Custom Rust engine runs 524k context on a single RX 7900 XTX

A developer ran Inkling-2-Mini with 524,288-token context on a single 24GB RX 7900 XTX using a custom Rust/Vulkan engine, achieving 3-5x the speed of llama.cpp.

2026-11-22 ~ 2026-11-22 · 2 related posts