7.6B 中文总指挥模型：把活落到 341 位原生能力 / 30 域，写盘·花钱·出网一律先过人闸，每次调用都留审计留痕 | A 7.6B Chinese "commander" LLM that routes natural language to 341 audited capability bits behind one single gate. 335-item benchmark PASS45 99.10%; per-bit live-run proof included. Python client + 30s demo, zero deps.
benchmark robotics chinese-nlp audit-log teleoperation ai-agents tool-use model-card bionic-arm ai-agent llm vllm qwen2 tool-calling openai-compatible dexterous-hand capability-routing mirror-space omni-plug reverse-control
-
Updated
Oct 26, 2026 - Python