Debito's picture
Upload utils.py
4372b35 verified
raw
history blame
27.9 kB
# =============================================================================
# utils/utils.py - Utility Functions for Mamba Encoder Swarm Architecture
# =============================================================================
import torch
import torch.nn as nn
import torch.nn.functional as F
import numpy as np
import time
import json
import logging
import os
import psutil
import gc
from typing import Dict, List, Tuple, Optional, Union, Any
from collections import defaultdict, deque
from datetime import datetime, timedelta
import threading
import warnings
from functools import wraps, lru_cache
import hashlib
import pickle
# Setup logging
logger = logging.getLogger(__name__)
# =============================================================================
# PERFORMANCE MONITORING UTILITIES
# =============================================================================
class PerformanceMonitor:
"""Monitor and track performance metrics for the swarm architecture"""
def __init__(self, max_history: int = 1000):
self.metrics = defaultdict(list)
self.max_history = max_history
self.start_times = {}
self.counters = defaultdict(int)
self.lock = threading.Lock()
def start_timer(self, name: str) -> None:
"""Start timing an operation"""
with self.lock:
self.start_times[name] = time.time()
def end_timer(self, name: str) -> float:
"""End timing and record duration"""
with self.lock:
if name in self.start_times:
duration = time.time() - self.start_times[name]
self.record_metric(f"{name}_duration", duration)
del self.start_times[name]
return duration
return 0.0
def record_metric(self, name: str, value: float) -> None:
"""Record a metric value"""
with self.lock:
self.metrics[name].append({
'value': value,
'timestamp': time.time()
})
# Keep only recent history
if len(self.metrics[name]) > self.max_history:
self.metrics[name] = self.metrics[name][-self.max_history:]
def increment_counter(self, name: str, amount: int = 1) -> None:
"""Increment a counter"""
with self.lock:
self.counters[name] += amount
def get_stats(self, name: str) -> Dict[str, float]:
"""Get statistics for a metric"""
with self.lock:
if name not in self.metrics or not self.metrics[name]:
return {}
values = [m['value'] for m in self.metrics[name]]
return {
'count': len(values),
'mean': np.mean(values),
'std': np.std(values),
'min': np.min(values),
'max': np.max(values),
'median': np.median(values),
'recent': values[-10:] if len(values) >= 10 else values
}
def get_summary(self) -> Dict[str, Any]:
"""Get complete performance summary"""
with self.lock:
summary = {
'metrics': {name: self.get_stats(name) for name in self.metrics},
'counters': dict(self.counters),
'active_timers': list(self.start_times.keys()),
'timestamp': datetime.now().isoformat()
}
return summary
# Global performance monitor instance
perf_monitor = PerformanceMonitor()
def monitor_performance(func_name: str = None):
"""Decorator to monitor function performance"""
def decorator(func):
name = func_name or f"{func.__module__}.{func.__name__}"
@wraps(func)
def wrapper(*args, **kwargs):
perf_monitor.start_timer(name)
perf_monitor.increment_counter(f"{name}_calls")
try:
result = func(*args, **kwargs)
perf_monitor.increment_counter(f"{name}_success")
return result
except Exception as e:
perf_monitor.increment_counter(f"{name}_errors")
raise
finally:
perf_monitor.end_timer(name)
return wrapper
return decorator
# =============================================================================
# MEMORY MANAGEMENT UTILITIES
# =============================================================================
class MemoryTracker:
"""Track memory usage across the swarm system"""
@staticmethod
def get_memory_info() -> Dict[str, float]:
"""Get current memory information"""
process = psutil.Process()
memory_info = process.memory_info()
virtual_memory = psutil.virtual_memory()
gpu_memory = {}
if torch.cuda.is_available():
for i in range(torch.cuda.device_count()):
gpu_memory[f'gpu_{i}'] = {
