MetroLLM-Bench Demo
🚇
1
Transit kiosk run by a 9B local model (arXiv:2609.10016)
MetroLLM-Bench v24
Qwen3.5 students (2B/4B/9B/27B) fine-tuned for MetroLLM-Bench, plus the live demo. A 2.6 GB student matches a frontier API offline.
-
-
MetroLLM-Bench: Evaluating Language Models as Transit Kiosk Runtimes
Paper • 2609.10016 • Published • 32 -
continker/Qwen3.5-2B-metro-v24
Text Generation • 2B • Updated • 49 -
continker/Qwen3.5-4B-metro-v24
Text Generation • 4B • Updated • 33 -
continker/Qwen3.5-9B-metro-v24
Text Generation • 9B • Updated • 33 -
continker/Qwen3.5-27B-metro-v24
Text Generation • 27B • Updated • 27
continker/metrollm-bench
Viewer • Updated • 955 • 27 • 2Note The 955 cases with train/holdout splits