Spaces:
Running
on
Zero
Running
on
Zero
File size: 12,971 Bytes
a7d8c02 |
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 |
"""
Model management for FLUX Prompt Optimizer
Handles Florence-2 and Bagel model integration
"""
import logging
import requests
import spaces
import torch
from typing import Optional, Dict, Any, Tuple
from PIL import Image
from transformers import AutoProcessor, AutoModelForCausalLM
from config import MODEL_CONFIG, get_device_config
from utils import clean_memory, safe_execute
logger = logging.getLogger(__name__)
class BaseImageAnalyzer:
"""Base class for image analysis models"""
def __init__(self):
self.model = None
self.processor = None
self.device_config = get_device_config()
self.is_initialized = False
def initialize(self) -> bool:
"""Initialize the model"""
raise NotImplementedError
def analyze_image(self, image: Image.Image) -> Tuple[str, Dict[str, Any]]:
"""Analyze image and return description"""
raise NotImplementedError
def cleanup(self) -> None:
"""Clean up model resources"""
if self.model is not None:
del self.model
self.model = None
if self.processor is not None:
del self.processor
self.processor = None
clean_memory()
class Florence2Analyzer(BaseImageAnalyzer):
"""Florence-2 model for image analysis"""
def __init__(self):
super().__init__()
self.config = MODEL_CONFIG["florence2"]
def initialize(self) -> bool:
"""Initialize Florence-2 model"""
if self.is_initialized:
return True
try:
logger.info("Initializing Florence-2 model...")
model_id = self.config["model_id"]
# Load processor
self.processor = AutoProcessor.from_pretrained(
model_id,
trust_remote_code=self.config["trust_remote_code"]
)
# Load model
self.model = AutoModelForCausalLM.from_pretrained(
model_id,
trust_remote_code=self.config["trust_remote_code"],
torch_dtype=self.config["torch_dtype"] if self.device_config["use_gpu"] else torch.float32
)
# Move to appropriate device
if self.device_config["use_gpu"]:
self.model = self.model.to(self.device_config["device"])
else:
self.model = self.model.to("cpu")
self.model.eval()
self.is_initialized = True
logger.info(f"Florence-2 initialized on {self.device_config['device']}")
return True
except Exception as e:
logger.error(f"Florence-2 initialization failed: {e}")
self.cleanup()
return False
@spaces.GPU(duration=60)
def _gpu_inference(self, image: Image.Image, task_prompt: str) -> str:
"""Run inference on GPU with spaces decorator"""
try:
# Move model to GPU for inference
if self.device_config["use_gpu"]:
self.model = self.model.to("cuda")
# Prepare inputs
inputs = self.processor(text=task_prompt, images=image, return_tensors="pt")
# Move inputs to device
device = "cuda" if self.device_config["use_gpu"] else self.device_config["device"]
inputs = {k: v.to(device) for k, v in inputs.items()}
# Generate response
with torch.no_grad():
if self.device_config["use_gpu"]:
with torch.cuda.amp.autocast(dtype=torch.float16):
generated_ids = self.model.generate(
input_ids=inputs["input_ids"],
pixel_values=inputs["pixel_values"],
max_new_tokens=self.config["max_new_tokens"],
num_beams=3,
do_sample=False
)
else:
generated_ids = self.model.generate(
input_ids=inputs["input_ids"],
pixel_values=inputs["pixel_values"],
max_new_tokens=self.config["max_new_tokens"],
num_beams=3,
do_sample=False
)
# Decode response
generated_text = self.processor.batch_decode(generated_ids, skip_special_tokens=False)[0]
parsed = self.processor.post_process_generation(
generated_text,
task=task_prompt,
image_size=(image.width, image.height)
)
# Extract caption
if task_prompt in parsed:
return parsed[task_prompt]
else:
return str(parsed) if parsed else ""
except Exception as e:
logger.error(f"Florence-2 GPU inference failed: {e}")
return ""
finally:
# Move model back to CPU to free GPU memory
if self.device_config["use_gpu"]:
self.model = self.model.to("cpu")
clean_memory()
def analyze_image(self, image: Image.Image) -> Tuple[str, Dict[str, Any]]:
"""Analyze image using Florence-2"""
if not self.is_initialized:
success = self.initialize()
if not success:
return "Model initialization failed", {"error": "Florence-2 not available"}
try:
# Define analysis tasks
tasks = {
"detailed": "<DETAILED_CAPTION>",
"more_detailed": "<MORE_DETAILED_CAPTION>",
"caption": "<CAPTION>"
}
results = {}
# Run analysis for each task
for task_name, task_prompt in tasks.items():
if self.device_config["use_gpu"]:
result = self._gpu_inference(image, task_prompt)
else:
result = self._cpu_inference(image, task_prompt)
results[task_name] = result
# Choose best result
if results["more_detailed"]:
main_description = results["more_detailed"]
elif results["detailed"]:
main_description = results["detailed"]
else:
