File size: 13,242 Bytes
3cb0dc4 |
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 |
"""
Fast Batch Processing Module
Optimized batch processing that works around generate_batch issues
"""
import json
import time
import logging
from collections import defaultdict, Counter
from pathlib import Path
from typing import List, Dict, Tuple, Any
import torch
class FastBatchProcessor:
"""Fast batch processing using optimized individual calls"""
def __init__(self, model, tolerance=0.05, min_batch_size=2, max_batch_size=8):
"""
Initialize fast batch processor
Args:
model: ChatterboxTTS model instance
tolerance: Parameter tolerance for grouping (0.05 = 5% variation allowed)
min_batch_size: Minimum chunks to form a batch
max_batch_size: Maximum chunks per batch (memory/performance limit)
"""
self.model = model
self.tolerance = tolerance
self.min_batch_size = min_batch_size
self.max_batch_size = max_batch_size
self.logger = logging.getLogger(__name__)
# Performance tracking
self.stats = {
'total_chunks': 0,
'batched_chunks': 0,
'individual_chunks': 0,
'batch_groups': 0,
'total_time': 0,
'batch_time': 0,
'individual_time': 0,
'parameter_switches': 0
}
def analyze_chunk_distribution(self, chunks: List[Dict]) -> Dict:
"""Analyze parameter distribution and batching potential"""
if not chunks:
return {'error': 'No chunks provided'}
# Extract parameters
param_combos = []
for chunk in chunks:
if 'tts_params' not in chunk:
continue
params = chunk['tts_params']
combo = (
round(params.get('exaggeration', 0.5), 3),
round(params.get('cfg_weight', 0.5), 3),
round(params.get('temperature', 0.8), 3),
round(params.get('min_p', 0.05), 3),
round(params.get('repetition_penalty', 1.2), 3)
)
param_combos.append(combo)
# Count combinations
combo_counts = Counter(param_combos)
unique_combos = len(combo_counts)
# Calculate batching potential (consecutive chunks with same params)
consecutive_groups = self._find_consecutive_groups(param_combos)
batchable = sum(len(group) for group in consecutive_groups if len(group) >= self.min_batch_size)
# Estimate speedup from reduced parameter switches
total_param_switches = len(set(param_combos))
optimized_switches = len(consecutive_groups)
switch_reduction = (total_param_switches - optimized_switches) / max(total_param_switches, 1)
analysis = {
'total_chunks': len(chunks),
'unique_combinations': unique_combos,
'consecutive_batchable': batchable,
'batch_percentage': (batchable / len(chunks)) * 100,
'consecutive_groups': len(consecutive_groups),
'parameter_switch_reduction': switch_reduction * 100,
'most_common_combos': combo_counts.most_common(5),
'estimated_speedup': self._estimate_speedup(len(chunks), batchable, switch_reduction)
}
return analysis
def _find_consecutive_groups(self, param_combos: List[Tuple]) -> List[List[int]]:
"""Find consecutive chunks with same parameters"""
if not param_combos:
return []
groups = []
current_group = [0]
current_params = param_combos[0]
for i in range(1, len(param_combos)):
if self._params_within_tolerance(current_params, param_combos[i]):
current_group.append(i)
else:
groups.append(current_group)
current_group = [i]
current_params = param_combos[i]
# Add the last group
groups.append(current_group)
return groups
def _params_within_tolerance(self, params1: Tuple, params2: Tuple) -> bool:
"""Check if two parameter sets are within tolerance"""
for p1, p2 in zip(params1, params2):
if abs(p1 - p2) > self.tolerance:
return False
return True
def _estimate_speedup(self, total_chunks: int, batchable_chunks: int, switch_reduction: float) -> Dict:
"""Estimate performance improvements"""
# Parameter switching overhead reduction
switch_speedup = 1.0 + (switch_reduction * 0.3) # 30% speedup from fewer switches
# Memory optimization speedup (fewer allocations/deallocations)
memory_speedup = 1.0 + (batchable_chunks / total_chunks * 0.4) # Up to 40% from memory optimization
# Combined speedup
combined_speedup = switch_speedup * memory_speedup
return {
'parameter_switch_speedup': switch_speedup,
'memory_optimization_speedup': memory_speedup,
'combined_speedup': combined_speedup,
'estimated_time_saving': ((combined_speedup - 1.0) / combined_speedup) * 100
}
def process_chunks_fast_batch(self, chunks: List[Dict], use_tolerance: bool = True) -> List[torch.Tensor]:
"""
Process chunks using fast batch optimization
Args:
chunks: List of chunk dictionaries from JSON
use_tolerance: Whether to use parameter tolerance
Returns:
List of audio tensors in original chunk order
"""
if not chunks:
return []
start_time = time.time()
self.stats['total_chunks'] = len(chunks)
# Group consecutive chunks with similar parameters
consecutive_groups = self._group_consecutive_chunks(chunks, use_tolerance)
# Initialize results array
results = [None] * len(chunks)
# Process each group with optimized parameter handling
current_params = None
param_switches = 0
for group in consecutive_groups:
group_start = time.time()
# Extract parameters for this group (use first chunk's params as representative)
_, first_chunk = group['chunks'][0]
