Upload folder using huggingface_hub
Browse files- evals/json_object_1.json +56 -0
- evals/json_object_2.json +25 -0
- evals/json_object_3.json +25 -0
- evals/json_object_4.json +25 -0
- evals/json_object_5.json +417 -0
evals/json_object_1.json
ADDED
|
@@ -0,0 +1,56 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"results": {
|
| 3 |
+
"arc_easy": {
|
| 4 |
+
"acc": 0.43813131313131315,
|
| 5 |
+
"acc_stderr": 0.010180937100600052,
|
| 6 |
+
"acc_norm": 0.4019360269360269,
|
| 7 |
+
"acc_norm_stderr": 0.010060521220920566
|
| 8 |
+
},
|
| 9 |
+
"boolq": {
|
| 10 |
+
"acc": 0.617737003058104,
|
| 11 |
+
"acc_stderr": 0.00849914969044927
|
| 12 |
+
},
|
| 13 |
+
"lambada_openai": {
|
| 14 |
+
"ppl": 64.94966274873535,
|
| 15 |
+
"ppl_stderr": 2.5466406639926897,
|
| 16 |
+
"acc": 0.26470017465554047,
|
| 17 |
+
"acc_stderr": 0.006146408462993569
|
| 18 |
+
},
|
| 19 |
+
"openbookqa": {
|
| 20 |
+
"acc": 0.166,
|
| 21 |
+
"acc_stderr": 0.016656616876531142,
|
| 22 |
+
"acc_norm": 0.28,
|
| 23 |
+
"acc_norm_stderr": 0.020099950647503237
|
| 24 |
+
},
|
| 25 |
+
"piqa": {
|
| 26 |
+
"acc": 0.5973884657236126,
|
| 27 |
+
"acc_stderr": 0.011442395233488702,
|
| 28 |
+
"acc_norm": 0.6088139281828074,
|
| 29 |
+
"acc_norm_stderr": 0.0113862156067287
|
| 30 |
+
},
|
| 31 |
+
"winogrande": {
|
| 32 |
+
"acc": 0.5098658247829518,
|
| 33 |
+
"acc_stderr": 0.014049749833367589
|
| 34 |
+
}
|
| 35 |
+
},
|
| 36 |
+
"versions": {
|
| 37 |
+
"arc_easy": 0,
|
| 38 |
+
"boolq": 1,
|
| 39 |
+
"lambada_openai": 0,
|
| 40 |
+
"openbookqa": 0,
|
| 41 |
+
"piqa": 0,
|
| 42 |
+
"winogrande": 0
|
| 43 |
+
},
|
| 44 |
+
"config": {
|
| 45 |
+
"model": "hf-causal-experimental",
|
| 46 |
+
"model_args": "pretrained=BEE-spoke-data/smol_llama-220M-GQA,revision=main,trust_remote_code=True,dtype='bfloat16'",
|
| 47 |
+
"num_fewshot": 0,
|
| 48 |
+
"batch_size": "8",
|
| 49 |
+
"batch_sizes": [],
|
| 50 |
+
"device": "cuda",
|
| 51 |
+
"no_cache": false,
|
| 52 |
+
"limit": null,
|
| 53 |
+
"bootstrap_iters": 100000,
|
| 54 |
+
"description_dict": {}
|
| 55 |
+
}
|
| 56 |
+
}
|
evals/json_object_2.json
ADDED
|
@@ -0,0 +1,25 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"results": {
|
| 3 |
+
"arc_challenge": {
|
| 4 |
+
"acc": 0.20392491467576793,
|
| 5 |
+
"acc_stderr": 0.01177426247870226,
|
| 6 |
+
"acc_norm": 0.25,
|
| 7 |
+
"acc_norm_stderr": 0.012653835621466646
|
| 8 |
+
}
|
| 9 |
+
},
|
| 10 |
+
"versions": {
|
| 11 |
+
"arc_challenge": 0
|
| 12 |
+
},
|
| 13 |
+
"config": {
|
| 14 |
+
"model": "hf-causal-experimental",
