56 lines
1.4 KiB
JSON
56 lines
1.4 KiB
JSON
{
|
|
"results": {
|
|
"arc_easy": {
|
|
"acc": 0.41624579124579125,
|
|
"acc_stderr": 0.010114819404500878,
|
|
"acc_norm": 0.38846801346801346,
|
|
"acc_norm_stderr": 0.01000127604448523
|
|
},
|
|
"boolq": {
|
|
"acc": 0.5831804281345566,
|
|
"acc_stderr": 0.00862319210884368
|
|
},
|
|
"lambada_openai": {
|
|
"ppl": 79.45218123817662,
|
|
"ppl_stderr": 3.1355336623454866,
|
|
"acc": 0.2522802251115855,
|
|
"acc_stderr": 0.006050943684570117
|
|
},
|
|
"openbookqa": {
|
|
"acc": 0.154,
|
|
"acc_stderr": 0.016158285192455334,
|
|
"acc_norm": 0.278,
|
|
"acc_norm_stderr": 0.02005583388807091
|
|
},
|
|
"piqa": {
|
|
"acc": 0.6050054406964092,
|
|
"acc_stderr": 0.011405665187969021,
|
|
"acc_norm": 0.5897714907508161,
|
|
"acc_norm_stderr": 0.011476256036359109
|
|
},
|
|
"winogrande": {
|
|
"acc": 0.5272296764009471,
|
|
"acc_stderr": 0.014031631629827708
|
|
}
|
|
},
|
|
"versions": {
|
|
"arc_easy": 0,
|
|
"boolq": 1,
|
|
"lambada_openai": 0,
|
|
"openbookqa": 0,
|
|
"piqa": 0,
|
|
"winogrande": 0
|
|
},
|
|
"config": {
|
|
"model": "hf-causal-experimental",
|
|
"model_args": "pretrained=BEE-spoke-data/smol_llama-81M-tied,revision=main,trust_remote_code=True,dtype='float'",
|
|
"num_fewshot": 0,
|
|
"batch_size": "16",
|
|
"batch_sizes": [],
|
|
"device": "cuda",
|
|
"no_cache": false,
|
|
"limit": null,
|
|
"bootstrap_iters": 100000,
|
|
"description_dict": {}
|
|
}
|
|
} |