Spaces:
Running
on
Zero
Running
on
Zero
File size: 17,318 Bytes
55d584b 0ccf2f0 55d584b ec38897 55d584b ec38897 55d584b 0ccf2f0 55d584b 0ccf2f0 55d584b 0ccf2f0 55d584b 0ccf2f0 55d584b |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 |
"""
Summarization Ladder - Hierarchical Memory Compression with Micro and Macro Distillation.
Implements rolling N-window micro-summaries and pipeline macro distillation
for the Cognitive Geo-Thermal Lore Engine v0.3.
"""
from typing import List, Dict, Any, Optional
import time
import hashlib
from dataclasses import dataclass
from collections import deque
@dataclass
class MicroSummary:
"""Rolling window micro-summary with provenance."""
summary_id: str
window_fragments: List[str] # Original fragment IDs in this window
compressed_text: str
window_size: int
creation_timestamp: float
heat_aggregate: float
semantic_centroid: Optional[List[float]] = None
def get_age_seconds(self) -> float:
"""Get summary age in seconds."""
return time.time() - self.creation_timestamp
@dataclass
class MacroDistillation:
"""Macro distillation from N micro-summaries."""
distillation_id: str
source_micro_summaries: List[str] # Micro-summary IDs
distilled_essence: str
consolidation_ratio: float # Original fragments / distilled size
provenance_chain: List[Dict[str, Any]]
creation_timestamp: float
anchor_reinforcements: List[str] # Anchor IDs that were reinforced
class SummarizationLadder:
"""
Hierarchical summarization system with micro-summaries and macro distillation.
Architecture:
- Micro-summaries: Rolling N-window summaries of recent fragments
- Macro distillation: Pipeline processing after N micro-summaries accumulated
- Recovery distillation: Anchor reinforcement during distillation process
"""
def __init__(self, config: Optional[Dict[str, Any]] = None, embedding_provider=None):
"""Initialize the summarization ladder."""
self.config = config or {}
self.embedding_provider = embedding_provider
# Configuration parameters
self.micro_window_size = self.config.get("micro_window_size", 5)
self.macro_trigger_count = self.config.get("macro_trigger_count", 3)
self.max_micro_summaries = self.config.get("max_micro_summaries", 20)
# Storage
self.micro_summaries: deque = deque(maxlen=self.max_micro_summaries)
self.macro_distillations: List[MacroDistillation] = []
self.fragment_buffer: deque = deque(maxlen=self.micro_window_size)
# State tracking
self.total_fragments_processed = 0
self.micro_summaries_created = 0
self.macro_distillations_created = 0
# Metrics
self.metrics = {
"total_fragments": 0,
"micro_summaries_created": 0,
"macro_distillations_created": 0,
"compression_ratio": 0.0,
"processing_time_ms": 0.0,
}
def process_fragments(self, fragments: List[Dict[str, Any]]) -> Dict[str, Any]:
"""
Process incoming fragments through the summarization ladder.
Returns processing report with micro-summaries and any macro distillations.
"""
start_time = time.time()
processing_report = {
"fragments_processed": len(fragments),
"micro_summaries_created": 0,
"macro_distillations_created": 0,
"new_micro_summaries": [],
"new_macro_distillations": [],
}
for fragment in fragments:
self.fragment_buffer.append(fragment)
self.total_fragments_processed += 1
# Check if we should create a micro-summary
if len(self.fragment_buffer) >= self.micro_window_size:
micro_summary = self._create_micro_summary()
if micro_summary:
self.micro_summaries.append(micro_summary)
processing_report["micro_summaries_created"] += 1
processing_report["new_micro_summaries"].append(
{
"summary_id": micro_summary.summary_id,
"compressed_text": micro_summary.compressed_text[:100] + "...",
"window_size": micro_summary.window_size,
"heat_aggregate": micro_summary.heat_aggregate,
}
)
# Check if we should trigger macro distillation
if len(self.micro_summaries) >= self.macro_trigger_count:
macro_distillation = self._create_macro_distillation()
if macro_distillation:
self.macro_distillations.append(macro_distillation)
processing_report["macro_distillations_created"] += 1
processing_report["new_macro_distillations"].append(
{
"distillation_id": macro_distillation.distillation_id,
"distilled_essence": macro_distillation.distilled_essence[:100]
+ "...",
"consolidation_ratio": macro_distillation.consolidation_ratio,
"source_count": len(macro_distillation.source_micro_summaries),
}
)
# Update metrics
elapsed_ms = (time.time() - start_time) * 1000
self.metrics["total_fragments"] = self.total_fragments_processed
self.metrics["micro_summaries_created"] = self.micro_summaries_created
self.metrics["macro_distillations_created"] = self.macro_distillations_created
self.metrics["processing_time_ms"] += elapsed_ms
# Calculate compression ratio
if self.total_fragments_processed > 0:
compressed_units = len(self.micro_summaries) + len(self.macro_distillations)
self.metrics["compression_ratio"] = self.total_fragments_processed / max(
compressed_units, 1
)
processing_report["elapsed_ms"] = elapsed_ms
processing_report["total_micro_summaries"] = len(self.micro_summaries)
processing_report["total_macro_distillations"] = len(self.macro_distillations)
return processing_report
def get_recovery_context(self, anchor_id: str, context_size: int = 3) -> Dict[str, Any]:
"""
Get recovery distillation context for anchor reinforcement.
