Published July 3, 2026 | Version v17
Technical note Open

Madhava V17 — BIGANN-100M Streaming Stress Test & Benchmark (Verified, int8, Zero Bound Violations)

Authors/Creators

  • 1. Winnex AI

Description

Madhava V17 — BIGANN-100M Streaming Stress Test & Benchmark (Verified, int8, Zero Bound Violations)

C++ implementation with Modified Gram-Schmidt orthogonal projections, int8 quantization with dynamic dequantization margin, and empirical bound violation verification against exact cosine similarity.

Results (28-thread CPU, AVX2+FMA, 100M vectors)

  • Build: 21.5s
  • Query latency: 1.9s
  • Memory: 18.6GB (int8 projection cache)
  • NDCG@10: 1.000 • R@10: 1.000
  • Bound violations: 0% (verified empirically)
  • Early exit: 124 cossenos exatos por query

Kaggle Notebook

https://www.kaggle.com/code/kleniopadilha/madhava-v12-bigann-verified

BSL 1.1 | pay@winnex.ai

Files

madhava_v12_verified.ipynb

Files (346.2 kB)

Name Size Download all
md5:e525c6be26e7fd43c0ffa8863d1e1ee8
10.6 kB Download
md5:650031d6574397d3e3fe65726ea52119
15.3 kB Download
md5:895ad450126e2d3c54ac315ca51cd92c
29.4 kB Preview Download
md5:04c15fa3e0efab7c7ea0da0c525d2fdd
290.9 kB Preview Download