Upload eval_results/HuggingFaceH4/mistral-7b-kto/v0.0/bbh/results_2024-03-22T16-15-08.088486.json with huggingface_hub
Browse files
eval_results/HuggingFaceH4/mistral-7b-kto/v0.0/bbh/results_2024-03-22T16-15-08.088486.json
ADDED
|
@@ -0,0 +1,1184 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 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": 2723.723505233,
|
| 9 |
+
"end_time": 2949.089739925,
|
| 10 |
+
"total_evaluation_time_secondes": "225.36623469199958",
|
| 11 |
+
"model_name": "HuggingFaceH4/mistral-7b-kto",
|
| 12 |
+
"model_sha": "23199c67ea7e0f8a1a635b7f918c25425289aeeb",
|
| 13 |
+
"model_dtype": "torch.bfloat16",
|
| 14 |
+
"model_size": "13.99 GB",
|
| 15 |
+
"config": null
|
| 16 |
+
},
|
| 17 |
+
"results": {
|
| 18 |
+
"harness|bbh:causal_judgment|3": {
|
| 19 |
+
"em": 0.5614973262032086,
|
| 20 |
+
"em_stderr": 0.03638341809400996,
|
| 21 |
+
"qem": 0.5614973262032086,
|
| 22 |
+
"qem_stderr": 0.03638341809400996,
|
| 23 |
+
"pem": 0.5775401069518716,
|
| 24 |
+
"pem_stderr": 0.03621824020753356,
|
| 25 |
+
"pqem": 0.5775401069518716,
|
| 26 |
+
"pqem_stderr": 0.03621824020753356,
|
| 27 |
+
"perfect_em": 0.5614973262032086,
|
| 28 |
+
"perfect_em_stderr": 0.03638341809400996
|
| 29 |
+
},
|
| 30 |
+
"harness|bbh:date_understanding|3": {
|
| 31 |
+
"em": 0.464,
|
| 32 |
+
"em_stderr": 0.031603975145223735,
|
| 33 |
+
"qem": 0.464,
|
| 34 |
+
"qem_stderr": 0.031603975145223735,
|
| 35 |
+
"pem": 0.464,
|
| 36 |
+
"pem_stderr": 0.031603975145223735,
|
| 37 |
+
"pqem": 0.552,
|
| 38 |
+
"pqem_stderr": 0.031514387611153515,
|
| 39 |
+
"perfect_em": 0.464,
|
| 40 |
+
"perfect_em_stderr": 0.031603975145223735
|
| 41 |
+
},
|
| 42 |
+
"harness|bbh:disambiguation_qa|3": {
|
| 43 |
+
"em": 0.604,
|
| 44 |
+
"em_stderr": 0.03099319785457785,
|
| 45 |
+
"qem": 0.604,
|
| 46 |
+
"qem_stderr": 0.03099319785457785,
|
| 47 |
+
"pem": 0.604,
|
| 48 |
+
"pem_stderr": 0.03099319785457785,
|
| 49 |
+
"pqem": 0.704,
|
| 50 |
+
"pqem_stderr": 0.02892893938837963,
|
| 51 |
+
"perfect_em": 0.604,
|
| 52 |
+
"perfect_em_stderr": 0.03099319785457785
|
| 53 |
+
},
|
| 54 |
+
"harness|bbh:geometric_shapes|3": {
|
| 55 |
+
"em": 0.22,
|
| 56 |
+
"em_stderr": 0.02625179282460584,
|
| 57 |
+
"qem": 0.22,
|
| 58 |
+
"qem_stderr": 0.02625179282460584,
|
| 59 |
+
"pem": 0.22,
|
| 60 |
+
"pem_stderr": 0.02625179282460584,
|
| 61 |
+
"pqem": 0.22,
|
| 62 |
+
"pqem_stderr": 0.02625179282460584,
|
| 63 |
+
"perfect_em": 0.22,
|
| 64 |
+
"perfect_em_stderr": 0.02625179282460584
|
| 65 |
+
},
|
| 66 |
+
"harness|bbh:logical_deduction_five_objects|3": {
|
| 67 |
+
"em": 0.396,
|
| 68 |
+
"em_stderr": 0.030993197854577853,
|
| 69 |
+
"qem": 0.396,
|
| 70 |
+
"qem_stderr": 0.030993197854577853,
|
| 71 |
+
"pem": 0.396,
|
| 72 |
+
"pem_stderr": 0.030993197854577853,
|
| 73 |
+
"pqem": 0.516,
|
| 74 |
+
"pqem_stderr": 0.03166998503010742,
|
| 75 |
+
"perfect_em": 0.396,
|
| 76 |
+
"perfect_em_stderr": 0.030993197854577853
|
| 77 |
+
},
|
| 78 |
+
"harness|bbh:logical_deduction_seven_objects|3": {
|
| 79 |
+
"em": 0.336,
|
| 80 |
+
"em_stderr": 0.029933259094191516,
|
| 81 |
+
"qem": 0.336,
|
| 82 |
+
"qem_stderr": 0.029933259094191516,
|
| 83 |
+
"pem": 0.336,
|
| 84 |
+
"pem_stderr": 0.029933259094191516,
|
| 85 |
+
"pqem": 0.448,
|
| 86 |
+
"pqem_stderr": 0.03151438761115355,
|
| 87 |
+
"perfect_em": 0.336,
|
| 88 |
+
"perfect_em_stderr": 0.029933259094191516
|
| 89 |
+
},
|
| 90 |
+
"harness|bbh:logical_deduction_three_objects|3": {
|
| 91 |
+
"em": 0.512,
|
| 92 |
+
"em_stderr": 0.03167708558254709,
|
| 93 |
+
"qem": 0.512,
|
| 94 |
+
"qem_stderr": 0.03167708558254709,
|
| 95 |
+
"pem": 0.512,
|
| 96 |
+
"pem_stderr": 0.03167708558254709,
|
| 97 |
+
"pqem": 0.736,
|
| 98 |
+
"pqem_stderr": 0.027934518957690908,
|
| 99 |
+
"perfect_em": 0.512,
|
| 100 |
+
"perfect_em_stderr": 0.03167708558254709
|
| 101 |
+
},
|
| 102 |
+
"harness|bbh:movie_recommendation|3": {
|
| 103 |
+
"em": 0.42168674698795183,