'allocated': torch.cuda.memory_allocated(i) / 1024**3,
'cached': torch.cuda.memory_reserved(i) / 1024**3,
'max_allocated': torch.cuda.max_memory_allocated(i) / 1024**3
}
return {
'process_memory_gb': memory_info.rss / 1024**3,
'system_memory_percent': virtual_memory.percent,
'system_memory_available_gb': virtual_memory.available / 1024**3,
'gpu_memory': gpu_memory
}
@staticmethod
def clear_gpu_cache():
"""Clear GPU memory cache"""
if torch.cuda.is_available():
torch.cuda.empty_cache()
gc.collect()
@staticmethod
def optimize_memory():
"""Perform memory optimization"""
gc.collect()
if torch.cuda.is_available():
torch.cuda.empty_cache()
def memory_efficient(clear_cache: bool = True):
"""Decorator for memory-efficient functions"""
def decorator(func):
@wraps(func)
def wrapper(*args, **kwargs):
if clear_cache:
MemoryTracker.clear_gpu_cache()
try:
result = func(*args, **kwargs)
return result
finally:
if clear_cache:
MemoryTracker.clear_gpu_cache()
return wrapper
return decorator
# =============================================================================
# TENSOR UTILITIES
# =============================================================================
class TensorUtils:
"""Utility functions for tensor operations"""
@staticmethod
def safe_tensor_to_device(tensor: torch.Tensor, device: torch.device) -> torch.Tensor:
"""Safely move tensor to device with error handling"""
try:
if tensor.device != device:
return tensor.to(device)
return tensor
except RuntimeError as e:
logger.warning(f"Failed to move tensor to {device}: {e}")
return tensor
@staticmethod
def get_tensor_info(tensor: torch.Tensor) -> Dict[str, Any]:
"""Get comprehensive tensor information"""
return {
'shape': list(tensor.shape),
'dtype': str(tensor.dtype),
'device': str(tensor.device),
'requires_grad': tensor.requires_grad,
'memory_mb': tensor.numel() * tensor.element_size() / 1024**2,
'is_contiguous': tensor.is_contiguous(),
'stride': tensor.stride() if tensor.dim() > 0 else []
}
@staticmethod
def batch_tensors(tensors: List[torch.Tensor], pad_value: float = 0.0) -> torch.Tensor:
"""Batch tensors with padding to same length"""
if not tensors:
return torch.empty(0)
max_len = max(t.size(-1) for t in tensors)
batch_size = len(tensors)
if len(tensors[0].shape) == 1:
batched = torch.full((batch_size, max_len), pad_value, dtype=tensors[0].dtype, device=tensors[0].device)
else:
feature_dim = tensors[0].size(-2)
batched = torch.full((batch_size, feature_dim, max_len), pad_value, dtype=tensors[0].dtype, device=tensors[0].device)
for i, tensor in enumerate(tensors):
if len(tensor.shape) == 1:
batched[i, :tensor.size(0)] = tensor
else:
batched[i, :, :tensor.size(-1)] = tensor
return batched
@staticmethod
def split_tensor_by_chunks(tensor: torch.Tensor, chunk_size: int) -> List[torch.Tensor]:
"""Split tensor into chunks of specified size"""
if tensor.size(0) <= chunk_size:
return [tensor]
return [tensor[i:i + chunk_size] for i in range(0, tensor.size(0), chunk_size)]
# =============================================================================
# ROUTING UTILITIES
# =============================================================================
class RoutingUtils:
"""Utilities for encoder routing and load balancing"""
@staticmethod
def calculate_load_balance_loss(routing_weights: torch.Tensor, epsilon: float = 1e-8) -> torch.Tensor:
"""Calculate load balance loss to encourage equal encoder usage"""
# routing_weights: [batch_size, seq_len, num_encoders]
avg_routing = routing_weights.mean(dim=[0, 1]) # [num_encoders]
# Variance penalty to encourage uniform distribution
target_load = 1.0 / routing_weights.size(-1)
load_balance_loss = torch.var(avg_routing) / (target_load ** 2 + epsilon)
return load_balance_loss
@staticmethod
def apply_top_k_routing(logits: torch.Tensor, k: int) -> Tuple[torch.Tensor, torch.Tensor]:
"""Apply top-k routing with Gumbel softmax"""
# Get top-k indices
top_k_logits, top_k_indices = torch.topk(logits, k, dim=-1)
# Create mask for top-k
mask = torch.zeros_like(logits)
mask.scatter_(-1, top_k_indices, 1.0)
