Spaces:
Restarting
Restarting
Upload eval_results/orpo-explorers/argilla-mistral-orpo-OpenHermesPreferences-50k-beta-0.2-ckpt-epoch-2/main/agieval/results_2024-05-09T18-58-33.971890.json with huggingface_hub
Browse files
eval_results/orpo-explorers/argilla-mistral-orpo-OpenHermesPreferences-50k-beta-0.2-ckpt-epoch-2/main/agieval/results_2024-05-09T18-58-33.971890.json
ADDED
|
@@ -0,0 +1,934 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"config_general": {
|
| 3 |
+
"lighteval_sha": "?",
|
| 4 |
+
"num_fewshot_seeds": 1,
|
| 5 |
+
"override_batch_size": 4,
|
| 6 |
+
"max_samples": null,
|
| 7 |
+
"job_id": "",
|
| 8 |
+
"start_time": 833244.635872818,
|
| 9 |
+
"end_time": 833433.874554734,
|
| 10 |
+
"total_evaluation_time_secondes": "189.23868191603106",
|
| 11 |
+
"model_name": "orpo-explorers/argilla-mistral-orpo-OpenHermesPreferences-50k-beta-0.2-ckpt-epoch-2",
|
| 12 |
+
"model_sha": "153b3124eb16cf41c5f0d98cd51706c385ca1ed5",
|
| 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": 0.026228687798575897,
|
| 21 |
+
"acc_norm": 0.2204724409448819,
|
| 22 |
+
"acc_norm_stderr": 0.026063493749639587
|
| 23 |
+
},
|
| 24 |
+
"lighteval|agieval:gaokao-biology|0": {
|
| 25 |
+
"acc": 0.28095238095238095,
|
| 26 |
+
"acc_stderr": 0.03109009446934461,
|
| 27 |
+
"acc_norm": 0.30952380952380953,
|
| 28 |
+
"acc_norm_stderr": 0.03197777494209474
|
| 29 |
+
},
|
| 30 |
+
"lighteval|agieval:gaokao-chemistry|0": {
|
| 31 |
+
"acc": 0.2028985507246377,
|
| 32 |
+
"acc_stderr": 0.028019647132494573,
|
| 33 |
+
"acc_norm": 0.2318840579710145,
|
| 34 |
+
"acc_norm_stderr": 0.029404596565406543
|
| 35 |
+
},
|
| 36 |
+
"lighteval|agieval:gaokao-chinese|0": {
|
| 37 |
+
"acc": 0.2886178861788618,
|
| 38 |
+
"acc_stderr": 0.028948765576340286,
|
| 39 |
+
"acc_norm": 0.2804878048780488,
|
| 40 |
+
"acc_norm_stderr": 0.028700735693673953
|
| 41 |
+
},
|
| 42 |
+
"lighteval|agieval:gaokao-english|0": {
|
| 43 |
+
"acc": 0.5522875816993464,
|
| 44 |
+
"acc_stderr": 0.02847293847803353,
|
| 45 |
+
"acc_norm": 0.5424836601307189,
|
| 46 |
+
"acc_norm_stderr": 0.028526383452142645
|
| 47 |
+
},
|
| 48 |
+
"lighteval|agieval:gaokao-geography|0": {
|
| 49 |
+
"acc": 0.31155778894472363,
|
| 50 |
+
"acc_stderr": 0.03291322637124229,
|
| 51 |
+
"acc_norm": 0.3015075376884422,
|
| 52 |
+
"acc_norm_stderr": 0.03261349587454653
|
| 53 |
+
},
|
| 54 |
+
"lighteval|agieval:gaokao-history|0": {
|
| 55 |
+
"acc": 0.2851063829787234,
|
| 56 |
+
"acc_stderr": 0.02951319662553935,
|
| 57 |
+
"acc_norm": 0.2723404255319149,
|
| 58 |
+
"acc_norm_stderr": 0.0291012906983867
|
| 59 |
+
},
|
| 60 |
+
"lighteval|agieval:gaokao-mathqa|0": {
|
| 61 |
+
"acc": 0.2706552706552707,
|
| 62 |
+
"acc_stderr": 0.023748744034266786,
|
| 63 |
+
"acc_norm": 0.28205128205128205,
|
| 64 |
+
"acc_norm_stderr": 0.024053414152940704
|
| 65 |
+
},
|
| 66 |
+
"lighteval|agieval:gaokao-physics|0": {
|
| 67 |
+
"acc": 0.285,
|
| 68 |
+
"acc_stderr": 0.03199992148231578,
|
| 69 |
+
"acc_norm": 0.325,
|
| 70 |
+
"acc_norm_stderr": 0.03320221279784479
|
| 71 |
+
},
|
| 72 |
+
"lighteval|agieval:logiqa-en|0": {
|
| 73 |
+
"acc": 0.3241167434715822,
|
| 74 |
+
"acc_stderr": 0.01835819163513243,
|
| 75 |
+
"acc_norm": 0.3241167434715822,
|
| 76 |
+
"acc_norm_stderr": 0.01835819163513243
|
| 77 |
+
},
|
| 78 |
+
"lighteval|agieval:logiqa-zh|0": {
|
| 79 |
+
"acc": 0.31490015360983103,
|
| 80 |
+
"acc_stderr": 0.018218251493671685,
|
| 81 |
+
"acc_norm": 0.3317972350230415,
|
| 82 |
+
"acc_norm_stderr": 0.0184685941264168
|
| 83 |
+
},
|
| 84 |
+
"lighteval|agieval:lsat-ar|0": {
|
| 85 |
+
"acc": 0.17391304347826086,
|
| 86 |
+
"acc_stderr": 0.025047317386049713,
|
| 87 |
+
"acc_norm": 0.2,
|
| 88 |
+
"acc_norm_stderr": 0.026432744018203558
|
| 89 |
+
},
|
| 90 |
+
"lighteval|agieval:lsat-lr|0": {
|
| 91 |
+
"acc": 0.2784313725490196,