|
| 15 |
+
"model_args": "pretrained=BEE-spoke-data/smol_llama-220M-GQA,revision=main,trust_remote_code=True,dtype='bfloat16'",
|
| 16 |
+
"num_fewshot": 25,
|
| 17 |
+
"batch_size": "8",
|
| 18 |
+
"batch_sizes": [],
|
| 19 |
+
"device": "cuda",
|
| 20 |
+
"no_cache": false,
|
| 21 |
+
"limit": null,
|
| 22 |
+
"bootstrap_iters": 100000,
|
| 23 |
+
"description_dict": {}
|
| 24 |
+
}
|
| 25 |
+
}
|
evals/json_object_3.json
ADDED
|
@@ -0,0 +1,25 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"results": {
|
| 3 |
+
"hellaswag": {
|
| 4 |
+
"acc": 0.2752988047808765,
|
| 5 |
+
"acc_stderr": 0.008917257773359156,
|
| 6 |
+
"acc_norm": 0.2968127490039841,
|
| 7 |
+
"acc_norm_stderr": 0.009120663626901691
|
| 8 |
+
}
|
| 9 |
+
},
|
| 10 |
+
"versions": {
|
| 11 |
+
"hellaswag": 0
|
| 12 |
+
},
|
| 13 |
+
"config": {
|
| 14 |
+
"model": "hf-causal-experimental",
|
| 15 |
+
"model_args": "pretrained=BEE-spoke-data/smol_llama-220M-GQA,revision=main,trust_remote_code=True,dtype='bfloat16'",
|
| 16 |
+
"num_fewshot": 10,
|
| 17 |
+
"batch_size": "8",
|
| 18 |
+
"batch_sizes": [],
|
| 19 |
+
"device": "cuda",
|
| 20 |
+
"no_cache": false,
|
| 21 |
+
"limit": 0.25,
|
| 22 |
+
"bootstrap_iters": 100000,
|
| 23 |
+
"description_dict": {}
|
| 24 |
+
}
|
| 25 |
+
}
|
evals/json_object_4.json
ADDED
|
@@ -0,0 +1,25 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"results": {
|
| 3 |
+
"truthfulqa_mc": {
|
| 4 |
+
"mc1": 0.23745410036719705,
|
| 5 |
+
"mc1_stderr": 0.014896277441041836,
|
| 6 |
+
"mc2": 0.4402813457518687,
|
| 7 |
+
"mc2_stderr": 0.015339681556915718
|
| 8 |
+
}
|
| 9 |
+
},
|
| 10 |
+
"versions": {
|
| 11 |
+
"truthfulqa_mc": 1
|
| 12 |
+
},
|
| 13 |
+
"config": {
|
| 14 |
+
"model": "hf-causal-experimental",
|
| 15 |
+
"model_args": "pretrained=BEE-spoke-data/smol_llama-220M-GQA,revision=main,trust_remote_code=True,dtype='bfloat16'",
|
| 16 |
+
"num_fewshot": 0,
|
| 17 |
+
"batch_size": "8",
|
| 18 |
+
"batch_sizes": [],
|
| 19 |
+
"device": "cuda",
|
| 20 |
+
"no_cache": false,
|
| 21 |
+
"limit": null,
|
| 22 |
+
"bootstrap_iters": 100000,
|
| 23 |
+
"description_dict": {}
|
| 24 |
+
}
|
| 25 |
+
}
|
evals/json_object_5.json
ADDED
|
@@ -0,0 +1,417 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"results": {
|
| 3 |
+
"hendrycksTest-abstract_algebra": {
|
| 4 |
+
"acc": 0.2,
|
| 5 |
+
"acc_stderr": 0.08164965809277261,
|
| 6 |
+
"acc_norm": 0.2,
|
| 7 |
+
"acc_norm_stderr": 0.08164965809277261
|
| 8 |
+
},
|
| 9 |
+
"hendrycksTest-anatomy": {
|
| 10 |
+
"acc": 0.4,
|
| 11 |
