1 repository on SrcLog
Add it to a GitHub README or profile β the card updates as the ranking changes:
Native MLX runtime for Laya typed decision models β 7β14 ms short decisions on M3 Max. No text generation, PyTorch, or cloud API.