|
| 104 |
+
"em_stderr": 0.031358169846784814,
|
| 105 |
+
"qem": 0.42168674698795183,
|
| 106 |
+
"qem_stderr": 0.031358169846784814,
|
| 107 |
+
"pem": 0.42168674698795183,
|
| 108 |
+
"pem_stderr": 0.031358169846784814,
|
| 109 |
+
"pqem": 0.5622489959839357,
|
| 110 |
+
"pqem_stderr": 0.03150301204870893,
|
| 111 |
+
"perfect_em": 0.42168674698795183,
|
| 112 |
+
"perfect_em_stderr": 0.031358169846784814
|
| 113 |
+
},
|
| 114 |
+
"harness|bbh:navigate|3": {
|
| 115 |
+
"em": 0.624,
|
| 116 |
+
"em_stderr": 0.03069633626739458,
|
| 117 |
+
"qem": 0.624,
|
| 118 |
+
"qem_stderr": 0.03069633626739458,
|
| 119 |
+
"pem": 0.624,
|
| 120 |
+
"pem_stderr": 0.03069633626739458,
|
| 121 |
+
"pqem": 0.624,
|
| 122 |
+
"pqem_stderr": 0.03069633626739458,
|
| 123 |
+
"perfect_em": 0.624,
|
| 124 |
+
"perfect_em_stderr": 0.03069633626739458
|
| 125 |
+
},
|
| 126 |
+
"harness|bbh:reasoning_about_colored_objects|3": {
|
| 127 |
+
"em": 0.352,
|
| 128 |
+
"em_stderr": 0.03026628805735994,
|
| 129 |
+
"qem": 0.352,
|
| 130 |
+
"qem_stderr": 0.03026628805735994,
|
| 131 |
+
"pem": 0.368,
|
| 132 |
+
"pem_stderr": 0.030562070620993167,
|
| 133 |
+
"pqem": 0.496,
|
| 134 |
+
"pqem_stderr": 0.031685198551199154,
|
| 135 |
+
"perfect_em": 0.352,
|
| 136 |
+
"perfect_em_stderr": 0.03026628805735994
|
| 137 |
+
},
|
| 138 |
+
"harness|bbh:ruin_names|3": {
|
| 139 |
+
"em": 0.375,
|
| 140 |
+
"em_stderr": 0.03080400363063401,
|
| 141 |
+
"qem": 0.375,
|
| 142 |
+
"qem_stderr": 0.03080400363063401,
|
| 143 |
+
"pem": 0.375,
|
| 144 |
+
"pem_stderr": 0.03080400363063401,
|
| 145 |
+
"pqem": 0.5645161290322581,
|
| 146 |
+
"pqem_stderr": 0.031548283738756754,
|
| 147 |
+
"perfect_em": 0.375,
|
| 148 |
+
"perfect_em_stderr": 0.03080400363063401
|
| 149 |
+
},
|
| 150 |
+
"harness|bbh:salient_translation_error_detection|3": {
|
| 151 |
+
"em": 0.384,
|
| 152 |
+
"em_stderr": 0.03082167911737538,
|
| 153 |
+
"qem": 0.384,
|
| 154 |
+
"qem_stderr": 0.03082167911737538,
|
| 155 |
+
"pem": 0.384,
|
| 156 |
+
"pem_stderr": 0.03082167911737538,
|
| 157 |
+
"pqem": 0.516,
|
| 158 |
+
"pqem_stderr": 0.03166998503010742,
|
| 159 |
+
"perfect_em": 0.384,
|
| 160 |
+
"perfect_em_stderr": 0.03082167911737538
|
| 161 |
+
},
|
| 162 |
+
"harness|bbh:snarks|3": {
|
| 163 |
+
"em": 0.7303370786516854,
|
| 164 |
+
"em_stderr": 0.03335689818443928,
|
| 165 |
+
"qem": 0.7303370786516854,
|
| 166 |
+
"qem_stderr": 0.03335689818443928,
|
| 167 |
+
"pem": 0.7359550561797753,
|
| 168 |
+
"pem_stderr": 0.0331343107765884,
|
| 169 |
+
"pqem": 0.7752808988764045,
|
| 170 |
+
"pqem_stderr": 0.031373495121251,
|
| 171 |
+
"perfect_em": 0.7303370786516854,
|
| 172 |
+
"perfect_em_stderr": 0.03335689818443928
|
| 173 |
+
},
|
| 174 |
+
"harness|bbh:sports_understanding|3": {
|
| 175 |
+
"em": 0.536,
|
| 176 |
+
"em_stderr": 0.03160397514522374,
|
| 177 |
+
"qem": 0.536,
|
| 178 |
+
"qem_stderr": 0.03160397514522374,
|
| 179 |
+
"pem": 0.772,
|
| 180 |
+
"pem_stderr": 0.026587432487268473,
|
| 181 |
+
"pqem": 0.772,
|
| 182 |
+
"pqem_stderr": 0.026587432487268473,
|
| 183 |
+
"perfect_em": 0.536,
|
| 184 |
+
"perfect_em_stderr": 0.03160397514522374
|
| 185 |
+
},
|
| 186 |
+
"harness|bbh:temporal_sequences|3": {
|
| 187 |
+
"em": 0.264,
|
| 188 |
+
"em_stderr": 0.027934518957690908,
|
| 189 |
+
"qem": 0.264,
|
| 190 |
+
"qem_stderr": 0.027934518957690908,
|
| 191 |
+
"pem": 0.264,
|
| 192 |
+
"pem_stderr": 0.027934518957690908,
|
| 193 |
+
"pqem": 0.496,
|
| 194 |
+
"pqem_stderr": 0.03168519855119917,
|
| 195 |
+
"perfect_em": 0.264,
|
| 196 |
+
"perfect_em_stderr": 0.027934518957690908
|
| 197 |
+
},
|
| 198 |
+
"harness|bbh:tracking_shuffled_objects_five_objects|3": {
|
| 199 |
+
"em": 0.152,
|
| 200 |
+
"em_stderr": 0.022752024491765464,
|
| 201 |
+
"qem": 0.152,
|
| 202 |
+
"qem_stderr": 0.022752024491765464,
|
| 203 |
+
"pem": 0.152,
|
| 204 |
+
"pem_stderr": 0.022752024491765464,
|
| 205 |
+
"pqem": 0.352,
|
| 206 |
+
