ADM-1 / benchmark_comparison.json
Aurley's picture
Add benchmark comparison data
0ae5db0 verified
Raw History Blame Contribute Delete
1.51 kB
{
"source": "Previous 512-example run vs expanded 3,072-example run",
"previous": {
"commonsense_qa": {
"accuracy_top1": 0.5827996340347666,
"accuracy_top3": 0.9048490393412626,
"mean_nll": 1.3366108938601398,
"mean_brier": 0.6047167307695813,
"ece_10_bins": 0.1994374088758955
},
"banking77": {
"accuracy_top1": 0.5800865800865801,
"accuracy_top5": 0.8701298701298701,
"mean_nll": 1.5664262868874264,
"mean_brier": 0.5760736580374399,
"ece_10_bins": 0.09326245064859266
}
},
"expanded": {
"commonsense_qa": {
"accuracy_top1": 0.6020128087831657,
"accuracy_top3": 0.9277218664226898,
"mean_nll": 1.1035297405179396,
"mean_brier": 0.5445770138152226,
"ece_10_bins": 0.13095674654920497
},
"banking77": {
"accuracy_top1": 0.645021645021645,
"accuracy_top5": 0.9523809523809523,
"mean_nll": 1.329584948457993,
"mean_brier": 0.5151303437000476,
"ece_10_bins": 0.16459291234805987
}
},
"examples": {
"commonsense_qa": 1093,
"banking77": 231
},
"chance_top1": {
"commonsense_qa_5_choice": 0.2,
"banking77_77_intent": 0.012987012987012988
},
"notes": [
"CommonsenseQA uses 1,093 official validation examples.",
"BANKING77 is a stratified sample of 231 official test examples, three per intent.",
"These task-specific measurements are not a general intelligence score."
]
}