# Apply Gumbel softmax to top-k
gumbel_noise = -torch.log(-torch.log(torch.rand_like(top_k_logits) + 1e-8) + 1e-8)
top_k_weights = F.softmax((top_k_logits + gumbel_noise) / 1.0, dim=-1)
# Reconstruct full weights
weights = torch.zeros_like(logits)
weights.scatter_(-1, top_k_indices, top_k_weights)
return weights, mask
@staticmethod
def entropy_regularization(routing_weights: torch.Tensor) -> torch.Tensor:
"""Add entropy regularization to encourage exploration"""
# Avoid log(0)
routing_weights = torch.clamp(routing_weights, min=1e-8)
entropy = -torch.sum(routing_weights * torch.log(routing_weights), dim=-1)
return -entropy.mean() # Negative because we want to maximize entropy
# =============================================================================
# TEXT PROCESSING UTILITIES
# =============================================================================
class TextUtils:
"""Utilities for text processing and analysis"""
@staticmethod
def chunk_text(text: str, chunk_size: int = 512, overlap: int = 50) -> List[str]:
"""Split text into overlapping chunks"""
words = text.split()
if len(words) <= chunk_size:
return [text]
chunks = []
start = 0
while start < len(words):
end = min(start + chunk_size, len(words))
chunk = ' '.join(words[start:end])
chunks.append(chunk)
if end >= len(words):
break
start = end - overlap
return chunks
@staticmethod
def estimate_tokens(text: str, chars_per_token: float = 4.0) -> int:
"""Estimate number of tokens in text"""
return max(1, int(len(text) / chars_per_token))
@staticmethod
def clean_text(text: str) -> str:
"""Clean and normalize text"""
if not text:
return ""
# Remove excessive whitespace
text = ' '.join(text.split())
# Remove control characters
text = ''.join(char for char in text if ord(char) >= 32 or char in '\n\t')
return text.strip()
@staticmethod
def detect_language(text: str) -> str:
"""Simple language detection based on character patterns"""
# This is a simplified version - for production, use langdetect library
if not text:
return "unknown"
# Count character types
ascii_count = sum(1 for c in text if ord(c) < 128)
total_chars = len(text)
if total_chars == 0:
return "unknown"
ascii_ratio = ascii_count / total_chars
if ascii_ratio > 0.9:
return "en" # Likely English
elif ascii_ratio > 0.7:
return "mixed"
else:
return "non-latin"
# =============================================================================
# CONFIGURATION UTILITIES
# =============================================================================
class ConfigUtils:
"""Utilities for configuration management"""
@staticmethod
def load_config(config_path: str) -> Dict[str, Any]:
"""Load configuration from JSON file"""
try:
with open(config_path, 'r', encoding='utf-8') as f:
config = json.load(f)
logger.info(f"Loaded configuration from {config_path}")
return config
except Exception as e:
logger.error(f"Failed to load config from {config_path}: {e}")
return {}
@staticmethod
def save_config(config: Dict[str, Any], config_path: str) -> bool:
"""Save configuration to JSON file"""
try:
os.makedirs(os.path.dirname(config_path), exist_ok=True)
with open(config_path, 'w', encoding='utf-8') as f:
json.dump(config, f, indent=2, ensure_ascii=False)
logger.info(f"Saved configuration to {config_path}")
return True
except Exception as e:
logger.error(f"Failed to save config to {config_path}: {e}")
return False
@staticmethod
def merge_configs(base_config: Dict[str, Any], override_config: Dict[str, Any]) -> Dict[str, Any]:
"""Merge two configuration dictionaries"""
merged = base_config.copy()
for key, value in override_config.items():
if key in merged and isinstance(merged[key], dict) and isinstance(value, dict):
merged[key] = ConfigUtils.merge_configs(merged[key], value)
else:
merged[key] = value
return merged
@staticmethod
def validate_config(config: Dict[str, Any], required_keys: List[str]) -> List[str]:
"""Validate configuration has required keys"""
missing_keys = []
for key in required_keys:
if '.' in key:
# Handle nested keys
keys = key.split('.')