|
| 92 |
+
"acc_stderr": 0.019867307525414934,
|
| 93 |
+
"acc_norm": 0.30196078431372547,
|
| 94 |
+
"acc_norm_stderr": 0.020349619453119146
|
| 95 |
+
},
|
| 96 |
+
"lighteval|agieval:lsat-rc|0": {
|
| 97 |
+
"acc": 0.34944237918215615,
|
| 98 |
+
"acc_stderr": 0.02912482161970038,
|
| 99 |
+
"acc_norm": 0.3048327137546468,
|
| 100 |
+
"acc_norm_stderr": 0.02811952967561346
|
| 101 |
+
},
|
| 102 |
+
"lighteval|agieval:sat-en|0": {
|
| 103 |
+
"acc": 0.46116504854368934,
|
| 104 |
+
"acc_stderr": 0.03481602144131183,
|
| 105 |
+
"acc_norm": 0.4223300970873786,
|
| 106 |
+
"acc_norm_stderr": 0.03449760586825818
|
| 107 |
+
},
|
| 108 |
+
"lighteval|agieval:sat-en-without-passage|0": {
|
| 109 |
+
"acc": 0.2912621359223301,
|
| 110 |
+
"acc_stderr": 0.031732764025283834,
|
| 111 |
+
"acc_norm": 0.2912621359223301,
|
| 112 |
+
"acc_norm_stderr": 0.031732764025283834
|
| 113 |
+
},
|
| 114 |
+
"lighteval|agieval:sat-math|0": {
|
| 115 |
+
"acc": 0.2409090909090909,
|
| 116 |
+
"acc_stderr": 0.02889691178349409,
|
| 117 |
+
"acc_norm": 0.21363636363636362,
|
| 118 |
+
"acc_norm_stderr": 0.027696649960503875
|
| 119 |
+
},
|
| 120 |
+
"lighteval|agieval:_average|0": {
|
| 121 |
+
"acc": 0.30209560344816483,
|
| 122 |
+
"acc_stderr": 0.027470400522247765,
|
| 123 |
+
"acc_norm": 0.30327571128995184,
|
| 124 |
+
"acc_norm_stderr": 0.027605829217012205
|
| 125 |
+
},
|
| 126 |
+
"all": {
|
| 127 |
+
"acc": 0.30209560344816483,
|
| 128 |
+
"acc_stderr": 0.027470400522247765,
|
| 129 |
+
"acc_norm": 0.30327571128995184,
|
| 130 |
+
"acc_norm_stderr": 0.027605829217012205
|
| 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 |
+
"num_samples": null,
|
| 174 |
+
"frozen": false,
|
| 175 |
+
"suite": [
|
| 176 |
+
"lighteval"
|
| 177 |
+
],
|
| 178 |
+
"original_num_docs": 254,
|
| 179 |
+
"effective_num_docs": 254,
|
| 180 |
+
"trust_dataset": true,
|
| 181 |
+
"must_remove_duplicate_docs": null,
|
| 182 |
+
"version": 0
|
| 183 |
+
},
|
| 184 |
+
"lighteval|agieval:gaokao-biology": {
|
| 185 |
+
"name": "agieval:gaokao-biology",
|
| 186 |
+
"prompt_function": "agieval",
|
| 187 |
+
"hf_repo": "dmayhem93/agieval-gaokao-biology",
|
| 188 |
+
"hf_subset": "default",
|
| 189 |
+
"metric": [
|
| 190 |
+
"loglikelihood_acc",
|
| 191 |
+
"loglikelihood_acc_norm_nospace"
|
| 192 |
+
],
|
| 193 |
+
"hf_avail_splits": [
|
| 194 |
+
"test"
|
| 195 |
+
],
|
| 196 |
+
"evaluation_splits": [
|
| 197 |
+
"test"
|
| 198 |
+
],
|
| 199 |
+
"few_shots_split": null,
|
| 200 |
+
"few_shots_select": "random_sampling",
|
| 201 |
+
"generation_size": 1,
|
| 202 |
+
"stop_sequence": null,
|
| 203 |
+
"output_regex": null,
|
| 204 |
+
"num_samples": null,
|
| 205 |
+
"frozen": false,
|
| 206 |
+
"suite": [
|
| 207 |
+
"lighteval"
|
| 208 |
+
],
|
| 209 |
+
"original_num_docs": 210,
|
| 210 |
+
"effective_num_docs": 210,
|
| 211 |
+
"trust_dataset": true,
|
| 212 |
+
"must_remove_duplicate_docs": null,
|
| 213 |
+
"version": 0
|
| 214 |
+
},
|
| 215 |
+
"lighteval|agieval:gaokao-chemistry": {
|
| 216 |
+
"name": "agieval:gaokao-chemistry",
|
| 217 |
+
"prompt_function": "agieval",
|
| 218 |
+
"hf_repo": "dmayhem93/agieval-gaokao-chemistry",
|
| 219 |
+
"hf_subset": "default",
|
| 220 |
+
"metric": [
|
| 221 |
+
"loglikelihood_acc",
|
| 222 |
+
"loglikelihood_acc_norm_nospace"
|
| 223 |
+
],
|
| 224 |
+
"hf_avail_splits": [
|
| 225 |
+
"test"
|
| 226 |
+
],
|
| 227 |
+
"evaluation_splits": [
|
| 228 |
+
"test"
|
| 229 |
+
],
|
| 230 |
+
"few_shots_split": null,
|
| 231 |
+
"few_shots_select": "random_sampling",
|
| 232 |
+
"generation_size": 1,
|
| 233 |
+
"stop_sequence": null,
|
| 234 |
+
"output_regex": null,
|
| 235 |
+
"num_samples": null,
|
| 236 |
+
"frozen": false,
|
| 237 |
+
"suite": [
|
| 238 |
+
"lighteval"
|
| 239 |
+
],
|
| 240 |
+
"original_num_docs": 207,
|
| 241 |
+