"pqem_stderr": 0.030266288057359945,
|
| 207 |
+
"perfect_em": 0.152,
|
| 208 |
+
"perfect_em_stderr": 0.022752024491765464
|
| 209 |
+
},
|
| 210 |
+
"harness|bbh:tracking_shuffled_objects_seven_objects|3": {
|
| 211 |
+
"em": 0.144,
|
| 212 |
+
"em_stderr": 0.022249407735450213,
|
| 213 |
+
"qem": 0.144,
|
| 214 |
+
"qem_stderr": 0.022249407735450213,
|
| 215 |
+
"pem": 0.144,
|
| 216 |
+
"pem_stderr": 0.022249407735450213,
|
| 217 |
+
"pqem": 0.288,
|
| 218 |
+
"pqem_stderr": 0.028697004587398215,
|
| 219 |
+
"perfect_em": 0.144,
|
| 220 |
+
"perfect_em_stderr": 0.022249407735450213
|
| 221 |
+
},
|
| 222 |
+
"harness|bbh:tracking_shuffled_objects_three_objects|3": {
|
| 223 |
+
"em": 0.336,
|
| 224 |
+
"em_stderr": 0.029933259094191512,
|
| 225 |
+
"qem": 0.336,
|
| 226 |
+
"qem_stderr": 0.029933259094191512,
|
| 227 |
+
"pem": 0.336,
|
| 228 |
+
"pem_stderr": 0.029933259094191512,
|
| 229 |
+
"pqem": 0.648,
|
| 230 |
+
"pqem_stderr": 0.030266288057359928,
|
| 231 |
+
"perfect_em": 0.336,
|
| 232 |
+
"perfect_em_stderr": 0.029933259094191512
|
| 233 |
+
},
|
| 234 |
+
"harness|bbh:_average|3": {
|
| 235 |
+
"em": 0.411806730657936,
|
| 236 |
+
"em_stderr": 0.029978471498780204,
|
| 237 |
+
"qem": 0.411806730657936,
|
| 238 |
+
"qem_stderr": 0.029978471498780204,
|
| 239 |
+
"pem": 0.4270101061177556,
|
| 240 |
+
"pem_stderr": 0.029694664532744133,
|
| 241 |
+
"pqem": 0.5470881183802484,
|
| 242 |
+
"pqem_stderr": 0.030556154118257114,
|
| 243 |
+
"perfect_em": 0.411806730657936,
|
| 244 |
+
"perfect_em_stderr": 0.029978471498780204
|
| 245 |
+
}
|
| 246 |
+
},
|
| 247 |
+
"versions": {
|
| 248 |
+
"harness|bbh:causal_judgment|3": 0,
|
| 249 |
+
"harness|bbh:date_understanding|3": 0,
|
| 250 |
+
"harness|bbh:disambiguation_qa|3": 0,
|
| 251 |
+
"harness|bbh:geometric_shapes|3": 0,
|
| 252 |
+
"harness|bbh:logical_deduction_five_objects|3": 0,
|
| 253 |
+
"harness|bbh:logical_deduction_seven_objects|3": 0,
|
| 254 |
+
"harness|bbh:logical_deduction_three_objects|3": 0,
|
| 255 |
+
"harness|bbh:movie_recommendation|3": 0,
|
| 256 |
+
"harness|bbh:navigate|3": 0,
|
| 257 |
+
"harness|bbh:reasoning_about_colored_objects|3": 0,
|
| 258 |
+
"harness|bbh:ruin_names|3": 0,
|
| 259 |
+
"harness|bbh:salient_translation_error_detection|3": 0,
|
| 260 |
+
"harness|bbh:snarks|3": 0,
|
| 261 |
+
"harness|bbh:sports_understanding|3": 0,
|
| 262 |
+
"harness|bbh:temporal_sequences|3": 0,
|
| 263 |
+
"harness|bbh:tracking_shuffled_objects_five_objects|3": 0,
|
| 264 |
+
"harness|bbh:tracking_shuffled_objects_seven_objects|3": 0,
|
| 265 |
+
"harness|bbh:tracking_shuffled_objects_three_objects|3": 0
|
| 266 |
+
},
|
| 267 |
+
"config_tasks": {
|
| 268 |
+
"harness|bbh:causal_judgment": {
|
| 269 |
+
"name": "bbh:causal_judgment",
|
| 270 |
+
"prompt_function": "bbh_causal_judgment",
|
| 271 |
+
"hf_repo": "lukaemon/bbh",
|
| 272 |
+
"hf_subset": "causal_judgement",
|
| 273 |
+
"metric": [
|
| 274 |
+
"exact_match",
|
| 275 |
+
"quasi_exact_match",
|
| 276 |
+
"prefix_exact_match",
|
| 277 |
+
"prefix_quasi_exact_match",
|
| 278 |
+
"perfect_exact_match"
|
| 279 |
+
],
|
| 280 |
+
"hf_avail_splits": [
|
| 281 |
+
"test"
|
| 282 |
+
],
|
| 283 |
+
"evaluation_splits": [
|
| 284 |
+
"test"
|
| 285 |
+
],
|
| 286 |
+
"few_shots_split": null,
|
| 287 |
+
"few_shots_select": null,
|
| 288 |
+
"generation_size": 20,
|
| 289 |
+
"stop_sequence": [
|
| 290 |
+
"</s>",
|
| 291 |
+
"Q:",
|
| 292 |
+
"\n\n"
|
| 293 |
+
],
|
| 294 |
+
"output_regex": null,
|
| 295 |
+
"frozen": false,
|
| 296 |
+
"suite": [
|
| 297 |
+
"harness"
|
| 298 |
+
],
|
| 299 |
+
"original_num_docs": 187,
|
| 300 |
+
"effective_num_docs": 187,
|
| 301 |
+
"trust_dataset": true,
|
| 302 |
+
"must_remove_duplicate_docs": null
|
| 303 |
+
},
|
| 304 |
+
"harness|bbh:date_understanding": {
|
| 305 |
+
"name": "bbh:date_understanding",
|
| 306 |
+
"prompt_function": "bbh_date_understanding",
|
| 307 |
+
"hf_repo": "lukaemon/bbh",
|
| 308 |
+
