Upload eval_results/orpo-explorers/Mistral-7B-v0.1-ORPO-Capybara-TRL/main/agieval/results_2024-04-01T12-53-30.544870.json with huggingface_hub
Browse files
eval_results/orpo-explorers/Mistral-7B-v0.1-ORPO-Capybara-TRL/main/agieval/results_2024-04-01T12-53-30.544870.json
ADDED
|
@@ -0,0 +1,900 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"config_general": {
|
| 3 |
+
"lighteval_sha": "?",
|
| 4 |
+
"num_fewshot_seeds": 1,
|
| 5 |
+
"override_batch_size": 1,
|
| 6 |
+
"max_samples": null,
|
| 7 |
+
"job_id": "",
|
| 8 |
+
"start_time": 2125556.329311793,
|
| 9 |
+
"end_time": 2125848.827267762,
|
| 10 |
+
"total_evaluation_time_secondes": "292.4979559686035",
|
| 11 |
+
"model_name": "orpo-explorers/Mistral-7B-v0.1-ORPO-Capybara-TRL",
|
| 12 |
+
"model_sha": "168947bd058019e99cdfc20694c62fb497c365ba",
|
| 13 |
+
"model_dtype": "torch.bfloat16",
|
| 14 |
+
"model_size": "13.99 GB",
|
| 15 |
+
"config": null
|
| 16 |
+
},
|
| 17 |
+
"results": {
|
| 18 |
+
"lighteval|agieval:aqua-rat|0": {
|
| 19 |
+
"acc": 0.22440944881889763,
|
| 20 |
+
"acc_stderr": 0.026228687798575887,
|
| 21 |
+
"acc_norm": 0.20866141732283464,
|
| 22 |
+
"acc_norm_stderr": 0.02554712225493389
|
| 23 |
+
},
|
| 24 |
+
"lighteval|agieval:gaokao-biology|0": {
|
| 25 |
+
"acc": 0.21428571428571427,
|
| 26 |
+
"acc_stderr": 0.028382836222822366,
|
| 27 |
+
"acc_norm": 0.2761904761904762,
|
| 28 |
+
"acc_norm_stderr": 0.030927395843275758
|
| 29 |
+
},
|
| 30 |
+
"lighteval|agieval:gaokao-chemistry|0": {
|
| 31 |
+
"acc": 0.24154589371980675,
|
| 32 |
+
"acc_stderr": 0.02982159533435382,
|
| 33 |
+
"acc_norm": 0.2608695652173913,
|
| 34 |
+
"acc_norm_stderr": 0.030594167471991694
|
| 35 |
+
},
|
| 36 |
+
"lighteval|agieval:gaokao-chinese|0": {
|
| 37 |
+
"acc": 0.17479674796747968,
|
| 38 |
+
"acc_stderr": 0.024264094987893572,
|
| 39 |
+
"acc_norm": 0.17073170731707318,
|
| 40 |
+
"acc_norm_stderr": 0.024039286844125873
|
| 41 |
+
},
|
| 42 |
+
"lighteval|agieval:gaokao-english|0": {
|
| 43 |
+
"acc": 0.5915032679738562,
|
| 44 |
+
"acc_stderr": 0.02814640599309636,
|
| 45 |
+
"acc_norm": 0.5522875816993464,
|
| 46 |
+
"acc_norm_stderr": 0.02847293847803353
|
| 47 |
+
},
|
| 48 |
+
"lighteval|agieval:gaokao-geography|0": {
|
| 49 |
+
"acc": 0.2562814070351759,
|
| 50 |
+
"acc_stderr": 0.031026320485108765,
|
| 51 |
+
"acc_norm": 0.27638190954773867,
|
| 52 |
+
"acc_norm_stderr": 0.03178168502681786
|
| 53 |
+
},
|
| 54 |
+
"lighteval|agieval:gaokao-history|0": {
|
| 55 |
+
"acc": 0.2765957446808511,
|
| 56 |
+
"acc_stderr": 0.02924188386962884,
|
| 57 |
+
"acc_norm": 0.2553191489361702,
|
| 58 |
+
"acc_norm_stderr": 0.02850485647051421
|
| 59 |
+
},
|
| 60 |
+
"lighteval|agieval:gaokao-mathqa|0": {
|
| 61 |
+
"acc": 0.27635327635327633,
|
| 62 |
+
"acc_stderr": 0.023903505003127216,
|
| 63 |
+
"acc_norm": 0.2849002849002849,
|
| 64 |
+
"acc_norm_stderr": 0.024126577672411737
|
| 65 |
+
},
|
| 66 |
+
"lighteval|agieval:gaokao-physics|0": {
|
| 67 |
+
"acc": 0.275,
|
| 68 |
+
"acc_stderr": 0.03165255790786193,
|
| 69 |
+
"acc_norm": 0.305,
|
| 70 |
+
"acc_norm_stderr": 0.03263741725420572
|
| 71 |
+
},
|
| 72 |
+
"lighteval|agieval:logiqa-en|0": {
|
| 73 |
+
"acc": 0.29339477726574503,
|
| 74 |
+
"acc_stderr": 0.017859032704399497,
|
| 75 |
+
"acc_norm": 0.2995391705069124,
|
| 76 |
+
"acc_norm_stderr": 0.017966441188587944
|
| 77 |
+
},
|
| 78 |
+
"lighteval|agieval:logiqa-zh|0": {
|
| 79 |
+
"acc": 0.25806451612903225,
|
| 80 |
+
"acc_stderr": 0.017162894755127077,
|
| 81 |
+
"acc_norm": 0.31797235023041476,
|
| 82 |
+
"acc_norm_stderr": 0.01826581231613446
|
| 83 |
+
},
|
| 84 |
+
"lighteval|agieval:lsat-ar|0": {
|
| 85 |
+
"acc": 0.17391304347826086,
|
| 86 |
+
"acc_stderr": 0.025047317386049716,
|
| 87 |
+
"acc_norm": 0.19130434782608696,
|
| 88 |
+
"acc_norm_stderr": 0.025991852462828483
|
| 89 |
+
},
|
| 90 |
+
"lighteval|agieval:lsat-lr|0": {
|
| 91 |
+
"acc": 0.3235294117647059,
|
| 92 |
+
