turbovec

name	turbovec
description	High-performance vector index built on TurboQuant with Rust core and Python bindings
version	1.0.0
author	Hermes Agent
license	MIT
platforms	["linux","macos","windows"]
metadata	{"hermes":{"tags":["vector-database","similarity-search","ml","rust","python"],"related_skills":["llama-cpp","serving-llms-vllm"]}}

turbovec

A fast, lightweight vector index for similarity search, ideal for recommendation systems, semantic search, and RAG pipelines.

Prerequisites

• Python 3.8+
• pip

Installation

pip install turbovec

Usage

Basic Vector Operations

import turbovec
import numpy as np

# Create an index with dimension 128
index = turbovec.Index(dim=128)

# Generate some random vectors
vectors = np.random.randn(1000, 128).astype(np.float32)
ids = list(range(1000))

# Add vectors to the index
index.add(vectors, ids)

# Search for similar vectors
query = np.random.randn(1, 128).astype(np.float32)
results = index.search(query, k=10)

print(f"Found {len(results[0])} similar vectors")

Persistence

# Save index to disk
index.save("my_index.tvec")

# Load index from disk
loaded_index = turbovec.Index.load("my_index.tvec")

Integration with sentence-transformers

from sentence_transformers import SentenceTransformer
import turbovec

# Encode sentences
model = SentenceTransformer('all-MiniLM-L6-v2')
sentences = ["This is a test", "Another sentence", "Third example"]
embeddings = model.encode(sentences)

# Create and populate index
index = turbovec.Index(dim=384)
index.add(embeddings, list(range(len(sentences))))

# Search
query_embedding = model.encode(["Find similar sentences"])
results = index.search(query_embedding, k=2)

Common Pitfalls

1. Memory usage: Large indices require sufficient RAM. Consider indexing strategies.
2. Dimension consistency: All vectors must have the same dimension.
3. Update limitations: Turbovec is optimized for append-only workloads. Frequent updates may require rebuilding.
4. Distance metrics: Currently supports L2 distance. For cosine similarity, normalize vectors first.

Verification

# Test installation
python -c "import turbovec; print('turbovec imported successfully')"

# Run a simple test
python -c "
import turbovec
import numpy as np

index = turbovec.Index(dim=64)
vectors = np.random.randn(100, 64).astype(np.float32)
index.add(vectors, list(range(100)))
query = np.random.randn(1, 64).astype(np.float32)
results = index.search(query, k=5)
print(f'Test passed: found {len(results[0])} results')
"