Spaces:
Build error
Build error
File size: 13,446 Bytes
7a05808 e4a41fa cdde22e e4a41fa 7a05808 cdde22e 7a05808 cdde22e 7a05808 bf779b5 7a05808 c8388c1 7a05808 c8388c1 7a05808 e4a41fa 7a05808 bf779b5 7a05808 e4a41fa 7a05808 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 | """
Semantic Cache Layer.
Provides query-level semantic caching using embedding similarity.
This is COMPLEMENTARY to provider prompt caching - it caches complete
responses for semantically similar queries.
How it works:
1. When a query comes in, compute its embedding
2. Search for similar queries in the cache (cosine similarity)
3. If similarity > threshold, return cached response
4. Otherwise, proceed with normal optimization
Key difference from Prompt Caching:
- Prompt Caching: Provider caches KV-cache for prefix (same prompt = faster)
- Semantic Caching: We cache responses for similar queries (similar query = cached answer)
Usage:
from headroom.cache import SemanticCacheLayer, CacheOptimizerRegistry
# Get provider optimizer
provider_optimizer = CacheOptimizerRegistry.get("anthropic")
# Wrap with semantic layer
semantic = SemanticCacheLayer(
provider_optimizer,
similarity_threshold=0.95,
)
result = semantic.process(messages, context)
if result.semantic_cache_hit:
# Use result.cached_response directly
pass
"""
from __future__ import annotations
import hashlib
import time
from collections import OrderedDict
from collections.abc import Callable
from dataclasses import dataclass, field
from typing import Any
from headroom.models.config import ML_MODEL_DEFAULTS
from .base import (
BaseCacheOptimizer,
CacheConfig,
CacheMetrics,
CacheResult,
OptimizationContext,
)
@dataclass
class CacheEntry:
"""Entry in the semantic cache."""
# Query embedding
embedding: list[float]
# Original query text
query: str
# Cached response
response: Any
# Metadata
created_at: float
last_accessed: float
access_count: int = 1
# Hash of the full messages for exact matching
messages_hash: str = ""
@dataclass
class SemanticCacheConfig:
"""Configuration for semantic caching."""
# Similarity threshold for cache hit (0.0 - 1.0)
similarity_threshold: float = 0.95
# Maximum entries in cache
max_entries: int = 1000
# TTL in seconds (0 = no expiry)
ttl_seconds: int = 300
# Whether to use exact hash matching as fallback
use_exact_matching: bool = True
# Embedding model (if using embeddings)
embedding_model: str = field(default_factory=lambda: ML_MODEL_DEFAULTS.sentence_transformer)
class SemanticCache:
"""
In-memory semantic cache with LRU eviction.
Stores query embeddings and responses, supporting both
semantic similarity search and exact hash matching.
"""
def __init__(
self,
config: SemanticCacheConfig | None = None,
embedding_fn: Callable[[str], list[float]] | None = None,
):
"""
Initialize the semantic cache.
Args:
config: Cache configuration
embedding_fn: Optional custom embedding function.
If not provided, uses simple hash-based matching.
"""
self.config = config or SemanticCacheConfig()
self._embedding_fn = embedding_fn
# LRU cache: key -> CacheEntry
self._cache: OrderedDict[str, CacheEntry] = OrderedDict()
# Exact hash index: messages_hash -> key
self._hash_index: dict[str, str] = {}
# Statistics
self._hits = 0
self._misses = 0
self._evictions = 0
def get(
self,
query: str,
messages_hash: str | None = None,
) -> CacheEntry | None:
"""
Look up a cached entry.
Args:
query: Query text to search for
messages_hash: Optional exact hash for fast lookup
Returns:
CacheEntry if found, None otherwise
"""
self._cleanup_expired()
# Try exact hash match first
if messages_hash and self.config.use_exact_matching:
key = self._hash_index.get(messages_hash)
if key and key in self._cache:
entry = self._cache[key]
self._touch(key)
self._hits += 1
return entry
# Try semantic similarity if we have embedding function
if self._embedding_fn:
query_embedding = self._embedding_fn(query)
best_match, best_similarity = self._find_similar(query_embedding)
if best_similarity >= self.config.similarity_threshold:
self._touch(best_match)
self._hits += 1
return self._cache[best_match]
self._misses += 1
return None
def put(
self,
query: str,
response: Any,
messages_hash: str | None = None,
) -> str:
"""
Store a response in the cache.
Args:
query: Query text
response: Response to cache
messages_hash: Optional exact hash for fast lookup
Returns:
Cache key for the entry
"""
self._cleanup_expired()
# Evict if at capacity
while len(self._cache) >= self.config.max_entries:
self._evict_oldest()
# Generate embedding if available
embedding: list[float] = []
if self._embedding_fn:
embedding = self._embedding_fn(query)
# Create cache key
key = self._generate_key(query)
now = time.time()
entry = CacheEntry(
embedding=embedding,
query=query,
response=response,
created_at=now,
last_accessed=now,
messages_hash=messages_hash or "",
)
self._cache[key] = entry
# Index by hash for fast exact matching
if messages_hash:
self._hash_index[messages_hash] = key
return key
def invalidate(self, key: str) -> bool:
"""Invalidate a cache entry by key."""
if key in self._cache:
entry = self._cache.pop(key)
if entry.messages_hash:
self._hash_index.pop(entry.messages_hash, None)
return True
return False
def clear(self) -> None:
"""Clear all cache entries."""
self._cache.clear()
self._hash_index.clear()
def get_stats(self) -> dict[str, Any]:
"""Get cache statistics."""