main_description = results["caption"] or "A photograph"
# Prepare metadata
metadata = {
"model": "Florence-2",
"device": self.device_config["device"],
"all_results": results,
"confidence": 0.85 # Florence-2 generally reliable
}
logger.info(f"Florence-2 analysis complete: {len(main_description)} chars")
return main_description, metadata
except Exception as e:
logger.error(f"Florence-2 analysis failed: {e}")
return "Analysis failed", {"error": str(e)}
def _cpu_inference(self, image: Image.Image, task_prompt: str) -> str:
"""Run inference on CPU"""
try:
inputs = self.processor(text=task_prompt, images=image, return_tensors="pt")
with torch.no_grad():
generated_ids = self.model.generate(
input_ids=inputs["input_ids"],
pixel_values=inputs["pixel_values"],
max_new_tokens=self.config["max_new_tokens"],
num_beams=2, # Reduced for CPU
do_sample=False
)
generated_text = self.processor.batch_decode(generated_ids, skip_special_tokens=False)[0]
parsed = self.processor.post_process_generation(
generated_text,
task=task_prompt,
image_size=(image.width, image.height)
)
if task_prompt in parsed:
return parsed[task_prompt]
else:
return str(parsed) if parsed else ""
except Exception as e:
logger.error(f"Florence-2 CPU inference failed: {e}")
return ""
class BagelAnalyzer(BaseImageAnalyzer):
"""Bagel-7B model analyzer via API"""
def __init__(self):
super().__init__()
self.config = MODEL_CONFIG["bagel"]
self.session = requests.Session()
def initialize(self) -> bool:
"""Initialize Bagel analyzer (API-based)"""
try:
# Test API connectivity
test_response = self.session.get(
self.config["api_url"],
timeout=self.config["timeout"]
)
if test_response.status_code == 200:
self.is_initialized = True
logger.info("Bagel API connection established")
return True
else:
logger.error(f"Bagel API not accessible: {test_response.status_code}")
return False
except Exception as e:
logger.error(f"Bagel initialization failed: {e}")
return False
def analyze_image(self, image: Image.Image) -> Tuple[str, Dict[str, Any]]:
"""Analyze image using Bagel-7B API"""
if not self.is_initialized:
success = self.initialize()
if not success:
return "Bagel API not available", {"error": "API connection failed"}
try:
# Convert image to base64 or prepare for API call
# Note: This is a placeholder - actual implementation would depend on Bagel API format
# For now, return a placeholder response
# In real implementation, you would:
# 1. Convert image to required format
# 2. Make API call to Bagel endpoint
# 3. Parse response
description = "Detailed image analysis via Bagel-7B (API implementation needed)"
metadata = {
"model": "Bagel-7B",
"method": "API",
"confidence": 0.8
}
logger.info("Bagel analysis complete (placeholder)")
return description, metadata
except Exception as e:
logger.error(f"Bagel analysis failed: {e}")
return "Analysis failed", {"error": str(e)}
class ModelManager:
"""Manager for handling multiple analysis models"""
def __init__(self, preferred_model: str = None):
self.preferred_model = preferred_model or MODEL_CONFIG["primary_model"]
self.analyzers = {}
self.current_analyzer = None
def get_analyzer(self, model_name: str = None) -> Optional[BaseImageAnalyzer]:
"""Get or create analyzer for specified model"""
model_name = model_name or self.preferred_model
if model_name not in self.analyzers:
if model_name == "florence2":
self.analyzers[model_name] = Florence2Analyzer()
elif model_name == "bagel":
self.analyzers[model_name] = BagelAnalyzer()
else:
logger.error(f"Unknown model: {model_name}")
return None
return self.analyzers[model_name]
def analyze_image(self, image: Image.Image, model_name: str = None) -> Tuple[str, Dict[str, Any]]:
"""Analyze image with specified or preferred model"""
analyzer = self.get_analyzer(model_name)
if analyzer is None:
return "No analyzer available", {"error": "Model not found"}
success, result = safe_execute(analyzer.analyze_image, image)
if success:
return result
else:
return "Analysis failed", {"error": result}
def cleanup_all(self) -> None:
"""Clean up all model resources"""
for analyzer in self.analyzers.values():
analyzer.cleanup()
self.analyzers.clear()
clean_memory()
# Global model manager instance
model_manager = ModelManager()
def analyze_image(image: Image.Image, model_name: str = None) -> Tuple[str, Dict[str, Any]]:
"""
Convenience function for image analysis
Args:
image: PIL Image to analyze
model_name: Optional model name ("florence2" or "bagel")
Returns:
Tuple of (description, metadata)
"""
return model_manager.analyze_image(image, model_name)
# Export main components
__all__ = [
"BaseImageAnalyzer",
"Florence2Analyzer",
"BagelAnalyzer",
"ModelManager",
"model_manager",
"analyze_image"
] |