current = config
for k in keys:
if not isinstance(current, dict) or k not in current:
missing_keys.append(key)
break
current = current[k]
elif key not in config:
missing_keys.append(key)
return missing_keys
# =============================================================================
# CACHING UTILITIES
# =============================================================================
class CacheManager:
"""Intelligent caching for model outputs and computations"""
def __init__(self, max_size: int = 1000, ttl_seconds: int = 3600):
self.max_size = max_size
self.ttl_seconds = ttl_seconds
self.cache = {}
self.access_times = {}
self.lock = threading.Lock()
def _generate_key(self, *args, **kwargs) -> str:
"""Generate cache key from arguments"""
key_data = {
'args': args,
'kwargs': sorted(kwargs.items())
}
return hashlib.md5(pickle.dumps(key_data)).hexdigest()
def get(self, key: str) -> Optional[Any]:
"""Get item from cache"""
with self.lock:
if key not in self.cache:
return None
# Check TTL
if time.time() - self.cache[key]['timestamp'] > self.ttl_seconds:
self._remove_key(key)
return None
self.access_times[key] = time.time()
return self.cache[key]['value']
def put(self, key: str, value: Any) -> None:
"""Put item in cache"""
with self.lock:
# Clean up if cache is full
if len(self.cache) >= self.max_size:
self._evict_lru()
self.cache[key] = {
'value': value,
'timestamp': time.time()
}
self.access_times[key] = time.time()
def _remove_key(self, key: str) -> None:
"""Remove key from cache"""
if key in self.cache:
del self.cache[key]
if key in self.access_times:
del self.access_times[key]
def _evict_lru(self) -> None:
"""Evict least recently used item"""
if not self.access_times:
return
lru_key = min(self.access_times.keys(), key=lambda k: self.access_times[k])
self._remove_key(lru_key)
def clear(self) -> None:
"""Clear all cached items"""
with self.lock:
self.cache.clear()
self.access_times.clear()
def stats(self) -> Dict[str, Any]:
"""Get cache statistics"""
with self.lock:
return {
'size': len(self.cache),
'max_size': self.max_size,
'hit_ratio': getattr(self, '_hits', 0) / max(getattr(self, '_requests', 1), 1),
'ttl_seconds': self.ttl_seconds
}
# Global cache manager
cache_manager = CacheManager()
def cached(ttl_seconds: int = 3600):
"""Decorator for caching function results"""
def decorator(func):
@wraps(func)
def wrapper(*args, **kwargs):
cache_key = cache_manager._generate_key(func.__name__, *args, **kwargs)
# Try to get from cache
result = cache_manager.get(cache_key)
if result is not None:
return result
# Compute and cache
result = func(*args, **kwargs)
cache_manager.put(cache_key, result)
return result
return wrapper
return decorator
# =============================================================================
# DEBUGGING AND LOGGING UTILITIES
# =============================================================================
class DebugUtils:
"""Utilities for debugging the swarm architecture"""
@staticmethod
def log_tensor_stats(tensor: torch.Tensor, name: str) -> None:
"""Log comprehensive tensor statistics"""
if not tensor.numel():
logger.debug(f"{name}: Empty tensor")
return
stats = {
'shape': list(tensor.shape),
'dtype': str(tensor.dtype),
'device': str(tensor.device),
'mean': tensor.float().mean().item(),
'std': tensor.float().std().item(),
'min': tensor.min().item(),
'max': tensor.max().item(),
'has_nan': torch.isnan(tensor).any().item(),
'has_inf': torch.isinf(tensor).any().item()
}
logger.debug(f"{name} stats: {stats}")
@staticmethod
def validate_tensor(tensor: torch.Tensor, name: str, check_finite: bool = True) -> bool:
"""Validate tensor for common issues"""
if not isinstance(tensor, torch.Tensor):
logger.error(f"{name}: Not a tensor, got {type(tensor)}")
return False
if tensor.numel() == 0:
logger.warning(f"{name}: Empty tensor")
return False
if check_finite:
if torch.isnan(tensor).any():
logger.error(f"{name}: Contains NaN values")
return False
if torch.isinf(tensor).any():
logger.error(f"{name}: Contains infinite values")
return False
return True
@staticmethod
def trace_function_calls(func):
"""Decorator to trace function calls"""
@wraps(func)
def wrapper(*args, **kwargs):
logger.debug(f"Calling {func.__name__} with args: {len(args)}, kwargs: {list(kwargs.keys())}")
start_time = time.time()
try:
result = func(*args, **kwargs)
duration = time.time() - start_time
logger.debug(f"{func.__name__} completed in {duration:.4f}s")
return result
except Exception as e:
duration = time.time() - start_time