"effective_num_docs": 207,
|
| 242 |
+
"trust_dataset": true,
|
| 243 |
+
"must_remove_duplicate_docs": null,
|
| 244 |
+
"version": 0
|
| 245 |
+
},
|
| 246 |
+
"lighteval|agieval:gaokao-chinese": {
|
| 247 |
+
"name": "agieval:gaokao-chinese",
|
| 248 |
+
"prompt_function": "agieval",
|
| 249 |
+
"hf_repo": "dmayhem93/agieval-gaokao-chinese",
|
| 250 |
+
"hf_subset": "default",
|
| 251 |
+
"metric": [
|
| 252 |
+
"loglikelihood_acc",
|
| 253 |
+
"loglikelihood_acc_norm_nospace"
|
| 254 |
+
],
|
| 255 |
+
"hf_avail_splits": [
|
| 256 |
+
"test"
|
| 257 |
+
],
|
| 258 |
+
"evaluation_splits": [
|
| 259 |
+
"test"
|
| 260 |
+
],
|
| 261 |
+
"few_shots_split": null,
|
| 262 |
+
"few_shots_select": "random_sampling",
|
| 263 |
+
"generation_size": 1,
|
| 264 |
+
"stop_sequence": null,
|
| 265 |
+
"output_regex": null,
|
| 266 |
+
"num_samples": null,
|
| 267 |
+
"frozen": false,
|
| 268 |
+
"suite": [
|
| 269 |
+
"lighteval"
|
| 270 |
+
],
|
| 271 |
+
"original_num_docs": 246,
|
| 272 |
+
"effective_num_docs": 246,
|
| 273 |
+
"trust_dataset": true,
|
| 274 |
+
"must_remove_duplicate_docs": null,
|
| 275 |
+
"version": 0
|
| 276 |
+
},
|
| 277 |
+
"lighteval|agieval:gaokao-english": {
|
| 278 |
+
"name": "agieval:gaokao-english",
|
| 279 |
+
"prompt_function": "agieval",
|
| 280 |
+
"hf_repo": "dmayhem93/agieval-gaokao-english",
|
| 281 |
+
"hf_subset": "default",
|
| 282 |
+
"metric": [
|
| 283 |
+
"loglikelihood_acc",
|
| 284 |
+
"loglikelihood_acc_norm_nospace"
|
| 285 |
+
],
|
| 286 |
+
"hf_avail_splits": [
|
| 287 |
+
"test"
|
| 288 |
+
],
|
| 289 |
+
"evaluation_splits": [
|
| 290 |
+
"test"
|
| 291 |
+
],
|
| 292 |
+
"few_shots_split": null,
|
| 293 |
+
"few_shots_select": "random_sampling",
|
| 294 |
+
"generation_size": 1,
|
| 295 |
+
"stop_sequence": null,
|
| 296 |
+
"output_regex": null,
|
| 297 |
+
"num_samples": null,
|
| 298 |
+
"frozen": false,
|
| 299 |
+
"suite": [
|
| 300 |
+
"lighteval"
|
| 301 |
+
],
|
| 302 |
+
"original_num_docs": 306,
|
| 303 |
+
"effective_num_docs": 306,
|
| 304 |
+
"trust_dataset": true,
|
| 305 |
+
"must_remove_duplicate_docs": null,
|
| 306 |
+
"version": 0
|
| 307 |
+
},
|
| 308 |
+
"lighteval|agieval:gaokao-geography": {
|
| 309 |
+
"name": "agieval:gaokao-geography",
|
| 310 |
+
"prompt_function": "agieval",
|
| 311 |
+
"hf_repo": "dmayhem93/agieval-gaokao-geography",
|
| 312 |
+
"hf_subset": "default",
|
| 313 |
+
"metric": [
|
| 314 |
+
"loglikelihood_acc",
|
| 315 |
+
"loglikelihood_acc_norm_nospace"
|
| 316 |
+
],
|
| 317 |
+
"hf_avail_splits": [
|
| 318 |
+
"test"
|
| 319 |
+
],
|
| 320 |
+
"evaluation_splits": [
|
| 321 |
+
"test"
|
| 322 |
+
],
|
| 323 |
+
"few_shots_split": null,
|
| 324 |
+
"few_shots_select": "random_sampling",
|
| 325 |
+
"generation_size": 1,
|
| 326 |
+
"stop_sequence": null,
|
| 327 |
+
"output_regex": null,
|
| 328 |
+
"num_samples": null,
|
| 329 |
+
"frozen": false,
|
| 330 |
+
"suite": [
|
| 331 |
+
"lighteval"
|
| 332 |
+
],
|
| 333 |
+
"original_num_docs": 199,
|
| 334 |
+
"effective_num_docs": 199,
|
| 335 |
+
"trust_dataset": true,
|
| 336 |
+
"must_remove_duplicate_docs": null,
|
| 337 |
+
"version": 0
|
| 338 |
+
},
|
| 339 |
+
"lighteval|agieval:gaokao-history": {
|
| 340 |
+
"name": "agieval:gaokao-history",
|
| 341 |
+
"prompt_function": "agieval",
|
| 342 |
+
"hf_repo": "dmayhem93/agieval-gaokao-history",
|
| 343 |
+
"hf_subset": "default",
|
| 344 |
+
"metric": [
|
| 345 |
+
"loglikelihood_acc",
|
| 346 |
+
"loglikelihood_acc_norm_nospace"
|
| 347 |
+
],
|
| 348 |
+
"hf_avail_splits": [
|
| 349 |
+
"test"
|
| 350 |
+
],
|
| 351 |
+
"evaluation_splits": [
|
| 352 |
+
"test"
|
| 353 |
+
],
|
| 354 |
+
"few_shots_split": null,
|
| 355 |
+
"few_shots_select": "random_sampling",
|
| 356 |
+
"generation_size": 1,
|
| 357 |
+
"stop_sequence": null,
|
| 358 |
+
"output_regex": null,
|
| 359 |
