model_id string | task string | benchmark_language string | total_samples int64 | groups dict | groups_evaluated list | average dict | timestamp string | model_type string | organization string | hf_model_id string | base_architecture string | model_size_b float64 | model_size_category string | submitter string | submitted_time string |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
openai/whisper-small-duplicate-test | asr | mos_Latn | 1,012 | {
"general": {
"n_samples": 1012,
"wer": 3.065483986188832,
"cer": 2.3236588557154754,
"mer": 0.993555607007795
}
} | [
"general"
] | {
"wer": 3.065483986188832,
"cer": 2.3236588557154754,
"mer": 0.993555607007795
} | 2026-07-04T23:59:28.559688+00:00 | Open Source | OpenAI | openai/whisper-small | Whisper | 0 | <0.5B | nabla-w | 2026-04-08T01:07:07.157868+00:00 |
goai-bench-submissions
Submissions to the GO AI Bench Leaderboard: evaluation requests and result files for Mooré and Dioula models.
%20%7C%20Dioula%20(dyu)-022871){kind=icon}
This dataset stores what people submit to the GO AI Bench Leaderboard. The leaderboard Space writes each submitted result here, and the backend Space syncs it and restarts the leaderboard so new results show. It replaces the two former repos goaicorp/goai-bench-results (which now redirects here) and goaicorp/goai-bench-requests.
Résumé en français. Ce jeu de données contient les soumissions au GO AI Bench Leaderboard : les demandes d'évaluation (
requests/) et les fichiers de résultats (results/) pour des modèles en mooré et en dioula. Il remplace les anciens dépôtsgoai-bench-resultsetgoai-bench-requests.
Layout
results/{submitter}/{task}_{language}_{model}_{mt_pair}_{timestamp}.json
requests/
results/: one JSON file per submission, in the GO AI Bench output format (model_id,task,benchmark_language,groups,average, ...) plus the metadata the submitter gave (organization,model_type,base_architecture,model_size_b,submitter,submitted_time).requests/: reserved for evaluation requests. It is empty: the leaderboard takes self-reported results only, and no service evaluates models yet.
Limits
- Scores are self-reported. Nobody re-runs the model or re-scores its outputs.
- The 26 baseline results shown on the leaderboard live in the Space itself (
data/results_final.json), not here. - Result files have different metric keys per task, so the dataset viewer may not show every file.
License
Apache 2.0, like the leaderboard and the GO AI Bench toolkit.
Developed and maintained by
- Wendyellé A. Alban NYANTUDRE: nyantudre@gmail.com
- Aristide BANDAOGO: aristide@goaicorporation.org
- GO AI Corp
- Downloads last month
- 60