Why the E8 lattice is the perfect quantizer for KV cachesMost quantizers are chosen for convenience. E8 was chosen because the math demanded it — and then it surprised us. What Makes E8 Special The E8 lattice is a root lattice in 8 dimensions with 240 nearest neighbors. Its kissing number (240) is the high...Apr 7, 2026·3 min read
KV cache memory calculator: how much does your LLM actually use?Before you can compress something, you need to know how big it is. Most engineers know the KV cache is "large" but few have actually calculated the exact number. This post gives you the formula, a table for popular models, and a one-liner to compute ...Apr 7, 2026·3 min read
How to benchmark NexusQuant on your own modelRunning benchmarks on someone else's hardware tells you very little. This guide shows you how to measure NexusQuant's impact on your model, your data, and your hardware in under 15 minutes. Prerequisites pip install nexusquant-kv transformers torch ...Apr 7, 2026·2 min read
What I Learned Testing 12 Compression Approaches That FailedNegative results from building NexusQuant: PCA rotation, per-head eviction, token merging, adaptive bitwidth, and 8 more dead ends.Apr 7, 2026·8 min read
The Math Behind E8 Lattice Quantization (with Code)Conway-Sloane algorithm, integer and half-integer cosets, relaxed parity discovery, and the full nearest_point() walkthrough.Apr 7, 2026·7 min read
How Much GPU Memory Does NexusQuant Actually Save?A concrete memory calculator: real GB numbers for A100, A10G, RTX 3090, with Python code for any model.Apr 7, 2026·5 min read
How to deploy NexusQuant in production (and what's missing)This post is a practical deployment guide. Install, configuration, how to pick the right eviction rate, domain testing, and an honest list of what does not work yet. Install pip install nexusquant Requires Python 3.9+, PyTorch 2.1+, and Transformer...Apr 7, 2026·5 min read