"hf_subset": "date_understanding",
|
| 309 |
+
"metric": [
|
| 310 |
+
"exact_match",
|
| 311 |
+
"quasi_exact_match",
|
| 312 |
+
"prefix_exact_match",
|
| 313 |
+
"prefix_quasi_exact_match",
|
| 314 |
+
"perfect_exact_match"
|
| 315 |
+
],
|
| 316 |
+
"hf_avail_splits": [
|
| 317 |
+
"test"
|
| 318 |
+
],
|
| 319 |
+
"evaluation_splits": [
|
| 320 |
+
"test"
|
| 321 |
+
],
|
| 322 |
+
"few_shots_split": null,
|
| 323 |
+
"few_shots_select": null,
|
| 324 |
+
"generation_size": 20,
|
| 325 |
+
"stop_sequence": [
|
| 326 |
+
"</s>",
|
| 327 |
+
"Q:",
|
| 328 |
+
"\n\n"
|
| 329 |
+
],
|
| 330 |
+
"output_regex": null,
|
| 331 |
+
"frozen": false,
|
| 332 |
+
"suite": [
|
| 333 |
+
"harness"
|
| 334 |
+
],
|
| 335 |
+
"original_num_docs": 250,
|
| 336 |
+
"effective_num_docs": 250,
|
| 337 |
+
"trust_dataset": true,
|
| 338 |
+
"must_remove_duplicate_docs": null
|
| 339 |
+
},
|
| 340 |
+
"harness|bbh:disambiguation_qa": {
|
| 341 |
+
"name": "bbh:disambiguation_qa",
|
| 342 |
+
"prompt_function": "bbh_disambiguation_qa",
|
| 343 |
+
"hf_repo": "lukaemon/bbh",
|
| 344 |
+
"hf_subset": "disambiguation_qa",
|
| 345 |
+
"metric": [
|
| 346 |
+
"exact_match",
|
| 347 |
+
"quasi_exact_match",
|
| 348 |
+
"prefix_exact_match",
|
| 349 |
+
"prefix_quasi_exact_match",
|
| 350 |
+
"perfect_exact_match"
|
| 351 |
+
],
|
| 352 |
+
"hf_avail_splits": [
|
| 353 |
+
"test"
|
| 354 |
+
],
|
| 355 |
+
"evaluation_splits": [
|
| 356 |
+
"test"
|
| 357 |
+
],
|
| 358 |
+
"few_shots_split": null,
|
| 359 |
+
"few_shots_select": null,
|
| 360 |
+
"generation_size": 20,
|
| 361 |
+
"stop_sequence": [
|
| 362 |
+
"</s>",
|
| 363 |
+
"Q:",
|
| 364 |
+
"\n\n"
|
| 365 |
+
],
|
| 366 |
+
"output_regex": null,
|
| 367 |
+
"frozen": false,
|
| 368 |
+
"suite": [
|
| 369 |
+
"harness"
|
| 370 |
+
],
|
| 371 |
+
"original_num_docs": 250,
|
| 372 |
+
"effective_num_docs": 250,
|
| 373 |
+
"trust_dataset": true,
|
| 374 |
+
"must_remove_duplicate_docs": null
|
| 375 |
+
},
|
| 376 |
+
"harness|bbh:geometric_shapes": {
|
| 377 |
+
"name": "bbh:geometric_shapes",
|
| 378 |
+
"prompt_function": "bbh_geometric_shapes",
|
| 379 |
+
"hf_repo": "lukaemon/bbh",
|
| 380 |
+
"hf_subset": "geometric_shapes",
|
| 381 |
+
"metric": [
|
| 382 |
+
"exact_match",
|
| 383 |
+
"quasi_exact_match",
|
| 384 |
+
"prefix_exact_match",
|
| 385 |
+
"prefix_quasi_exact_match",
|
| 386 |
+
"perfect_exact_match"
|
| 387 |
+
],
|
| 388 |
+
"hf_avail_splits": [
|
| 389 |
+
"test"
|
| 390 |
+
],
|
| 391 |
+
"evaluation_splits": [
|
| 392 |
+
"test"
|
| 393 |
+
],
|
| 394 |
+
"few_shots_split": null,
|
| 395 |
+
"few_shots_select": null,
|
| 396 |
+
"generation_size": 20,
|
| 397 |
+
"stop_sequence": [
|
| 398 |
+
"</s>",
|
| 399 |
+
"Q:",
|
| 400 |
+
"\n\n"
|
| 401 |
+
],
|
| 402 |
+
"output_regex": null,
|
| 403 |
+
"frozen": false,
|
| 404 |
+
"suite": [
|
| 405 |
+
"harness"
|
| 406 |
+
],
|
| 407 |
+
"original_num_docs": 250,
|
| 408 |
+
"effective_num_docs": 250,
|
| 409 |
+
"trust_dataset": true,
|
| 410 |
+
"must_remove_duplicate_docs": null
|
| 411 |
+
},
|
| 412 |
+
"harness|bbh:logical_deduction_five_objects": {
|
| 413 |
+
"name": "bbh:logical_deduction_five_objects",
|
| 414 |
+
"prompt_function": "bbh_logical_deduction_five_objects",
|
| 415 |
+
"hf_repo": "lukaemon/bbh",
|
| 416 |
+
"hf_subset": "logical_deduction_five_objects",
|
| 417 |
+
"metric": [
|
| 418 |
+
"exact_match",
|
| 419 |
+
"quasi_exact_match",
|
| 420 |
+
"prefix_exact_match",
|
| 421 |
+
"prefix_quasi_exact_match",
|
| 422 |
+
"perfect_exact_match"
|
| 423 |
+
],
|
| 424 |
+
"hf_avail_splits": [
|
| 425 |
+
"test"
|
| 426 |
+
],
|
| 427 |
+
"evaluation_splits": [
|
| 428 |
+
"test"
|
| 429 |
+
],
|
| 430 |
+
"few_shots_split": null,
|
| 431 |
+
"few_shots_select": null,
|
| 432 |
+
"generation_size": 20,
|
| 433 |
+
"stop_sequence": [
|
| 434 |
+
"</s>",
|
| 435 |
+
"Q:",
|
| 436 |
+
"\n\n"
|
| 437 |
+
],
|
| 438 |
+
"output_regex": null,
|
| 439 |
+
"frozen": false,
|
