File size: 10,753 Bytes
4c5e9dd |
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 |
"""
Helion-V1-Embeddings Training Script
Train a lightweight embedding model for semantic similarity and retrieval
"""
import json
import logging
from typing import List, Dict, Tuple
from pathlib import Path
from datetime import datetime
logging.basicConfig(
level=logging.INFO,
format='%(asctime)s - %(levelname)s - %(message)s'
)
logger = logging.getLogger(__name__)
class EmbeddingsTrainer:
"""Train embeddings model for Helion-V1-Embeddings."""
def __init__(
self,
base_model: str = "sentence-transformers/all-MiniLM-L6-v2",
output_path: str = "./helion-embeddings-output"
):
self.base_model = base_model
self.output_path = Path(output_path)
self.output_path.mkdir(parents=True, exist_ok=True)
def prepare_training_data(self) -> List[Dict]:
"""
Prepare training data for embeddings.
Format: sentence pairs with similarity scores.
"""
training_examples = [
# High similarity pairs
{
"sentence1": "How do I reset my password?",
"sentence2": "What's the password reset process?",
"score": 0.95
},
{
"sentence1": "Machine learning training methods",
"sentence2": "How to train ML models",
"score": 0.90
},
{
"sentence1": "Python programming tutorial",
"sentence2": "Learn Python coding",
"score": 0.88
},
# Medium similarity pairs
{
"sentence1": "Install Python on Windows",
"sentence2": "Python setup guide",
"score": 0.70
},
{
"sentence1": "Best restaurants in Paris",
"sentence2": "Where to eat in France",
"score": 0.65
},
# Low similarity pairs
{
"sentence1": "How to bake cookies",
"sentence2": "Machine learning algorithms",
"score": 0.10
},
{
"sentence1": "Weather forecast tomorrow",
"sentence2": "Stock market analysis",
"score": 0.05
}
]
logger.info(f"Prepared {len(training_examples)} training examples")
return training_examples
def create_contrastive_pairs(self) -> List[Tuple[str, str]]:
"""
Create pairs for contrastive learning.
Format: (anchor, positive) pairs.
"""
pairs = [
("What is machine learning?", "Machine learning explained simply"),
("How to learn Python?", "Python learning resources"),
("Best coding practices", "Software development best practices"),
("Data science tutorial", "Learn data science basics"),
("Natural language processing", "NLP fundamentals guide"),
("Deep learning introduction", "Getting started with deep learning"),
("Web development guide", "How to build websites"),
("Database design principles", "SQL database design tutorial"),
("Cloud computing basics", "Introduction to cloud services"),
("API development guide", "How to create REST APIs"),
]
logger.info(f"Created {len(pairs)} contrastive pairs")
return pairs
def train_model(
self,
train_examples: List[Dict] = None,
epochs: int = 3,
batch_size: int = 16,
warmup_steps: int = 100
):
"""
Train the embeddings model.
Args:
train_examples: Training data (if None, uses default)
epochs: Number of training epochs
batch_size: Batch size for training
warmup_steps: Warmup steps for learning rate
"""
try:
from sentence_transformers import (
SentenceTransformer,
InputExample,
losses,
evaluation
)
from torch.utils.data import DataLoader
logger.info("Loading base model...")
model = SentenceTransformer(self.base_model)
# Prepare data
if train_examples is None:
train_examples = self.prepare_training_data()
# Convert to InputExample format
train_data = []
for example in train_examples:
train_data.append(InputExample(
texts=[example["sentence1"], example["sentence2"]],
label=example["score"]
))
# Create DataLoader
train_dataloader = DataLoader(
train_data,
shuffle=True,
batch_size=batch_size
)
# Define loss function
train_loss = losses.CosineSimilarityLoss(model)
# Training
logger.info("Starting training...")
model.fit(
train_objectives=[(train_dataloader, train_loss)],
epochs=epochs,
warmup_steps=warmup_steps,
output_path=str(self.output_path),
show_progress_bar=True,
save_best_model=True
)
logger.info(f"✅ Training complete! Model saved to {self.output_path}")
return model
except ImportError:
logger.error("sentence-transformers not installed. Install with: pip install sentence-transformers")
return None
except Exception as e:
logger.error(f"Training failed: {e}")
return None
def evaluate_model(self, model, test_pairs: List[Tuple[str, str, float]] = None):
"""
Evaluate the trained model.
