Upload eval_results/orpo-explorers/argilla-mistral-orpo-OpenHermesPreferences-50k-beta-0.2/main/bbh/results_2024-05-09T19-42-29.906304.json with huggingface_hub
Browse files
eval_results/orpo-explorers/argilla-mistral-orpo-OpenHermesPreferences-50k-beta-0.2/main/bbh/results_2024-05-09T19-42-29.906304.json
ADDED
|
@@ -0,0 +1,1000 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 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": 2423635.741185222,
|
| 9 |
+
"end_time": 2423731.717094719,
|
| 10 |
+
"total_evaluation_time_secondes": "95.97590949712321",
|
| 11 |
+
"model_name": "orpo-explorers/argilla-mistral-orpo-OpenHermesPreferences-50k-beta-0.2",
|
| 12 |
+
"model_sha": "8e456b857961e00f2e1b076756e84b97a3b88ea5",
|
| 13 |
+
"model_dtype": "torch.bfloat16",
|
| 14 |
+
"model_size": "13.99 GB",
|
| 15 |
+
"config": null
|
| 16 |
+
},
|
| 17 |
+
"results": {
|
| 18 |
+
"lighteval|bigbench:causal_judgment|0": {
|
| 19 |
+
"acc": 0.5421052631578948,
|
| 20 |
+
"acc_stderr": 0.03624046284425957
|
| 21 |
+
},
|
| 22 |
+
"lighteval|bigbench:date_understanding|0": {
|
| 23 |
+
"acc": 0.4173441734417344,
|
| 24 |
+
"acc_stderr": 0.025705692903559226
|
| 25 |
+
},
|
| 26 |
+
"lighteval|bigbench:disambiguation_qa|0": {
|
| 27 |
+
"acc": 0.6007751937984496,
|
| 28 |
+
"acc_stderr": 0.03054908408103616
|
| 29 |
+
},
|
| 30 |
+
"lighteval|bigbench:geometric_shapes|0": {
|
| 31 |
+
"acc": 0.19166666666666668,
|
| 32 |
+
"acc_stderr": 0.020774056803809365
|
| 33 |
+
},
|
| 34 |
+
"lighteval|bigbench:logical_deduction_five_objects|0": {
|
| 35 |
+
"acc": 0.2,
|
| 36 |
+
"acc_stderr": 0.01790645924143384
|
| 37 |
+
},
|
| 38 |
+
"lighteval|bigbench:logical_deduction_seven_objects|0": {
|
| 39 |
+
"acc": 0.14285714285714285,
|
| 40 |
+
"acc_stderr": 0.013235458703202278
|
| 41 |
+
},
|
| 42 |
+
"lighteval|bigbench:logical_deduction_three_objects|0": {
|
| 43 |
+
"acc": 0.3333333333333333,
|
| 44 |
+
"acc_stderr": 0.027262027336984396
|
| 45 |
+
},
|
| 46 |
+
"lighteval|bigbench:movie_recommendation|0": {
|
| 47 |
+
"acc": 0.498,
|
| 48 |
+
"acc_stderr": 0.02238289498648353
|
| 49 |
+
},
|
| 50 |
+
"lighteval|bigbench:navigate|0": {
|
| 51 |
+
"acc": 0.524,
|
| 52 |
+
"acc_stderr": 0.015801065586651755
|
| 53 |
+
},
|
| 54 |
+
"lighteval|bigbench:reasoning_about_colored_objects|0": {
|
| 55 |
+
"acc": 0.263,
|
| 56 |
+
"acc_stderr": 0.009847029094655511
|
| 57 |
+
},
|
| 58 |
+
"lighteval|bigbench:ruin_names|0": {
|
| 59 |
+
"acc": 0.26339285714285715,
|
| 60 |
+
"acc_stderr": 0.02083369001657861
|
| 61 |
+
},
|
| 62 |
+
"lighteval|bigbench:salient_translation_error_detection|0": {
|
| 63 |
+
"acc": 0.28857715430861725,
|
| 64 |
+
"acc_stderr": 0.014349847898982434
|
| 65 |
+
},
|
| 66 |
+
"lighteval|bigbench:snarks|0": {
|
| 67 |
+
"acc": 0.4696132596685083,
|
| 68 |
+
"acc_stderr": 0.037198913216803256
|
| 69 |
+
},
|
| 70 |
+
"lighteval|bigbench:sports_understanding|0": {
|
| 71 |
+
"acc": 0.552,
|
| 72 |
+
"acc_stderr": 0.01573351656634783
|
| 73 |
+
},
|
| 74 |
+
"lighteval|bigbench:temporal_sequences|0": {
|
| 75 |
+
"acc": 0.069,
|
| 76 |
+
"acc_stderr": 0.008018934050315151
|
| 77 |
+
},
|
| 78 |
+
"lighteval|bigbench:tracking_shuffled_objects_five_objects|0": {
|
| 79 |
+
"acc": 0.2,
|
| 80 |
+
"acc_stderr": 0.011318236699485788
|
| 81 |
+
},
|
| 82 |
+
"lighteval|bigbench:tracking_shuffled_objects_seven_objects|0": {
|
| 83 |
+
"acc": 0.14285714285714285,
|
| 84 |
+
"acc_stderr": 0.008367248752248818
|
| 85 |
+
},
|
| 86 |
+
"lighteval|bigbench:tracking_shuffled_objects_three_objects|0": {
|
| 87 |
+
"acc": 0.3333333333333333,
|
| 88 |
+
"acc_stderr": 0.027262027336984396
|
| 89 |
+
},
|
| 90 |
+
"lighteval|bigbench:_average|0": {
|
| 91 |
+
"acc": 0.3351030844758711,
|
| 92 |
+
"acc_stderr": 0.02015481367332344
|
| 93 |
+
},
|
| 94 |
+
"all": {
|
| 95 |
+
"acc": 0.3351030844758711,
|
| 96 |
+
"acc_stderr": 0.02015481367332344
|
| 97 |
+
}
|
| 98 |
+
},
|
| 99 |
+
"versions": {
|
| 100 |
+
"lighteval|bigbench:causal_judgment|0": 0,
|
| 101 |
+
"lighteval|bigbench:date_understanding|0": 0,
|
| 102 |
+
"lighteval|bigbench:disambiguation_qa|0": 0,
|
| 103 |
+
"lighteval|bigbench:geometric_shapes|0": 0,
|
| 104 |
+
"lighteval|bigbench:logical_deduction_five_objects|0": 0,
|