+
"num_samples": null,
|
| 360 |
+
"frozen": false,
|
| 361 |
+
"suite": [
|
| 362 |
+
"lighteval"
|
| 363 |
+
],
|
| 364 |
+
"original_num_docs": 235,
|
| 365 |
+
"effective_num_docs": 235,
|
| 366 |
+
"trust_dataset": true,
|
| 367 |
+
"must_remove_duplicate_docs": null,
|
| 368 |
+
"version": 0
|
| 369 |
+
},
|
| 370 |
+
"lighteval|agieval:gaokao-mathqa": {
|
| 371 |
+
"name": "agieval:gaokao-mathqa",
|
| 372 |
+
"prompt_function": "agieval",
|
| 373 |
+
"hf_repo": "dmayhem93/agieval-gaokao-mathqa",
|
| 374 |
+
"hf_subset": "default",
|
| 375 |
+
"metric": [
|
| 376 |
+
"loglikelihood_acc",
|
| 377 |
+
"loglikelihood_acc_norm_nospace"
|
| 378 |
+
],
|
| 379 |
+
"hf_avail_splits": [
|
| 380 |
+
"test"
|
| 381 |
+
],
|
| 382 |
+
"evaluation_splits": [
|
| 383 |
+
"test"
|
| 384 |
+
],
|
| 385 |
+
"few_shots_split": null,
|
| 386 |
+
"few_shots_select": "random_sampling",
|
| 387 |
+
"generation_size": 1,
|
| 388 |
+
"stop_sequence": null,
|
| 389 |
+
"output_regex": null,
|
| 390 |
+
"num_samples": null,
|
| 391 |
+
"frozen": false,
|
| 392 |
+
"suite": [
|
| 393 |
+
"lighteval"
|
| 394 |
+
],
|
| 395 |
+
"original_num_docs": 351,
|
| 396 |
+
"effective_num_docs": 351,
|
| 397 |
+
"trust_dataset": true,
|
| 398 |
+
"must_remove_duplicate_docs": null,
|
| 399 |
+
"version": 0
|
| 400 |
+
},
|
| 401 |
+
"lighteval|agieval:gaokao-physics": {
|
| 402 |
+
"name": "agieval:gaokao-physics",
|
| 403 |
+
"prompt_function": "agieval",
|
| 404 |
+
"hf_repo": "dmayhem93/agieval-gaokao-physics",
|
| 405 |
+
"hf_subset": "default",
|
| 406 |
+
"metric": [
|
| 407 |
+
"loglikelihood_acc",
|
| 408 |
+
"loglikelihood_acc_norm_nospace"
|
| 409 |
+
],
|
| 410 |
+
"hf_avail_splits": [
|
| 411 |
+
"test"
|
| 412 |
+
],
|
| 413 |
+
"evaluation_splits": [
|
| 414 |
+
"test"
|
| 415 |
+
],
|
| 416 |
+
"few_shots_split": null,
|
| 417 |
+
"few_shots_select": "random_sampling",
|
| 418 |
+
"generation_size": 1,
|
| 419 |
+
"stop_sequence": null,
|
| 420 |
+
"output_regex": null,
|
| 421 |
+
"num_samples": null,
|
| 422 |
+
"frozen": false,
|
| 423 |
+
"suite": [
|
| 424 |
+
"lighteval"
|
| 425 |
+
],
|
| 426 |
+
"original_num_docs": 200,
|
| 427 |
+
"effective_num_docs": 200,
|
| 428 |
+
"trust_dataset": true,
|
| 429 |
+
"must_remove_duplicate_docs": null,
|
| 430 |
+
"version": 0
|
| 431 |
+
},
|
| 432 |
+
"lighteval|agieval:logiqa-en": {
|
| 433 |
+
"name": "agieval:logiqa-en",
|
| 434 |
+
"prompt_function": "agieval",
|
| 435 |
+
"hf_repo": "dmayhem93/agieval-logiqa-en",
|
| 436 |
+
"hf_subset": "default",
|
| 437 |
+
"metric": [
|
| 438 |
+
"loglikelihood_acc",
|
| 439 |
+
"loglikelihood_acc_norm_nospace"
|
| 440 |
+
],
|
| 441 |
+
"hf_avail_splits": [
|
| 442 |
+
"test"
|
| 443 |
+
],
|
| 444 |
+
"evaluation_splits": [
|
| 445 |
+
"test"
|
| 446 |
+
],
|
| 447 |
+
"few_shots_split": null,
|
| 448 |
+
"few_shots_select": "random_sampling",
|
| 449 |
+
"generation_size": 1,
|
| 450 |
+
"stop_sequence": null,
|
| 451 |
+
"output_regex": null,
|
| 452 |
+
"num_samples": null,
|
| 453 |
+
"frozen": false,
|
| 454 |
+
"suite": [
|
| 455 |
+
"lighteval"
|
| 456 |
+
],
|
| 457 |
+
"original_num_docs": 651,
|
| 458 |
+
"effective_num_docs": 651,
|
| 459 |
+
"trust_dataset": true,
|
| 460 |
+
"must_remove_duplicate_docs": null,
|
| 461 |
+
"version": 0
|
| 462 |
+
},
|
| 463 |
+
"lighteval|agieval:logiqa-zh": {
|
| 464 |
+
"name": "agieval:logiqa-zh",
|
| 465 |
+
"prompt_function": "agieval",
|
| 466 |
+
"hf_repo": "dmayhem93/agieval-logiqa-zh",
|
| 467 |
+
"hf_subset": "default",
|
| 468 |
+
"metric": [
|
| 469 |
+
"loglikelihood_acc",
|
| 470 |
+
"loglikelihood_acc_norm_nospace"
|
| 471 |
+
],
|
| 472 |
+
"hf_avail_splits": [
|
| 473 |
+
"test"
|
| 474 |
+
],
|
| 475 |
+
"evaluation_splits": [
|
| 476 |
+
"test"
|
| 477 |
+
],
|
| 478 |
+
"few_shots_split": null,
|
| 479 |
+
