Spaces:
Running
Running
File size: 18,319 Bytes
168b0da |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 |
"""
Modal Vector Service - GPU-accelerated vector memory processing
This service provides:
- GPU-accelerated embedding generation using sentence-transformers
- FAISS with Modal Volume storage for scalable vector search
- FAISS for fast similarity search optimization
- Auto-scaling based on workload
"""
import os
import time
import json
import modal
import asyncio
from typing import List, Dict, Any, Optional
# Modal App Configuration
app = modal.App("memvid-vector-service")
# Docker image with all vector processing dependencies
vector_image = modal.Image.debian_slim().pip_install(
[
"sentence-transformers>=2.0.0",
"faiss-cpu>=1.8.0",
"numpy>=1.24.0",
"scikit-learn>=1.3.0", # For additional vector operations
]
)
# Volume for persistent model storage
models_volume = modal.Volume.from_name("vector-models", create_if_missing=True)
@app.function(
image=vector_image,
gpu="A100", # High-performance GPU for embedding generation
volumes={"/models": models_volume},
timeout=600, # 10 minutes timeout for large operations
)
def process_vector_memory(
text: str, client_id: str, metadata: Dict[str, Any]
) -> Dict[str, Any]:
"""
GPU-accelerated vector memory processing on Modal
Args:
text: Text content to store as vector embeddings
client_id: Unique identifier for the client/user
metadata: Additional metadata for the memory
Returns:
Dict with processing results and metrics
"""
import numpy as np
from sentence_transformers import SentenceTransformer
import json
start_time = time.time()
try:
# Load or download sentence transformer model (cached in volume)
model_path = "/models/sentence-transformer"
if not os.path.exists(model_path):
print("π₯ Downloading sentence transformer model...")
model = SentenceTransformer("all-MiniLM-L6-v2", device="cuda")
model.save(model_path)
else:
print("π Loading cached sentence transformer model...")
model = SentenceTransformer(model_path, device="cuda")
# Generate embeddings on GPU
print(f"π Generating embeddings for text: {text[:100]}...")
embeddings = model.encode([text], device="cuda")
embedding_vector = embeddings[0].tolist() # Convert to list for JSON storage
# Calculate processing metrics
embedding_time = time.time() - start_time
# Store vector in Modal Volume with FAISS index
import faiss
import pickle
storage_path = f"/models/vectors/{client_id}"
os.makedirs(storage_path, exist_ok=True)
# Load or create FAISS index
index_path = f"{storage_path}/faiss_index.bin"
metadata_path = f"{storage_path}/metadata.json"
if os.path.exists(index_path):
print("π Loading existing FAISS index...")
index = faiss.read_index(index_path)
with open(metadata_path, "r") as f:
all_metadata = json.load(f)
else:
print("π Creating new FAISS index...")
# Create FAISS index for 384-dimensional vectors
index = faiss.IndexFlatIP(384) # Inner product for cosine similarity
all_metadata = []
# Add vector to index
vector_array = np.array([embedding_vector], dtype=np.float32)
# Normalize for cosine similarity
faiss.normalize_L2(vector_array)
index.add(vector_array)
# Store metadata
memory_id = f"vector_{len(all_metadata)}"
memory_metadata = {
"id": memory_id,
"client_id": client_id,
"text": text,
"metadata": metadata,
"created_at": time.time(),
}
all_metadata.append(memory_metadata)
# Save updated index and metadata
faiss.write_index(index, index_path)
with open(metadata_path, "w") as f:
json.dump(all_metadata, f)
print(
f"β
Vector memory stored with ID: {memory_id} (FAISS index size: {index.ntotal})"
)
total_time = time.time() - start_time
return {
"success": True,
"memory_id": memory_id,
"client_id": client_id,
"embedding_dim": len(embedding_vector),
"embedding_preview": embedding_vector[:5], # First 5 dimensions for preview
"processing_metrics": {
"embedding_time": embedding_time,
"total_time": total_time,
"storage_size": len(embedding_vector) * 4, # 4 bytes per float32
"gpu_used": "A100",
"model_used": "all-MiniLM-L6-v2",
},
"metadata": metadata,
"infrastructure": "Modal + A100 GPU + FAISS + Volume Storage",
}
except Exception as e:
print(f"β Error in vector processing: {str(e)}")
return {
"success": False,
"error": str(e),
"processing_time": time.time() - start_time,
"infrastructure": "Modal + A100 GPU + FAISS + Volume Storage",
}
@app.function(
image=vector_image,
gpu="A100",
volumes={"/models": models_volume},
timeout=300, # 5 minutes timeout for search operations
)
def search_vector_memory(
query: str, client_id: str, memory_name: Optional[str] = None, top_k: int = 5
) -> Dict[str, Any]:
"""
Ultra-fast vector similarity search on Modal
Args:
query: Search query text
client_id: Client identifier to search within
memory_name: Optional specific memory name filter
top_k: Number of top results to return
Returns:
Dict with search results and metrics
"""
import numpy as np
from sentence_transformers import SentenceTransformer
import json
start_time = time.time()
try:
# Load model for query embedding
model_path = "/models/sentence-transformer"
model = SentenceTransformer(model_path, device="cuda")
# Generate query embedding
query_embedding = model.encode([query], device="cuda")[0].tolist()
embedding_time = time.time() - start_time
# Search in Modal Volume with FAISS
storage_path = f"/models/vectors/{client_id}"
index_path = f"{storage_path}/faiss_index.bin"
metadata_path = f"{storage_path}/metadata.json"
if os.path.exists(index_path) and os.path.exists(metadata_path):
print("π Searching in FAISS index...")