| 105 |
+
"lighteval|bigbench:logical_deduction_seven_objects|0": 0,
|
| 106 |
+
"lighteval|bigbench:logical_deduction_three_objects|0": 0,
|
| 107 |
+
"lighteval|bigbench:movie_recommendation|0": 0,
|
| 108 |
+
"lighteval|bigbench:navigate|0": 0,
|
| 109 |
+
"lighteval|bigbench:reasoning_about_colored_objects|0": 0,
|
| 110 |
+
"lighteval|bigbench:ruin_names|0": 0,
|
| 111 |
+
"lighteval|bigbench:salient_translation_error_detection|0": 0,
|
| 112 |
+
"lighteval|bigbench:snarks|0": 0,
|
| 113 |
+
"lighteval|bigbench:sports_understanding|0": 0,
|
| 114 |
+
"lighteval|bigbench:temporal_sequences|0": 0,
|
| 115 |
+
"lighteval|bigbench:tracking_shuffled_objects_five_objects|0": 0,
|
| 116 |
+
"lighteval|bigbench:tracking_shuffled_objects_seven_objects|0": 0,
|
| 117 |
+
"lighteval|bigbench:tracking_shuffled_objects_three_objects|0": 0
|
| 118 |
+
},
|
| 119 |
+
"config_tasks": {
|
| 120 |
+
"lighteval|bigbench:causal_judgment": {
|
| 121 |
+
"name": "bigbench:causal_judgment",
|
| 122 |
+
"prompt_function": "bbh_lighteval",
|
| 123 |
+
"hf_repo": "lighteval/bbh",
|
| 124 |
+
"hf_subset": "causal_judgement",
|
| 125 |
+
"metric": [
|
| 126 |
+
"loglikelihood_acc_single_token"
|
| 127 |
+
],
|
| 128 |
+
"hf_avail_splits": [
|
| 129 |
+
"train"
|
| 130 |
+
],
|
| 131 |
+
"evaluation_splits": [
|
| 132 |
+
"train"
|
| 133 |
+
],
|
| 134 |
+
"few_shots_split": null,
|
| 135 |
+
"few_shots_select": null,
|
| 136 |
+
"generation_size": -1,
|
| 137 |
+
"stop_sequence": [
|
| 138 |
+
"</s>",
|
| 139 |
+
"Q:",
|
| 140 |
+
"\n\n"
|
| 141 |
+
],
|
| 142 |
+
"output_regex": null,
|
| 143 |
+
"num_samples": null,
|
| 144 |
+
"frozen": false,
|
| 145 |
+
"suite": [
|
| 146 |
+
"lighteval"
|
| 147 |
+
],
|
| 148 |
+
"original_num_docs": 190,
|
| 149 |
+
"effective_num_docs": 190,
|
| 150 |
+
"trust_dataset": true,
|
| 151 |
+
"must_remove_duplicate_docs": null,
|
| 152 |
+
"version": 0
|
| 153 |
+
},
|
| 154 |
+
"lighteval|bigbench:date_understanding": {
|
| 155 |
+
"name": "bigbench:date_understanding",
|
| 156 |
+
"prompt_function": "bbh_lighteval",
|
| 157 |
+
"hf_repo": "lighteval/bbh",
|
| 158 |
+
"hf_subset": "date_understanding",
|
| 159 |
+
"metric": [
|
| 160 |
+
"loglikelihood_acc_single_token"
|
| 161 |
+
],
|
| 162 |
+
"hf_avail_splits": [
|
| 163 |
+
"train"
|
| 164 |
+
],
|
| 165 |
+
"evaluation_splits": [
|
| 166 |
+
"train"
|
| 167 |
+
],
|
| 168 |
+
"few_shots_split": null,
|
| 169 |
+
"few_shots_select": null,
|
| 170 |
+
"generation_size": -1,
|
| 171 |
+
"stop_sequence": [
|
| 172 |
+
"</s>",
|
| 173 |
+
"Q:",
|
| 174 |
+
"\n\n"
|
| 175 |
+
],
|
| 176 |
+
"output_regex": null,
|
| 177 |
+
"num_samples": null,
|
| 178 |
+
"frozen": false,
|
| 179 |
+
"suite": [
|
| 180 |
+
"lighteval"
|
| 181 |
+
],
|
| 182 |
+
"original_num_docs": 369,
|
| 183 |
+
"effective_num_docs": 369,
|
| 184 |
+
"trust_dataset": true,
|
| 185 |
+
"must_remove_duplicate_docs": null,
|
| 186 |
+
"version": 0
|
| 187 |
+
},
|
| 188 |
+
"lighteval|bigbench:disambiguation_qa": {
|
| 189 |
+
"name": "bigbench:disambiguation_qa",
|
| 190 |
+
"prompt_function": "bbh_lighteval",
|
| 191 |
+
"hf_repo": "lighteval/bbh",
|
| 192 |
+
"hf_subset": "disambiguation_qa",
|
| 193 |
+
"metric": [
|
| 194 |
+
"loglikelihood_acc_single_token"
|
| 195 |
+
],
|
| 196 |
+
"hf_avail_splits": [
|
| 197 |
+
"train"
|
| 198 |
+
],
|
| 199 |
+
"evaluation_splits": [
|
| 200 |
+
"train"
|
| 201 |
+
],
|
| 202 |
+
"few_shots_split": null,
|
| 203 |
+
"few_shots_select": null,
|
| 204 |
+
"generation_size": -1,
|
| 205 |
+
"stop_sequence": [
|
| 206 |
+
"</s>",
|
| 207 |
+
"Q:",
|
| 208 |
+
"\n\n"
|
| 209 |
+
],
|
| 210 |
+
"output_regex": null,
|
| 211 |
+
"num_samples": null,
|
| 212 |
+
"frozen": false,
|
| 213 |
+
"suite": [
|
| 214 |
+
"lighteval"
|
| 215 |
+
],
|
| 216 |
+
"original_num_docs": 258,
|
| 217 |
+
"effective_num_docs": 258,
|
| 218 |
+
"trust_dataset": true,
|
| 219 |
+
"must_remove_duplicate_docs": null,
|
| 220 |
+
"version": 0
|
| 221 |
+
},
|
| 222 |
+
"lighteval|bigbench:geometric_shapes": {
|
| 223 |
+
"name": "bigbench:geometric_shapes",
|
| 224 |
+
"prompt_function": "bbh_lighteval",
|
| 225 |
+
"hf_repo": "lighteval/bbh",
|
| 226 |
+
"hf_subset": "geometric_shapes",
|
| 227 |
+
"metric": [
|
| 228 |