"acc_stderr": 0.02073587911537405,
|
| 93 |
+
"acc_norm": 0.30980392156862746,
|
| 94 |
+
"acc_norm_stderr": 0.020496080019546087
|
| 95 |
+
},
|
| 96 |
+
"lighteval|agieval:lsat-rc|0": {
|
| 97 |
+
"acc": 0.34944237918215615,
|
| 98 |
+
"acc_stderr": 0.029124821619700387,
|
| 99 |
+
"acc_norm": 0.30111524163568776,
|
| 100 |
+
"acc_norm_stderr": 0.0280221695876122
|
| 101 |
+
},
|
| 102 |
+
"lighteval|agieval:sat-en|0": {
|
| 103 |
+
"acc": 0.5728155339805825,
|
| 104 |
+
"acc_stderr": 0.03454921537431907,
|
| 105 |
+
"acc_norm": 0.47572815533980584,
|
| 106 |
+
"acc_norm_stderr": 0.03488034442356183
|
| 107 |
+
},
|
| 108 |
+
"lighteval|agieval:sat-en-without-passage|0": {
|
| 109 |
+
"acc": 0.3446601941747573,
|
| 110 |
+
"acc_stderr": 0.0331934128585908,
|
| 111 |
+
"acc_norm": 0.2766990291262136,
|
| 112 |
+
"acc_norm_stderr": 0.031245423189279947
|
| 113 |
+
},
|
| 114 |
+
"lighteval|agieval:sat-math|0": {
|
| 115 |
+
"acc": 0.2818181818181818,
|
| 116 |
+
"acc_stderr": 0.030400424640665242,
|
| 117 |
+
"acc_norm": 0.21818181818181817,
|
| 118 |
+
"acc_norm_stderr": 0.02790873191099663
|
| 119 |
+
},
|
| 120 |
+
"lighteval|agieval:_average|0": {
|
| 121 |
+
"acc": 0.3016711493310871,
|
| 122 |
+
"acc_stderr": 0.027102405062158497,
|
| 123 |
+
"acc_norm": 0.2929815367968754,
|
| 124 |
+
"acc_norm_stderr": 0.027141664847932817
|
| 125 |
+
},
|
| 126 |
+
"all": {
|
| 127 |
+
"acc": 0.3016711493310871,
|
| 128 |
+
"acc_stderr": 0.027102405062158497,
|
| 129 |
+
"acc_norm": 0.2929815367968754,
|
| 130 |
+
"acc_norm_stderr": 0.027141664847932817
|
| 131 |
+
}
|
| 132 |
+
},
|
| 133 |
+
"versions": {
|
| 134 |
+
"lighteval|agieval:aqua-rat|0": 0,
|
| 135 |
+
"lighteval|agieval:gaokao-biology|0": 0,
|
| 136 |
+
"lighteval|agieval:gaokao-chemistry|0": 0,
|
| 137 |
+
"lighteval|agieval:gaokao-chinese|0": 0,
|
| 138 |
+
"lighteval|agieval:gaokao-english|0": 0,
|
| 139 |
+
"lighteval|agieval:gaokao-geography|0": 0,
|
| 140 |
+
"lighteval|agieval:gaokao-history|0": 0,
|
| 141 |
+
"lighteval|agieval:gaokao-mathqa|0": 0,
|
| 142 |
+
"lighteval|agieval:gaokao-physics|0": 0,
|
| 143 |
+
"lighteval|agieval:logiqa-en|0": 0,
|
| 144 |
+
"lighteval|agieval:logiqa-zh|0": 0,
|
| 145 |
+
"lighteval|agieval:lsat-ar|0": 0,
|
| 146 |
+
"lighteval|agieval:lsat-lr|0": 0,
|
| 147 |
+
"lighteval|agieval:lsat-rc|0": 0,
|
| 148 |
+
"lighteval|agieval:sat-en|0": 0,
|
| 149 |
+
"lighteval|agieval:sat-en-without-passage|0": 0,
|
| 150 |
+
"lighteval|agieval:sat-math|0": 0
|
| 151 |
+
},
|
| 152 |
+
"config_tasks": {
|
| 153 |
+
"lighteval|agieval:aqua-rat": {
|
| 154 |
+
"name": "agieval:aqua-rat",
|
| 155 |
+
"prompt_function": "agieval",
|
| 156 |
+
"hf_repo": "dmayhem93/agieval-aqua-rat",
|
| 157 |
+
"hf_subset": "default",
|
| 158 |
+
"metric": [
|
| 159 |
+
"loglikelihood_acc",
|
| 160 |
+
"loglikelihood_acc_norm_nospace"
|
| 161 |
+
],
|
| 162 |
+
"hf_avail_splits": [
|
| 163 |
+
"test"
|
| 164 |
+
],
|
| 165 |
+
"evaluation_splits": [
|
| 166 |
+
"test"
|
| 167 |
+
],
|
| 168 |
+
"few_shots_split": null,
|
| 169 |
+
"few_shots_select": "random_sampling",
|
| 170 |
+
"generation_size": 1,
|
| 171 |
+
"stop_sequence": null,
|
| 172 |
+
"output_regex": null,
|
| 173 |
+
"frozen": false,
|
| 174 |
+
"suite": [
|
| 175 |
+
"lighteval"
|
| 176 |
+
],
|
| 177 |
+
"original_num_docs": 254,
|
| 178 |
+
"effective_num_docs": 254,
|
| 179 |
+
"trust_dataset": true,
|
| 180 |
+
"must_remove_duplicate_docs": null
|
| 181 |
+
},
|
| 182 |
+
"lighteval|agieval:gaokao-biology": {
|
| 183 |
+
"name": "agieval:gaokao-biology",
|
| 184 |
+
"prompt_function": "agieval",
|
| 185 |
+
"hf_repo": "dmayhem93/agieval-gaokao-biology",
|
| 186 |
+
"hf_subset": "default",
|
| 187 |
+
"metric": [
|
| 188 |
+
"loglikelihood_acc",
|
| 189 |
+
"loglikelihood_acc_norm_nospace"
|
| 190 |
+
],
|
| 191 |
+
"hf_avail_splits": [
|
| 192 |
+
"test"
|
| 193 |
+
],
|
| 194 |
+
"evaluation_splits": [
|
| 195 |
+
"test"
|
| 196 |
+
],
|
