LexiEngine: Rust inference engine for AMD runs 27B at full 262k context 47% faster than llama.cpp

W61k3r · reddit · 2026-11-20

A developer released LexiEngine, a Rust-based inference engine focused on long-context stability and AMD GPUs, with benchmarks on a single RX 7900 XTX against llama.cpp b12531 and Strata 0.2.16.

Full measurement methodology is published; the engine is part of a self-hosted suite with LexiPanel and a bootable LexiOS image.

Original post →

More from coding & agent

coding & agent channel →