sharp-421m-mlx / evaluation_report.json
ya9919's picture
Model inference release
eb9d073
Raw History Blame Contribute Delete
1.33 kB
{
"report_scope": "Aggregate evaluation values reported in the paper",
"paper": {
"title": "S²-MAS: A Society-Scale Multi-Agent Simulation Framework with 500 Million Household Persona Agents",
"version": "S²-MAS manuscript",
"date": "2026-10-20",
"project_url": "https://github.com/KataDavidXD/s2-mas",
"manuscript_url": "https://github.com/KataDavidXD/s2-mas/blob/main/paper/S2_MAS_Paper.pdf"
},
"repository_role": "MLX deployment of the six-round model",
"model_reference": {
"model": "Fin-Laya 6-round",
"runtime": "MLX CUDA",
"precision": "FP16"
},
"training": {
"continuation_examples_per_language": 2925263,
"paper_reference": "Supplementary Methods M6"
},
"metric_definition": "Accuracy pools records; question-macro F1 averages candidate-class F1 within each question version, then weights versions equally",
"shared_test": {
"records": 240458,
"question_versions": 119,
"language": "English",
"accuracy_pct": 74.5544,
"question_macro_f1_pct": 38.3746,
"paper_reference": "Supplementary Methods M5"
},
"expanded_test": {
"records": 418597,
"language": "English",
"accuracy_pct": 67.2726,
"question_macro_f1_pct": 27.8955,
"paper_reference": "Supplementary Methods M6"
}
}