import faiss
# Load FAISS index and metadata
index = faiss.read_index(index_path)
with open(metadata_path, "r") as f:
all_metadata = json.load(f)
# Prepare query vector
query_vector = np.array([query_embedding], dtype=np.float32)
faiss.normalize_L2(query_vector)
# Perform similarity search
scores, indices = index.search(query_vector, min(top_k, index.ntotal))
# Format results
formatted_results = []
for i, (score, idx) in enumerate(zip(scores[0], indices[0])):
if idx < len(all_metadata): # Valid index
metadata_item = all_metadata[idx]
formatted_results.append(
{
"memory_id": metadata_item["id"],
"text": metadata_item["text"],
"metadata": metadata_item.get("metadata", {}),
"similarity_score": float(score),
"distance": 1 - float(score),
}
)
else:
# No stored vectors yet
formatted_results = []
search_time = time.time() - start_time
return {
"success": True,
"query": query,
"client_id": client_id,
"results": formatted_results,
"total_results": len(formatted_results),
"processing_metrics": {
"embedding_time": embedding_time,
"search_time": search_time - embedding_time,
"total_time": search_time,
"gpu_used": "A100",
"model_used": "all-MiniLM-L6-v2",
},
"infrastructure": "Modal + A100 GPU + FAISS + Volume Storage",
}
except Exception as e:
print(f"β Error in vector search: {str(e)}")
return {
"success": False,
"error": str(e),
"processing_time": time.time() - start_time,
"results": [],
"infrastructure": "Modal + A100 GPU + FAISS + Volume Storage",
}
@app.function(
image=vector_image,
volumes={"/models": models_volume},
timeout=60,
)
def get_vector_stats(client_id: str) -> Dict[str, Any]:
"""
Get statistics for vector storage
Args:
client_id: Client identifier
Returns:
Dict with storage statistics
"""
import json
import os
try:
storage_path = f"/models/vectors/{client_id}"
index_path = f"{storage_path}/faiss_index.bin"
metadata_path = f"{storage_path}/metadata.json"
if os.path.exists(index_path) and os.path.exists(metadata_path):
import faiss
# Load FAISS index and metadata
index = faiss.read_index(index_path)
with open(metadata_path, "r") as f:
all_metadata = json.load(f)
# Calculate stats
memory_count = len(all_metadata)
first_memory = (
min(item["created_at"] for item in all_metadata)
if all_metadata
else None
)
last_memory = (
max(item["created_at"] for item in all_metadata)
if all_metadata
else None
)
return {
"client_id": client_id,
"storage_type": "modal_vector_faiss",
"memory_count": memory_count,
"avg_embedding_dim": 384, # all-MiniLM-L6-v2 dimension
"index_size": index.ntotal,
"first_memory": (
time.strftime("%Y-%m-%dT%H:%M:%S", time.localtime(first_memory))
if first_memory
else None
),
"last_memory": (
time.strftime("%Y-%m-%dT%H:%M:%S", time.localtime(last_memory))
if last_memory
else None
),
"infrastructure": "Modal + A100 GPU + FAISS + Volume Storage",
}
else:
return {
"client_id": client_id,
"storage_type": "modal_vector_faiss",
"memory_count": 0,
"infrastructure": "Modal + A100 GPU + FAISS + Volume Storage",
"note": "No vectors stored yet",
}
except Exception as e:
return {
"client_id": client_id,
"storage_type": "modal_vector_faiss",
"error": str(e),
"infrastructure": "Modal + A100 GPU + FAISS + Volume Storage",
}
# Client class for easy integration with DualStorageManager
class ModalVectorClient:
"""Client for interacting with Modal Vector Service"""
def __init__(self, modal_token: Optional[str] = None):
"""
Initialize Modal Vector Client
Args:
modal_token: Optional Modal token (uses environment if not provided)
"""
if modal_token:
os.environ["MODAL_TOKEN"] = modal_token
# Test Modal connection
try:
import modal
print("β
Modal Vector Client initialized successfully")
except Exception as e:
print(f"β οΈ Modal Vector Client initialization warning: {e}")
def store_memory(
self, text: str, client_id: str, metadata: Dict[str, Any]
) -> Dict[str, Any]:
"""Store memory using Modal vector service"""
try:
# Use the deployed app's function with correct Modal calling pattern
import modal
func = modal.Function.from_name(
"memvid-vector-service", "process_vector_memory"
)
return func.remote(text, client_id, metadata)
except Exception as e:
return {"success": False, "error": f"Modal vector storage failed: {e}"}
def search_memory(
self,
query: str,
client_id: str,
memory_name: Optional[str] = None,
top_k: int = 5,
) -> Dict[str, Any]:
