Add benchmark artifacts tab

#70

Adds a Benchmark artifacts tab for two diagnostics: reference-error agreement on corrected VoxPopuli errors and reference-rendering agreement across equivalent spelling, initialism, number, and lexical forms. The dataset selector controls rendering results; reference-error agreement remains fixed to VoxPopuli, and rendering rates below 30 eligible cases are suppressed. Following the other tabs, the Space expects benchmark_artifacts_latest.csv in hf-audio/open-asr-leaderboard-results; scores reflect the published prediction manifests, including decoding and post-processing.

Scorers: PR #199 and PR #200.

tlebryk02 changed pull request title from Add benchmark optimization tab to Add benchmark artifacts tab
Hugging Face for Audio org

Merged through separate commits

bezzam changed pull request status to closed

Sign up or log in to comment