Performance

Benchmarks

Independent results on public MTEB / BEIR / OMB datasets. SEMQ preserves accuracy while competitors collapse.

Dataset

banking77 (MTEB) · Model: all-MiniLM-L6-v2

MethodAccuracyΔ vs FP32Pass
FP3292.26%
SEMQ·B92.23%−0.03 pp
SEMQ·A92.27%+0.01 pp
PQ 4-bit56.05%−36.22 ppcatastrophic collapse
OPQ 4-bit~60%~−32 ppcatastrophic collapse

PQ / OPQ collapse: −36 pp accuracy

Quantization methods that discard angular structure lose over a third of classification accuracy. SEMQ preserves the full semantic geometry — accuracy loss is statistically negligible.