Reproducible high-speed Qwen3.6-35B-A3B inference on NVIDIA GB10 - View it on GitHub