Materialize API
Instant data materialization from any source directly into .cvc optimized formats.
Version 2.1 Native Stable
GPU-NATIVE
VECTOR STORAGE
Eliminate CPU bottlenecks. Compress embeddings by 4x and decompress them directly in VRAM with zero latency penalty.
user@decompressed:~$
# 1. Install Library
pip install decompressed[gpu]
# 2. Pack Heterogeneous Data
pack_cvc(
embeddings,
"data.cvc",
compression="int8"
)
# 3. Direct GPU Decompression
vectors = load_cvc(
"data.cvc",
device="cuda"
)PineconeWeaviateMilvusQdrantFaiss
HARDWARE-ACCELERATED
INFRASTRUCTURE
GPU Session MGMT
Auto-scaling GPU instances for high-throughput batch decompression.
Selective Filters
Load only specific columns or sections. 5x faster than standard loading.
Quota & Billing
Enterprise-grade usage tracking and billing integration for multi-tenant apps.
Client SDK
Framework agnostic. Works with NumPy, PyTorch, CuPy, and pure C++.
Index Helpers
Native batching helpers for Pinecone, Weaviate, Milvus, and Faiss.
BENCHMARK
VS BASELINE
Speed Optimization
2.2X Faster I/O
Zero-copy memory mapping ensures your GPU is never waiting for the disk.
Storage Efficiency
75% REDUCTION
Native INT8 quantization kernels optimized for modern GPU architectures.
System Metric: Retrieval Latency
Time in Seconds (Lower is Better)
Decompressed INT8
0.632s
PyTorch FP16
0.654s
NumPy (.npy)
0.855s
Decompressed FP16
0.946s
PyTorch INT8
1.431s
Infrastructure Ready
READY TO SCALE
PRODUCTION VECTORS?
Latency: 0.2msThroughput: 1.2M QPSUptime: 99.99%