+
"acc_stderr": 0.1,
|
| 12 |
+
"acc_norm": 0.4,
|
| 13 |
+
"acc_norm_stderr": 0.1
|
| 14 |
+
},
|
| 15 |
+
"hendrycksTest-astronomy": {
|
| 16 |
+
"acc": 0.32,
|
| 17 |
+
"acc_stderr": 0.09521904571390466,
|
| 18 |
+
"acc_norm": 0.32,
|
| 19 |
+
"acc_norm_stderr": 0.09521904571390466
|
| 20 |
+
},
|
| 21 |
+
"hendrycksTest-business_ethics": {
|
| 22 |
+
"acc": 0.2,
|
| 23 |
+
"acc_stderr": 0.08164965809277261,
|
| 24 |
+
"acc_norm": 0.2,
|
| 25 |
+
"acc_norm_stderr": 0.08164965809277261
|
| 26 |
+
},
|
| 27 |
+
"hendrycksTest-clinical_knowledge": {
|
| 28 |
+
"acc": 0.24,
|
| 29 |
+
"acc_stderr": 0.08717797887081345,
|
| 30 |
+
"acc_norm": 0.24,
|
| 31 |
+
"acc_norm_stderr": 0.08717797887081345
|
| 32 |
+
},
|
| 33 |
+
"hendrycksTest-college_biology": {
|
| 34 |
+
"acc": 0.2,
|
| 35 |
+
"acc_stderr": 0.08164965809277262,
|
| 36 |
+
"acc_norm": 0.2,
|
| 37 |
+
"acc_norm_stderr": 0.08164965809277262
|
| 38 |
+
},
|
| 39 |
+
"hendrycksTest-college_chemistry": {
|
| 40 |
+
"acc": 0.16,
|
| 41 |
+
"acc_stderr": 0.0748331477354788,
|
| 42 |
+
"acc_norm": 0.16,
|
| 43 |
+
"acc_norm_stderr": 0.0748331477354788
|
| 44 |
+
},
|
| 45 |
+
"hendrycksTest-college_computer_science": {
|
| 46 |
+
"acc": 0.32,
|
| 47 |
+
"acc_stderr": 0.09521904571390466,
|
| 48 |
+
"acc_norm": 0.32,
|
| 49 |
+
"acc_norm_stderr": 0.09521904571390466
|
| 50 |
+
},
|
| 51 |
+
"hendrycksTest-college_mathematics": {
|
| 52 |
+
"acc": 0.4,
|
| 53 |
+
"acc_stderr": 0.1,
|
| 54 |
+
"acc_norm": 0.4,
|
| 55 |
+
"acc_norm_stderr": 0.1
|
| 56 |
+
},
|
| 57 |
+
"hendrycksTest-college_medicine": {
|
| 58 |
+
"acc": 0.2,
|
| 59 |
+
"acc_stderr": 0.08164965809277261,
|
| 60 |
+
"acc_norm": 0.2,
|
| 61 |
+
"acc_norm_stderr": 0.08164965809277261
|
| 62 |
+
},
|
| 63 |
+
"hendrycksTest-college_physics": {
|
| 64 |
+
"acc": 0.16,
|
| 65 |
+
"acc_stderr": 0.0748331477354788,
|
| 66 |
+
"acc_norm": 0.16,
|
| 67 |
+
"acc_norm_stderr": 0.0748331477354788
|
| 68 |
+
},
|
| 69 |
+
"hendrycksTest-computer_security": {
|
| 70 |
+
"acc": 0.52,
|
| 71 |
+
"acc_stderr": 0.10198039027185572,
|
| 72 |
+
"acc_norm": 0.52,
|
| 73 |
+
"acc_norm_stderr": 0.10198039027185572
|
| 74 |
+
},
|
| 75 |
+
"hendrycksTest-conceptual_physics": {
|
| 76 |
+
"acc": 0.24,
|
| 77 |
+
"acc_stderr": 0.08717797887081347,
|
| 78 |
+
"acc_norm": 0.24,
|
| 79 |
+
"acc_norm_stderr": 0.08717797887081347
|
| 80 |
+
},
|
| 81 |
+
"hendrycksTest-econometrics": {
|
| 82 |
+
"acc": 0.2,
|
| 83 |
+
"acc_stderr": 0.08164965809277261,
|
| 84 |
+
"acc_norm": 0.2,
|
| 85 |
+
"acc_norm_stderr": 0.08164965809277261