"few_shots_select": "random_sampling",
|
| 480 |
+
"generation_size": 1,
|
| 481 |
+
"stop_sequence": null,
|
| 482 |
+
"output_regex": null,
|
| 483 |
+
"num_samples": null,
|
| 484 |
+
"frozen": false,
|
| 485 |
+
"suite": [
|
| 486 |
+
"lighteval"
|
| 487 |
+
],
|
| 488 |
+
"original_num_docs": 651,
|
| 489 |
+
"effective_num_docs": 651,
|
| 490 |
+
"trust_dataset": true,
|
| 491 |
+
"must_remove_duplicate_docs": null,
|
| 492 |
+
"version": 0
|
| 493 |
+
},
|
| 494 |
+
"lighteval|agieval:lsat-ar": {
|
| 495 |
+
"name": "agieval:lsat-ar",
|
| 496 |
+
"prompt_function": "agieval",
|
| 497 |
+
"hf_repo": "dmayhem93/agieval-lsat-ar",
|
| 498 |
+
"hf_subset": "default",
|
| 499 |
+
"metric": [
|
| 500 |
+
"loglikelihood_acc",
|
| 501 |
+
"loglikelihood_acc_norm_nospace"
|
| 502 |
+
],
|
| 503 |
+
"hf_avail_splits": [
|
| 504 |
+
"test"
|
| 505 |
+
],
|
| 506 |
+
"evaluation_splits": [
|
| 507 |
+
"test"
|
| 508 |
+
],
|
| 509 |
+
"few_shots_split": null,
|
| 510 |
+
"few_shots_select": "random_sampling",
|
| 511 |
+
"generation_size": 1,
|
| 512 |
+
"stop_sequence": null,
|
| 513 |
+
"output_regex": null,
|
| 514 |
+
"num_samples": null,
|
| 515 |
+
"frozen": false,
|
| 516 |
+
"suite": [
|
| 517 |
+
"lighteval"
|
| 518 |
+
],
|
| 519 |
+
"original_num_docs": 230,
|
| 520 |
+
"effective_num_docs": 230,
|
| 521 |
+
"trust_dataset": true,
|
| 522 |
+
"must_remove_duplicate_docs": null,
|
| 523 |
+
"version": 0
|
| 524 |
+
},
|
| 525 |
+
"lighteval|agieval:lsat-lr": {
|
| 526 |
+
"name": "agieval:lsat-lr",
|
| 527 |
+
"prompt_function": "agieval",
|
| 528 |
+
"hf_repo": "dmayhem93/agieval-lsat-lr",
|
| 529 |
+
"hf_subset": "default",
|
| 530 |
+
"metric": [
|
| 531 |
+
"loglikelihood_acc",
|
| 532 |
+
"loglikelihood_acc_norm_nospace"
|
| 533 |
+
],
|
| 534 |
+
"hf_avail_splits": [
|
| 535 |
+
"test"
|
| 536 |
+
],
|
| 537 |
+
"evaluation_splits": [
|
| 538 |
+
"test"
|
| 539 |
+
],
|
| 540 |
+
"few_shots_split": null,
|
| 541 |
+
"few_shots_select": "random_sampling",
|
| 542 |
+
"generation_size": 1,
|
| 543 |
+
"stop_sequence": null,
|
| 544 |
+
"output_regex": null,
|
| 545 |
+
"num_samples": null,
|
| 546 |
+
"frozen": false,
|
| 547 |
+
"suite": [
|
| 548 |
+
"lighteval"
|
| 549 |
+
],
|
| 550 |
+
"original_num_docs": 510,
|
| 551 |
+
"effective_num_docs": 510,
|
| 552 |
+
"trust_dataset": true,
|
| 553 |
+
"must_remove_duplicate_docs": null,
|
| 554 |
+
"version": 0
|
| 555 |
+
},
|
| 556 |
+
"lighteval|agieval:lsat-rc": {
|
| 557 |
+
"name": "agieval:lsat-rc",
|
| 558 |
+
"prompt_function": "agieval",
|
| 559 |
+
"hf_repo": "dmayhem93/agieval-lsat-rc",
|
| 560 |
+
"hf_subset": "default",
|
| 561 |
+
"metric": [
|
| 562 |
+
"loglikelihood_acc",
|
| 563 |
+
"loglikelihood_acc_norm_nospace"
|
| 564 |
+
],
|
| 565 |
+
"hf_avail_splits": [
|
| 566 |
+
"test"
|
| 567 |
+
],
|
| 568 |
+
"evaluation_splits": [
|
| 569 |
+
"test"
|
| 570 |
+
],
|
| 571 |
+
"few_shots_split": null,
|
| 572 |
+
"few_shots_select": "random_sampling",
|
| 573 |
+
"generation_size": 1,
|
| 574 |
+
"stop_sequence": null,
|
| 575 |
+
"output_regex": null,
|
| 576 |
+
"num_samples": null,
|
| 577 |
+
"frozen": false,
|
| 578 |
+
"suite": [
|
| 579 |
+
"lighteval"
|
| 580 |
+
],
|
| 581 |
+
"original_num_docs": 269,
|
| 582 |
+
"effective_num_docs": 269,
|
| 583 |
+
"trust_dataset": true,
|
| 584 |
+
"must_remove_duplicate_docs": null,
|
| 585 |
+
"version": 0
|
| 586 |
+
},
|
| 587 |
+
"lighteval|agieval:sat-en": {
|
| 588 |
+
"name": "agieval:sat-en",
|
| 589 |
+
"prompt_function": "agieval",
|
| 590 |
+
"hf_repo": "dmayhem93/agieval-sat-en",
|
| 591 |
+
"hf_subset": "default",
|
| 592 |
+
"metric": [
|
| 593 |
+
"loglikelihood_acc",
|
| 594 |
+
"loglikelihood_acc_norm_nospace"
|
| 595 |
+
],
|
| 596 |
+
"hf_avail_splits": [
|
| 597 |
+
"test"
|
| 598 |
+
],
|
| 599 |
+
"evaluation_splits": [
|
| 600 |
+
"test"
|