| 440 |
+
"suite": [
|
| 441 |
+
"harness"
|
| 442 |
+
],
|
| 443 |
+
"original_num_docs": 250,
|
| 444 |
+
"effective_num_docs": 250,
|
| 445 |
+
"trust_dataset": true,
|
| 446 |
+
"must_remove_duplicate_docs": null
|
| 447 |
+
},
|
| 448 |
+
"harness|bbh:logical_deduction_seven_objects": {
|
| 449 |
+
"name": "bbh:logical_deduction_seven_objects",
|
| 450 |
+
"prompt_function": "bbh_logical_deduction_seven_objects",
|
| 451 |
+
"hf_repo": "lukaemon/bbh",
|
| 452 |
+
"hf_subset": "logical_deduction_seven_objects",
|
| 453 |
+
"metric": [
|
| 454 |
+
"exact_match",
|
| 455 |
+
"quasi_exact_match",
|
| 456 |
+
"prefix_exact_match",
|
| 457 |
+
"prefix_quasi_exact_match",
|
| 458 |
+
"perfect_exact_match"
|
| 459 |
+
],
|
| 460 |
+
"hf_avail_splits": [
|
| 461 |
+
"test"
|
| 462 |
+
],
|
| 463 |
+
"evaluation_splits": [
|
| 464 |
+
"test"
|
| 465 |
+
],
|
| 466 |
+
"few_shots_split": null,
|
| 467 |
+
"few_shots_select": null,
|
| 468 |
+
"generation_size": 20,
|
| 469 |
+
"stop_sequence": [
|
| 470 |
+
"</s>",
|
| 471 |
+
"Q:",
|
| 472 |
+
"\n\n"
|
| 473 |
+
],
|
| 474 |
+
"output_regex": null,
|
| 475 |
+
"frozen": false,
|
| 476 |
+
"suite": [
|
| 477 |
+
"harness"
|
| 478 |
+
],
|
| 479 |
+
"original_num_docs": 250,
|
| 480 |
+
"effective_num_docs": 250,
|
| 481 |
+
"trust_dataset": true,
|
| 482 |
+
"must_remove_duplicate_docs": null
|
| 483 |
+
},
|
| 484 |
+
"harness|bbh:logical_deduction_three_objects": {
|
| 485 |
+
"name": "bbh:logical_deduction_three_objects",
|
| 486 |
+
"prompt_function": "bbh_logical_deduction_three_objects",
|
| 487 |
+
"hf_repo": "lukaemon/bbh",
|
| 488 |
+
"hf_subset": "logical_deduction_three_objects",
|
| 489 |
+
"metric": [
|
| 490 |
+
"exact_match",
|
| 491 |
+
"quasi_exact_match",
|
| 492 |
+
"prefix_exact_match",
|
| 493 |
+
"prefix_quasi_exact_match",
|
| 494 |
+
"perfect_exact_match"
|
| 495 |
+
],
|
| 496 |
+
"hf_avail_splits": [
|
| 497 |
+
"test"
|
| 498 |
+
],
|
| 499 |
+
"evaluation_splits": [
|
| 500 |
+
"test"
|
| 501 |
+
],
|
| 502 |
+
"few_shots_split": null,
|
| 503 |
+
"few_shots_select": null,
|
| 504 |
+
"generation_size": 20,
|
| 505 |
+
"stop_sequence": [
|
| 506 |
+
"</s>",
|
| 507 |
+
"Q:",
|
| 508 |
+
"\n\n"
|
| 509 |
+
],
|
| 510 |
+
"output_regex": null,
|
| 511 |
+
"frozen": false,
|
| 512 |
+
"suite": [
|
| 513 |
+
"harness"
|
| 514 |
+
],
|
| 515 |
+
"original_num_docs": 250,
|
| 516 |
+
"effective_num_docs": 250,
|
| 517 |
+
"trust_dataset": true,
|
| 518 |
+
"must_remove_duplicate_docs": null
|
| 519 |
+
},
|
| 520 |
+
"harness|bbh:movie_recommendation": {
|
| 521 |
+
"name": "bbh:movie_recommendation",
|
| 522 |
+
"prompt_function": "bbh_movie_recommendation",
|
| 523 |
+
"hf_repo": "lukaemon/bbh",
|
| 524 |
+
"hf_subset": "movie_recommendation",
|
| 525 |
+
"metric": [
|
| 526 |
+
"exact_match",
|
| 527 |
+
"quasi_exact_match",
|
| 528 |
+
"prefix_exact_match",
|
| 529 |
+
"prefix_quasi_exact_match",
|
| 530 |
+
"perfect_exact_match"
|
| 531 |
+
],
|
| 532 |
+
"hf_avail_splits": [
|
| 533 |
+
"test"
|
| 534 |
+
],
|
| 535 |
+
"evaluation_splits": [
|
| 536 |
+
"test"
|
| 537 |
+
],
|
| 538 |
+
"few_shots_split": null,
|
| 539 |
+
"few_shots_select": null,
|
| 540 |
+
"generation_size": 20,
|
| 541 |
+
"stop_sequence": [
|
| 542 |
+
"</s>",
|
| 543 |
+
"Q:",
|
| 544 |
+
"\n\n"
|
| 545 |
+
],
|
| 546 |
+
"output_regex": null,
|
| 547 |
+
"frozen": false,
|
| 548 |
+
"suite": [
|
| 549 |
+
"harness"
|
| 550 |
+
],
|
| 551 |
+
"original_num_docs": 249,
|
| 552 |
+
"effective_num_docs": 249,
|
| 553 |
+
"trust_dataset": true,
|
| 554 |
+
"must_remove_duplicate_docs": null
|
| 555 |
+
},
|
| 556 |
+
"harness|bbh:navigate": {
|
| 557 |
+
"name": "bbh:navigate",
|
| 558 |
+
"prompt_function": "bbh_navigate",
|
| 559 |
+
"hf_repo": "lukaemon/bbh",
|
| 560 |
+
"hf_subset": "navigate",
|
| 561 |
+
"metric": [
|
| 562 |
+
"exact_match",
|
| 563 |
+
"quasi_exact_match",
|
| 564 |
+
"prefix_exact_match",
|
| 565 |
+
"prefix_quasi_exact_match",