| 197 |
+
"few_shots_split": null,
|
| 198 |
+
"few_shots_select": "random_sampling",
|
| 199 |
+
"generation_size": 1,
|
| 200 |
+
"stop_sequence": null,
|
| 201 |
+
"output_regex": null,
|
| 202 |
+
"frozen": false,
|
| 203 |
+
"suite": [
|
| 204 |
+
"lighteval"
|
| 205 |
+
],
|
| 206 |
+
"original_num_docs": 210,
|
| 207 |
+
"effective_num_docs": 210,
|
| 208 |
+
"trust_dataset": true,
|
| 209 |
+
"must_remove_duplicate_docs": null
|
| 210 |
+
},
|
| 211 |
+
"lighteval|agieval:gaokao-chemistry": {
|
| 212 |
+
"name": "agieval:gaokao-chemistry",
|
| 213 |
+
"prompt_function": "agieval",
|
| 214 |
+
"hf_repo": "dmayhem93/agieval-gaokao-chemistry",
|
| 215 |
+
"hf_subset": "default",
|
| 216 |
+
"metric": [
|
| 217 |
+
"loglikelihood_acc",
|
| 218 |
+
"loglikelihood_acc_norm_nospace"
|
| 219 |
+
],
|
| 220 |
+
"hf_avail_splits": [
|
| 221 |
+
"test"
|
| 222 |
+
],
|
| 223 |
+
"evaluation_splits": [
|
| 224 |
+
"test"
|
| 225 |
+
],
|
| 226 |
+
"few_shots_split": null,
|
| 227 |
+
"few_shots_select": "random_sampling",
|
| 228 |
+
"generation_size": 1,
|
| 229 |
+
"stop_sequence": null,
|
| 230 |
+
"output_regex": null,
|
| 231 |
+
"frozen": false,
|
| 232 |
+
"suite": [
|
| 233 |
+
"lighteval"
|
| 234 |
+
],
|
| 235 |
+
"original_num_docs": 207,
|
| 236 |
+
"effective_num_docs": 207,
|
| 237 |
+
"trust_dataset": true,
|
| 238 |
+
"must_remove_duplicate_docs": null
|
| 239 |
+
},
|
| 240 |
+
"lighteval|agieval:gaokao-chinese": {
|
| 241 |
+
"name": "agieval:gaokao-chinese",
|
| 242 |
+
"prompt_function": "agieval",
|
| 243 |
+
"hf_repo": "dmayhem93/agieval-gaokao-chinese",
|
| 244 |
+
"hf_subset": "default",
|
| 245 |
+
"metric": [
|
| 246 |
+
"loglikelihood_acc",
|
| 247 |
+
"loglikelihood_acc_norm_nospace"
|
| 248 |
+
],
|
| 249 |
+
"hf_avail_splits": [
|
| 250 |
+
"test"
|
| 251 |
+
],
|
| 252 |
+
"evaluation_splits": [
|
| 253 |
+
"test"
|
| 254 |
+
],
|
| 255 |
+
"few_shots_split": null,
|
| 256 |
+
"few_shots_select": "random_sampling",
|
| 257 |
+
"generation_size": 1,
|
| 258 |
+
"stop_sequence": null,
|
| 259 |
+
"output_regex": null,
|
| 260 |
+
"frozen": false,
|
| 261 |
+
"suite": [
|
| 262 |
+
"lighteval"
|
| 263 |
+
],
|
| 264 |
+
"original_num_docs": 246,
|
| 265 |
+
"effective_num_docs": 246,
|
| 266 |
+
"trust_dataset": true,
|
| 267 |
+
"must_remove_duplicate_docs": null
|
| 268 |
+
},
|
| 269 |
+
"lighteval|agieval:gaokao-english": {
|
| 270 |
+
"name": "agieval:gaokao-english",
|
| 271 |
+
"prompt_function": "agieval",
|
| 272 |
+
"hf_repo": "dmayhem93/agieval-gaokao-english",
|
| 273 |
+
"hf_subset": "default",
|
| 274 |
+
"metric": [
|
| 275 |
+
"loglikelihood_acc",
|
| 276 |
+
"loglikelihood_acc_norm_nospace"
|
| 277 |
+
],
|
| 278 |
+
"hf_avail_splits": [
|
| 279 |
+
"test"
|
| 280 |
+
],
|
| 281 |
+
"evaluation_splits": [
|
| 282 |
+
"test"
|
| 283 |
+
],
|
| 284 |
+
"few_shots_split": null,
|
| 285 |
+
"few_shots_select": "random_sampling",
|
| 286 |
+
"generation_size": 1,
|
| 287 |
+
"stop_sequence": null,
|
| 288 |
+
"output_regex": null,
|
| 289 |
+
"frozen": false,
|
| 290 |
+
"suite": [
|
| 291 |
+
"lighteval"
|
| 292 |
+
],
|
| 293 |
+
"original_num_docs": 306,
|
| 294 |
+
"effective_num_docs": 306,
|
| 295 |
+
"trust_dataset": true,
|
| 296 |
+
"must_remove_duplicate_docs": null
|
| 297 |
+
},
|
| 298 |
+
"lighteval|agieval:gaokao-geography": {
|
| 299 |
+
"name": "agieval:gaokao-geography",
|
| 300 |
+
"prompt_function": "agieval",
|
| 301 |
+
"hf_repo": "dmayhem93/agieval-gaokao-geography",
|
| 302 |
+
"hf_subset": "default",
|
| 303 |
+
"metric": [
|
| 304 |
+
"loglikelihood_acc",
|
| 305 |
+
"loglikelihood_acc_norm_nospace"
|
| 306 |
+
],
|
| 307 |
+
"hf_avail_splits": [
|
| 308 |
+
"test"
|
| 309 |
+
],
|
| 310 |
+
"evaluation_splits": [
|
| 311 |
+
"test"
|
| 312 |
+
],
|
| 313 |
+
"few_shots_split": null,
|
| 314 |
+
"few_shots_select": "random_sampling",
|
| 315 |
+
"generation_size": 1,
|
| 316 |
+
"stop_sequence": null,
|
| 317 |
+