|
| 86 |
+
},
|
| 87 |
+
"hendrycksTest-electrical_engineering": {
|
| 88 |
+
"acc": 0.16,
|
| 89 |
+
"acc_stderr": 0.07483314773547882,
|
| 90 |
+
"acc_norm": 0.16,
|
| 91 |
+
"acc_norm_stderr": 0.07483314773547882
|
| 92 |
+
},
|
| 93 |
+
"hendrycksTest-elementary_mathematics": {
|
| 94 |
+
"acc": 0.16,
|
| 95 |
+
"acc_stderr": 0.0748331477354788,
|
| 96 |
+
"acc_norm": 0.16,
|
| 97 |
+
"acc_norm_stderr": 0.0748331477354788
|
| 98 |
+
},
|
| 99 |
+
"hendrycksTest-formal_logic": {
|
| 100 |
+
"acc": 0.24,
|
| 101 |
+
"acc_stderr": 0.08717797887081345,
|
| 102 |
+
"acc_norm": 0.24,
|
| 103 |
+
"acc_norm_stderr": 0.08717797887081345
|
| 104 |
+
},
|
| 105 |
+
"hendrycksTest-global_facts": {
|
| 106 |
+
"acc": 0.4,
|
| 107 |
+
"acc_stderr": 0.1,
|
| 108 |
+
"acc_norm": 0.4,
|
| 109 |
+
"acc_norm_stderr": 0.1
|
| 110 |
+
},
|
| 111 |
+
"hendrycksTest-high_school_biology": {
|
| 112 |
+
"acc": 0.32,
|
| 113 |
+
"acc_stderr": 0.09521904571390465,
|
| 114 |
+
"acc_norm": 0.32,
|
| 115 |
+
"acc_norm_stderr": 0.09521904571390465
|
| 116 |
+
},
|
| 117 |
+
"hendrycksTest-high_school_chemistry": {
|
| 118 |
+
"acc": 0.2,
|
| 119 |
+
"acc_stderr": 0.08164965809277261,
|
| 120 |
+
"acc_norm": 0.2,
|
| 121 |
+
"acc_norm_stderr": 0.08164965809277261
|
| 122 |
+
},
|
| 123 |
+
"hendrycksTest-high_school_computer_science": {
|
| 124 |
+
"acc": 0.2,
|
| 125 |
+
"acc_stderr": 0.08164965809277261,
|
| 126 |
+
"acc_norm": 0.2,
|
| 127 |
+
"acc_norm_stderr": 0.08164965809277261
|
| 128 |
+
},
|
| 129 |
+
"hendrycksTest-high_school_european_history": {
|
| 130 |
+
"acc": 0.16,
|
| 131 |
+
"acc_stderr": 0.0748331477354788,
|
| 132 |
+
"acc_norm": 0.16,
|
| 133 |
+
"acc_norm_stderr": 0.0748331477354788
|
| 134 |
+
},
|
| 135 |
+
"hendrycksTest-high_school_geography": {
|
| 136 |
+
"acc": 0.32,
|
| 137 |
+
"acc_stderr": 0.09521904571390466,
|
| 138 |
+
"acc_norm": 0.32,
|
| 139 |
+
"acc_norm_stderr": 0.09521904571390466
|
| 140 |
+
},
|
| 141 |
+
"hendrycksTest-high_school_government_and_politics": {
|
| 142 |
+
"acc": 0.28,
|
| 143 |
+
"acc_stderr": 0.09165151389911677,
|
| 144 |
+
"acc_norm": 0.28,
|
| 145 |
+
"acc_norm_stderr": 0.09165151389911677
|
| 146 |
+
},
|
| 147 |
+
"hendrycksTest-high_school_macroeconomics": {
|
| 148 |
+
"acc": 0.28,
|
| 149 |
+
"acc_stderr": 0.09165151389911681,
|
| 150 |
+
"acc_norm": 0.28,
|
| 151 |
+
"acc_norm_stderr": 0.09165151389911681
|
| 152 |
+
},
|
| 153 |
+
"hendrycksTest-high_school_mathematics": {
|
| 154 |
+
"acc": 0.32,
|
| 155 |
+
"acc_stderr": 0.09521904571390467,
