Spaces:
Sleeping
Sleeping
File size: 10,217 Bytes
e6341fe 1377fb1 e6341fe 1377fb1 e6341fe 1377fb1 b08ba59 1377fb1 e6341fe 1377fb1 e6341fe 1377fb1 e6341fe |
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 |
"""
Model Manager for Fine-Tuned Model Deployment and Versioning
Handles loading, deploying, and rolling back fine-tuned models.
"""
import os
import json
import shutil
from typing import Optional, Dict
from datetime import datetime
import logging
from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch
logger = logging.getLogger(__name__)
class ModelManager:
"""Manage fine-tuned model deployment and versioning"""
def __init__(self, models_dir: str = None):
"""
Initialize ModelManager.
Args:
models_dir: Base directory for storing fine-tuned models
(defaults to MODELS_DIR env var or './models/finetuned')
"""
if models_dir is None:
# Use environment variable or /data path for HF Spaces
models_dir = os.getenv('MODELS_DIR', '/data/models/finetuned')
self.models_dir = models_dir
self.base_model_name = "facebook/bart-large-mnli"
# Create directory if it doesn't exist
try:
os.makedirs(models_dir, exist_ok=True)
except PermissionError:
logger.error(f"Permission denied creating models directory: {models_dir}")
raise
def get_model_path(self, run_id: int) -> str:
"""Get path to model for a specific training run"""
return os.path.join(self.models_dir, f"run_{run_id}")
def load_model(self, run_id: Optional[int] = None):
"""
Load a fine-tuned model or base model.
Args:
run_id: Training run ID (None for base model)
Returns:
Tuple of (model, tokenizer)
"""
if run_id is None:
logger.info("Loading base model")
model_name = self.base_model_name
else:
model_path = self.get_model_path(run_id)
if not os.path.exists(model_path):
raise FileNotFoundError(f"Model not found: {model_path}")
logger.info(f"Loading fine-tuned model from run {run_id}")
model_name = model_path
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSequenceClassification.from_pretrained(
model_name,
ignore_mismatched_sizes=True
)
return model, tokenizer
def deploy_model(self, run_id: int, db_session) -> Dict:
"""
Deploy a fine-tuned model (set as active).
Args:
run_id: Training run ID to deploy
db_session: Database session for updating FineTuningRun
Returns:
Dict with deployment info
"""
from app.models.models import FineTuningRun
logger.info(f"Deploying model from run {run_id}")
# Verify model exists
model_path = self.get_model_path(run_id)
if not os.path.exists(model_path):
raise FileNotFoundError(f"Model not found: {model_path}")
# Get the run record
run = db_session.query(FineTuningRun).filter_by(id=run_id).first()
if not run:
raise ValueError(f"Training run {run_id} not found")
if run.status != 'completed':
raise ValueError(f"Cannot deploy non-completed run (status: {run.status})")
# Deactivate all other models
db_session.query(FineTuningRun).update({'is_active_model': False})
# Activate this model
run.is_active_model = True
db_session.commit()
logger.info(f"Model from run {run_id} is now active")
return {
'run_id': run_id,
'deployed_at': datetime.utcnow().isoformat(),
'model_path': model_path
}
def rollback_to_baseline(self, db_session) -> Dict:
"""
Rollback to base model (deactivate all fine-tuned models).
Args:
db_session: Database session
Returns:
Dict with rollback info
"""
from app.models.models import FineTuningRun
logger.info("Rolling back to base model")
# Deactivate all fine-tuned models
active_count = db_session.query(FineTuningRun).filter_by(is_active_model=True).count()
db_session.query(FineTuningRun).update({'is_active_model': False})
db_session.commit()
logger.info(f"Deactivated {active_count} fine-tuned model(s)")
return {
'rolled_back_at': datetime.utcnow().isoformat(),
'deactivated_models': active_count,
'active_model': 'base'
}
def get_active_model_info(self, db_session) -> Optional[Dict]:
"""
Get information about the currently active model.