+
"loglikelihood_acc_single_token"
|
| 229 |
+
],
|
| 230 |
+
"hf_avail_splits": [
|
| 231 |
+
"train"
|
| 232 |
+
],
|
| 233 |
+
"evaluation_splits": [
|
| 234 |
+
"train"
|
| 235 |
+
],
|
| 236 |
+
"few_shots_split": null,
|
| 237 |
+
"few_shots_select": null,
|
| 238 |
+
"generation_size": -1,
|
| 239 |
+
"stop_sequence": [
|
| 240 |
+
"</s>",
|
| 241 |
+
"Q:",
|
| 242 |
+
"\n\n"
|
| 243 |
+
],
|
| 244 |
+
"output_regex": null,
|
| 245 |
+
"num_samples": null,
|
| 246 |
+
"frozen": false,
|
| 247 |
+
"suite": [
|
| 248 |
+
"lighteval"
|
| 249 |
+
],
|
| 250 |
+
"original_num_docs": 360,
|
| 251 |
+
"effective_num_docs": 360,
|
| 252 |
+
"trust_dataset": true,
|
| 253 |
+
"must_remove_duplicate_docs": null,
|
| 254 |
+
"version": 0
|
| 255 |
+
},
|
| 256 |
+
"lighteval|bigbench:logical_deduction_five_objects": {
|
| 257 |
+
"name": "bigbench:logical_deduction_five_objects",
|
| 258 |
+
"prompt_function": "bbh_lighteval",
|
| 259 |
+
"hf_repo": "lighteval/bbh",
|
| 260 |
+
"hf_subset": "logical_deduction_five_objects",
|
| 261 |
+
"metric": [
|
| 262 |
+
"loglikelihood_acc_single_token"
|
| 263 |
+
],
|
| 264 |
+
"hf_avail_splits": [
|
| 265 |
+
"train"
|
| 266 |
+
],
|
| 267 |
+
"evaluation_splits": [
|
| 268 |
+
"train"
|
| 269 |
+
],
|
| 270 |
+
"few_shots_split": null,
|
| 271 |
+
"few_shots_select": null,
|
| 272 |
+
"generation_size": -1,
|
| 273 |
+
"stop_sequence": [
|
| 274 |
+
"</s>",
|
| 275 |
+
"Q:",
|
| 276 |
+
"\n\n"
|
| 277 |
+
],
|
| 278 |
+
"output_regex": null,
|
| 279 |
+
"num_samples": null,
|
| 280 |
+
"frozen": false,
|
| 281 |
+
"suite": [
|
| 282 |
+
"lighteval"
|
| 283 |
+
],
|
| 284 |
+
"original_num_docs": 500,
|
| 285 |
+
"effective_num_docs": 500,
|
| 286 |
+
"trust_dataset": true,
|
| 287 |
+
"must_remove_duplicate_docs": null,
|
| 288 |
+
"version": 0
|
| 289 |
+
},
|
| 290 |
+
"lighteval|bigbench:logical_deduction_seven_objects": {
|
| 291 |
+
"name": "bigbench:logical_deduction_seven_objects",
|
| 292 |
+
"prompt_function": "bbh_lighteval",
|
| 293 |
+
"hf_repo": "lighteval/bbh",
|
| 294 |
+
"hf_subset": "logical_deduction_seven_objects",
|
| 295 |
+
"metric": [
|
| 296 |
+
"loglikelihood_acc_single_token"
|
| 297 |
+
],
|
| 298 |
+
"hf_avail_splits": [
|
| 299 |
+
"train"
|
| 300 |
+
],
|
| 301 |
+
"evaluation_splits": [
|
| 302 |
+
"train"
|
| 303 |
+
],
|
| 304 |
+
"few_shots_split": null,
|
| 305 |
+
"few_shots_select": null,
|
| 306 |
+
"generation_size": -1,
|
| 307 |
+
"stop_sequence": [
|
| 308 |
+
"</s>",
|
| 309 |
+
"Q:",
|
| 310 |
+
"\n\n"
|
| 311 |
+
],
|
| 312 |
+
"output_regex": null,
|
| 313 |
+
"num_samples": null,
|
| 314 |
+
"frozen": false,
|
| 315 |
+
"suite": [
|
| 316 |
+
"lighteval"
|
| 317 |
+
],
|
| 318 |
+
"original_num_docs": 700,
|
| 319 |
+
"effective_num_docs": 700,
|
| 320 |
+
"trust_dataset": true,
|
| 321 |
+
"must_remove_duplicate_docs": null,
|
| 322 |
+
"version": 0
|
| 323 |
+
},
|
| 324 |
+
"lighteval|bigbench:logical_deduction_three_objects": {
|
| 325 |
+
"name": "bigbench:logical_deduction_three_objects",
|
| 326 |
+
"prompt_function": "bbh_lighteval",
|
| 327 |
+
"hf_repo": "lighteval/bbh",
|
| 328 |
+
"hf_subset": "logical_deduction_three_objects",
|
| 329 |
+
"metric": [
|
| 330 |
+
"loglikelihood_acc_single_token"
|
| 331 |
+
],
|
| 332 |
+
"hf_avail_splits": [
|
| 333 |
+
"train"
|
| 334 |
+
],
|
| 335 |
+
"evaluation_splits": [
|
| 336 |
+
"train"
|
| 337 |
+
],
|
| 338 |
+
"few_shots_split": null,
|
| 339 |
+
"few_shots_select": null,
|
| 340 |
+
"generation_size": -1,
|
| 341 |
+
"stop_sequence": [
|
| 342 |
+
"</s>",
|
| 343 |
+
"Q:",
|
| 344 |
+
"\n\n"
|
| 345 |
+
],
|
| 346 |
+
"output_regex": null,
|
| 347 |
+
"num_samples": null,
|
| 348 |
+
"frozen": false,
|
| 349 |
+
"suite": [
|
| 350 |
+
"lighteval"
|
| 351 |
+
],
|
| 352 |
+
"original_num_docs": 300,
|
| 353 |
+
"effective_num_docs": 300,
|
| 354 |
+
"trust_dataset": true,
|
| 355 |
+
"must_remove_duplicate_docs": null,
|
| 356 |
+
"version": 0
|
| 357 |
+
},
|
| 358 |
+
"lighteval|bigbench:movie_recommendation": {
|
| 359 |
+
"name": "bigbench:movie_recommendation",
|
| 360 |
+
"prompt_function": "bbh_lighteval",
|
| 361 |
+
"hf_repo": "lighteval/bbh",
|
| 362 |
+
"hf_subset": "movie_recommendation",
|
| 363 |
+
"metric": [
|