| 601 |
+
],
|
| 602 |
+
"few_shots_split": null,
|
| 603 |
+
"few_shots_select": "random_sampling",
|
| 604 |
+
"generation_size": 1,
|
| 605 |
+
"stop_sequence": null,
|
| 606 |
+
"output_regex": null,
|
| 607 |
+
"num_samples": 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 |
+
"version": 0
|
| 617 |
+
},
|
| 618 |
+
"lighteval|agieval:sat-en-without-passage": {
|
| 619 |
+
"name": "agieval:sat-en-without-passage",
|
| 620 |
+
"prompt_function": "agieval",
|
| 621 |
+
"hf_repo": "dmayhem93/agieval-sat-en-without-passage",
|
| 622 |
+
"hf_subset": "default",
|
| 623 |
+
"metric": [
|
| 624 |
+
"loglikelihood_acc",
|
| 625 |
+
"loglikelihood_acc_norm_nospace"
|
| 626 |
+
],
|
| 627 |
+
"hf_avail_splits": [
|
| 628 |
+
"test"
|
| 629 |
+
],
|
| 630 |
+
"evaluation_splits": [
|
| 631 |
+
"test"
|
| 632 |
+
],
|
| 633 |
+
"few_shots_split": null,
|
| 634 |
+
"few_shots_select": "random_sampling",
|
| 635 |
+
"generation_size": 1,
|
| 636 |
+
"stop_sequence": null,
|
| 637 |
+
"output_regex": null,
|
| 638 |
+
"num_samples": null,
|
| 639 |
+
"frozen": false,
|
| 640 |
+
"suite": [
|
| 641 |
+
"lighteval"
|
| 642 |
+
],
|
| 643 |
+
"original_num_docs": 206,
|
| 644 |
+
"effective_num_docs": 206,
|
| 645 |
+
"trust_dataset": true,
|
| 646 |
+
"must_remove_duplicate_docs": null,
|
| 647 |
+
"version": 0
|
| 648 |
+
},
|
| 649 |
+
"lighteval|agieval:sat-math": {
|
| 650 |
+
"name": "agieval:sat-math",
|
| 651 |
+
"prompt_function": "agieval",
|
| 652 |
+
"hf_repo": "dmayhem93/agieval-sat-math",
|
| 653 |
+
"hf_subset": "default",
|
| 654 |
+
"metric": [
|
| 655 |
+
"loglikelihood_acc",
|
| 656 |
+
"loglikelihood_acc_norm_nospace"
|
| 657 |
+
],
|
| 658 |
+
"hf_avail_splits": [
|
| 659 |
+
"test"
|
| 660 |
+
],
|
| 661 |
+
"evaluation_splits": [
|
| 662 |
+
"test"
|
| 663 |
+
],
|
| 664 |
+
"few_shots_split": null,
|
| 665 |
+
"few_shots_select": "random_sampling",
|
| 666 |
+
"generation_size": 1,
|
| 667 |
+
"stop_sequence": null,
|
| 668 |
+
"output_regex": null,
|
| 669 |
+
"num_samples": null,
|
| 670 |
+
"frozen": false,
|
| 671 |
+
"suite": [
|
| 672 |
+
"lighteval"
|
| 673 |
+
],
|
| 674 |
+
"original_num_docs": 220,
|
| 675 |
+
"effective_num_docs": 220,
|
| 676 |
+
"trust_dataset": true,
|
| 677 |
+
"must_remove_duplicate_docs": null,
|
| 678 |
+
"version": 0
|
| 679 |
+
}
|
| 680 |
+
},
|
| 681 |
+
"summary_tasks": {
|
| 682 |
+
"lighteval|agieval:aqua-rat|0": {
|
| 683 |
+
"hashes": {
|
| 684 |
+
"hash_examples": "f09607f69e5b7525",
|
| 685 |
+
"hash_full_prompts": "8b913655a6fea4ab",
|
| 686 |
+
"hash_input_tokens": "293a61e5163aa27e",
|
| 687 |
+
"hash_cont_tokens": "6e9004a0164bb799"
|
| 688 |
+
},
|
| 689 |
+
"truncated": 0,
|
| 690 |
+
"non_truncated": 254,
|
| 691 |
+
"padded": 1270,
|
| 692 |
+
"non_padded": 0,
|
| 693 |
+
"effective_few_shots": 0.0,
|
| 694 |
+
"num_truncated_few_shots": 0
|
| 695 |
+
},
|
| 696 |
+
"lighteval|agieval:gaokao-biology|0": {
|
| 697 |
+
"hashes": {
|
| 698 |
+
"hash_examples": "f262eaf4a72db963",
|
| 699 |
+
"hash_full_prompts": "c7078ace868f7ee8",
|
| 700 |
+
"hash_input_tokens": "4abbc3b98bca27c3",
|
| 701 |
+
"hash_cont_tokens": "1246a476ed0cbfc5"
|
| 702 |
+
},
|
| 703 |
+
"truncated": 0,
|
| 704 |
+
"non_truncated": 210,
|
| 705 |
+
"padded": 840,
|
| 706 |
+
"non_padded": 0,
|
| 707 |
+
"effective_few_shots": 0.0,
|
| 708 |
+
"num_truncated_few_shots": 0
|
| 709 |
+
},
|
| 710 |
+
"lighteval|agieval:gaokao-chemistry|0": {
|
| 711 |
+
"hashes": {
|
| 712 |
+
"hash_examples": "47f2e649f58d9da5",
|
| 713 |
+
"hash_full_prompts": "bd066d6d8c807f39",
|
| 714 |
+
"hash_input_tokens": "3d9728247bf04ac3",
|
| 715 |
+
"hash_cont_tokens": "a9eb294028dc6ef3"
|
| 716 |
+
},
|
| 717 |
+
"truncated": 0,
|
| 718 |
+
"non_truncated": 207,
|
| 719 |
+
"padded": 831,
|
| 720 |
+
"non_padded": 0,
|
| 721 |
+
"effective_few_shots": 0.0,
