mirror of
https://github.com/hanzoai/trolley-bench.git
synced 2026-10-30 15:00:49 +00:00
* Packaged the benchmark's working copy with its frozen prompts, runners, raw answers, analysis and built page as committed evidence * Rewrote run-machine paths and left out share credentials, drafts and operator notes; PACKAGING.md lists every exclusion and change * Fixed the models.yaml generator listing rerun models twice, and let the Jev runner take its key from TYPESAFE_API_KEY
1 line
550 B
JSON
1 line
550 B
JSON
{"model":"clm-latest","answers":{"urgency":{"type":"noul","noul":0.8285061788283119},"department":{"type":"choice","choice":"billing","confidence":0.9750985939065437,"probabilities":{"billing":0.9875492969532719,"technical":0.012450703046728092}},"frustration":{"type":"score","score":1.9999755595297797,"confidence":0.9999644121451241,"legend":{"0":"Calm","1":"Frustrated","2":"Very angry"},"probabilities":{"0":7.152336361410804e-07,"1":2.301000294771923e-05,"2":0.999976274763416}}},"usage":{"billing_units":3,"input_tokens":98,"output_tokens":0}} |