Spaces:
Sleeping
Sleeping
File size: 8,261 Bytes
6510698 |
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 |
"""
Model loading module with robust error handling and environment adaptation.
"""
import logging
import torch
from typing import Optional, Tuple, Any
from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline
from .config.model_config import ModelConfig, EnvironmentDetector, DependencyValidator
logger = logging.getLogger(__name__)
class ModelLoader:
"""Handles model loading with environment-specific optimizations."""
def __init__(self):
self.config: Optional[ModelConfig] = None
self.model: Optional[Any] = None
self.tokenizer: Optional[Any] = None
self.pipeline: Optional[Any] = None
self._is_loaded = False
def validate_environment(self) -> bool:
"""Validate that the environment is ready for model loading."""
logger.info("🔍 Validating environment...")
# Check dependencies
if not DependencyValidator.is_environment_ready():
logger.error("❌ Environment validation failed - missing dependencies")
return False
# Log environment info
env_info = EnvironmentDetector.detect_environment()
logger.info(f"📊 Environment info: {env_info}")
return True
def create_config(
self,
model_id: Optional[str] = None,
revision: Optional[str] = None
) -> ModelConfig:
"""Create model configuration based on environment."""
logger.info("⚙️ Creating model configuration...")
self.config = EnvironmentDetector.create_model_config(model_id, revision)
logger.info(f"📋 Model config created:")
logger.info(f" Model ID: {self.config.model_id}")
logger.info(f" Revision: {self.config.revision or 'latest'}")
logger.info(f" Device: {self.config.device_map}")
logger.info(f" Dtype: {self.config.dtype}")
logger.info(f" Attention: {self.config.attn_implementation}")
return self.config
def load_tokenizer(self) -> bool:
"""Load the tokenizer."""
if not self.config:
logger.error("❌ No configuration available")
return False
try:
logger.info("📝 Loading tokenizer...")
self.tokenizer = AutoTokenizer.from_pretrained(
self.config.model_id,
trust_remote_code=self.config.trust_remote_code,
revision=self.config.revision
)
logger.info("✅ Tokenizer loaded successfully")
return True
except Exception as e:
logger.error(f"❌ Failed to load tokenizer: {e}")
return False
def load_model(self) -> bool:
"""Load the model with environment-specific configuration."""
if not self.config:
logger.error("❌ No configuration available")
return False
try:
logger.info("🤖 Loading model...")
logger.info(f" This may take several minutes for {self.config.model_id}")
# Load model with configuration
self.model = AutoModelForCausalLM.from_pretrained(
self.config.model_id,
trust_remote_code=self.config.trust_remote_code,
revision=self.config.revision,
attn_implementation=self.config.attn_implementation,
dtype=self.config.dtype, # Use dtype instead of deprecated torch_dtype
device_map=self.config.device_map,
low_cpu_mem_usage=self.config.low_cpu_mem_usage
).eval()
logger.info("✅ Model loaded successfully")
# Log model info
if hasattr(self.model, 'config'):
logger.info(f"📊 Model info:")
logger.info(f" Architecture: {getattr(self.model.config, 'architectures', 'unknown')}")
logger.info(f" Parameters: ~{self.model.num_parameters() / 1e9:.1f}B")
return True
except Exception as e:
logger.error(f"❌ Failed to load model: {e}")
return False
def create_pipeline(self) -> bool:
"""Create inference pipeline."""
if not self.model or not self.tokenizer:
logger.error("❌ Model or tokenizer not loaded")
return False
try:
logger.info("🔧 Creating inference pipeline...")
self.pipeline = pipeline(
"text-generation",
model=self.model,
tokenizer=self.tokenizer,
dtype=self.config.dtype, # Use dtype instead of deprecated torch_dtype
device_map=self.config.device_map,
trust_remote_code=self.config.trust_remote_code
)
logger.info("✅ Pipeline created successfully")
return True
except Exception as e:
logger.error(f"❌ Failed to create pipeline: {e}")
return False
def load_complete_model(
self,
model_id: Optional[str] = None,
revision: Optional[str] = None
) -> bool:
"""Load complete model (tokenizer + model + pipeline)."""
logger.info("🚀 Starting complete model loading process...")
try:
# Validate environment
if not self.validate_environment():
return False
# Create configuration
self.create_config(model_id, revision)
# Load components in order
if not self.load_tokenizer():
return False
if not self.load_model():
return False
if not self.create_pipeline():
return False
# Run smoke test
if not self.smoke_test():
logger.warning("⚠️ Smoke test failed, but model appears loaded")
self._is_loaded = True
logger.info("🎉 Model loading completed successfully!")
return True
except Exception as e:
logger.error(f"❌ Complete model loading failed: {e}")
return False
def smoke_test(self) -> bool:
"""Run a quick smoke test to verify model works."""
if not self.pipeline:
return False
try:
logger.info("🧪 Running smoke test...")
# Simple test generation
test_input = "Hello"
result = self.pipeline(
test_input,
max_new_tokens=4,
do_sample=False,
pad_token_id=self.tokenizer.eos_token_id
)
if result and len(result) > 0:
logger.info("✅ Smoke test passed")
return True
else:
logger.warning("⚠️ Smoke test returned empty result")
return False
except Exception as e:
logger.warning(f"⚠️ Smoke test failed: {e}")
return False
@property
def is_loaded(self) -> bool:
"""Check if model is fully loaded and ready."""
return self._is_loaded and self.pipeline is not None
def get_model_info(self) -> dict:
"""Get information about the loaded model."""
if not self.is_loaded:
return {"status": "not_loaded"}
info = {
"status": "loaded",
"model_id": self.config.model_id,
"revision": self.config.revision,
"device": self.config.device_map,
"dtype": str(self.config.dtype),
"attention": self.config.attn_implementation,
"device_info": self.config.device_info
}
if hasattr(self.model, 'config'):
info["architecture"] = getattr(self.model.config, 'architectures', 'unknown')
info["parameters"] = f"~{self.model.num_parameters() / 1e9:.1f}B"
return info
|