|
| 566 |
+
"perfect_exact_match"
|
| 567 |
+
],
|
| 568 |
+
"hf_avail_splits": [
|
| 569 |
+
"test"
|
| 570 |
+
],
|
| 571 |
+
"evaluation_splits": [
|
| 572 |
+
"test"
|
| 573 |
+
],
|
| 574 |
+
"few_shots_split": null,
|
| 575 |
+
"few_shots_select": null,
|
| 576 |
+
"generation_size": 20,
|
| 577 |
+
"stop_sequence": [
|
| 578 |
+
"</s>",
|
| 579 |
+
"Q:",
|
| 580 |
+
"\n\n"
|
| 581 |
+
],
|
| 582 |
+
"output_regex": null,
|
| 583 |
+
"frozen": false,
|
| 584 |
+
"suite": [
|
| 585 |
+
"harness"
|
| 586 |
+
],
|
| 587 |
+
"original_num_docs": 250,
|
| 588 |
+
"effective_num_docs": 250,
|
| 589 |
+
"trust_dataset": true,
|
| 590 |
+
"must_remove_duplicate_docs": null
|
| 591 |
+
},
|
| 592 |
+
"harness|bbh:reasoning_about_colored_objects": {
|
| 593 |
+
"name": "bbh:reasoning_about_colored_objects",
|
| 594 |
+
"prompt_function": "bbh_reasoning_about_colored_objects",
|
| 595 |
+
"hf_repo": "lukaemon/bbh",
|
| 596 |
+
"hf_subset": "reasoning_about_colored_objects",
|
| 597 |
+
"metric": [
|
| 598 |
+
"exact_match",
|
| 599 |
+
"quasi_exact_match",
|
| 600 |
+
"prefix_exact_match",
|
| 601 |
+
"prefix_quasi_exact_match",
|
| 602 |
+
"perfect_exact_match"
|
| 603 |
+
],
|
| 604 |
+
"hf_avail_splits": [
|
| 605 |
+
"test"
|
| 606 |
+
],
|
| 607 |
+
"evaluation_splits": [
|
| 608 |
+
"test"
|
| 609 |
+
],
|
| 610 |
+
"few_shots_split": null,
|
| 611 |
+
"few_shots_select": null,
|
| 612 |
+
"generation_size": 20,
|
| 613 |
+
"stop_sequence": [
|
| 614 |
+
"</s>",
|
| 615 |
+
"Q:",
|
| 616 |
+
"\n\n"
|
| 617 |
+
],
|
| 618 |
+
"output_regex": null,
|
| 619 |
+
"frozen": false,
|
| 620 |
+
"suite": [
|
| 621 |
+
"harness"
|
| 622 |
+
],
|
| 623 |
+
"original_num_docs": 250,
|
| 624 |
+
"effective_num_docs": 250,
|
| 625 |
+
"trust_dataset": true,
|
| 626 |
+
"must_remove_duplicate_docs": null
|
| 627 |
+
},
|
| 628 |
+
"harness|bbh:ruin_names": {
|
| 629 |
+
"name": "bbh:ruin_names",
|
| 630 |
+
"prompt_function": "bbh_ruin_names",
|
| 631 |
+
"hf_repo": "lukaemon/bbh",
|
| 632 |
+
"hf_subset": "ruin_names",
|
| 633 |
+
"metric": [
|
| 634 |
+
"exact_match",
|
| 635 |
+
"quasi_exact_match",
|
| 636 |
+
"prefix_exact_match",
|
| 637 |
+
"prefix_quasi_exact_match",
|
| 638 |
+
"perfect_exact_match"
|
| 639 |
+
],
|
| 640 |
+
"hf_avail_splits": [
|
| 641 |
+
"test"
|
| 642 |
+
],
|
| 643 |
+
"evaluation_splits": [
|
| 644 |
+
"test"
|
| 645 |
+
],
|
| 646 |
+
"few_shots_split": null,
|
| 647 |
+
"few_shots_select": null,
|
| 648 |
+
"generation_size": 20,
|
| 649 |
+
"stop_sequence": [
|
| 650 |
+
"</s>",
|
| 651 |
+
"Q:",
|
| 652 |
+
"\n\n"
|
| 653 |
+
],
|
| 654 |
+
"output_regex": null,
|
| 655 |
+
"frozen": false,
|
| 656 |
+
"suite": [
|
| 657 |
+
"harness"
|
| 658 |
+
],
|
| 659 |
+
"original_num_docs": 248,
|
| 660 |
+
"effective_num_docs": 248,
|
| 661 |
+
"trust_dataset": true,
|
| 662 |
+
"must_remove_duplicate_docs": null
|
| 663 |
+
},
|
| 664 |
+
"harness|bbh:salient_translation_error_detection": {
|
| 665 |
+
"name": "bbh:salient_translation_error_detection",
|
| 666 |
+
"prompt_function": "bbh_salient_translation_error_detection",
|
| 667 |
+
"hf_repo": "lukaemon/bbh",
|
| 668 |
+
"hf_subset": "salient_translation_error_detection",
|
| 669 |
+
"metric": [
|
| 670 |
+
"exact_match",
|
| 671 |
+
"quasi_exact_match",
|
| 672 |
+
"prefix_exact_match",
|
| 673 |
+
"prefix_quasi_exact_match",
|
| 674 |
+
"perfect_exact_match"
|
| 675 |
+
],
|
| 676 |
+
"hf_avail_splits": [
|
| 677 |
+
"test"
|
| 678 |
+
],
|
| 679 |
+
"evaluation_splits": [
|
| 680 |
+
"test"
|
| 681 |
+
],
|
| 682 |
+
"few_shots_split": null,
|
| 683 |
+
"few_shots_select": null,
|
| 684 |
+
"generation_size": 20,
|
| 685 |
+
"stop_sequence": [
|
| 686 |
+
"</s>",
|
| 687 |
+
"Q:",
|
| 688 |
+
"\n\n"
|
| 689 |
+
],
|
| 690 |
+
"output_regex": null,
|
| 691 |
+
"frozen": false,
|
| 692 |
+
"suite": [
|
| 693 |
+
"harness"
|
| 694 |
+
],
|
| 695 |
+
"original_num_docs": 250,
|
| 696 |
+
"effective_num_docs": 250,