Returns relevant micro-summaries and macro distillations that relate to the anchor.
"""
recovery_context = {
"anchor_id": anchor_id,
"related_micro_summaries": [],
"related_macro_distillations": [],
"temporal_sequence": [],
"consolidation_path": [],
}
# Find micro-summaries that might relate to the anchor
# (In a full implementation, this would use semantic similarity)
recent_micros = list(self.micro_summaries)[-context_size:]
for micro in recent_micros:
recovery_context["related_micro_summaries"].append(
{
"summary_id": micro.summary_id,
"compressed_text": micro.compressed_text,
"heat_aggregate": micro.heat_aggregate,
"age_seconds": micro.get_age_seconds(),
}
)
# Find relevant macro distillations
recent_macros = self.macro_distillations[-context_size:] if self.macro_distillations else []
for macro in recent_macros:
if anchor_id in macro.anchor_reinforcements:
recovery_context["related_macro_distillations"].append(
{
"distillation_id": macro.distillation_id,
"distilled_essence": macro.distilled_essence,
"consolidation_ratio": macro.consolidation_ratio,
"anchor_reinforcements": macro.anchor_reinforcements,
}
)
# Build temporal sequence showing information flow
all_items = []
for micro in recent_micros:
all_items.append(("micro", micro.creation_timestamp, micro.summary_id))
for macro in recent_macros:
all_items.append(("macro", macro.creation_timestamp, macro.distillation_id))
all_items.sort(key=lambda x: x[1])
recovery_context["temporal_sequence"] = [
{"type": item[0], "timestamp": item[1], "id": item[2]} for item in all_items
]
return recovery_context
def get_compression_metrics(self) -> Dict[str, Any]:
"""Get comprehensive compression and performance metrics."""
return {
"summarization_ladder_metrics": self.metrics.copy(),
"current_state": {
"micro_summaries_active": len(self.micro_summaries),
"macro_distillations_total": len(self.macro_distillations),
"fragment_buffer_size": len(self.fragment_buffer),
"compression_ratio": self.metrics["compression_ratio"],
},
"ladder_health": {
"processing_efficiency": self._calculate_processing_efficiency(),
"compression_effectiveness": self._calculate_compression_effectiveness(),
"temporal_coverage_hours": self._calculate_temporal_coverage(),
},
}
def _create_micro_summary(self) -> Optional[MicroSummary]:
"""Create a micro-summary from the current fragment buffer."""
if len(self.fragment_buffer) < self.micro_window_size:
return None
fragments = list(self.fragment_buffer)
# Extract fragment IDs and text
fragment_ids = [f.get("id", f"frag_{i}") for i, f in enumerate(fragments)]
fragment_texts = [f.get("text", "") for f in fragments]
# Simple summarization (in production, would use more sophisticated methods)
compressed_text = self._compress_fragment_texts(fragment_texts)
# Calculate aggregate heat
heat_aggregate = sum(f.get("heat", 0.1) for f in fragments) / len(fragments)
# Generate semantic centroid if embedding provider available
semantic_centroid = None
if self.embedding_provider and fragment_texts:
try:
embeddings = [self.embedding_provider.embed_text(text) for text in fragment_texts]
if embeddings:
# Calculate centroid
dim = len(embeddings[0])
semantic_centroid = [
sum(emb[i] for emb in embeddings) / len(embeddings) for i in range(dim)
]
except Exception:
# Fallback to None if embedding fails
pass
# Create micro-summary
summary_id = self._generate_summary_id(compressed_text)
micro_summary = MicroSummary(
summary_id=summary_id,
window_fragments=fragment_ids,
compressed_text=compressed_text,
window_size=len(fragments),
creation_timestamp=time.time(),
heat_aggregate=heat_aggregate,
semantic_centroid=semantic_centroid,
)
self.micro_summaries_created += 1
# Clear part of buffer to allow for overlap
overlap_size = max(1, self.micro_window_size // 3)
for _ in range(len(self.fragment_buffer) - overlap_size):
self.fragment_buffer.popleft()
return micro_summary
def _create_macro_distillation(self) -> Optional[MacroDistillation]:
"""Create a macro distillation from recent micro-summaries."""