|
| 156 |
+
"acc_norm": 0.32,
|
| 157 |
+
"acc_norm_stderr": 0.09521904571390467
|
| 158 |
+
},
|
| 159 |
+
"hendrycksTest-high_school_microeconomics": {
|
| 160 |
+
"acc": 0.28,
|
| 161 |
+
"acc_stderr": 0.09165151389911678,
|
| 162 |
+
"acc_norm": 0.28,
|
| 163 |
+
"acc_norm_stderr": 0.09165151389911678
|
| 164 |
+
},
|
| 165 |
+
"hendrycksTest-high_school_physics": {
|
| 166 |
+
"acc": 0.16,
|
| 167 |
+
"acc_stderr": 0.0748331477354788,
|
| 168 |
+
"acc_norm": 0.16,
|
| 169 |
+
"acc_norm_stderr": 0.0748331477354788
|
| 170 |
+
},
|
| 171 |
+
"hendrycksTest-high_school_psychology": {
|
| 172 |
+
"acc": 0.28,
|
| 173 |
+
"acc_stderr": 0.09165151389911678,
|
| 174 |
+
"acc_norm": 0.28,
|
| 175 |
+
"acc_norm_stderr": 0.09165151389911678
|
| 176 |
+
},
|
| 177 |
+
"hendrycksTest-high_school_statistics": {
|
| 178 |
+
"acc": 0.4,
|
| 179 |
+
"acc_stderr": 0.10000000000000002,
|
| 180 |
+
"acc_norm": 0.4,
|
| 181 |
+
"acc_norm_stderr": 0.10000000000000002
|
| 182 |
+
},
|
| 183 |
+
"hendrycksTest-high_school_us_history": {
|
| 184 |
+
"acc": 0.28,
|
| 185 |
+
"acc_stderr": 0.0916515138991168,
|
| 186 |
+
"acc_norm": 0.28,
|
| 187 |
+
"acc_norm_stderr": 0.0916515138991168
|
| 188 |
+
},
|
| 189 |
+
"hendrycksTest-high_school_world_history": {
|
| 190 |
+
"acc": 0.28,
|
| 191 |
+
"acc_stderr": 0.0916515138991168,
|
| 192 |
+
"acc_norm": 0.28,
|
| 193 |
+
"acc_norm_stderr": 0.0916515138991168
|
| 194 |
+
},
|
| 195 |
+
"hendrycksTest-human_aging": {
|
| 196 |
+
"acc": 0.2,
|
| 197 |
+
"acc_stderr": 0.08164965809277261,
|
| 198 |
+
"acc_norm": 0.2,
|
| 199 |
+
"acc_norm_stderr": 0.08164965809277261
|
| 200 |
+
},
|
| 201 |
+
"hendrycksTest-human_sexuality": {
|
| 202 |
+
"acc": 0.44,
|
| 203 |
+
"acc_stderr": 0.10132456102380442,
|
| 204 |
+
"acc_norm": 0.44,
|
| 205 |
+
"acc_norm_stderr": 0.10132456102380442
|
| 206 |
+
},
|
| 207 |
+
"hendrycksTest-international_law": {
|
| 208 |
+
"acc": 0.32,
|
| 209 |
+
"acc_stderr": 0.09521904571390466,
|
| 210 |
+
"acc_norm": 0.32,
|
| 211 |
+
"acc_norm_stderr": 0.09521904571390466
|
| 212 |
+
},
|
| 213 |
+
"hendrycksTest-jurisprudence": {
|
| 214 |
+
"acc": 0.36,
|
| 215 |
+
"acc_stderr": 0.09797958971132713,
|
| 216 |
+
"acc_norm": 0.36,
|
| 217 |
+
"acc_norm_stderr": 0.09797958971132713
|
| 218 |
+
},
|
| 219 |
+
"hendrycksTest-logical_fallacies": {
|
| 220 |
+
"acc": 0.24,
|
| 221 |
+
"acc_stderr": 0.08717797887081345,
|
| 222 |
+
"acc_norm": 0.24,
|
| 223 |
+
"acc_norm_stderr": 0.08717797887081345
|
| 224 |
+
},
|
| 225 |
+
"hendrycksTest-machine_learning": {