| 364 |
+
"loglikelihood_acc_single_token"
|
| 365 |
+
],
|
| 366 |
+
"hf_avail_splits": [
|
| 367 |
+
"train"
|
| 368 |
+
],
|
| 369 |
+
"evaluation_splits": [
|
| 370 |
+
"train"
|
| 371 |
+
],
|
| 372 |
+
"few_shots_split": null,
|
| 373 |
+
"few_shots_select": null,
|
| 374 |
+
"generation_size": -1,
|
| 375 |
+
"stop_sequence": [
|
| 376 |
+
"</s>",
|
| 377 |
+
"Q:",
|
| 378 |
+
"\n\n"
|
| 379 |
+
],
|
| 380 |
+
"output_regex": null,
|
| 381 |
+
"num_samples": null,
|
| 382 |
+
"frozen": false,
|
| 383 |
+
"suite": [
|
| 384 |
+
"lighteval"
|
| 385 |
+
],
|
| 386 |
+
"original_num_docs": 500,
|
| 387 |
+
"effective_num_docs": 500,
|
| 388 |
+
"trust_dataset": true,
|
| 389 |
+
"must_remove_duplicate_docs": null,
|
| 390 |
+
"version": 0
|
| 391 |
+
},
|
| 392 |
+
"lighteval|bigbench:navigate": {
|
| 393 |
+
"name": "bigbench:navigate",
|
| 394 |
+
"prompt_function": "bbh_lighteval",
|
| 395 |
+
"hf_repo": "lighteval/bbh",
|
| 396 |
+
"hf_subset": "navigate",
|
| 397 |
+
"metric": [
|
| 398 |
+
"loglikelihood_acc_single_token"
|
| 399 |
+
],
|
| 400 |
+
"hf_avail_splits": [
|
| 401 |
+
"train"
|
| 402 |
+
],
|
| 403 |
+
"evaluation_splits": [
|
| 404 |
+
"train"
|
| 405 |
+
],
|
| 406 |
+
"few_shots_split": null,
|
| 407 |
+
"few_shots_select": null,
|
| 408 |
+
"generation_size": -1,
|
| 409 |
+
"stop_sequence": [
|
| 410 |
+
"</s>",
|
| 411 |
+
"Q:",
|
| 412 |
+
"\n\n"
|
| 413 |
+
],
|
| 414 |
+
"output_regex": null,
|
| 415 |
+
"num_samples": null,
|
| 416 |
+
"frozen": false,
|
| 417 |
+
"suite": [
|
| 418 |
+
"lighteval"
|
| 419 |
+
],
|
| 420 |
+
"original_num_docs": 1000,
|
| 421 |
+
"effective_num_docs": 1000,
|
| 422 |
+
"trust_dataset": true,
|
| 423 |
+
"must_remove_duplicate_docs": null,
|
| 424 |
+
"version": 0
|
| 425 |
+
},
|
| 426 |
+
"lighteval|bigbench:reasoning_about_colored_objects": {
|
| 427 |
+
"name": "bigbench:reasoning_about_colored_objects",
|
| 428 |
+
"prompt_function": "bbh_lighteval",
|
| 429 |
+
"hf_repo": "lighteval/bbh",
|
| 430 |
+
"hf_subset": "reasoning_about_colored_objects",
|
| 431 |
+
"metric": [
|
| 432 |
+
"loglikelihood_acc_single_token"
|
| 433 |
+
],
|
| 434 |
+
"hf_avail_splits": [
|
| 435 |
+
"train"
|
| 436 |
+
],
|
| 437 |
+
"evaluation_splits": [
|
| 438 |
+
"train"
|
| 439 |
+
],
|
| 440 |
+
"few_shots_split": null,
|
| 441 |
+
"few_shots_select": null,
|
| 442 |
+
"generation_size": -1,
|
| 443 |
+
"stop_sequence": [
|
| 444 |
+
"</s>",
|
| 445 |
+
"Q:",
|
| 446 |
+
"\n\n"
|
| 447 |
+
],
|
| 448 |
+
"output_regex": null,
|
| 449 |
+
"num_samples": null,
|
| 450 |
+
"frozen": false,
|
| 451 |
+
"suite": [
|
| 452 |
+
"lighteval"
|
| 453 |
+
],
|
| 454 |
+
"original_num_docs": 2000,
|
| 455 |
+
"effective_num_docs": 2000,
|
| 456 |
+
"trust_dataset": true,
|
| 457 |
+
"must_remove_duplicate_docs": null,
|
| 458 |
+
"version": 0
|
| 459 |
+
},
|
| 460 |
+
"lighteval|bigbench:ruin_names": {
|
| 461 |
+
"name": "bigbench:ruin_names",
|
| 462 |
+
"prompt_function": "bbh_lighteval",
|
| 463 |
+
"hf_repo": "lighteval/bbh",
|
| 464 |
+
"hf_subset": "ruin_names",
|
| 465 |
+
"metric": [
|
| 466 |
+
"loglikelihood_acc_single_token"
|
| 467 |
+
],
|
| 468 |
+
"hf_avail_splits": [
|
| 469 |
+
"train"
|
| 470 |
+
],
|
| 471 |
+
"evaluation_splits": [
|
| 472 |
+
"train"
|
| 473 |
+
],
|
| 474 |
+
"few_shots_split": null,
|
| 475 |
+
"few_shots_select": null,
|
| 476 |
+
"generation_size": -1,
|
| 477 |
+
"stop_sequence": [
|
| 478 |
+
"</s>",
|
| 479 |
+
"Q:",
|
| 480 |
+
"\n\n"
|
| 481 |
+
],
|
| 482 |
+
"output_regex": null,
|
| 483 |
+
"num_samples": null,
|
| 484 |
+
"frozen": false,
|
| 485 |
+
"suite": [
|
| 486 |
+
"lighteval"
|
| 487 |
+
],
|
| 488 |
+
"original_num_docs": 448,
|
| 489 |
+
"effective_num_docs": 448,
|
| 490 |
+
"trust_dataset": true,
|
| 491 |
+
"must_remove_duplicate_docs": null,
|
| 492 |
+
"version": 0
|
| 493 |
+
},
|
| 494 |
+
"lighteval|bigbench:salient_translation_error_detection": {
|
| 495 |
+
"name": "bigbench:salient_translation_error_detection",
|
| 496 |
+
"prompt_function": "bbh_lighteval",
|
| 497 |
+
"hf_repo": "lighteval/bbh",
|
| 498 |
+
"hf_subset": "salient_translation_error_detection",
|
| 499 |
+
"metric": [
|
| 500 |
+
"loglikelihood_acc_single_token"
|
| 501 |
+
],
|
| 502 |
+