|
| 697 |
+
"trust_dataset": true,
|
| 698 |
+
"must_remove_duplicate_docs": null
|
| 699 |
+
},
|
| 700 |
+
"harness|bbh:snarks": {
|
| 701 |
+
"name": "bbh:snarks",
|
| 702 |
+
"prompt_function": "bbh_snarks",
|
| 703 |
+
"hf_repo": "lukaemon/bbh",
|
| 704 |
+
"hf_subset": "snarks",
|
| 705 |
+
"metric": [
|
| 706 |
+
"exact_match",
|
| 707 |
+
"quasi_exact_match",
|
| 708 |
+
"prefix_exact_match",
|
| 709 |
+
"prefix_quasi_exact_match",
|
| 710 |
+
"perfect_exact_match"
|
| 711 |
+
],
|
| 712 |
+
"hf_avail_splits": [
|
| 713 |
+
"test"
|
| 714 |
+
],
|
| 715 |
+
"evaluation_splits": [
|
| 716 |
+
"test"
|
| 717 |
+
],
|
| 718 |
+
"few_shots_split": null,
|
| 719 |
+
"few_shots_select": null,
|
| 720 |
+
"generation_size": 20,
|
| 721 |
+
"stop_sequence": [
|
| 722 |
+
"</s>",
|
| 723 |
+
"Q:",
|
| 724 |
+
"\n\n"
|
| 725 |
+
],
|
| 726 |
+
"output_regex": null,
|
| 727 |
+
"frozen": false,
|
| 728 |
+
"suite": [
|
| 729 |
+
"harness"
|
| 730 |
+
],
|
| 731 |
+
"original_num_docs": 178,
|
| 732 |
+
"effective_num_docs": 178,
|
| 733 |
+
"trust_dataset": true,
|
| 734 |
+
"must_remove_duplicate_docs": null
|
| 735 |
+
},
|
| 736 |
+
"harness|bbh:sports_understanding": {
|
| 737 |
+
"name": "bbh:sports_understanding",
|
| 738 |
+
"prompt_function": "bbh_sports_understanding",
|
| 739 |
+
"hf_repo": "lukaemon/bbh",
|
| 740 |
+
"hf_subset": "sports_understanding",
|
| 741 |
+
"metric": [
|
| 742 |
+
"exact_match",
|
| 743 |
+
"quasi_exact_match",
|
| 744 |
+
"prefix_exact_match",
|
| 745 |
+
"prefix_quasi_exact_match",
|
| 746 |
+
"perfect_exact_match"
|
| 747 |
+
],
|
| 748 |
+
"hf_avail_splits": [
|
| 749 |
+
"test"
|
| 750 |
+
],
|
| 751 |
+
"evaluation_splits": [
|
| 752 |
+
"test"
|
| 753 |
+
],
|
| 754 |
+
"few_shots_split": null,
|
| 755 |
+
"few_shots_select": null,
|
| 756 |
+
"generation_size": 20,
|
| 757 |
+
"stop_sequence": [
|
| 758 |
+
"</s>",
|
| 759 |
+
"Q:",
|
| 760 |
+
"\n\n"
|
| 761 |
+
],
|
| 762 |
+
"output_regex": null,
|
| 763 |
+
"frozen": false,
|
| 764 |
+
"suite": [
|
| 765 |
+
"harness"
|
| 766 |
+
],
|
| 767 |
+
"original_num_docs": 250,
|
| 768 |
+
"effective_num_docs": 250,
|
| 769 |
+
"trust_dataset": true,
|
| 770 |
+
"must_remove_duplicate_docs": null
|
| 771 |
+
},
|
| 772 |
+
"harness|bbh:temporal_sequences": {
|
| 773 |
+
"name": "bbh:temporal_sequences",
|
| 774 |
+
"prompt_function": "bbh_temporal_sequences",
|
| 775 |
+
"hf_repo": "lukaemon/bbh",
|
| 776 |
+
"hf_subset": "temporal_sequences",
|
| 777 |
+
"metric": [
|
| 778 |
+
"exact_match",
|
| 779 |
+
"quasi_exact_match",
|
| 780 |
+
"prefix_exact_match",
|
| 781 |
+
"prefix_quasi_exact_match",
|
| 782 |
+
"perfect_exact_match"
|
| 783 |
+
],
|
| 784 |
+
"hf_avail_splits": [
|
| 785 |
+
"test"
|
| 786 |
+
],
|
| 787 |
+
"evaluation_splits": [
|
| 788 |
+
"test"
|
| 789 |
+
],
|
| 790 |
+
"few_shots_split": null,
|
| 791 |
+
"few_shots_select": null,
|
| 792 |
+
"generation_size": 20,
|
| 793 |
+
"stop_sequence": [
|
| 794 |
+
"</s>",
|
| 795 |
+
"Q:",
|
| 796 |
+
"\n\n"
|
| 797 |
+
],
|
| 798 |
+
"output_regex": null,
|
| 799 |
+
"frozen": false,
|
| 800 |
+
"suite": [
|
| 801 |
+
"harness"
|
| 802 |
+
],
|
| 803 |
+
"original_num_docs": 250,
|
| 804 |
+
"effective_num_docs": 250,
|
| 805 |
+
"trust_dataset": true,
|
| 806 |
+
"must_remove_duplicate_docs": null
|
| 807 |
+
},
|
| 808 |
+
"harness|bbh:tracking_shuffled_objects_five_objects": {
|
| 809 |
+
"name": "bbh:tracking_shuffled_objects_five_objects",
|
| 810 |
+
"prompt_function": "bbh_tracking_shuffled_objects_five_objects",
|
| 811 |
+
"hf_repo": "lukaemon/bbh",
|
| 812 |
+
"hf_subset": "tracking_shuffled_objects_five_objects",
|
| 813 |
+
"metric": [
|
| 814 |
+
"exact_match",
|
| 815 |
+
"quasi_exact_match",
|
| 816 |
+
"prefix_exact_match",
|
| 817 |
+
"prefix_quasi_exact_match",
|
| 818 |
+
"perfect_exact_match"
|
| 819 |
+
],
|
| 820 |
+
"hf_avail_splits": [
|
| 821 |
+
"test"
|
| 822 |
+
],
|
| 823 |
+
"evaluation_splits": [