Args:
model: Trained SentenceTransformer model
test_pairs: List of (sentence1, sentence2, expected_similarity)
"""
from sentence_transformers import util
if test_pairs is None:
# Default test pairs
test_pairs = [
("How to code?", "Coding tutorial", 0.85),
("Weather today", "Stock prices", 0.1),
("Machine learning", "AI and ML", 0.95),
]
logger.info("Evaluating model...")
total_error = 0
for sent1, sent2, expected in test_pairs:
emb1 = model.encode(sent1)
emb2 = model.encode(sent2)
similarity = float(util.cos_sim(emb1, emb2)[0][0])
error = abs(similarity - expected)
total_error += error
logger.info(f"'{sent1}' <-> '{sent2}'")
logger.info(f" Expected: {expected:.2f}, Got: {similarity:.2f}, Error: {error:.2f}")
avg_error = total_error / len(test_pairs)
logger.info(f"Average error: {avg_error:.3f}")
return avg_error
def create_config_files(self):
"""Create necessary configuration files."""
# Sentence transformers config
config = {
"__version__": {
"sentence_transformers": "2.2.2",
"transformers": "4.36.0",
"pytorch": "2.0.0"
},
"prompts": {},
"default_prompt_name": None,
"similarity_fn_name": "cosine",
"max_seq_length": 256,
"do_lower_case": False
}
with open(self.output_path / "config_sentence_transformers.json", 'w') as f:
json.dump(config, f, indent=2)
# Modules configuration
modules = [
{
"idx": 0,
"name": "0",
"path": "",
"type": "sentence_transformers.models.Transformer"
},
{
"idx": 1,
"name": "1",
"path": "1_Pooling",
"type": "sentence_transformers.models.Pooling"
},
{
"idx": 2,
"name": "2",
"path": "2_Normalize",
"type": "sentence_transformers.models.Normalize"
}
]
with open(self.output_path / "modules.json", 'w') as f:
json.dump(modules, f, indent=2)
logger.info("✅ Configuration files created")
def main():
"""Main training function."""
import argparse
parser = argparse.ArgumentParser(
description="Train Helion-V1-Embeddings model"
)
parser.add_argument(
"--base-model",
default="sentence-transformers/all-MiniLM-L6-v2",
help="Base model to fine-tune"
)
parser.add_argument(
"--output",
default="./helion-embeddings-output",
help="Output directory"
)
parser.add_argument(
"--epochs",
type=int,
default=3,
help="Number of training epochs"
)
parser.add_argument(
"--batch-size",
type=int,
default=16,
help="Batch size"
)
parser.add_argument(
"--data-file",
type=str,
help="Path to training data JSON file"
)
args = parser.parse_args()
# Create trainer
trainer = EmbeddingsTrainer(
base_model=args.base_model,
output_path=args.output
)
# Load custom data if provided
train_examples = None
if args.data_file:
with open(args.data_file, 'r') as f:
train_examples = json.load(f)
logger.info(f"Loaded {len(train_examples)} examples from {args.data_file}")
# Train model
model = trainer.train_model(
train_examples=train_examples,
epochs=args.epochs,
batch_size=args.batch_size
)
if model:
# Evaluate
trainer.evaluate_model(model)
# Create config files
trainer.create_config_files()
print("\n" + "="*60)
print("✅ Helion-V1-Embeddings Training Complete!")
print("="*60)
print(f"📁 Model saved to: {args.output}")
print("\n💡 Test your model:")
print("```python")
print("from sentence_transformers import SentenceTransformer")
print(f"model = SentenceTransformer('{args.output}')")
print("embeddings = model.encode(['Hello world'])")
print("```")
print("="*60)
if __name__ == "__main__":
main() |