mirror of
https://github.com/hanzoai/trolley-bench.git
synced 2026-11-02 13:22:03 +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
284 B
JSON
1 line
284 B
JSON
{"model":"clm-latest","answers":{"rank":{"type":"choice","choice":"a","confidence":0.997909408386072,"probabilities":{"a":0.998432056289554,"b":0.001557211690167138,"c":4.8811550855986875e-06,"d":5.850865193222095e-06}}},"usage":{"billing_units":1,"input_tokens":0,"output_tokens":0}} |