total = self._hits + self._misses
hit_rate = self._hits / total if total > 0 else 0.0
return {
"entries": len(self._cache),
"max_entries": self.config.max_entries,
"hits": self._hits,
"misses": self._misses,
"hit_rate": hit_rate,
"evictions": self._evictions,
}
def _find_similar(
self,
query_embedding: list[float],
) -> tuple[str, float]:
"""Find the most similar cached entry."""
best_key = ""
best_similarity = -1.0
for key, entry in self._cache.items():
if not entry.embedding:
continue
similarity = self._cosine_similarity(query_embedding, entry.embedding)
if similarity > best_similarity:
best_similarity = similarity
best_key = key
return best_key, best_similarity
def _cosine_similarity(
self,
a: list[float],
b: list[float],
) -> float:
"""Compute cosine similarity between two vectors."""
if len(a) != len(b) or not a:
return 0.0
dot_product = sum(x * y for x, y in zip(a, b))
norm_a = sum(x * x for x in a) ** 0.5
norm_b = sum(x * x for x in b) ** 0.5
if norm_a == 0 or norm_b == 0:
return 0.0
return float(dot_product / (norm_a * norm_b))
def _touch(self, key: str) -> None:
"""Update access time and move to end of LRU."""
try:
entry = self._cache.pop(key)
except KeyError:
return
entry.last_accessed = time.time()
entry.access_count += 1
self._cache[key] = entry
def _evict_oldest(self) -> None:
"""Evict the oldest (least recently used) entry."""
if self._cache:
key, entry = self._cache.popitem(last=False)
if entry.messages_hash:
self._hash_index.pop(entry.messages_hash, None)
self._evictions += 1
def _cleanup_expired(self) -> None:
"""Remove expired entries."""
if self.config.ttl_seconds <= 0:
return
now = time.time()
expired = [
key
for key, entry in self._cache.items()
if now - entry.created_at > self.config.ttl_seconds
]
for key in expired:
entry = self._cache.pop(key)
if entry.messages_hash:
self._hash_index.pop(entry.messages_hash, None)
def _generate_key(self, query: str) -> str:
"""Generate a cache key for a query."""
return hashlib.sha256(query.encode()).hexdigest()[:16]
class SemanticCacheLayer:
"""
Layer that adds semantic caching on top of provider optimizers.
This layer checks for semantically similar queries before
delegating to the underlying provider optimizer.
"""
def __init__(
self,
provider_optimizer: BaseCacheOptimizer,
similarity_threshold: float = 0.95,
max_entries: int = 1000,
ttl_seconds: int = 300,
embedding_fn: Callable[[str], list[float]] | None = None,
):
"""
Initialize the semantic cache layer.
Args:
provider_optimizer: Underlying provider optimizer
similarity_threshold: Similarity threshold for cache hits
max_entries: Maximum cache entries
ttl_seconds: Cache TTL in seconds
embedding_fn: Optional embedding function
"""
self.provider_optimizer = provider_optimizer
cache_config = SemanticCacheConfig(
similarity_threshold=similarity_threshold,
max_entries=max_entries,
ttl_seconds=ttl_seconds,
)
self.cache = SemanticCache(cache_config, embedding_fn)
def process(
self,
messages: list[dict[str, Any]],
context: OptimizationContext,
config: CacheConfig | None = None,
) -> CacheResult:
"""
Process messages through semantic cache and provider optimizer.
Args:
messages: Messages to process
context: Optimization context
config: Optional configuration override
Returns:
CacheResult with semantic_cache_hit=True if cache hit
"""
# Extract query for semantic matching
query = context.query or self._extract_query(messages)
messages_hash = self._compute_messages_hash(messages)
# Check semantic cache
cached = self.cache.get(query, messages_hash)
if cached:
return CacheResult(
messages=messages,
semantic_cache_hit=True,
cached_response=cached.response,
metrics=CacheMetrics(
estimated_cache_hit=True,
estimated_savings_percent=100.0,
),
transforms_applied=["semantic_cache_hit"],
)
# Delegate to provider optimizer
result = self.provider_optimizer.optimize(messages, context, config)
return result
def store_response(
self,
messages: list[dict[str, Any]],
response: Any,
context: OptimizationContext | None = None,
) -> str:
"""
Store a response in the semantic cache.
Call this after receiving a response from the LLM to enable
future cache hits.
Args:
messages: Original messages
response: Response from LLM
context: Optional context with query
Returns:
Cache key
"""
query = (context.query if context else None) or self._extract_query(messages)
messages_hash = self._compute_messages_hash(messages)
return self.cache.put(query, response, messages_hash)
def get_stats(self) -> dict[str, Any]:
"""Get combined statistics."""
return {
"semantic_cache": self.cache.get_stats(),
"provider_optimizer": self.provider_optimizer.name,
}
def _extract_query(self, messages: list[dict[str, Any]]) -> str:
"""Extract the last user query from messages."""
for msg in reversed(messages):
if msg.get("role") == "user":
content = msg.get("content", "")
if isinstance(content, str):
return content
elif isinstance(content, list):
for block in content:
if isinstance(block, dict) and block.get("type") == "text":
text_val = block.get("text", "")
return str(text_val) if text_val else ""
return ""
def _compute_messages_hash(self, messages: list[dict[str, Any]]) -> str:
"""Compute a hash of all messages."""
import json
try:
content = json.dumps(messages, sort_keys=True)
return hashlib.sha256(content.encode()).hexdigest()[:24]
except (TypeError, ValueError):
return ""
|