zlaya: a Zig-based CPU-only inference engine runs Laya on CDN edge via WebAssembly
jedisct1 · x · 2026-09-24
zlaya is a new CPU-only inference engine for Laya written in Zig. It needs no GPU, runs entirely in WebAssembly (including on CDN edge apps), and answers choice/score/yes-no questions about text or JSON using native Zig tokenization, safetensors loading, ModernBERT, and trained decision/action heads. Weights take 804 MiB on disk, expand to float32 in memory, and inference works fully offline after download.
More from coding & agent
- Opus 5.5 drives Chrome to repaint the Mona Lisa in JS paint in 20 minutes — _sholtodouglas · 2026-09-24
- Cloud tools whose moat was human-friendly UX are in trouble as agents go direct — emeka_boris · 2026-09-24
- ESP32-S3 connects to Cloudflare over WebRTC, with an open-source pocket radio example — irvinebroque · 2026-09-24
- NVIDIA releases SWE-Serve: 53 tasks testing if agents can build inference engines — NVIDIAAI · 2026-09-24
- Hamel Husain ships eval skills: run /eval-audit to find low-hanging fruit in your pipeline — HamelHusain · 2026-09-24
- franken_code_browser: open-source spatial Rust source browser for 5M+ line repos on Mac — doodlestein · 2026-09-24