|
| 226 |
+
"acc": 0.16,
|
| 227 |
+
"acc_stderr": 0.0748331477354788,
|
| 228 |
+
"acc_norm": 0.16,
|
| 229 |
+
"acc_norm_stderr": 0.0748331477354788
|
| 230 |
+
},
|
| 231 |
+
"hendrycksTest-management": {
|
| 232 |
+
"acc": 0.32,
|
| 233 |
+
"acc_stderr": 0.09521904571390465,
|
| 234 |
+
"acc_norm": 0.32,
|
| 235 |
+
"acc_norm_stderr": 0.09521904571390465
|
| 236 |
+
},
|
| 237 |
+
"hendrycksTest-marketing": {
|
| 238 |
+
"acc": 0.08,
|
| 239 |
+
"acc_stderr": 0.05537749241945382,
|
| 240 |
+
"acc_norm": 0.08,
|
| 241 |
+
"acc_norm_stderr": 0.05537749241945382
|
| 242 |
+
},
|
| 243 |
+
"hendrycksTest-medical_genetics": {
|
| 244 |
+
"acc": 0.28,
|
| 245 |
+
"acc_stderr": 0.09165151389911678,
|
| 246 |
+
"acc_norm": 0.28,
|
| 247 |
+
"acc_norm_stderr": 0.09165151389911678
|
| 248 |
+
},
|
| 249 |
+
"hendrycksTest-miscellaneous": {
|
| 250 |
+
"acc": 0.08,
|
| 251 |
+
"acc_stderr": 0.05537749241945382,
|
| 252 |
+
"acc_norm": 0.08,
|
| 253 |
+
"acc_norm_stderr": 0.05537749241945382
|
| 254 |
+
},
|
| 255 |
+
"hendrycksTest-moral_disputes": {
|
| 256 |
+
"acc": 0.12,
|
| 257 |
+
"acc_stderr": 0.066332495807108,
|
| 258 |
+
"acc_norm": 0.12,
|
| 259 |
+
"acc_norm_stderr": 0.066332495807108
|
| 260 |
+
},
|
| 261 |
+
"hendrycksTest-moral_scenarios": {
|
| 262 |
+
"acc": 0.2,
|
| 263 |
+
"acc_stderr": 0.08164965809277261,
|
| 264 |
+
"acc_norm": 0.2,
|
| 265 |
+
"acc_norm_stderr": 0.08164965809277261
|
| 266 |
+
},
|
| 267 |
+
"hendrycksTest-nutrition": {
|
| 268 |
+
"acc": 0.52,
|
| 269 |
+
"acc_stderr": 0.10198039027185572,
|
| 270 |
+
"acc_norm": 0.52,
|
| 271 |
+
"acc_norm_stderr": 0.10198039027185572
|
| 272 |
+
},
|
| 273 |
+
"hendrycksTest-philosophy": {
|
| 274 |
+
"acc": 0.2,
|
| 275 |
+
"acc_stderr": 0.08164965809277261,
|
| 276 |
+
"acc_norm": 0.2,
|
| 277 |
+
"acc_norm_stderr": 0.08164965809277261
|
| 278 |
+
},
|
| 279 |
+
"hendrycksTest-prehistory": {
|
| 280 |
+
"acc": 0.16,
|
| 281 |
+
"acc_stderr": 0.07483314773547882,
|
| 282 |
+
"acc_norm": 0.16,
|
| 283 |
+
"acc_norm_stderr": 0.07483314773547882
|
| 284 |
+
},
|
| 285 |
+
"hendrycksTest-professional_accounting": {
|
| 286 |
+
"acc": 0.32,
|
| 287 |
+
"acc_stderr": 0.09521904571390465,
|
| 288 |
+
"acc_norm": 0.32,
|
| 289 |
+
"acc_norm_stderr": 0.09521904571390465
|
| 290 |
+
},
|
| 291 |
+
"hendrycksTest-professional_law": {
|
| 292 |
+
"acc": 0.16,
|
| 293 |
+
"acc_stderr": 0.07483314773547882,
|
| 294 |
+
"acc_norm": 0.16,
|
| 295 |
+
"acc_norm_stderr": 0.07483314773547882
|
| 296 |
+
},
|
| 297 |
+
"hendrycksTest-professional_medicine": {