"output_regex": null,
|
| 318 |
+
"frozen": false,
|
| 319 |
+
"suite": [
|
| 320 |
+
"lighteval"
|
| 321 |
+
],
|
| 322 |
+
"original_num_docs": 199,
|
| 323 |
+
"effective_num_docs": 199,
|
| 324 |
+
"trust_dataset": true,
|
| 325 |
+
"must_remove_duplicate_docs": null
|
| 326 |
+
},
|
| 327 |
+
"lighteval|agieval:gaokao-history": {
|
| 328 |
+
"name": "agieval:gaokao-history",
|
| 329 |
+
"prompt_function": "agieval",
|
| 330 |
+
"hf_repo": "dmayhem93/agieval-gaokao-history",
|
| 331 |
+
"hf_subset": "default",
|
| 332 |
+
"metric": [
|
| 333 |
+
"loglikelihood_acc",
|
| 334 |
+
"loglikelihood_acc_norm_nospace"
|
| 335 |
+
],
|
| 336 |
+
"hf_avail_splits": [
|
| 337 |
+
"test"
|
| 338 |
+
],
|
| 339 |
+
"evaluation_splits": [
|
| 340 |
+
"test"
|
| 341 |
+
],
|
| 342 |
+
"few_shots_split": null,
|
| 343 |
+
"few_shots_select": "random_sampling",
|
| 344 |
+
"generation_size": 1,
|
| 345 |
+
"stop_sequence": null,
|
| 346 |
+
"output_regex": null,
|
| 347 |
+
"frozen": false,
|
| 348 |
+
"suite": [
|
| 349 |
+
"lighteval"
|
| 350 |
+
],
|
| 351 |
+
"original_num_docs": 235,
|
| 352 |
+
"effective_num_docs": 235,
|
| 353 |
+
"trust_dataset": true,
|
| 354 |
+
"must_remove_duplicate_docs": null
|
| 355 |
+
},
|
| 356 |
+
"lighteval|agieval:gaokao-mathqa": {
|
| 357 |
+
"name": "agieval:gaokao-mathqa",
|
| 358 |
+
"prompt_function": "agieval",
|
| 359 |
+
"hf_repo": "dmayhem93/agieval-gaokao-mathqa",
|
| 360 |
+
"hf_subset": "default",
|
| 361 |
+
"metric": [
|
| 362 |
+
"loglikelihood_acc",
|
| 363 |
+
"loglikelihood_acc_norm_nospace"
|
| 364 |
+
],
|
| 365 |
+
"hf_avail_splits": [
|
| 366 |
+
"test"
|
| 367 |
+
],
|
| 368 |
+
"evaluation_splits": [
|
| 369 |
+
"test"
|
| 370 |
+
],
|
| 371 |
+
"few_shots_split": null,
|
| 372 |
+
"few_shots_select": "random_sampling",
|
| 373 |
+
"generation_size": 1,
|
| 374 |
+
"stop_sequence": null,
|
| 375 |
+
"output_regex": null,
|
| 376 |
+
"frozen": false,
|
| 377 |
+
"suite": [
|
| 378 |
+
"lighteval"
|
| 379 |
+
],
|
| 380 |
+
"original_num_docs": 351,
|
| 381 |
+
"effective_num_docs": 351,
|
| 382 |
+
"trust_dataset": true,
|
| 383 |
+
"must_remove_duplicate_docs": null
|
| 384 |
+
},
|
| 385 |
+
"lighteval|agieval:gaokao-physics": {
|
| 386 |
+
"name": "agieval:gaokao-physics",
|
| 387 |
+
"prompt_function": "agieval",
|
| 388 |
+
"hf_repo": "dmayhem93/agieval-gaokao-physics",
|
| 389 |
+
"hf_subset": "default",
|
| 390 |
+
"metric": [
|
| 391 |
+
"loglikelihood_acc",
|
| 392 |
+
"loglikelihood_acc_norm_nospace"
|
| 393 |
+
],
|
| 394 |
+
"hf_avail_splits": [
|
| 395 |
+
"test"
|
| 396 |
+
],
|
| 397 |
+
"evaluation_splits": [
|
| 398 |
+
"test"
|
| 399 |
+
],
|
| 400 |
+
"few_shots_split": null,
|
| 401 |
+
"few_shots_select": "random_sampling",
|
| 402 |
+
"generation_size": 1,
|
| 403 |
+
"stop_sequence": null,
|
| 404 |
+
"output_regex": null,
|
| 405 |
+
"frozen": false,
|
| 406 |
+
"suite": [
|
| 407 |
+
"lighteval"
|
| 408 |
+
],
|
| 409 |
+
"original_num_docs": 200,
|
| 410 |
+
"effective_num_docs": 200,
|
| 411 |
+
"trust_dataset": true,
|
| 412 |
+
"must_remove_duplicate_docs": null
|
| 413 |
+
},
|
| 414 |
+
"lighteval|agieval:logiqa-en": {
|
| 415 |
+
"name": "agieval:logiqa-en",
|
| 416 |
+
"prompt_function": "agieval",
|
| 417 |
+
"hf_repo": "dmayhem93/agieval-logiqa-en",
|
| 418 |
+
"hf_subset": "default",
|
| 419 |
+
"metric": [
|
| 420 |
+
"loglikelihood_acc",
|
| 421 |
+
"loglikelihood_acc_norm_nospace"
|
| 422 |
+
],
|
| 423 |
+
"hf_avail_splits": [
|
| 424 |
+
"test"
|
| 425 |
+
],
|
| 426 |
+
"evaluation_splits": [
|
| 427 |
+
"test"
|
| 428 |
+
],
|
| 429 |
+
"few_shots_split": null,
|
| 430 |
+
"few_shots_select": "random_sampling",
|
| 431 |
+
"generation_size": 1,
|
| 432 |
+
"stop_sequence": null,
|
| 433 |
+
"output_regex": null,
|
| 434 |
+
"frozen": false,
|
| 435 |
+
"suite": [
|
| 436 |
+
"lighteval"
|
| 437 |
+
],
|
| 438 |
+
"original_num_docs": 651,
|
| 439 |
+
"effective_num_docs": 651,
|