"""Search memory using Modal vector service"""
try:
# Use the deployed app's function with correct Modal calling pattern
import modal
func = modal.Function.from_name(
"memvid-vector-service", "search_vector_memory"
)
return func.remote(query, client_id, memory_name, top_k)
except Exception as e:
return {
"success": False,
"error": f"Modal vector search failed: {e}",
"results": [],
}
def get_stats(self, client_id: str) -> Dict[str, Any]:
"""Get statistics using Modal vector service"""
try:
# Use the deployed app's function with correct Modal calling pattern
import modal
func = modal.Function.from_name("memvid-vector-service", "get_vector_stats")
return func.remote(client_id)
except Exception as e:
return {"success": False, "error": f"Modal vector stats failed: {e}"}
def list_memories(self, client_id: str) -> str:
"""List memories for client (Modal vector implementation)"""
try:
stats = self.get_stats(client_id)
if stats.get(
"success", True
): # Modal stats don't have success field currently
memory_list = {
"client_id": client_id,
"storage_type": "modal_vector",
"memory_count": stats.get("memory_count", 0),
"memories": [], # Modal doesn't currently track individual memory names
"avg_embedding_dim": stats.get("avg_embedding_dim", 0),
"infrastructure": "Modal + A100 GPU + PostgreSQL + pgvector",
}
return json.dumps(memory_list, indent=2)
else:
return json.dumps(
{
"error": f"Failed to list memories: {stats.get('error', 'Unknown error')}"
}
)
except Exception as e:
return json.dumps({"error": f"Modal vector list_memories failed: {e}"})
def build_memory_video(self, client_id: str, memory_name: str) -> str:
"""Build memory video (not applicable for vector storage)"""
return f"Memory videos are not applicable for vector storage. Client: {client_id}, Memory: {memory_name}"
def chat_with_memory(self, query: str, client_id: str, memory_name: str) -> str:
"""Chat with memory using Modal vector service"""
try:
# Use search as basis for chat
search_results = self.search_memory(query, client_id, memory_name, top_k=3)
if search_results.get("success", False):
results = search_results.get("results", [])
if results:
# Simple chat response based on search results
context = "\n".join(
[result.get("text", "") for result in results[:2]]
)
response = f"Based on your vector memories: {context}\n\nYour query '{query}' relates to the stored information above."
return response
else:
return f"I couldn't find any relevant memories for '{query}' in your vector storage."
else:
return f"Error accessing memories: {search_results.get('error', 'Unknown error')}"
except Exception as e:
return f"Modal vector chat failed: {e}"
def delete_memory(self, client_id: str, memory_name: str) -> str:
"""Delete memory (Modal vector implementation)"""
# Modal currently doesn't support selective deletion
return f"Memory deletion not yet implemented in Modal vector storage for client {client_id}, memory {memory_name}"
def get_memory_stats(self, client_id: str) -> str:
"""Get memory statistics as JSON string"""
try:
stats = self.get_stats(client_id)
return json.dumps(stats, indent=2)
except Exception as e:
return json.dumps({"error": f"Modal vector get_memory_stats failed: {e}"})
# For compatibility with the dual storage manager method calls
def store_embedding(
self, text: str, client_id: str, metadata: Dict[str, Any]
) -> str:
"""Alias for store_memory for backward compatibility"""
result = self.store_memory(text, client_id, metadata)
return json.dumps(result) if isinstance(result, dict) else str(result)
def search_embeddings(self, query: str, client_id: str, top_k: int = 5) -> str:
"""Alias for search_memory for backward compatibility"""
result = self.search_memory(query, client_id, top_k=top_k)
return json.dumps(result) if isinstance(result, dict) else str(result)
if __name__ == "__main__":
# Test the Modal functions locally
print("π§ͺ Testing Modal Vector Service...")
# Test client
client = ModalVectorClient()
# Test storage
result = client.store_memory(
"This is a test memory for Modal vector storage",
"test_client",
{"test": True, "timestamp": time.time()},
)
print(f"π₯ Storage result: {result}")
# Test search
search_result = client.search_memory("test memory", "test_client", top_k=3)
print(f"π Search result: {search_result}")
# Test stats
stats = client.get_stats("test_client")
print(f" Stats: {stats}")
|