if len(self.micro_summaries) < self.macro_trigger_count:
return None
# Take the oldest micro-summaries for distillation
source_summaries = []
source_summary_ids = []
for _ in range(self.macro_trigger_count):
if self.micro_summaries:
micro = self.micro_summaries.popleft()
source_summaries.append(micro)
source_summary_ids.append(micro.summary_id)
if not source_summaries:
return None
# Distill the essence from micro-summaries
distilled_essence = self._distill_macro_essence(source_summaries)
# Calculate consolidation ratio
total_original_fragments = sum(len(micro.window_fragments) for micro in source_summaries)
consolidation_ratio = total_original_fragments / 1.0 # 1 distillation from N fragments
# Build provenance chain
provenance_chain = [
{
"micro_summary_id": micro.summary_id,
"original_fragments": len(micro.window_fragments),
"heat_contribution": micro.heat_aggregate,
"creation_timestamp": micro.creation_timestamp,
}
for micro in source_summaries
]
# Mock anchor reinforcements (in production, would integrate with SemanticAnchorGraph)
anchor_reinforcements = [f"anchor_reinforce_{i}" for i in range(len(source_summaries))]
# Create macro distillation
distillation_id = self._generate_distillation_id(distilled_essence)
macro_distillation = MacroDistillation(
distillation_id=distillation_id,
source_micro_summaries=source_summary_ids,
distilled_essence=distilled_essence,
consolidation_ratio=consolidation_ratio,
provenance_chain=provenance_chain,
creation_timestamp=time.time(),
anchor_reinforcements=anchor_reinforcements,
)
self.macro_distillations_created += 1
return macro_distillation
def _compress_fragment_texts(self, texts: List[str]) -> str:
"""Compress multiple fragment texts into a micro-summary."""
if not texts:
return "(empty window)"
# Simple compression: take key phrases from each text
key_phrases = []
for text in texts:
# Extract first meaningful phrase (up to 30 chars)
clean_text = text.strip()
if clean_text:
phrase = clean_text[:30]
if len(clean_text) > 30:
phrase += "..."
key_phrases.append(phrase)
# Combine into micro-summary
if len(key_phrases) == 1:
return f"[Micro] {key_phrases[0]}"
else:
return f"[Micro] {' • '.join(key_phrases[:3])}" # Limit to 3 phrases
def _distill_macro_essence(self, micro_summaries: List[MicroSummary]) -> str:
"""Distill macro essence from multiple micro-summaries."""
if not micro_summaries:
return "(empty distillation)"
# Extract key themes from micro-summaries
themes = []
for micro in micro_summaries:
# Extract meaningful content from micro-summary
content = micro.compressed_text.replace("[Micro]", "").strip()
if content:
themes.append(content)
# Create macro distillation
if len(themes) == 1:
return f"[Macro] {themes[0]}"
else:
# Combine themes into higher-level abstraction
combined = " ⟶ ".join(themes[:2]) # Show progression
return f"[Macro] {combined}"
def _generate_summary_id(self, content: str) -> str:
"""Generate unique ID for micro-summary."""
timestamp = str(int(time.time() * 1000))
content_hash = hashlib.md5(content.encode()).hexdigest()[:8]
return f"micro_{timestamp}_{content_hash}"
def _generate_distillation_id(self, essence: str) -> str:
"""Generate unique ID for macro distillation."""
timestamp = str(int(time.time() * 1000))
essence_hash = hashlib.md5(essence.encode()).hexdigest()[:8]
return f"macro_{timestamp}_{essence_hash}"
def _calculate_processing_efficiency(self) -> float:
"""Calculate processing efficiency metric."""
if self.metrics["total_fragments"] == 0:
return 1.0
total_time_seconds = self.metrics["processing_time_ms"] / 1000.0
if total_time_seconds == 0:
return 1.0
return self.metrics["total_fragments"] / total_time_seconds
def _calculate_compression_effectiveness(self) -> float:
"""Calculate how effectively we're compressing information."""
return min(self.metrics["compression_ratio"] / 10.0, 1.0) # Normalize to 0-1
def _calculate_temporal_coverage(self) -> float:
"""Calculate temporal coverage in hours."""
if not self.micro_summaries and not self.macro_distillations:
return 0.0
oldest_time = time.time()
newest_time = 0
for micro in self.micro_summaries:
oldest_time = min(oldest_time, micro.creation_timestamp)
newest_time = max(newest_time, micro.creation_timestamp)
for macro in self.macro_distillations:
oldest_time = min(oldest_time, macro.creation_timestamp)
newest_time = max(newest_time, macro.creation_timestamp)
if newest_time > oldest_time:
return (newest_time - oldest_time) / 3600.0 # Convert to hours
return 0.0
|