File size: 7,055 Bytes
964f845 | 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 | """
Progress Tracking Utilities with tqdm
Provides:
- Epoch-level progress bars
- Step-level metrics display
- ETA calculation
- Formatted logging
"""
import time
from typing import Dict, Optional
from tqdm import tqdm
class TrainingProgressTracker:
"""
Manages training progress with tqdm bars and ETA.
"""
def __init__(
self,
total_epochs: int,
steps_per_epoch: int,
log_interval: int = 100,
val_interval: int = 2000,
):
self.total_epochs = total_epochs
self.steps_per_epoch = steps_per_epoch
self.log_interval = log_interval
self.val_interval = val_interval
# Progress bars
self.epoch_bar = None
self.step_bar = None
# Timing
self.epoch_start_time = None
self.training_start_time = time.time()
# Metrics
self.current_epoch = 0
self.global_step = 0
def start_epoch(self, epoch: int):
"""Start a new epoch"""
self.current_epoch = epoch
self.epoch_start_time = time.time()
# Create epoch-level progress bar
if self.epoch_bar is not None:
self.epoch_bar.close()
self.epoch_bar = tqdm(
total=self.steps_per_epoch,
desc=f"Epoch {epoch}/{self.total_epochs}",
position=0,
leave=True,
bar_format="{desc}: {percentage:3.0f}%|{bar}| {n}/{total} [{elapsed}<{remaining}, {rate_fmt}]"
)
def update_step(
self,
metrics: Dict[str, float],
global_step: int,
):
"""
Update progress for a single step.
Args:
metrics: Dict of metrics to display (loss, lr, etc.)
global_step: Global training step
"""
self.global_step = global_step
# Format metrics for display
metric_str = {}
for key, value in metrics.items():
if 'loss' in key.lower():
metric_str[key] = f"{value:.4f}"
elif 'lr' in key.lower():
metric_str[key] = f"{value:.2e}"
elif 'scale' in key.lower():
metric_str[key] = f"{value:.2f}"
else:
metric_str[key] = f"{value:.4f}"
# Add global step
metric_str['step'] = f"{global_step}"
# Update epoch bar
if self.epoch_bar is not None:
self.epoch_bar.set_postfix(metric_str)
self.epoch_bar.update(1)
def end_epoch(self):
"""End current epoch"""
if self.epoch_bar is not None:
self.epoch_bar.close()
self.epoch_bar = None
# Print epoch summary
epoch_time = time.time() - self.epoch_start_time
total_time = time.time() - self.training_start_time
print(f"\n{'='*60}")
print(f"Epoch {self.current_epoch} complete!")
print(f" Epoch time: {self.format_time(epoch_time)}")
print(f" Total training time: {self.format_time(total_time)}")
print(f" Global step: {self.global_step}")
print(f"{'='*60}\n")
def close(self):
"""Close all progress bars"""
if self.epoch_bar is not None:
self.epoch_bar.close()
total_time = time.time() - self.training_start_time
print(f"\n{'='*60}")
print(f"Training complete!")
print(f" Total time: {self.format_time(total_time)}")
print(f" Total steps: {self.global_step}")
print(f" Average steps/sec: {self.global_step / total_time:.2f}")
print(f"{'='*60}\n")
@staticmethod
def format_time(seconds: float) -> str:
"""Format seconds into human-readable string"""
if seconds < 60:
return f"{seconds:.1f}s"
elif seconds < 3600:
minutes = int(seconds // 60)
secs = int(seconds % 60)
return f"{minutes}m {secs}s"
else:
hours = int(seconds // 3600)
minutes = int((seconds % 3600) // 60)
return f"{hours}h {minutes}m"
def get_eta(self, current_step: int, total_steps: int) -> str:
"""Calculate ETA for remaining steps"""
if current_step == 0:
return "calculating..."
elapsed = time.time() - self.training_start_time
steps_per_sec = current_step / elapsed
remaining_steps = total_steps - current_step
eta_seconds = remaining_steps / steps_per_sec
return self.format_time(eta_seconds)
def format_metrics_for_wandb(metrics: Dict[str, float], prefix: str = "") -> Dict[str, float]:
"""
Format metrics for W&B logging.
Args:
metrics: Raw metrics dict
prefix: Prefix to add to keys (e.g., "train/", "val/")
Returns:
formatted_metrics: Formatted for W&B
"""
formatted = {}
for key, value in metrics.items():
if isinstance(value, (int, float)):
formatted[f"{prefix}{key}"] = value
return formatted
def print_training_header(config: Dict):
"""Print formatted training configuration"""
print("\n" + "="*80)
print("FROZEN VLM-IP2P TRAINING".center(80))
print("="*80)
print("\nConfiguration:")
print("-" * 80)
# Group configs
groups = {
"Data": ["data_path", "image_folder", "batch_size", "num_workers"],
"Model": ["resolution", "num_vlm_tokens", "resampler_depth"],
"Training": ["num_epochs", "lr_perceiver", "lr_adapter", "weight_decay"],
"CFG": ["cfg_dropout", "image_guidance_scale", "text_guidance_scale", "vlm_guidance_scale"],
"Validation": ["val_interval", "num_val_samples", "val_scheduler", "num_inference_steps"],
"Output": ["output_dir", "save_interval", "log_interval"],
}
for group_name, keys in groups.items():
print(f"\n{group_name}:")
for key in keys:
if key in config:
value = config[key]
print(f" {key:30s}: {value}")
print("\n" + "="*80 + "\n")
if __name__ == "__main__":
# Test progress tracker
print("Testing progress tracker...")
tracker = TrainingProgressTracker(
total_epochs=3,
steps_per_epoch=100,
log_interval=10,
)
# Simulate training
import random
for epoch in range(1, 4):
tracker.start_epoch(epoch)
for step in range(100):
global_step = (epoch - 1) * 100 + step
# Simulate metrics
metrics = {
'loss': random.uniform(0.1, 0.5),
'lr': 1e-4 * (0.99 ** global_step),
'ema_decay': 0.9999,
}
tracker.update_step(metrics, global_step)
time.sleep(0.01) # Simulate training time
tracker.end_epoch()
tracker.close()
print("\n✓ Progress tracker test complete!")
|