Args:
db_session: Database session
Returns:
Dict with active model info, or None if base model is active
"""
from app.models.models import FineTuningRun
active_run = db_session.query(FineTuningRun).filter_by(is_active_model=True).first()
if not active_run:
return None
return {
'run_id': active_run.id,
'model_path': self.get_model_path(active_run.id),
'created_at': active_run.created_at.isoformat() if active_run.created_at else None,
'results': active_run.get_results(),
'config': active_run.get_config()
}
def export_model(self, run_id: int, export_path: str) -> str:
"""
Export model for backup or sharing.
Args:
run_id: Training run ID
export_path: Destination path for export
Returns:
Path to exported model
"""
logger.info(f"Exporting model from run {run_id}")
model_path = self.get_model_path(run_id)
if not os.path.exists(model_path):
raise FileNotFoundError(f"Model not found: {model_path}")
# Create export directory
os.makedirs(export_path, exist_ok=True)
# Copy all model files
export_model_path = os.path.join(export_path, f"model_run_{run_id}")
shutil.copytree(model_path, export_model_path, dirs_exist_ok=True)
# Create model card
model_card = {
'run_id': run_id,
'export_date': datetime.utcnow().isoformat(),
'base_model': self.base_model_name,
'model_type': 'BART with LoRA fine-tuning',
'task': 'Multi-class text classification',
'categories': ['Vision', 'Problem', 'Objectives', 'Directives', 'Values', 'Actions']
}
with open(os.path.join(export_model_path, 'model_card.json'), 'w') as f:
json.dump(model_card, f, indent=2)
logger.info(f"Model exported to {export_model_path}")
return export_model_path
def import_model(self, import_path: str, run_id: int) -> str:
"""
Import a previously exported model.
Args:
import_path: Path to imported model directory
run_id: Training run ID to assign
Returns:
Path to imported model in models directory
"""
logger.info(f"Importing model to run {run_id}")
if not os.path.exists(import_path):
raise FileNotFoundError(f"Import path not found: {import_path}")
# Verify it's a valid model directory
required_files = ['config.json', 'pytorch_model.bin'] # or adapter_model.bin for LoRA
has_required = any(os.path.exists(os.path.join(import_path, f)) for f in required_files)
if not has_required:
raise ValueError(f"Import path does not contain a valid model")
# Copy to models directory
model_path = self.get_model_path(run_id)
shutil.copytree(import_path, model_path, dirs_exist_ok=True)
logger.info(f"Model imported to {model_path}")
return model_path
def delete_model(self, run_id: int) -> None:
"""
Delete a fine-tuned model from disk.
Args:
run_id: Training run ID
"""
logger.info(f"Deleting model from run {run_id}")
model_path = self.get_model_path(run_id)
if os.path.exists(model_path):
shutil.rmtree(model_path)
logger.info(f"Model deleted: {model_path}")
else:
logger.warning(f"Model not found: {model_path}")
def get_model_size(self, run_id: int) -> Dict:
"""
Get size information for a model.
Args:
run_id: Training run ID
Returns:
Dict with size info
"""
model_path = self.get_model_path(run_id)
if not os.path.exists(model_path):
return {'exists': False}
# Calculate directory size
total_size = 0
file_count = 0
for dirpath, dirnames, filenames in os.walk(model_path):
for filename in filenames:
filepath = os.path.join(dirpath, filename)
total_size += os.path.getsize(filepath)
file_count += 1
return {
'exists': True,
'total_size_bytes': total_size,
'total_size_mb': round(total_size / (1024 * 1024), 2),
'file_count': file_count,
'path': model_path
}
def list_available_models(self, db_session) -> list:
"""
List all available fine-tuned models.
Args:
db_session: Database session
Returns:
List of dicts with model info
"""
from app.models.models import FineTuningRun
runs = db_session.query(FineTuningRun).filter_by(status='completed').all()
models = []
for run in runs:
model_path = self.get_model_path(run.id)
size_info = self.get_model_size(run.id)
models.append({
'run_id': run.id,
'created_at': run.created_at.isoformat() if run.created_at else None,
'is_active': run.is_active_model,
'results': run.get_results(),
'model_exists': size_info.get('exists', False),
'size_mb': size_info.get('total_size_mb', 0)
})
return models
|