"hf_avail_splits": [
|
| 503 |
+
"train"
|
| 504 |
+
],
|
| 505 |
+
"evaluation_splits": [
|
| 506 |
+
"train"
|
| 507 |
+
],
|
| 508 |
+
"few_shots_split": null,
|
| 509 |
+
"few_shots_select": null,
|
| 510 |
+
"generation_size": -1,
|
| 511 |
+
"stop_sequence": [
|
| 512 |
+
"</s>",
|
| 513 |
+
"Q:",
|
| 514 |
+
"\n\n"
|
| 515 |
+
],
|
| 516 |
+
"output_regex": null,
|
| 517 |
+
"num_samples": null,
|
| 518 |
+
"frozen": false,
|
| 519 |
+
"suite": [
|
| 520 |
+
"lighteval"
|
| 521 |
+
],
|
| 522 |
+
"original_num_docs": 998,
|
| 523 |
+
"effective_num_docs": 998,
|
| 524 |
+
"trust_dataset": true,
|
| 525 |
+
"must_remove_duplicate_docs": null,
|
| 526 |
+
"version": 0
|
| 527 |
+
},
|
| 528 |
+
"lighteval|bigbench:snarks": {
|
| 529 |
+
"name": "bigbench:snarks",
|
| 530 |
+
"prompt_function": "bbh_lighteval",
|
| 531 |
+
"hf_repo": "lighteval/bbh",
|
| 532 |
+
"hf_subset": "snarks",
|
| 533 |
+
"metric": [
|
| 534 |
+
"loglikelihood_acc_single_token"
|
| 535 |
+
],
|
| 536 |
+
"hf_avail_splits": [
|
| 537 |
+
"train"
|
| 538 |
+
],
|
| 539 |
+
"evaluation_splits": [
|
| 540 |
+
"train"
|
| 541 |
+
],
|
| 542 |
+
"few_shots_split": null,
|
| 543 |
+
"few_shots_select": null,
|
| 544 |
+
"generation_size": -1,
|
| 545 |
+
"stop_sequence": [
|
| 546 |
+
"</s>",
|
| 547 |
+
"Q:",
|
| 548 |
+
"\n\n"
|
| 549 |
+
],
|
| 550 |
+
"output_regex": null,
|
| 551 |
+
"num_samples": null,
|
| 552 |
+
"frozen": false,
|
| 553 |
+
"suite": [
|
| 554 |
+
"lighteval"
|
| 555 |
+
],
|
| 556 |
+
"original_num_docs": 181,
|
| 557 |
+
"effective_num_docs": 181,
|
| 558 |
+
"trust_dataset": true,
|
| 559 |
+
"must_remove_duplicate_docs": null,
|
| 560 |
+
"version": 0
|
| 561 |
+
},
|
| 562 |
+
"lighteval|bigbench:sports_understanding": {
|
| 563 |
+
"name": "bigbench:sports_understanding",
|
| 564 |
+
"prompt_function": "bbh_lighteval",
|
| 565 |
+
"hf_repo": "lighteval/bbh",
|
| 566 |
+
"hf_subset": "sports_understanding",
|
| 567 |
+
"metric": [
|
| 568 |
+
"loglikelihood_acc_single_token"
|
| 569 |
+
],
|
| 570 |
+
"hf_avail_splits": [
|
| 571 |
+
"train"
|
| 572 |
+
],
|
| 573 |
+
"evaluation_splits": [
|
| 574 |
+
"train"
|
| 575 |
+
],
|
| 576 |
+
"few_shots_split": null,
|
| 577 |
+
"few_shots_select": null,
|
| 578 |
+
"generation_size": -1,
|
| 579 |
+
"stop_sequence": [
|
| 580 |
+
"</s>",
|
| 581 |
+
"Q:",
|
| 582 |
+
"\n\n"
|
| 583 |
+
],
|
| 584 |
+
"output_regex": null,
|
| 585 |
+
"num_samples": null,
|
| 586 |
+
"frozen": false,
|
| 587 |
+
"suite": [
|
| 588 |
+
"lighteval"
|
| 589 |
+
],
|
| 590 |
+
"original_num_docs": 1000,
|
| 591 |
+
"effective_num_docs": 1000,
|
| 592 |
+
"trust_dataset": true,
|
| 593 |
+
"must_remove_duplicate_docs": null,
|
| 594 |
+
"version": 0
|
| 595 |
+
},
|
| 596 |
+
"lighteval|bigbench:temporal_sequences": {
|
| 597 |
+
"name": "bigbench:temporal_sequences",
|
| 598 |
+
"prompt_function": "bbh_lighteval",
|
| 599 |
+
"hf_repo": "lighteval/bbh",
|
| 600 |
+
"hf_subset": "temporal_sequences",
|
| 601 |
+
"metric": [
|
| 602 |
+
"loglikelihood_acc_single_token"
|
| 603 |
+
],
|
| 604 |
+
"hf_avail_splits": [
|
| 605 |
+
"train"
|
| 606 |
+
],
|
| 607 |
+
"evaluation_splits": [
|
| 608 |
+
"train"
|
| 609 |
+
],
|
| 610 |
+
"few_shots_split": null,
|
| 611 |
+
"few_shots_select": null,
|
| 612 |
+
"generation_size": -1,
|
| 613 |
+
"stop_sequence": [
|
| 614 |
+
"</s>",
|
| 615 |
+
"Q:",
|
| 616 |
+
"\n\n"
|
| 617 |
+
],
|
| 618 |
+
"output_regex": null,
|
| 619 |
+
"num_samples": null,
|
| 620 |
+
"frozen": false,
|
| 621 |
+
"suite": [
|
| 622 |
+
"lighteval"
|
| 623 |
+
],
|
| 624 |
+
"original_num_docs": 1000,
|
| 625 |
+
"effective_num_docs": 1000,
|
| 626 |
+
"trust_dataset": true,
|
| 627 |
+
"must_remove_duplicate_docs": null,
|
| 628 |
+
"version": 0
|
| 629 |
+
},
|
| 630 |
+
"lighteval|bigbench:tracking_shuffled_objects_five_objects": {
|
| 631 |
+
"name": "bigbench:tracking_shuffled_objects_five_objects",
|
| 632 |
+
"prompt_function": "bbh_lighteval",
|
| 633 |
+
"hf_repo": "lighteval/bbh",
|
| 634 |
+
"hf_subset": "tracking_shuffled_objects_five_objects",
|
| 635 |
+
"metric": [
|
| 636 |
+
"loglikelihood_acc_single_token"
|
| 637 |
+
],
|
| 638 |
+
"hf_avail_splits": [
|
| 639 |
+
"train"
|
| 640 |
+
],
|
| 641 |
+
"evaluation_splits": [
