NierRouter
This project is a highly optimized, dual-tier task evaluation pipeline designed specifically to operate under strict hackathon hardware constraints, including a hard ceiling of 4GB RAM and 2 vCPUs. At its core, the system utilizes a zero-token semantic router built on all-MiniLM-L6-v2 to intelligently classify incoming tasks—ranging from code debugging and named entity recognition to complex math execution and logical puzzles. Instead of relying purely on expensive API calls, the primary inference engine is a locally hosted, 4-bit quantized Qwen2.5-Coder-1.5B model. By eagerly loading the models into memory at startup and implementing strict concurrency locks, the pipeline completely avoids Out-Of-Memory (OOM) crashes and thread collisions. The local model handles the vast majority of operations, ensuring fast execution
Category tags:
.png&w=96&q=75){kind=small}
.png&w=96&q=75){kind=small}
For Innovators & Creators