| 440 |
+
"trust_dataset": true,
|
| 441 |
+
"must_remove_duplicate_docs": null
|
| 442 |
+
},
|
| 443 |
+
"lighteval|agieval:logiqa-zh": {
|
| 444 |
+
"name": "agieval:logiqa-zh",
|
| 445 |
+
"prompt_function": "agieval",
|
| 446 |
+
"hf_repo": "dmayhem93/agieval-logiqa-zh",
|
| 447 |
+
"hf_subset": "default",
|
| 448 |
+
"metric": [
|
| 449 |
+
"loglikelihood_acc",
|
| 450 |
+
"loglikelihood_acc_norm_nospace"
|
| 451 |
+
],
|
| 452 |
+
"hf_avail_splits": [
|
| 453 |
+
"test"
|
| 454 |
+
],
|
| 455 |
+
"evaluation_splits": [
|
| 456 |
+
"test"
|
| 457 |
+
],
|
| 458 |
+
"few_shots_split": null,
|
| 459 |
+
"few_shots_select": "random_sampling",
|
| 460 |
+
"generation_size": 1,
|
| 461 |
+
"stop_sequence": null,
|
| 462 |
+
"output_regex": null,
|
| 463 |
+
"frozen": false,
|
| 464 |
+
"suite": [
|
| 465 |
+
"lighteval"
|
| 466 |
+
],
|
| 467 |
+
"original_num_docs": 651,
|
| 468 |
+
"effective_num_docs": 651,
|
| 469 |
+
"trust_dataset": true,
|
| 470 |
+
"must_remove_duplicate_docs": null
|
| 471 |
+
},
|
| 472 |
+
"lighteval|agieval:lsat-ar": {
|
| 473 |
+
"name": "agieval:lsat-ar",
|
| 474 |
+
"prompt_function": "agieval",
|
| 475 |
+
"hf_repo": "dmayhem93/agieval-lsat-ar",
|
| 476 |
+
"hf_subset": "default",
|
| 477 |
+
"metric": [
|
| 478 |
+
"loglikelihood_acc",
|
| 479 |
+
"loglikelihood_acc_norm_nospace"
|
| 480 |
+
],
|
| 481 |
+
"hf_avail_splits": [
|
| 482 |
+
"test"
|
| 483 |
+
],
|
| 484 |
+
"evaluation_splits": [
|
| 485 |
+
"test"
|
| 486 |
+
],
|
| 487 |
+
"few_shots_split": null,
|
| 488 |
+
"few_shots_select": "random_sampling",
|
| 489 |
+
"generation_size": 1,
|
| 490 |
+
"stop_sequence": null,
|
| 491 |
+
"output_regex": null,
|
| 492 |
+
"frozen": false,
|
| 493 |
+
"suite": [
|
| 494 |
+
"lighteval"
|
| 495 |
+
],
|
| 496 |
+
"original_num_docs": 230,
|
| 497 |
+
"effective_num_docs": 230,
|
| 498 |
+
"trust_dataset": true,
|
| 499 |
+
"must_remove_duplicate_docs": null
|
| 500 |
+
},
|
| 501 |
+
"lighteval|agieval:lsat-lr": {
|
| 502 |
+
"name": "agieval:lsat-lr",
|
| 503 |
+
"prompt_function": "agieval",
|
| 504 |
+
"hf_repo": "dmayhem93/agieval-lsat-lr",
|
| 505 |
+
"hf_subset": "default",
|
| 506 |
+
"metric": [
|
| 507 |
+
"loglikelihood_acc",
|
| 508 |
+
"loglikelihood_acc_norm_nospace"
|
| 509 |
+
],
|
| 510 |
+
"hf_avail_splits": [
|
| 511 |
+
"test"
|
| 512 |
+
],
|
| 513 |
+
"evaluation_splits": [
|
| 514 |
+
"test"
|
| 515 |
+
],
|
| 516 |
+
"few_shots_split": null,
|
| 517 |
+
"few_shots_select": "random_sampling",
|
| 518 |
+
"generation_size": 1,
|
| 519 |
+
"stop_sequence": null,
|
| 520 |
+
"output_regex": null,
|
| 521 |
+
"frozen": false,
|
| 522 |
+
"suite": [
|
| 523 |
+
"lighteval"
|
| 524 |
+
],
|
| 525 |
+
"original_num_docs": 510,
|
| 526 |
+
"effective_num_docs": 510,
|
| 527 |
+
"trust_dataset": true,
|
| 528 |
+
"must_remove_duplicate_docs": null
|
| 529 |
+
},
|
| 530 |
+
"lighteval|agieval:lsat-rc": {
|
| 531 |
+
"name": "agieval:lsat-rc",
|
| 532 |
+
"prompt_function": "agieval",
|
| 533 |
+
"hf_repo": "dmayhem93/agieval-lsat-rc",
|
| 534 |
+
"hf_subset": "default",
|
| 535 |
+
"metric": [
|
| 536 |
+
"loglikelihood_acc",
|
| 537 |
+
"loglikelihood_acc_norm_nospace"
|
| 538 |
+
],
|
| 539 |
+
"hf_avail_splits": [
|
| 540 |
+
"test"
|
| 541 |
+
],
|
| 542 |
+
"evaluation_splits": [
|
| 543 |
+
"test"
|
| 544 |
+
],
|
| 545 |
+
"few_shots_split": null,
|
| 546 |
+
"few_shots_select": "random_sampling",
|
| 547 |
+
"generation_size": 1,
|
| 548 |
+
"stop_sequence": null,
|
| 549 |
+
"output_regex": null,
|
| 550 |
+
"frozen": false,
|
| 551 |
+
"suite": [
|
| 552 |
+
"lighteval"
|
| 553 |
+
],
|
| 554 |
+
"original_num_docs": 269,
|
| 555 |
+
"effective_num_docs": 269,
|
| 556 |
+
"trust_dataset": true,
|
| 557 |
+
"must_remove_duplicate_docs": null
|
| 558 |
+
},
|
| 559 |
+
"lighteval|agieval:sat-en": {
|
| 560 |
+
"name": "agieval:sat-en",
|
| 561 |
+
"prompt_function": "agieval",
|
| 562 |
+
"hf_repo": "dmayhem93/agieval-sat-en",
