Splash is Inco's local inference engine for Qwen3.8-27B, and this is my port for M1 and M2. Its kernels used bfloat16 and Apple's MPP matrix library, which those GPUs do not run well; I rewrote them to use formats the M1 runs natively.
A launcher and a set of Metal kernels that run ISTA-DASLab's small quants of a 126B mixture-of-experts model (6.7B active) with the full 262K context, MTP and vision, behind an OpenAI and Anthropic compatible server.
Makes LM Studio load Ternary Bonsai 2 GGUF models (PQ2_0 and PTQ1_0) through Prism's llama.cpp fork, with a reasoning-effort selector. Linux and Windows, double-click install.
An online judge for the CPE, the programming exam many Taiwanese universities require for graduation, made for Yuan Ze University: 299 problems from real past exams, code judged in sandboxed containers and timed mock exams. Akmal Agzamov and I build and run it together.