|
| 642 |
+
"train"
|
| 643 |
+
],
|
| 644 |
+
"few_shots_split": null,
|
| 645 |
+
"few_shots_select": null,
|
| 646 |
+
"generation_size": -1,
|
| 647 |
+
"stop_sequence": [
|
| 648 |
+
"</s>",
|
| 649 |
+
"Q:",
|
| 650 |
+
"\n\n"
|
| 651 |
+
],
|
| 652 |
+
"output_regex": null,
|
| 653 |
+
"num_samples": null,
|
| 654 |
+
"frozen": false,
|
| 655 |
+
"suite": [
|
| 656 |
+
"lighteval"
|
| 657 |
+
],
|
| 658 |
+
"original_num_docs": 1250,
|
| 659 |
+
"effective_num_docs": 1250,
|
| 660 |
+
"trust_dataset": true,
|
| 661 |
+
"must_remove_duplicate_docs": null,
|
| 662 |
+
"version": 0
|
| 663 |
+
},
|
| 664 |
+
"lighteval|bigbench:tracking_shuffled_objects_seven_objects": {
|
| 665 |
+
"name": "bigbench:tracking_shuffled_objects_seven_objects",
|
| 666 |
+
"prompt_function": "bbh_lighteval",
|
| 667 |
+
"hf_repo": "lighteval/bbh",
|
| 668 |
+
"hf_subset": "tracking_shuffled_objects_seven_objects",
|
| 669 |
+
"metric": [
|
| 670 |
+
"loglikelihood_acc_single_token"
|
| 671 |
+
],
|
| 672 |
+
"hf_avail_splits": [
|
| 673 |
+
"train"
|
| 674 |
+
],
|
| 675 |
+
"evaluation_splits": [
|
| 676 |
+
"train"
|
| 677 |
+
],
|
| 678 |
+
"few_shots_split": null,
|
| 679 |
+
"few_shots_select": null,
|
| 680 |
+
"generation_size": -1,
|
| 681 |
+
"stop_sequence": [
|
| 682 |
+
"</s>",
|
| 683 |
+
"Q:",
|
| 684 |
+
"\n\n"
|
| 685 |
+
],
|
| 686 |
+
"output_regex": null,
|
| 687 |
+
"num_samples": null,
|
| 688 |
+
"frozen": false,
|
| 689 |
+
"suite": [
|
| 690 |
+
"lighteval"
|
| 691 |
+
],
|
| 692 |
+
"original_num_docs": 1750,
|
| 693 |
+
"effective_num_docs": 1750,
|
| 694 |
+
"trust_dataset": true,
|
| 695 |
+
"must_remove_duplicate_docs": null,
|
| 696 |
+
"version": 0
|
| 697 |
+
},
|
| 698 |
+
"lighteval|bigbench:tracking_shuffled_objects_three_objects": {
|
| 699 |
+
"name": "bigbench:tracking_shuffled_objects_three_objects",
|
| 700 |
+
"prompt_function": "bbh_lighteval",
|
| 701 |
+
"hf_repo": "lighteval/bbh",
|
| 702 |
+
"hf_subset": "tracking_shuffled_objects_three_objects",
|
| 703 |
+
"metric": [
|
| 704 |
+
"loglikelihood_acc_single_token"
|
| 705 |
+
],
|
| 706 |
+
"hf_avail_splits": [
|
| 707 |
+
"train"
|
| 708 |
+
],
|
| 709 |
+
"evaluation_splits": [
|
| 710 |
+
"train"
|
| 711 |
+
],
|
| 712 |
+
"few_shots_split": null,
|
| 713 |
+
"few_shots_select": null,
|
| 714 |
+
"generation_size": -1,
|
| 715 |
+
"stop_sequence": [
|
| 716 |
+
"</s>",
|
| 717 |
+
"Q:",
|
| 718 |
+
"\n\n"
|
| 719 |
+
],
|
| 720 |
+
"output_regex": null,
|
| 721 |
+
"num_samples": null,
|
| 722 |
+
"frozen": false,
|
| 723 |
+
"suite": [
|
| 724 |
+
"lighteval"
|
| 725 |
+
],
|
| 726 |
+
"original_num_docs": 300,
|
| 727 |
+
"effective_num_docs": 300,
|
| 728 |
+
"trust_dataset": true,
|
| 729 |
+
"must_remove_duplicate_docs": null,
|
| 730 |
+
"version": 0
|
| 731 |
+
}
|
| 732 |
+
},
|
| 733 |
+
"summary_tasks": {
|
| 734 |
+
"lighteval|bigbench:causal_judgment|0": {
|
| 735 |
+
"hashes": {
|
| 736 |
+
"hash_examples": "dfb1ae47218f2850",
|
| 737 |
+
"hash_full_prompts": "92b1cf75ca896127",
|
| 738 |
+
"hash_input_tokens": "3fbebfde3354c6ac",
|
| 739 |
+
"hash_cont_tokens": "5f93270a3abb23cb"
|
| 740 |
+
},
|
| 741 |
+
"truncated": 0,
|
| 742 |
+
"non_truncated": 190,
|
| 743 |
+
"padded": 189,
|
| 744 |
+
"non_padded": 1,
|
| 745 |
+
"effective_few_shots": 0.0,
|
| 746 |
+
"num_truncated_few_shots": 0
|
| 747 |
+
},
|
| 748 |
+
"lighteval|bigbench:date_understanding|0": {
|
| 749 |
+
"hashes": {
|
| 750 |
+
"hash_examples": "2b823c41500a6ec2",
|
| 751 |
+
"hash_full_prompts": "a086589baadb24a5",
|
| 752 |
+
"hash_input_tokens": "262880ca53810c51",
|
| 753 |
+
"hash_cont_tokens": "a0df85f715ef8665"
|
| 754 |
+
},
|
| 755 |
+
"truncated": 0,
|
| 756 |
+
"non_truncated": 369,
|
| 757 |
+
"padded": 369,
|
| 758 |
+
"non_padded": 0,
|
| 759 |
+
"effective_few_shots": 0.0,
|
| 760 |
+
"num_truncated_few_shots": 0
|
| 761 |
+
},
|
| 762 |
+
"lighteval|bigbench:disambiguation_qa|0": {
|
| 763 |
+
"hashes": {
|
| 764 |
+
"hash_examples": "2a4c3d41db198cea",
|
| 765 |
+
"hash_full_prompts": "407f0b9a565699a7",
|
| 766 |
+
"hash_input_tokens": "7159e757a625c662",
|
| 767 |
+
"hash_cont_tokens": "ea8c6c727a0e9657"
|
| 768 |
+
},
|
| 769 |
+
"truncated": 0,
|
| 770 |
+
"non_truncated": 258,