|
| 563 |
+
"hf_subset": "default",
|
| 564 |
+
"metric": [
|
| 565 |
+
"loglikelihood_acc",
|
| 566 |
+
"loglikelihood_acc_norm_nospace"
|
| 567 |
+
],
|
| 568 |
+
"hf_avail_splits": [
|
| 569 |
+
"test"
|
| 570 |
+
],
|
| 571 |
+
"evaluation_splits": [
|
| 572 |
+
"test"
|
| 573 |
+
],
|
| 574 |
+
"few_shots_split": null,
|
| 575 |
+
"few_shots_select": "random_sampling",
|
| 576 |
+
"generation_size": 1,
|
| 577 |
+
"stop_sequence": null,
|
| 578 |
+
"output_regex": null,
|
| 579 |
+
"frozen": false,
|
| 580 |
+
"suite": [
|
| 581 |
+
"lighteval"
|
| 582 |
+
],
|
| 583 |
+
"original_num_docs": 206,
|
| 584 |
+
"effective_num_docs": 206,
|
| 585 |
+
"trust_dataset": true,
|
| 586 |
+
"must_remove_duplicate_docs": null
|
| 587 |
+
},
|
| 588 |
+
"lighteval|agieval:sat-en-without-passage": {
|
| 589 |
+
"name": "agieval:sat-en-without-passage",
|
| 590 |
+
"prompt_function": "agieval",
|
| 591 |
+
"hf_repo": "dmayhem93/agieval-sat-en-without-passage",
|
| 592 |
+
"hf_subset": "default",
|
| 593 |
+
"metric": [
|
| 594 |
+
"loglikelihood_acc",
|
| 595 |
+
"loglikelihood_acc_norm_nospace"
|
| 596 |
+
],
|
| 597 |
+
"hf_avail_splits": [
|
| 598 |
+
"test"
|
| 599 |
+
],
|
| 600 |
+
"evaluation_splits": [
|
| 601 |
+
"test"
|
| 602 |
+
],
|
| 603 |
+
"few_shots_split": null,
|
| 604 |
+
"few_shots_select": "random_sampling",
|
| 605 |
+
"generation_size": 1,
|
| 606 |
+
"stop_sequence": null,
|
| 607 |
+
"output_regex": null,
|
| 608 |
+
"frozen": false,
|
| 609 |
+
"suite": [
|
| 610 |
+
"lighteval"
|
| 611 |
+
],
|
| 612 |
+
"original_num_docs": 206,
|
| 613 |
+
"effective_num_docs": 206,
|
| 614 |
+
"trust_dataset": true,
|
| 615 |
+
"must_remove_duplicate_docs": null
|
| 616 |
+
},
|
| 617 |
+
"lighteval|agieval:sat-math": {
|
| 618 |
+
"name": "agieval:sat-math",
|
| 619 |
+
"prompt_function": "agieval",
|
| 620 |
+
"hf_repo": "dmayhem93/agieval-sat-math",
|
| 621 |
+
"hf_subset": "default",
|
| 622 |
+
"metric": [
|
| 623 |
+
"loglikelihood_acc",
|
| 624 |
+
"loglikelihood_acc_norm_nospace"
|
| 625 |
+
],
|
| 626 |
+
"hf_avail_splits": [
|
| 627 |
+
"test"
|
| 628 |
+
],
|
| 629 |
+
"evaluation_splits": [
|
| 630 |
+
"test"
|
| 631 |
+
],
|
| 632 |
+
"few_shots_split": null,
|
| 633 |
+
"few_shots_select": "random_sampling",
|
| 634 |
+
"generation_size": 1,
|
| 635 |
+
"stop_sequence": null,
|
| 636 |
+
"output_regex": null,
|
| 637 |
+
"frozen": false,
|
| 638 |
+
"suite": [
|
| 639 |
+
"lighteval"
|
| 640 |
+
],
|
| 641 |
+
"original_num_docs": 220,
|
| 642 |
+
"effective_num_docs": 220,
|
| 643 |
+
"trust_dataset": true,
|
| 644 |
+
"must_remove_duplicate_docs": null
|
| 645 |
+
}
|
| 646 |
+
},
|
| 647 |
+
"summary_tasks": {
|
| 648 |
+
"lighteval|agieval:aqua-rat|0": {
|
| 649 |
+
"hashes": {
|
| 650 |
+
"hash_examples": "f09607f69e5b7525",
|
| 651 |
+
"hash_full_prompts": "8b913655a6fea4ab",
|
| 652 |
+
"hash_input_tokens": "293a61e5163aa27e",
|
| 653 |
+
"hash_cont_tokens": "a12c4ac8996ba11d"
|
| 654 |
+
},
|
| 655 |
+
"truncated": 0,
|
| 656 |
+
"non_truncated": 254,
|
| 657 |
+
"padded": 1270,
|
| 658 |
+
"non_padded": 0,
|
| 659 |
+
"effective_few_shots": 0.0,
|
| 660 |
+
"num_truncated_few_shots": 0
|
| 661 |
+
},
|
| 662 |
+
"lighteval|agieval:gaokao-biology|0": {
|
| 663 |
+
"hashes": {
|
| 664 |
+
"hash_examples": "f262eaf4a72db963",
|
| 665 |
+
"hash_full_prompts": "c7078ace868f7ee8",
|
| 666 |
+
"hash_input_tokens": "4abbc3b98bca27c3",
|
| 667 |
+
"hash_cont_tokens": "22b786cf7aa6d1a9"
|
| 668 |
+
},
|
| 669 |
+
"truncated": 0,
|
| 670 |
+
"non_truncated": 210,
|
| 671 |
+
"padded": 840,
|
| 672 |
+
"non_padded": 0,
|
| 673 |
+
"effective_few_shots": 0.0,
|
| 674 |
+
"num_truncated_few_shots": 0
|
| 675 |
+
},
|
| 676 |
+
"lighteval|agieval:gaokao-chemistry|0": {
|
| 677 |
+
"hashes": {
|
| 678 |
+
"hash_examples": "47f2e649f58d9da5",
|
| 679 |
+
"hash_full_prompts": "bd066d6d8c807f39",
|
| 680 |
+
"hash_input_tokens": "3d9728247bf04ac3",
|
