File size: 11,822 Bytes
c8b77b5 |
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 |
"""
Model Wrapper
=============
This module provides a wrapper for neural network models to integrate
with the gradient descent training system, including support for LoRA
adapters and the MangoMAS agent system.
"""
import logging
import torch
import torch.nn as nn
from typing import Dict, List, Optional, Any
from pathlib import Path
import json
logger = logging.getLogger(__name__)
class ModelWrapper:
"""
Wrapper for neural network models to integrate with gradient descent training
Provides a unified interface for different model types and handles
LoRA adapter integration for the MangoMAS system.
"""
def __init__(self, model: nn.Module, model_type: str = 'transformer',
lora_config: Optional[Dict[str, Any]] = None):
self.model = model
self.model_type = model_type
self.lora_config = lora_config or {}
self.lora_params = []
# Initialize LoRA if configured
if lora_config:
self._setup_lora()
logger.info(f"Initialized ModelWrapper for {model_type} model")
def _setup_lora(self):
"""Setup LoRA adapters for the model"""
if not self.lora_config:
return
# Extract LoRA parameters
for name, param in self.model.named_parameters():
if 'lora' in name.lower() or 'adapter' in name.lower():
self.lora_params.append(name)
param.requires_grad = True
else:
param.requires_grad = False
logger.info(f"Setup LoRA with {len(self.lora_params)} adapter parameters")
def forward(self, inputs: torch.Tensor, **kwargs) -> torch.Tensor:
"""
Forward pass through the model
Args:
inputs: Input tensor
**kwargs: Additional arguments
Returns:
Model output tensor
"""
return self.model(inputs, **kwargs)
def get_trainable_parameters(self) -> List[torch.Tensor]:
"""
Get list of trainable parameters
Returns:
List of trainable parameter tensors
"""
if self.lora_params:
# Return only LoRA parameters
return [param for name, param in self.model.named_parameters()
if name in self.lora_params and param.requires_grad]
else:
# Return all trainable parameters
return [param for param in self.model.parameters() if param.requires_grad]
def get_parameter_info(self) -> Dict[str, Any]:
"""
Get information about model parameters
Returns:
Dictionary of parameter information
"""
info = {
'total_parameters': sum(p.numel() for p in self.model.parameters()),
'trainable_parameters': sum(p.numel() for p in self.get_trainable_parameters()),
'lora_parameters': len(self.lora_params),
'parameter_details': {}
}
for name, param in self.model.named_parameters():
info['parameter_details'][name] = {
'shape': list(param.shape),
'numel': param.numel(),
'requires_grad': param.requires_grad,
'is_lora': name in self.lora_params
}
return info
def save_model(self, save_path: str, metadata: Optional[Dict[str, Any]] = None):
"""
Save the model and metadata
Args:
save_path: Path to save the model
metadata: Additional metadata to save
"""
save_path = Path(save_path)
save_path.mkdir(parents=True, exist_ok=True)
# Save model state
model_path = save_path / 'model.pt'
torch.save(self.model.state_dict(), model_path)
# Save metadata
if metadata is None:
metadata = {}
metadata.update({
'model_type': self.model_type,
'lora_config': self.lora_config,
'lora_params': self.lora_params,
'parameter_info': self.get_parameter_info()
})
metadata_path = save_path / 'metadata.json'
with open(metadata_path, 'w') as f:
json.dump(metadata, f, indent=2)
logger.info(f"Model saved to {save_path}")
def load_model(self, load_path: str):
"""
Load the model from saved files
Args:
load_path: Path to load the model from
"""
load_path = Path(load_path)
# Load model state
model_path = load_path / 'model.pt'
if model_path.exists():
state_dict = torch.load(model_path, map_location='cpu')
self.model.load_state_dict(state_dict)
logger.info(f"Model loaded from {model_path}")
# Load metadata
metadata_path = load_path / 'metadata.json'
if metadata_path.exists():
with open(metadata_path, 'r') as f:
metadata = json.load(f)
self.model_type = metadata.get('model_type', self.model_type)
self.lora_config = metadata.get('lora_config', self.lora_config)
self.lora_params = metadata.get('lora_params', self.lora_params)
logger.info(f"Metadata loaded from {metadata_path}")
def to(self, device: torch.device):
"""Move model to device"""
self.model.to(device)
return self
def train(self):
"""Set model to training mode"""
self.model.train()
return self
def eval(self):
"""Set model to evaluation mode"""
self.model.eval()
return self
def __call__(self, *args, **kwargs):
"""Call the model"""
return self.forward(*args, **kwargs)
class LoRAModelWrapper(ModelWrapper):
"""
Specialized wrapper for LoRA (Low-Rank Adaptation) models
Provides enhanced functionality for LoRA adapter management
and integration with the MangoMAS system.