|
| 771 |
+
"padded": 258,
|
| 772 |
+
"non_padded": 0,
|
| 773 |
+
"effective_few_shots": 0.0,
|
| 774 |
+
"num_truncated_few_shots": 0
|
| 775 |
+
},
|
| 776 |
+
"lighteval|bigbench:geometric_shapes|0": {
|
| 777 |
+
"hashes": {
|
| 778 |
+
"hash_examples": "24aa261103911b72",
|
| 779 |
+
"hash_full_prompts": "c7a3189ee1642ab5",
|
| 780 |
+
"hash_input_tokens": "2bbd71a22a0967ff",
|
| 781 |
+
"hash_cont_tokens": "cfff7ac805aca5af"
|
| 782 |
+
},
|
| 783 |
+
"truncated": 0,
|
| 784 |
+
"non_truncated": 360,
|
| 785 |
+
"padded": 360,
|
| 786 |
+
"non_padded": 0,
|
| 787 |
+
"effective_few_shots": 0.0,
|
| 788 |
+
"num_truncated_few_shots": 0
|
| 789 |
+
},
|
| 790 |
+
"lighteval|bigbench:logical_deduction_five_objects|0": {
|
| 791 |
+
"hashes": {
|
| 792 |
+
"hash_examples": "cb5bdc92afc41f83",
|
| 793 |
+
"hash_full_prompts": "199211ceee2c8d60",
|
| 794 |
+
"hash_input_tokens": "0c67894c26b553e1",
|
| 795 |
+
"hash_cont_tokens": "1f4302f16658f321"
|
| 796 |
+
},
|
| 797 |
+
"truncated": 0,
|
| 798 |
+
"non_truncated": 500,
|
| 799 |
+
"padded": 500,
|
| 800 |
+
"non_padded": 0,
|
| 801 |
+
"effective_few_shots": 0.0,
|
| 802 |
+
"num_truncated_few_shots": 0
|
| 803 |
+
},
|
| 804 |
+
"lighteval|bigbench:logical_deduction_seven_objects|0": {
|
| 805 |
+
"hashes": {
|
| 806 |
+
"hash_examples": "b6805ea696739f9f",
|
| 807 |
+
"hash_full_prompts": "38f66304971bdfb0",
|
| 808 |
+
"hash_input_tokens": "fe8815876ceefc9f",
|
| 809 |
+
"hash_cont_tokens": "bc16fbeb3f267778"
|
| 810 |
+
},
|
| 811 |
+
"truncated": 0,
|
| 812 |
+
"non_truncated": 700,
|
| 813 |
+
"padded": 700,
|
| 814 |
+
"non_padded": 0,
|
| 815 |
+
"effective_few_shots": 0.0,
|
| 816 |
+
"num_truncated_few_shots": 0
|
| 817 |
+
},
|
| 818 |
+
"lighteval|bigbench:logical_deduction_three_objects|0": {
|
| 819 |
+
"hashes": {
|
| 820 |
+
"hash_examples": "0509e5712ab9bcdb",
|
| 821 |
+
"hash_full_prompts": "0a9439e283842405",
|
| 822 |
+
"hash_input_tokens": "bdaf9a4ad6d5308d",
|
| 823 |
+
"hash_cont_tokens": "17b92ec4d7f3b0a6"
|
| 824 |
+
},
|
| 825 |
+
"truncated": 0,
|
| 826 |
+
"non_truncated": 300,
|
| 827 |
+
"padded": 300,
|
| 828 |
+
"non_padded": 0,
|
| 829 |
+
"effective_few_shots": 0.0,
|
| 830 |
+
"num_truncated_few_shots": 0
|
| 831 |
+
},
|
| 832 |
+
"lighteval|bigbench:movie_recommendation|0": {
|
| 833 |
+
"hashes": {
|
| 834 |
+
"hash_examples": "530cc6f737830f45",
|
| 835 |
+
"hash_full_prompts": "e7d59f843d80e6ba",
|
| 836 |
+
"hash_input_tokens": "d107040c803cc28a",
|
| 837 |
+
"hash_cont_tokens": "1a79efad33e450d9"
|
| 838 |
+
},
|
| 839 |
+
"truncated": 0,
|
| 840 |
+
"non_truncated": 500,
|
| 841 |
+
"padded": 500,
|
| 842 |
+
"non_padded": 0,
|
| 843 |
+
"effective_few_shots": 0.0,
|
| 844 |
+
"num_truncated_few_shots": 0
|
| 845 |
+
},
|
| 846 |
+
"lighteval|bigbench:navigate|0": {
|
| 847 |
+
"hashes": {
|
| 848 |
+
"hash_examples": "7962ef85d0058b9a",
|
| 849 |
+
"hash_full_prompts": "d58b607419984968",
|
| 850 |
+
"hash_input_tokens": "f163e125b820f371",
|
| 851 |
+
"hash_cont_tokens": "4cba3a6b51df2f82"
|
| 852 |
+
},
|
| 853 |
+
"truncated": 0,
|
| 854 |
+
"non_truncated": 1000,
|
| 855 |
+
"padded": 988,
|
| 856 |
+
"non_padded": 12,
|
| 857 |
+
"effective_few_shots": 0.0,
|
| 858 |
+
"num_truncated_few_shots": 0
|
| 859 |
+
},
|
| 860 |
+
"lighteval|bigbench:reasoning_about_colored_objects|0": {
|
| 861 |
+
"hashes": {
|
| 862 |
+
"hash_examples": "39be1ab1677a651d",
|
| 863 |
+
"hash_full_prompts": "157347bb0e7fe3f1",
|
| 864 |
+
"hash_input_tokens": "562dab78f82dce11",
|
| 865 |
+
"hash_cont_tokens": "b460bf430f1c4128"
|
| 866 |
+
},
|
| 867 |
+
"truncated": 0,
|
| 868 |
+
"non_truncated": 2000,
|
| 869 |
+
"padded": 1969,
|
| 870 |
+
"non_padded": 31,
|
| 871 |
+
"effective_few_shots": 0.0,
|
| 872 |
+
"num_truncated_few_shots": 0
|
| 873 |
+
},
|
| 874 |
+
"lighteval|bigbench:ruin_names|0": {
|
| 875 |
+
"hashes": {
|
| 876 |
+
"hash_examples": "e9b96b31d2154941",
|
| 877 |
+
"hash_full_prompts": "9cb73d2fcaf5ee1e",
|
| 878 |
+
"hash_input_tokens": "07268ad2915fab08",
|
| 879 |
+
"hash_cont_tokens": "da425c7c151ed7c3"
|
| 880 |
+
},
|
| 881 |
+
"truncated": 0,
|
| 882 |
+
"non_truncated": 448,
|
| 883 |
+
"padded": 442,
|
| 884 |
+
"non_padded": 6,
|
| 885 |
+
