self._stop_background:
# Check every hour
time.sleep(3600)
if self._stop_background:
break
# Check if retraining needed
if self.should_retrain():
print("Auto-training triggered...")
self.train()
self._stop_background = False
thread = threading.Thread(target=auto_train_loop, daemon=True)
thread.start()
print("Auto-training scheduler started")
def stop_auto_training(self):
"""Stop automatic retraining"""
self._stop_background = True
def get_status(self) -> dict:
"""Get trainer status"""
return {
'model_loaded': self.model is not None,
'model_version': self.model_version,
'is_training': self.is_training,
'training_progress': self.training_progress,
'last_training': self.last_training_time.isoformat() if self.last_training_time else None,
'pending_samples': collector.get_pending_count(),
'min_samples_for_training': MIN_SAMPLES_FOR_TRAINING
}
def generate(
self,
prompt: str,
max_tokens: int = 100,
temperature: float = 0.7,
repetition_penalty: float = 1.2,
top_k: int = 50
) -> str:
"""Generate code using the model"""
if self.model is None or self.tokenizer is None:
raise ValueError("Model not loaded")
tokens = self.tokenizer.encode(prompt)
if len(tokens) == 0:
tokens = [ord(' ')]
generated = self.model.generate(
tokens,
max_new_tokens=max_tokens,
temperature=temperature,
top_k=top_k,
repetition_penalty=repetition_penalty
)
return self.tokenizer.decode(generated)
# Global trainer instance
trainer = ContinuousTrainer() |