|
| 722 |
+
"num_truncated_few_shots": 0
|
| 723 |
+
},
|
| 724 |
+
"lighteval|agieval:gaokao-chinese|0": {
|
| 725 |
+
"hashes": {
|
| 726 |
+
"hash_examples": "1010b21fde4726ab",
|
| 727 |
+
"hash_full_prompts": "3f53e9dd34c43d52",
|
| 728 |
+
"hash_input_tokens": "f2a1d4c848527f86",
|
| 729 |
+
"hash_cont_tokens": "0a9d93afd335aa3c"
|
| 730 |
+
},
|
| 731 |
+
"truncated": 0,
|
| 732 |
+
"non_truncated": 246,
|
| 733 |
+
"padded": 982,
|
| 734 |
+
"non_padded": 2,
|
| 735 |
+
"effective_few_shots": 0.0,
|
| 736 |
+
"num_truncated_few_shots": 0
|
| 737 |
+
},
|
| 738 |
+
"lighteval|agieval:gaokao-english|0": {
|
| 739 |
+
"hashes": {
|
| 740 |
+
"hash_examples": "4864e492a350ae93",
|
| 741 |
+
"hash_full_prompts": "59104cb8623f69e5",
|
| 742 |
+
"hash_input_tokens": "ff82bcaabb6cde43",
|
| 743 |
+
"hash_cont_tokens": "3215b688d7c88b2f"
|
| 744 |
+
},
|
| 745 |
+
"truncated": 0,
|
| 746 |
+
"non_truncated": 306,
|
| 747 |
+
"padded": 1224,
|
| 748 |
+
"non_padded": 0,
|
| 749 |
+
"effective_few_shots": 0.0,
|
| 750 |
+
"num_truncated_few_shots": 0
|
| 751 |
+
},
|
| 752 |
+
"lighteval|agieval:gaokao-geography|0": {
|
| 753 |
+
"hashes": {
|
| 754 |
+
"hash_examples": "ec3a021e37650e7d",
|
| 755 |
+
"hash_full_prompts": "d2456e0377df1973",
|
| 756 |
+
"hash_input_tokens": "b6593d42d60f9e65",
|
| 757 |
+
"hash_cont_tokens": "8fd995e0ac66be75"
|
| 758 |
+
},
|
| 759 |
+
"truncated": 0,
|
| 760 |
+
"non_truncated": 199,
|
| 761 |
+
"padded": 796,
|
| 762 |
+
"non_padded": 0,
|
| 763 |
+
"effective_few_shots": 0.0,
|
| 764 |
+
"num_truncated_few_shots": 0
|
| 765 |
+
},
|
| 766 |
+
"lighteval|agieval:gaokao-history|0": {
|
| 767 |
+
"hashes": {
|
| 768 |
+
"hash_examples": "b3fad1596f1ae1f9",
|
| 769 |
+
"hash_full_prompts": "faea8f291d9a0cd5",
|
| 770 |
+
"hash_input_tokens": "ffa388f05b4bfadf",
|
| 771 |
+
"hash_cont_tokens": "4b0a7458d3ea41a3"
|
| 772 |
+
},
|
| 773 |
+
"truncated": 0,
|
| 774 |
+
"non_truncated": 235,
|
| 775 |
+
"padded": 940,
|
| 776 |
+
"non_padded": 0,
|
| 777 |
+
"effective_few_shots": 0.0,
|
| 778 |
+
"num_truncated_few_shots": 0
|
| 779 |
+
},
|
| 780 |
+
"lighteval|agieval:gaokao-mathqa|0": {
|
| 781 |
+
"hashes": {
|
| 782 |
+
"hash_examples": "1d1088556861b0b0",
|
| 783 |
+
"hash_full_prompts": "de899bfeaaa61154",
|
| 784 |
+
"hash_input_tokens": "1e826e8ae60c9cf4",
|
| 785 |
+
"hash_cont_tokens": "9bbd22eae9aff3f4"
|
| 786 |
+
},
|
| 787 |
+
"truncated": 0,
|
| 788 |
+
"non_truncated": 351,
|
| 789 |
+
"padded": 1404,
|
| 790 |
+
"non_padded": 0,
|
| 791 |
+
"effective_few_shots": 0.0,
|
| 792 |
+
"num_truncated_few_shots": 0
|
| 793 |
+
},
|
| 794 |
+
"lighteval|agieval:gaokao-physics|0": {
|
| 795 |
+
"hashes": {
|
| 796 |
+
"hash_examples": "eb05f035c7bfca2f",
|
| 797 |
+
"hash_full_prompts": "08008e0300283edc",
|
| 798 |
+
"hash_input_tokens": "0a4cddbeea3c31fa",
|
| 799 |
+
"hash_cont_tokens": "d7e81c1c739c343d"
|
| 800 |
+
},
|
| 801 |
+
"truncated": 0,
|
| 802 |
+
"non_truncated": 200,
|
| 803 |
+
"padded": 800,
|
| 804 |
+
"non_padded": 0,
|
| 805 |
+
"effective_few_shots": 0.0,
|
| 806 |
+
"num_truncated_few_shots": 0
|
| 807 |
+
},
|
| 808 |
+
"lighteval|agieval:logiqa-en|0": {
|
| 809 |
+
"hashes": {
|
| 810 |
+
"hash_examples": "0a688a45f69c21e0",
|
| 811 |
+
"hash_full_prompts": "3405fd262d4b2d28",
|
| 812 |
+
"hash_input_tokens": "8672a68d080ba1fb",
|
| 813 |
+
"hash_cont_tokens": "c1f60bd55a215749"
|
| 814 |
+
},
|
| 815 |
+
"truncated": 0,
|
| 816 |
+
"non_truncated": 651,
|
| 817 |
+
"padded": 2604,
|
| 818 |
+
"non_padded": 0,
|
| 819 |
+
"effective_few_shots": 0.0,
|
| 820 |
+
"num_truncated_few_shots": 0
|
| 821 |
+
},
|
| 822 |
+
"lighteval|agieval:logiqa-zh|0": {
|
| 823 |
+
"hashes": {
|
| 824 |
+
"hash_examples": "620d6888b6012ea5",
|
| 825 |
+
"hash_full_prompts": "ac19dc4eaa56f5e0",
|
| 826 |
+