"effective_few_shots": 0.0,
|
| 886 |
+
"num_truncated_few_shots": 0
|
| 887 |
+
},
|
| 888 |
+
"lighteval|bigbench:salient_translation_error_detection|0": {
|
| 889 |
+
"hashes": {
|
| 890 |
+
"hash_examples": "951ac59f7ad0427d",
|
| 891 |
+
"hash_full_prompts": "3598ce066bb83298",
|
| 892 |
+
"hash_input_tokens": "d514272792c1a43b",
|
| 893 |
+
"hash_cont_tokens": "52a851b402ce7859"
|
| 894 |
+
},
|
| 895 |
+
"truncated": 0,
|
| 896 |
+
"non_truncated": 998,
|
| 897 |
+
"padded": 998,
|
| 898 |
+
"non_padded": 0,
|
| 899 |
+
"effective_few_shots": 0.0,
|
| 900 |
+
"num_truncated_few_shots": 0
|
| 901 |
+
},
|
| 902 |
+
"lighteval|bigbench:snarks|0": {
|
| 903 |
+
"hashes": {
|
| 904 |
+
"hash_examples": "3a53eb9b9d758534",
|
| 905 |
+
"hash_full_prompts": "fb6c17d84dd479d6",
|
| 906 |
+
"hash_input_tokens": "a714129b2023dfb4",
|
| 907 |
+
"hash_cont_tokens": "fcf8b70a05755b30"
|
| 908 |
+
},
|
| 909 |
+
"truncated": 0,
|
| 910 |
+
"non_truncated": 181,
|
| 911 |
+
"padded": 178,
|
| 912 |
+
"non_padded": 3,
|
| 913 |
+
"effective_few_shots": 0.0,
|
| 914 |
+
"num_truncated_few_shots": 0
|
| 915 |
+
},
|
| 916 |
+
"lighteval|bigbench:sports_understanding|0": {
|
| 917 |
+
"hashes": {
|
| 918 |
+
"hash_examples": "bd65741f00770373",
|
| 919 |
+
"hash_full_prompts": "467a508a87ae3ce4",
|
| 920 |
+
"hash_input_tokens": "c1c3655837258597",
|
| 921 |
+
"hash_cont_tokens": "3a6f302fe5d32bc4"
|
| 922 |
+
},
|
| 923 |
+
"truncated": 0,
|
| 924 |
+
"non_truncated": 1000,
|
| 925 |
+
"padded": 1000,
|
| 926 |
+
"non_padded": 0,
|
| 927 |
+
"effective_few_shots": 0.0,
|
| 928 |
+
"num_truncated_few_shots": 0
|
| 929 |
+
},
|
| 930 |
+
"lighteval|bigbench:temporal_sequences|0": {
|
| 931 |
+
"hashes": {
|
| 932 |
+
"hash_examples": "1d13139f47cb2df7",
|
| 933 |
+
"hash_full_prompts": "c1b66e439c72477b",
|
| 934 |
+
"hash_input_tokens": "d47a810e1658bcc6",
|
| 935 |
+
"hash_cont_tokens": "156d739a1e6ce872"
|
| 936 |
+
},
|
| 937 |
+
"truncated": 0,
|
| 938 |
+
"non_truncated": 1000,
|
| 939 |
+
"padded": 1000,
|
| 940 |
+
"non_padded": 0,
|
| 941 |
+
"effective_few_shots": 0.0,
|
| 942 |
+
"num_truncated_few_shots": 0
|
| 943 |
+
},
|
| 944 |
+
"lighteval|bigbench:tracking_shuffled_objects_five_objects|0": {
|
| 945 |
+
"hashes": {
|
| 946 |
+
"hash_examples": "8770a702a9646648",
|
| 947 |
+
"hash_full_prompts": "8c353eaa84f712ff",
|
| 948 |
+
"hash_input_tokens": "e940ffaad712c055",
|
| 949 |
+
"hash_cont_tokens": "bb8b1af3f905c2e3"
|
| 950 |
+
},
|
| 951 |
+
"truncated": 0,
|
| 952 |
+
"non_truncated": 1250,
|
| 953 |
+
"padded": 1180,
|
| 954 |
+
"non_padded": 70,
|
| 955 |
+
"effective_few_shots": 0.0,
|
| 956 |
+
"num_truncated_few_shots": 0
|
| 957 |
+
},
|
| 958 |
+
"lighteval|bigbench:tracking_shuffled_objects_seven_objects|0": {
|
| 959 |
+
"hashes": {
|
| 960 |
+
"hash_examples": "b469b7d073824a59",
|
| 961 |
+
"hash_full_prompts": "902f5b74467353eb",
|
| 962 |
+
"hash_input_tokens": "9d52e492d2f6cd34",
|
| 963 |
+
"hash_cont_tokens": "0453f99f28d935a3"
|
| 964 |
+
},
|
| 965 |
+
"truncated": 0,
|
| 966 |
+
"non_truncated": 1750,
|
| 967 |
+
"padded": 1701,
|
| 968 |
+
"non_padded": 49,
|
| 969 |
+
"effective_few_shots": 0.0,
|
| 970 |
+
"num_truncated_few_shots": 0
|
| 971 |
+
},
|
| 972 |
+
"lighteval|bigbench:tracking_shuffled_objects_three_objects|0": {
|
| 973 |
+
"hashes": {
|
| 974 |
+
"hash_examples": "0509e5712ab9bcdb",
|
| 975 |
+
"hash_full_prompts": "0a9439e283842405",
|
| 976 |
+
"hash_input_tokens": "5867c63542b2b703",
|
| 977 |
+
"hash_cont_tokens": "8062227d4df3c18a"
|
| 978 |
+
},
|
| 979 |
+
"truncated": 0,
|
| 980 |
+
"non_truncated": 300,
|
| 981 |
+
"padded": 294,
|
| 982 |
+
"non_padded": 6,
|
| 983 |
+
"effective_few_shots": 0.0,
|
| 984 |
+
"num_truncated_few_shots": 0
|
| 985 |
+
}
|
| 986 |
+
},
|
| 987 |
+
"summary_general": {
|
| 988 |
+
"hashes": {
|
| 989 |
+
"hash_examples": "51a30c4501ba4586",
|
| 990 |
+
"hash_full_prompts": "2f8425bef1f1b307",
|
| 991 |
+
"hash_input_tokens": "b84ccc3699968295",
|
| 992 |
+
"hash_cont_tokens": "c93c9cc4e6d49fcb"
|
| 993 |
+
},
|
| 994 |
+
"truncated": 0,
|
| 995 |
+
"non_truncated": 13104,
|
| 996 |
+
"padded": 12926,
|
| 997 |
+
"non_padded": 178,
|
| 998 |
+
"num_truncated_few_shots": 0
|
| 999 |
+
}
|
| 1000 |
+
}
|