Complete model card: intended use, benchmark links, citation, honest limits
#1
by benchgen - opened
Fixes and additions to the model card.
- Replace the placeholder docs link (
benchgen-docs-url) with the real guide on benchgen.com - Add Intended use and Out of scope
- State the held-out result as counts: 27 vs 26 of 33 graded calls, a difference of one call
- Add the Router Fidelity Benchmark link and a run-it-yourself section
- Add licence note for the base model, a citation and an explicit pilot-verdict limitation
Open before merging (details: company brain, gtm/marketing/huggingface/README.md)
- D1 licence of the reward labels (proposed CC-BY-4.0)
- D3 which held-out numbers belong to the published weights (card vs guide)
- D6 the Router Fidelity Benchmark page must be public for the link to work signed out
benchgen changed pull request status to merged