logger.error(f"{func.__name__} failed after {duration:.4f}s: {e}")
raise
return wrapper
# =============================================================================
# SYSTEM UTILITIES
# =============================================================================
class SystemUtils:
"""System-level utilities"""
@staticmethod
def get_system_info() -> Dict[str, Any]:
"""Get comprehensive system information"""
cpu_info = {
'cpu_count': psutil.cpu_count(),
'cpu_percent': psutil.cpu_percent(interval=1),
'load_average': os.getloadavg() if hasattr(os, 'getloadavg') else None
}
memory_info = psutil.virtual_memory()._asdict()
gpu_info = {}
if torch.cuda.is_available():
gpu_info = {
'device_count': torch.cuda.device_count(),
'current_device': torch.cuda.current_device(),
'devices': [
{
'name': torch.cuda.get_device_name(i),
'memory_total': torch.cuda.get_device_properties(i).total_memory,
'memory_allocated': torch.cuda.memory_allocated(i),
'memory_cached': torch.cuda.memory_reserved(i)
}
for i in range(torch.cuda.device_count())
]
}
return {
'cpu': cpu_info,
'memory': memory_info,
'gpu': gpu_info,
'python_version': f"{__import__('sys').version_info.major}.{__import__('sys').version_info.minor}",
'torch_version': torch.__version__,
'timestamp': datetime.now().isoformat()
}
@staticmethod
def ensure_directory(path: str) -> None:
"""Ensure directory exists"""
os.makedirs(path, exist_ok=True)
@staticmethod
def safe_file_write(content: str, filepath: str, backup: bool = True) -> bool:
"""Safely write content to file with backup"""
try:
# Create directory if needed
os.makedirs(os.path.dirname(filepath), exist_ok=True)
# Create backup if file exists
if backup and os.path.exists(filepath):
backup_path = f"{filepath}.backup"
import shutil
shutil.copy2(filepath, backup_path)
# Write content
with open(filepath, 'w', encoding='utf-8') as f:
f.write(content)
return True
except Exception as e:
logger.error(f"Failed to write file {filepath}: {e}")
return False
# =============================================================================
# EXPORT UTILITIES
# =============================================================================
def format_model_size(num_params: int) -> str:
"""Format model size in human-readable format"""
for unit in ['', 'K', 'M', 'B', 'T']:
if num_params < 1000:
return f"{num_params:.1f}{unit}"
num_params /= 1000
return f"{num_params:.1f}P"
def format_memory_size(bytes_size: int) -> str:
"""Format memory size in human-readable format"""
for unit in ['B', 'KB', 'MB', 'GB', 'TB']:
if bytes_size < 1024:
return f"{bytes_size:.1f}{unit}"
bytes_size /= 1024
return f"{bytes_size:.1f}PB"
def format_duration(seconds: float) -> str:
"""Format duration in human-readable format"""
if seconds < 1:
return f"{seconds*1000:.1f}ms"
elif seconds < 60:
return f"{seconds:.1f}s"
elif seconds < 3600:
minutes = seconds / 60
return f"{minutes:.1f}m"
else:
hours = seconds / 3600
return f"{hours:.1f}h"
# =============================================================================
# INITIALIZATION
# =============================================================================
def initialize_logging(log_level: str = "INFO", log_file: Optional[str] = None) -> None:
"""Initialize logging configuration"""
level = getattr(logging, log_level.upper(), logging.INFO)
handlers = [logging.StreamHandler()]
if log_file:
handlers.append(logging.FileHandler(log_file))
logging.basicConfig(
level=level,
format='%(asctime)s - %(name)s - %(levelname)s - %(message)s',
handlers=handlers
)
def setup_warnings() -> None:
"""Setup warning filters"""
# Filter out common warnings that don't affect functionality
warnings.filterwarnings("ignore", category=UserWarning, module="torch")
warnings.filterwarnings("ignore", category=FutureWarning, module="transformers")
# Initialize on import
setup_warnings()
# =============================================================================
# MAIN UTILITIES EXPORT
# =============================================================================
__all__ = [
# Performance monitoring
'PerformanceMonitor', 'perf_monitor', 'monitor_performance',
# Memory management
'MemoryTracker', 'memory_efficient',
# Tensor utilities
'TensorUtils',
# Routing utilities
'RoutingUtils',
# Text processing
'TextUtils',
# Configuration
'ConfigUtils',
# Caching
'CacheManager', 'cache_manager', 'cached',
# Debugging
'DebugUtils',
# System utilities
'SystemUtils',
# Formatting utilities
'format_model_size', 'format_memory_size', 'format_duration',
# Initialization
'initialize_logging', 'setup_warnings'
]