|
| 298 |
+
"acc": 0.4,
|
| 299 |
+
"acc_stderr": 0.10000000000000002,
|
| 300 |
+
"acc_norm": 0.4,
|
| 301 |
+
"acc_norm_stderr": 0.10000000000000002
|
| 302 |
+
},
|
| 303 |
+
"hendrycksTest-professional_psychology": {
|
| 304 |
+
"acc": 0.24,
|
| 305 |
+
"acc_stderr": 0.08717797887081345,
|
| 306 |
+
"acc_norm": 0.24,
|
| 307 |
+
"acc_norm_stderr": 0.08717797887081345
|
| 308 |
+
},
|
| 309 |
+
"hendrycksTest-public_relations": {
|
| 310 |
+
"acc": 0.2,
|
| 311 |
+
"acc_stderr": 0.08164965809277261,
|
| 312 |
+
"acc_norm": 0.2,
|
| 313 |
+
"acc_norm_stderr": 0.08164965809277261
|
| 314 |
+
},
|
| 315 |
+
"hendrycksTest-security_studies": {
|
| 316 |
+
"acc": 0.32,
|
| 317 |
+
"acc_stderr": 0.09521904571390466,
|
| 318 |
+
"acc_norm": 0.32,
|
| 319 |
+
"acc_norm_stderr": 0.09521904571390466
|
| 320 |
+
},
|
| 321 |
+
"hendrycksTest-sociology": {
|
| 322 |
+
"acc": 0.28,
|
| 323 |
+
"acc_stderr": 0.09165151389911678,
|
| 324 |
+
"acc_norm": 0.28,
|
| 325 |
+
"acc_norm_stderr": 0.09165151389911678
|
| 326 |
+
},
|
| 327 |
+
"hendrycksTest-us_foreign_policy": {
|
| 328 |
+
"acc": 0.24,
|
| 329 |
+
"acc_stderr": 0.08717797887081345,
|
| 330 |
+
"acc_norm": 0.24,
|
| 331 |
+
"acc_norm_stderr": 0.08717797887081345
|
| 332 |
+
},
|
| 333 |
+
"hendrycksTest-virology": {
|
| 334 |
+
"acc": 0.12,
|
| 335 |
+
"acc_stderr": 0.06633249580710801,
|
| 336 |
+
"acc_norm": 0.12,
|
| 337 |
+
"acc_norm_stderr": 0.06633249580710801
|
| 338 |
+
},
|
| 339 |
+
"hendrycksTest-world_religions": {
|
| 340 |
+
"acc": 0.28,
|
| 341 |
+
"acc_stderr": 0.09165151389911678,
|
| 342 |
+
"acc_norm": 0.28,
|
| 343 |
+
"acc_norm_stderr": 0.09165151389911678
|
| 344 |
+
}
|
| 345 |
+
},
|
| 346 |
+
"versions": {
|
| 347 |
+
"hendrycksTest-abstract_algebra": 1,
|
| 348 |
+
"hendrycksTest-anatomy": 1,
|
| 349 |
+
"hendrycksTest-astronomy": 1,
|
| 350 |
+
"hendrycksTest-business_ethics": 1,
|
| 351 |
+
"hendrycksTest-clinical_knowledge": 1,
|
| 352 |
+
"hendrycksTest-college_biology": 1,
|
| 353 |
+
"hendrycksTest-college_chemistry": 1,
|
| 354 |
+
"hendrycksTest-college_computer_science": 1,
|
| 355 |
+
"hendrycksTest-college_mathematics": 1,
|
| 356 |
+
"hendrycksTest-college_medicine": 1,
|
| 357 |
+
"hendrycksTest-college_physics": 1,
|
| 358 |
+
"hendrycksTest-computer_security": 1,
|
| 359 |
+
"hendrycksTest-conceptual_physics": 1,
|
| 360 |
+
"hendrycksTest-econometrics": 1,
|
| 361 |
+
"hendrycksTest-electrical_engineering": 1,
|
| 362 |
+
"hendrycksTest-elementary_mathematics": 1,
|
| 363 |
+