target_params = first_chunk.get('tts_params', {})
# Check if we need to update model parameters
if current_params != target_params:
self._update_model_parameters(target_params)
current_params = target_params.copy()
param_switches += 1
# Process all chunks in this group with same parameters
self.logger.info(f"π₯ Processing group of {len(group['chunks'])} chunks with same parameters")
for idx, chunk in group['chunks']:
try:
# Generate without parameter overhead (params already set)
audio = self.model.generate(
chunk['text'],
exaggeration=target_params.get('exaggeration', 0.5),
cfg_weight=target_params.get('cfg_weight', 0.5),
temperature=target_params.get('temperature', 0.8),
min_p=target_params.get('min_p', 0.05),
top_p=target_params.get('top_p', 1.0),
repetition_penalty=target_params.get('repetition_penalty', 1.2)
)
results[idx] = audio
self.stats['batched_chunks'] += 1
except Exception as e:
self.logger.error(f"β Chunk {idx} failed: {e}")
# Create silent audio as fallback
results[idx] = torch.zeros(1, 24000)
self.stats['individual_chunks'] += 1
group_time = time.time() - group_start
self.logger.info(f" β
Group completed in {group_time:.2f}s ({len(group['chunks'])} chunks)")
self.stats['parameter_switches'] = param_switches
self.stats['batch_groups'] = len(consecutive_groups)
self.stats['total_time'] = time.time() - start_time
# Log performance summary
self._log_performance_summary()
return results
def _group_consecutive_chunks(self, chunks: List[Dict], use_tolerance: bool = True) -> List[Dict]:
"""Group consecutive chunks with similar parameters"""
if not chunks:
return []
groups = []
current_group = []
current_params = None
for i, chunk in enumerate(chunks):
if 'tts_params' not in chunk:
# Handle chunks without parameters as individual
if current_group:
groups.append({'chunks': current_group, 'params': current_params})
current_group = []
groups.append({'chunks': [(i, chunk)], 'params': {}})
current_params = None
continue
chunk_params = chunk['tts_params']
# Convert to comparable format
if use_tolerance:
param_signature = (
round(chunk_params.get('exaggeration', 0.5) / self.tolerance) * self.tolerance,
round(chunk_params.get('cfg_weight', 0.5) / self.tolerance) * self.tolerance,
round(chunk_params.get('temperature', 0.8) / self.tolerance) * self.tolerance,
round(chunk_params.get('min_p', 0.05) / self.tolerance) * self.tolerance,
round(chunk_params.get('repetition_penalty', 1.2) / self.tolerance) * self.tolerance
)
else:
param_signature = (
chunk_params.get('exaggeration', 0.5),
chunk_params.get('cfg_weight', 0.5),
chunk_params.get('temperature', 0.8),
chunk_params.get('min_p', 0.05),
chunk_params.get('repetition_penalty', 1.2)
)
# Check if this chunk can be grouped with current group
if current_params is None or param_signature == current_params:
current_group.append((i, chunk))
current_params = param_signature
else:
# Start new group
if current_group:
groups.append({'chunks': current_group, 'params': current_params})
current_group = [(i, chunk)]
current_params = param_signature
# Add the last group
if current_group:
groups.append({'chunks': current_group, 'params': current_params})
return groups
def _update_model_parameters(self, params: Dict):
"""Update model with new parameters (placeholder for future optimization)"""
# For now, parameters are passed directly to generate()
# Future optimization: pre-configure model components with parameters
pass
def _log_performance_summary(self):
"""Log performance statistics"""
stats = self.stats
self.logger.info("π FAST BATCH PROCESSING SUMMARY")
self.logger.info("=" * 50)
self.logger.info(f"Total chunks processed: {stats['total_chunks']}")
self.logger.info(f"Batched chunks: {stats['batched_chunks']} ({stats['batched_chunks']/stats['total_chunks']*100:.1f}%)")
self.logger.info(f"Individual chunks: {stats['individual_chunks']} ({stats['individual_chunks']/stats['total_chunks']*100:.1f}%)")
self.logger.info(f"Batch groups: {stats['batch_groups']}")
self.logger.info(f"Parameter switches: {stats['parameter_switches']}")
self.logger.info(f"Total time: {stats['total_time']:.2f}s")
# Calculate optimization efficiency
if stats['total_chunks'] > 0:
switch_efficiency = (stats['total_chunks'] - stats['parameter_switches']) / stats['total_chunks'] * 100
self.logger.info(f"Parameter switch efficiency: {switch_efficiency:.1f}%")
# Estimate what naive processing would have taken (one switch per chunk)
estimated_naive_switches = stats['total_chunks']
switch_reduction = (estimated_naive_switches - stats['parameter_switches']) / estimated_naive_switches * 100
self.logger.info(f"Parameter switches reduced by: {switch_reduction:.1f}%")
def load_chunks_from_json(json_path: str) -> List[Dict]:
"""Load chunks from JSON file"""
try:
with open(json_path, 'r', encoding='utf-8') as f:
data = json.load(f)
# Filter out metadata entries
chunks = [item for item in data if isinstance(item, dict) and 'text' in item]
return chunks
except Exception as e:
logging.error(f"Failed to load JSON file {json_path}: {e}")
return [] |