"hash_input_tokens": "24c3bd34df395ee2",
|
| 827 |
+
"hash_cont_tokens": "d1987256a4a8ad62"
|
| 828 |
+
},
|
| 829 |
+
"truncated": 0,
|
| 830 |
+
"non_truncated": 651,
|
| 831 |
+
"padded": 2603,
|
| 832 |
+
"non_padded": 1,
|
| 833 |
+
"effective_few_shots": 0.0,
|
| 834 |
+
"num_truncated_few_shots": 0
|
| 835 |
+
},
|
| 836 |
+
"lighteval|agieval:lsat-ar|0": {
|
| 837 |
+
"hashes": {
|
| 838 |
+
"hash_examples": "627c8f5ccd5da209",
|
| 839 |
+
"hash_full_prompts": "9aed992c4bfa8dd7",
|
| 840 |
+
"hash_input_tokens": "042173fcbbb85776",
|
| 841 |
+
"hash_cont_tokens": "f1f82716a353907a"
|
| 842 |
+
},
|
| 843 |
+
"truncated": 0,
|
| 844 |
+
"non_truncated": 230,
|
| 845 |
+
"padded": 1137,
|
| 846 |
+
"non_padded": 13,
|
| 847 |
+
"effective_few_shots": 0.0,
|
| 848 |
+
"num_truncated_few_shots": 0
|
| 849 |
+
},
|
| 850 |
+
"lighteval|agieval:lsat-lr|0": {
|
| 851 |
+
"hashes": {
|
| 852 |
+
"hash_examples": "794641c86de172f5",
|
| 853 |
+
"hash_full_prompts": "6a36e90325996129",
|
| 854 |
+
"hash_input_tokens": "58c268b8b0f3c5c4",
|
| 855 |
+
"hash_cont_tokens": "9f8685be13aa273f"
|
| 856 |
+
},
|
| 857 |
+
"truncated": 0,
|
| 858 |
+
"non_truncated": 510,
|
| 859 |
+
"padded": 2532,
|
| 860 |
+
"non_padded": 18,
|
| 861 |
+
"effective_few_shots": 0.0,
|
| 862 |
+
"num_truncated_few_shots": 0
|
| 863 |
+
},
|
| 864 |
+
"lighteval|agieval:lsat-rc|0": {
|
| 865 |
+
"hashes": {
|
| 866 |
+
"hash_examples": "35981ed917ea01cf",
|
| 867 |
+
"hash_full_prompts": "15f0f342f9572c41",
|
| 868 |
+
"hash_input_tokens": "1eaa7331cf1a555e",
|
| 869 |
+
"hash_cont_tokens": "e39cea8f1c13694c"
|
| 870 |
+
},
|
| 871 |
+
"truncated": 0,
|
| 872 |
+
"non_truncated": 269,
|
| 873 |
+
"padded": 1345,
|
| 874 |
+
"non_padded": 0,
|
| 875 |
+
"effective_few_shots": 0.0,
|
| 876 |
+
"num_truncated_few_shots": 0
|
| 877 |
+
},
|
| 878 |
+
"lighteval|agieval:sat-en|0": {
|
| 879 |
+
"hashes": {
|
| 880 |
+
"hash_examples": "041c39c646536a1e",
|
| 881 |
+
"hash_full_prompts": "163217fd603b9352",
|
| 882 |
+
"hash_input_tokens": "e1e9417c28fa0db0",
|
| 883 |
+
"hash_cont_tokens": "bc6d627ee3fe275d"
|
| 884 |
+
},
|
| 885 |
+
"truncated": 0,
|
| 886 |
+
"non_truncated": 206,
|
| 887 |
+
"padded": 821,
|
| 888 |
+
"non_padded": 0,
|
| 889 |
+
"effective_few_shots": 0.0,
|
| 890 |
+
"num_truncated_few_shots": 0
|
| 891 |
+
},
|
| 892 |
+
"lighteval|agieval:sat-en-without-passage|0": {
|
| 893 |
+
"hashes": {
|
| 894 |
+
"hash_examples": "e4d9284367dff68f",
|
| 895 |
+
"hash_full_prompts": "bdd4c7065b87de8a",
|
| 896 |
+
"hash_input_tokens": "70bcab139ba874e5",
|
| 897 |
+
"hash_cont_tokens": "71939640e1182ae0"
|
| 898 |
+
},
|
| 899 |
+
"truncated": 0,
|
| 900 |
+
"non_truncated": 206,
|
| 901 |
+
"padded": 817,
|
| 902 |
+
"non_padded": 4,
|
| 903 |
+
"effective_few_shots": 0.0,
|
| 904 |
+
"num_truncated_few_shots": 0
|
| 905 |
+
},
|
| 906 |
+
"lighteval|agieval:sat-math|0": {
|
| 907 |
+
"hashes": {
|
| 908 |
+
"hash_examples": "01db7291603fc1a0",
|
| 909 |
+
"hash_full_prompts": "63ca65b2f0baebb5",
|
| 910 |
+
"hash_input_tokens": "e242b0a44e256fbb",
|
| 911 |
+
"hash_cont_tokens": "065259bbb09c8c6d"
|
| 912 |
+
},
|
| 913 |
+
"truncated": 0,
|
| 914 |
+
"non_truncated": 220,
|
| 915 |
+
"padded": 877,
|
| 916 |
+
"non_padded": 3,
|
| 917 |
+
"effective_few_shots": 0.0,
|
| 918 |
+
"num_truncated_few_shots": 0
|
| 919 |
+
}
|
| 920 |
+
},
|
| 921 |
+
"summary_general": {
|
| 922 |
+
"hashes": {
|
| 923 |
+
"hash_examples": "da3af66181f18ddf",
|
| 924 |
+
"hash_full_prompts": "f7c298d03686fa0e",
|
| 925 |
+
"hash_input_tokens": "84e02a40c0a714da",
|
| 926 |
+
"hash_cont_tokens": "a8617293ec34570e"
|
| 927 |
+
},
|
| 928 |
+
"truncated": 0,
|
| 929 |
+
"non_truncated": 5151,
|
| 930 |
+
"padded": 21823,
|
| 931 |
+
"non_padded": 41,
|
| 932 |
+
"num_truncated_few_shots": 0
|
| 933 |
+
}
|
| 934 |
+
}
|