"hendrycksTest-formal_logic": 1,
|
| 364 |
+
"hendrycksTest-global_facts": 1,
|
| 365 |
+
"hendrycksTest-high_school_biology": 1,
|
| 366 |
+
"hendrycksTest-high_school_chemistry": 1,
|
| 367 |
+
"hendrycksTest-high_school_computer_science": 1,
|
| 368 |
+
"hendrycksTest-high_school_european_history": 1,
|
| 369 |
+
"hendrycksTest-high_school_geography": 1,
|
| 370 |
+
"hendrycksTest-high_school_government_and_politics": 1,
|
| 371 |
+
"hendrycksTest-high_school_macroeconomics": 1,
|
| 372 |
+
"hendrycksTest-high_school_mathematics": 1,
|
| 373 |
+
"hendrycksTest-high_school_microeconomics": 1,
|
| 374 |
+
"hendrycksTest-high_school_physics": 1,
|
| 375 |
+
"hendrycksTest-high_school_psychology": 1,
|
| 376 |
+
"hendrycksTest-high_school_statistics": 1,
|
| 377 |
+
"hendrycksTest-high_school_us_history": 1,
|
| 378 |
+
"hendrycksTest-high_school_world_history": 1,
|
| 379 |
+
"hendrycksTest-human_aging": 1,
|
| 380 |
+
"hendrycksTest-human_sexuality": 1,
|
| 381 |
+
"hendrycksTest-international_law": 1,
|
| 382 |
+
"hendrycksTest-jurisprudence": 1,
|
| 383 |
+
"hendrycksTest-logical_fallacies": 1,
|
| 384 |
+
"hendrycksTest-machine_learning": 1,
|
| 385 |
+
"hendrycksTest-management": 1,
|
| 386 |
+
"hendrycksTest-marketing": 1,
|
| 387 |
+
"hendrycksTest-medical_genetics": 1,
|
| 388 |
+
"hendrycksTest-miscellaneous": 1,
|
| 389 |
+
"hendrycksTest-moral_disputes": 1,
|
| 390 |
+
"hendrycksTest-moral_scenarios": 1,
|
| 391 |
+
"hendrycksTest-nutrition": 1,
|
| 392 |
+
"hendrycksTest-philosophy": 1,
|
| 393 |
+
"hendrycksTest-prehistory": 1,
|
| 394 |
+
"hendrycksTest-professional_accounting": 1,
|
| 395 |
+
"hendrycksTest-professional_law": 1,
|
| 396 |
+
"hendrycksTest-professional_medicine": 1,
|
| 397 |
+
"hendrycksTest-professional_psychology": 1,
|
| 398 |
+
"hendrycksTest-public_relations": 1,
|
| 399 |
+
"hendrycksTest-security_studies": 1,
|
| 400 |
+
"hendrycksTest-sociology": 1,
|
| 401 |
+
"hendrycksTest-us_foreign_policy": 1,
|
| 402 |
+
"hendrycksTest-virology": 1,
|
| 403 |
+
"hendrycksTest-world_religions": 1
|
| 404 |
+
},
|
| 405 |
+
"config": {
|
| 406 |
+
"model": "hf-causal-experimental",
|
| 407 |
+
"model_args": "pretrained=BEE-spoke-data/smol_llama-220M-GQA,revision=main,trust_remote_code=True,dtype='bfloat16'",
|
| 408 |
+
"num_fewshot": 5,
|
| 409 |
+
"batch_size": "8",
|
| 410 |
+
"batch_sizes": [],
|
| 411 |
+
"device": "cuda",
|
| 412 |
+
"no_cache": false,
|
| 413 |
+
"limit": 0.25,
|
| 414 |
+
"bootstrap_iters": 100000,
|
| 415 |
+
"description_dict": {}
|
| 416 |
+
}
|
| 417 |
+
}
|