YuhuaBillChen/justfit-llm: Local Inference System for 27B Models on Apple Silicon

Yuhua Chen released JustFit v2, a local inference system designed to serve a 27B language-and-vision model on memory-constrained Macs. The system coordinates compressed context, resident speculative draft execution, and unified CPU, GPU, and Apple Neural Engine acceleration under a single unified-memory budget. According to the release notes and preprint, the software achieves significant capacity and speed improvements on M-series chips while supporting OpenAI-compatible APIs. Open article →