| 681 |
+
"hash_cont_tokens": "318562bcb4103fc4"
|
| 682 |
+
},
|
| 683 |
+
"truncated": 0,
|
| 684 |
+
"non_truncated": 207,
|
| 685 |
+
"padded": 831,
|
| 686 |
+
"non_padded": 0,
|
| 687 |
+
"effective_few_shots": 0.0,
|
| 688 |
+
"num_truncated_few_shots": 0
|
| 689 |
+
},
|
| 690 |
+
"lighteval|agieval:gaokao-chinese|0": {
|
| 691 |
+
"hashes": {
|
| 692 |
+
"hash_examples": "1010b21fde4726ab",
|
| 693 |
+
"hash_full_prompts": "3f53e9dd34c43d52",
|
| 694 |
+
"hash_input_tokens": "f2a1d4c848527f86",
|
| 695 |
+
"hash_cont_tokens": "7b177add04591cdb"
|
| 696 |
+
},
|
| 697 |
+
"truncated": 0,
|
| 698 |
+
"non_truncated": 246,
|
| 699 |
+
"padded": 982,
|
| 700 |
+
"non_padded": 2,
|
| 701 |
+
"effective_few_shots": 0.0,
|
| 702 |
+
"num_truncated_few_shots": 0
|
| 703 |
+
},
|
| 704 |
+
"lighteval|agieval:gaokao-english|0": {
|
| 705 |
+
"hashes": {
|
| 706 |
+
"hash_examples": "4864e492a350ae93",
|
| 707 |
+
"hash_full_prompts": "59104cb8623f69e5",
|
| 708 |
+
"hash_input_tokens": "ff82bcaabb6cde43",
|
| 709 |
+
"hash_cont_tokens": "c9ca0addab2a9327"
|
| 710 |
+
},
|
| 711 |
+
"truncated": 0,
|
| 712 |
+
"non_truncated": 306,
|
| 713 |
+
"padded": 1224,
|
| 714 |
+
"non_padded": 0,
|
| 715 |
+
"effective_few_shots": 0.0,
|
| 716 |
+
"num_truncated_few_shots": 0
|
| 717 |
+
},
|
| 718 |
+
"lighteval|agieval:gaokao-geography|0": {
|
| 719 |
+
"hashes": {
|
| 720 |
+
"hash_examples": "ec3a021e37650e7d",
|
| 721 |
+
"hash_full_prompts": "d2456e0377df1973",
|
| 722 |
+
"hash_input_tokens": "b6593d42d60f9e65",
|
| 723 |
+
"hash_cont_tokens": "e1bc87e81807da78"
|
| 724 |
+
},
|
| 725 |
+
"truncated": 0,
|
| 726 |
+
"non_truncated": 199,
|
| 727 |
+
"padded": 796,
|
| 728 |
+
"non_padded": 0,
|
| 729 |
+
"effective_few_shots": 0.0,
|
| 730 |
+
"num_truncated_few_shots": 0
|
| 731 |
+
},
|
| 732 |
+
"lighteval|agieval:gaokao-history|0": {
|
| 733 |
+
"hashes": {
|
| 734 |
+
"hash_examples": "b3fad1596f1ae1f9",
|
| 735 |
+
"hash_full_prompts": "faea8f291d9a0cd5",
|
| 736 |
+
"hash_input_tokens": "ffa388f05b4bfadf",
|
| 737 |
+
"hash_cont_tokens": "b3c6c60f59b08db4"
|
| 738 |
+
},
|
| 739 |
+
"truncated": 0,
|
| 740 |
+
"non_truncated": 235,
|
| 741 |
+
"padded": 940,
|
| 742 |
+
"non_padded": 0,
|
| 743 |
+
"effective_few_shots": 0.0,
|
| 744 |
+
"num_truncated_few_shots": 0
|
| 745 |
+
},
|
| 746 |
+
"lighteval|agieval:gaokao-mathqa|0": {
|
| 747 |
+
"hashes": {
|
| 748 |
+
"hash_examples": "1d1088556861b0b0",
|
| 749 |
+
"hash_full_prompts": "de899bfeaaa61154",
|
| 750 |
+
"hash_input_tokens": "1e826e8ae60c9cf4",
|
| 751 |
+
"hash_cont_tokens": "5d69ebf8391bf298"
|
| 752 |
+
},
|
| 753 |
+
"truncated": 0,
|
| 754 |
+
"non_truncated": 351,
|
| 755 |
+
"padded": 1404,
|
| 756 |
+
"non_padded": 0,
|
| 757 |
+
"effective_few_shots": 0.0,
|
| 758 |
+
"num_truncated_few_shots": 0
|
| 759 |
+
},
|
| 760 |
+
"lighteval|agieval:gaokao-physics|0": {
|
| 761 |
+
"hashes": {
|
| 762 |
+
"hash_examples": "eb05f035c7bfca2f",
|
| 763 |
+
"hash_full_prompts": "08008e0300283edc",
|
| 764 |
+
"hash_input_tokens": "0a4cddbeea3c31fa",
|
| 765 |
+
"hash_cont_tokens": "93b4c52fa838ace2"
|
| 766 |
+
},
|
| 767 |
+
"truncated": 0,
|
| 768 |
+
"non_truncated": 200,
|
| 769 |
+
"padded": 800,
|
| 770 |
+
"non_padded": 0,
|
| 771 |
+
"effective_few_shots": 0.0,
|
| 772 |
+
"num_truncated_few_shots": 0
|
| 773 |
+
},
|
| 774 |
+
"lighteval|agieval:logiqa-en|0": {
|
| 775 |
+
"hashes": {
|
| 776 |
+
"hash_examples": "0a688a45f69c21e0",
|
| 777 |
+
"hash_full_prompts": "3405fd262d4b2d28",
|
| 778 |
+
"hash_input_tokens": "8672a68d080ba1fb",
|
| 779 |
+
"hash_cont_tokens": "2624c1243afac3f2"
|
| 780 |
+
},
|
| 781 |
+
"truncated": 0,
|
| 782 |
+
"non_truncated": 651,
|
| 783 |
+
"padded": 2604,
|
| 784 |
+
"non_padded": 0,
|
| 785 |
+
"effective_few_shots": 0.0,
|
| 786 |
+
"num_truncated_few_shots": 0
|
| 787 |
+
},
|
| 788 |
+
"lighteval|agieval:logiqa-zh|0": {
|
| 789 |
+
"hashes": {
|
| 790 |
+
