JiRack — Ternary Transformers & Micro-Agent Architecture We build highly efficient large language models using 1.58-bit ternary weights {-1, 0, 1} for extreme compression and fast CPU/GPU inference.
Core focus:
JiRack Ternary Transformer Architecture — fresh Qwen base, trained on DeepSeek-style datasets, optimized for fast CPU inference (MIT License) JiRack Micro-Agent Deployment — specialized small models + smart router for low-cost agentic systems Production-ready ONNX Runtime & Docker inference stacks
Public Models ModelSizeStatusJiRackUltra series (1B / 7B / 14B / 32B)—Released
Mission Democratize frontier-scale language models through extreme efficiency. Train and run powerful models on accessible hardware without sacrificing quality.
Solved issues Benefits of JiRack Micro-Agent Architecture:
Solves catastrophic forgetting during training by using small, specialized models for each domain, managed by a smart router Enables extremely cheap inference using ternary models Significantly reduces cloud inference costs while maintaining high performance In classical architecture, an expensive model has to search for MCP-agents every time, while JiRack uses a very small model and cheap router for agent tasks, saving big money right from the start Considered one of the best approaches for enterprise AI deployments