"""
def __init__(self, base_model: nn.Module, lora_config: Dict[str, Any]):
super().__init__(base_model, 'lora_transformer', lora_config)
self.base_model = base_model
self.adapters = {}
# Initialize LoRA adapters
self._initialize_lora_adapters()
def _initialize_lora_adapters(self):
"""Initialize LoRA adapters based on configuration"""
rank = self.lora_config.get('rank', 16)
alpha = self.lora_config.get('alpha', 32)
dropout = self.lora_config.get('dropout', 0.1)
target_modules = self.lora_config.get('target_modules', ['c_attn', 'c_proj'])
# Add LoRA adapters to target modules
for name, module in self.base_model.named_modules():
if any(target in name for target in target_modules):
if isinstance(module, (nn.Linear, nn.Conv2d)):
# Add LoRA adapter
adapter = LoRAAdapter(module, rank, alpha, dropout)
self.adapters[name] = adapter
# Replace original module
self._replace_module(name, adapter)
logger.info(f"Initialized {len(self.adapters)} LoRA adapters")
def _replace_module(self, module_name: str, new_module: nn.Module):
"""Replace a module in the model"""
parts = module_name.split('.')
parent = self.base_model
for part in parts[:-1]:
parent = getattr(parent, part)
setattr(parent, parts[-1], new_module)
def get_lora_parameters(self) -> List[torch.Tensor]:
"""Get LoRA adapter parameters"""
lora_params = []
for adapter in self.adapters.values():
lora_params.extend(adapter.parameters())
return lora_params
def merge_adapters(self):
"""Merge LoRA adapters into base model"""
for adapter in self.adapters.values():
adapter.merge()
logger.info("LoRA adapters merged into base model")
def unmerge_adapters(self):
"""Unmerge LoRA adapters from base model"""
for adapter in self.adapters.values():
adapter.unmerge()
logger.info("LoRA adapters unmerged from base model")
class LoRAAdapter(nn.Module):
"""
LoRA (Low-Rank Adaptation) adapter module
Implements the LoRA technique for efficient fine-tuning of large models.
"""
def __init__(self, original_module: nn.Module, rank: int = 16,
alpha: float = 32, dropout: float = 0.1):
super().__init__()
self.original_module = original_module
self.rank = rank
self.alpha = alpha
self.dropout = dropout
# Get original module dimensions
if isinstance(original_module, nn.Linear):
in_features = original_module.in_features
out_features = original_module.out_features
elif isinstance(original_module, nn.Conv2d):
in_features = original_module.in_channels
out_features = original_module.out_channels
else:
raise ValueError(f"Unsupported module type: {type(original_module)}")
# Initialize LoRA matrices
self.lora_A = nn.Linear(in_features, rank, bias=False)
self.lora_B = nn.Linear(rank, out_features, bias=False)
self.dropout_layer = nn.Dropout(dropout)
# Initialize weights
nn.init.kaiming_uniform_(self.lora_A.weight)
nn.init.zeros_(self.lora_B.weight)
# Store original weights
self.original_weight = original_module.weight.data.clone()
self.merged = False
def forward(self, x: torch.Tensor) -> torch.Tensor:
"""Forward pass through LoRA adapter"""
if self.merged:
# Use merged weights
return F.linear(x, self.original_weight, self.original_module.bias)
else:
# Use LoRA adaptation
lora_output = self.lora_B(self.dropout_layer(self.lora_A(x)))
original_output = F.linear(x, self.original_weight, self.original_module.bias)
return original_output + (self.alpha / self.rank) * lora_output
def merge(self):
"""Merge LoRA weights into original weights"""
if not self.merged:
lora_weight = (self.alpha / self.rank) * torch.mm(
self.lora_B.weight, self.lora_A.weight
)
self.original_weight += lora_weight
self.merged = True
def unmerge(self):
"""Unmerge LoRA weights from original weights"""
if self.merged:
lora_weight = (self.alpha / self.rank) * torch.mm(
self.lora_B.weight, self.lora_A.weight
)
self.original_weight -= lora_weight
self.merged = False
class ModelFactory:
"""Factory class for creating model wrappers"""
@staticmethod
def create_model_wrapper(model_type: str, model: nn.Module,
**kwargs) -> ModelWrapper:
"""Create a model wrapper instance"""
if model_type.lower() == 'lora':
return LoRAModelWrapper(model, kwargs.get('lora_config', {}))
else:
return ModelWrapper(model, model_type, kwargs.get('lora_config'))
@staticmethod
def get_default_lora_config() -> Dict[str, Any]:
"""Get default LoRA configuration"""
return {
'rank': 16,
'alpha': 32,
'dropout': 0.1,
'target_modules': ['c_attn', 'c_proj']
}
|