"hash_examples": "620d6888b6012ea5",
|
| 791 |
+
"hash_full_prompts": "ac19dc4eaa56f5e0",
|
| 792 |
+
"hash_input_tokens": "24c3bd34df395ee2",
|
| 793 |
+
"hash_cont_tokens": "725ca2b921b6f8fe"
|
| 794 |
+
},
|
| 795 |
+
"truncated": 0,
|
| 796 |
+
"non_truncated": 651,
|
| 797 |
+
"padded": 2603,
|
| 798 |
+
"non_padded": 1,
|
| 799 |
+
"effective_few_shots": 0.0,
|
| 800 |
+
"num_truncated_few_shots": 0
|
| 801 |
+
},
|
| 802 |
+
"lighteval|agieval:lsat-ar|0": {
|
| 803 |
+
"hashes": {
|
| 804 |
+
"hash_examples": "627c8f5ccd5da209",
|
| 805 |
+
"hash_full_prompts": "9aed992c4bfa8dd7",
|
| 806 |
+
"hash_input_tokens": "042173fcbbb85776",
|
| 807 |
+
"hash_cont_tokens": "23c097e1d431f2b8"
|
| 808 |
+
},
|
| 809 |
+
"truncated": 0,
|
| 810 |
+
"non_truncated": 230,
|
| 811 |
+
"padded": 1137,
|
| 812 |
+
"non_padded": 13,
|
| 813 |
+
"effective_few_shots": 0.0,
|
| 814 |
+
"num_truncated_few_shots": 0
|
| 815 |
+
},
|
| 816 |
+
"lighteval|agieval:lsat-lr|0": {
|
| 817 |
+
"hashes": {
|
| 818 |
+
"hash_examples": "794641c86de172f5",
|
| 819 |
+
"hash_full_prompts": "6a36e90325996129",
|
| 820 |
+
"hash_input_tokens": "58c268b8b0f3c5c4",
|
| 821 |
+
"hash_cont_tokens": "b555f4319746d815"
|
| 822 |
+
},
|
| 823 |
+
"truncated": 0,
|
| 824 |
+
"non_truncated": 510,
|
| 825 |
+
"padded": 2532,
|
| 826 |
+
"non_padded": 18,
|
| 827 |
+
"effective_few_shots": 0.0,
|
| 828 |
+
"num_truncated_few_shots": 0
|
| 829 |
+
},
|
| 830 |
+
"lighteval|agieval:lsat-rc|0": {
|
| 831 |
+
"hashes": {
|
| 832 |
+
"hash_examples": "35981ed917ea01cf",
|
| 833 |
+
"hash_full_prompts": "15f0f342f9572c41",
|
| 834 |
+
"hash_input_tokens": "1eaa7331cf1a555e",
|
| 835 |
+
"hash_cont_tokens": "8c1c4fc8c9cabd97"
|
| 836 |
+
},
|
| 837 |
+
"truncated": 0,
|
| 838 |
+
"non_truncated": 269,
|
| 839 |
+
"padded": 1345,
|
| 840 |
+
"non_padded": 0,
|
| 841 |
+
"effective_few_shots": 0.0,
|
| 842 |
+
"num_truncated_few_shots": 0
|
| 843 |
+
},
|
| 844 |
+
"lighteval|agieval:sat-en|0": {
|
| 845 |
+
"hashes": {
|
| 846 |
+
"hash_examples": "041c39c646536a1e",
|
| 847 |
+
"hash_full_prompts": "163217fd603b9352",
|
| 848 |
+
"hash_input_tokens": "e1e9417c28fa0db0",
|
| 849 |
+
"hash_cont_tokens": "4837f17aae6c95e0"
|
| 850 |
+
},
|
| 851 |
+
"truncated": 0,
|
| 852 |
+
"non_truncated": 206,
|
| 853 |
+
"padded": 821,
|
| 854 |
+
"non_padded": 0,
|
| 855 |
+
"effective_few_shots": 0.0,
|
| 856 |
+
"num_truncated_few_shots": 0
|
| 857 |
+
},
|
| 858 |
+
"lighteval|agieval:sat-en-without-passage|0": {
|
| 859 |
+
"hashes": {
|
| 860 |
+
"hash_examples": "e4d9284367dff68f",
|
| 861 |
+
"hash_full_prompts": "bdd4c7065b87de8a",
|
| 862 |
+
"hash_input_tokens": "70bcab139ba874e5",
|
| 863 |
+
"hash_cont_tokens": "4837f17aae6c95e0"
|
| 864 |
+
},
|
| 865 |
+
"truncated": 0,
|
| 866 |
+
"non_truncated": 206,
|
| 867 |
+
"padded": 817,
|
| 868 |
+
"non_padded": 4,
|
| 869 |
+
"effective_few_shots": 0.0,
|
| 870 |
+
"num_truncated_few_shots": 0
|
| 871 |
+
},
|
| 872 |
+
"lighteval|agieval:sat-math|0": {
|
| 873 |
+
"hashes": {
|
| 874 |
+
"hash_examples": "01db7291603fc1a0",
|
| 875 |
+
"hash_full_prompts": "63ca65b2f0baebb5",
|
| 876 |
+
"hash_input_tokens": "e242b0a44e256fbb",
|
| 877 |
+
"hash_cont_tokens": "d959ef83452da9fe"
|
| 878 |
+
},
|
| 879 |
+
"truncated": 0,
|
| 880 |
+
"non_truncated": 220,
|
| 881 |
+
"padded": 877,
|
| 882 |
+
"non_padded": 3,
|
| 883 |
+
"effective_few_shots": 0.0,
|
| 884 |
+
"num_truncated_few_shots": 0
|
| 885 |
+
}
|
| 886 |
+
},
|
| 887 |
+
"summary_general": {
|
| 888 |
+
"hashes": {
|
| 889 |
+
"hash_examples": "da3af66181f18ddf",
|
| 890 |
+
"hash_full_prompts": "f7c298d03686fa0e",
|
| 891 |
+
"hash_input_tokens": "84e02a40c0a714da",
|
| 892 |
+
"hash_cont_tokens": "b3bace8c3199f6d8"
|
| 893 |
+
},
|
| 894 |
+
"truncated": 0,
|
| 895 |
+
"non_truncated": 5151,
|
| 896 |
+
"padded": 21823,
|
| 897 |
+
"non_padded": 41,
|
| 898 |
+
"num_truncated_few_shots": 0
|
| 899 |
+
}
|
| 900 |
+
}
|