|
| 824 |
+
"test"
|
| 825 |
+
],
|
| 826 |
+
"few_shots_split": null,
|
| 827 |
+
"few_shots_select": null,
|
| 828 |
+
"generation_size": 20,
|
| 829 |
+
"stop_sequence": [
|
| 830 |
+
"</s>",
|
| 831 |
+
"Q:",
|
| 832 |
+
"\n\n"
|
| 833 |
+
],
|
| 834 |
+
"output_regex": null,
|
| 835 |
+
"frozen": false,
|
| 836 |
+
"suite": [
|
| 837 |
+
"harness"
|
| 838 |
+
],
|
| 839 |
+
"original_num_docs": 250,
|
| 840 |
+
"effective_num_docs": 250,
|
| 841 |
+
"trust_dataset": true,
|
| 842 |
+
"must_remove_duplicate_docs": null
|
| 843 |
+
},
|
| 844 |
+
"harness|bbh:tracking_shuffled_objects_seven_objects": {
|
| 845 |
+
"name": "bbh:tracking_shuffled_objects_seven_objects",
|
| 846 |
+
"prompt_function": "bbh_tracking_shuffled_objects_seven_objects",
|
| 847 |
+
"hf_repo": "lukaemon/bbh",
|
| 848 |
+
"hf_subset": "tracking_shuffled_objects_seven_objects",
|
| 849 |
+
"metric": [
|
| 850 |
+
"exact_match",
|
| 851 |
+
"quasi_exact_match",
|
| 852 |
+
"prefix_exact_match",
|
| 853 |
+
"prefix_quasi_exact_match",
|
| 854 |
+
"perfect_exact_match"
|
| 855 |
+
],
|
| 856 |
+
"hf_avail_splits": [
|
| 857 |
+
"test"
|
| 858 |
+
],
|
| 859 |
+
"evaluation_splits": [
|
| 860 |
+
"test"
|
| 861 |
+
],
|
| 862 |
+
"few_shots_split": null,
|
| 863 |
+
"few_shots_select": null,
|
| 864 |
+
"generation_size": 20,
|
| 865 |
+
"stop_sequence": [
|
| 866 |
+
"</s>",
|
| 867 |
+
"Q:",
|
| 868 |
+
"\n\n"
|
| 869 |
+
],
|
| 870 |
+
"output_regex": null,
|
| 871 |
+
"frozen": false,
|
| 872 |
+
"suite": [
|
| 873 |
+
"harness"
|
| 874 |
+
],
|
| 875 |
+
"original_num_docs": 250,
|
| 876 |
+
"effective_num_docs": 250,
|
| 877 |
+
"trust_dataset": true,
|
| 878 |
+
"must_remove_duplicate_docs": null
|
| 879 |
+
},
|
| 880 |
+
"harness|bbh:tracking_shuffled_objects_three_objects": {
|
| 881 |
+
"name": "bbh:tracking_shuffled_objects_three_objects",
|
| 882 |
+
"prompt_function": "bbh_tracking_shuffled_objects_three_objects",
|
| 883 |
+
"hf_repo": "lukaemon/bbh",
|
| 884 |
+
"hf_subset": "tracking_shuffled_objects_three_objects",
|
| 885 |
+
"metric": [
|
| 886 |
+
"exact_match",
|
| 887 |
+
"quasi_exact_match",
|
| 888 |
+
"prefix_exact_match",
|
| 889 |
+
"prefix_quasi_exact_match",
|
| 890 |
+
"perfect_exact_match"
|
| 891 |
+
],
|
| 892 |
+
"hf_avail_splits": [
|
| 893 |
+
"test"
|
| 894 |
+
],
|
| 895 |
+
"evaluation_splits": [
|
| 896 |
+
"test"
|
| 897 |
+
],
|
| 898 |
+
"few_shots_split": null,
|
| 899 |
+
"few_shots_select": null,
|
| 900 |
+
"generation_size": 20,
|
| 901 |
+
"stop_sequence": [
|
| 902 |
+
"</s>",
|
| 903 |
+
"Q:",
|
| 904 |
+
"\n\n"
|
| 905 |
+
],
|
| 906 |
+
"output_regex": null,
|
| 907 |
+
"frozen": false,
|
| 908 |
+
"suite": [
|
| 909 |
+
"harness"
|
| 910 |
+
],
|
| 911 |
+
"original_num_docs": 250,
|
| 912 |
+
"effective_num_docs": 250,
|
| 913 |
+
"trust_dataset": true,
|
| 914 |
+
"must_remove_duplicate_docs": null
|
| 915 |
+
}
|
| 916 |
+
},
|
| 917 |
+
"summary_tasks": {
|
| 918 |
+
"harness|bbh:causal_judgment|3": {
|
| 919 |
+
"hashes": {
|
| 920 |
+
"hash_examples": "63218f5ae055ab2b",
|
| 921 |
+
"hash_full_prompts": "fa8168f39a475fb0",
|
| 922 |
+
"hash_input_tokens": "787f75e06fd43c0d",
|
| 923 |
+
"hash_cont_tokens": "284e857febd4be85"
|
| 924 |
+
},
|
| 925 |
+
"truncated": 187,
|
| 926 |
+
"non_truncated": 0,
|
| 927 |
+
"padded": 0,
|
| 928 |
+
"non_padded": 187,
|
| 929 |
+
"effective_few_shots": 3.0,
|
| 930 |
+
"num_truncated_few_shots": 0
|
| 931 |
+
},
|
| 932 |
+
"harness|bbh:date_understanding|3": {
|
| 933 |
+
"hashes": {
|
| 934 |
+
"hash_examples": "f145c7a06def3c8e",
|
| 935 |
+
"hash_full_prompts": "2cceeea606638d49",
|
| 936 |
+
"hash_input_tokens": "10c13d6fb8af7c22",
|
| 937 |
+
"hash_cont_tokens": "d3c601f01854a89e"
|
| 938 |
+
},
|
| 939 |
+
"truncated": 250,
|
| 940 |
+
"non_truncated": 0,
|
| 941 |
+
"padded": 0,
|
| 942 |
+
"non_padded": 250,
|
| 943 |
+
"effective_few_shots": 3.0,
|
| 944 |
+
"num_truncated_few_shots": 0
|
| 945 |
+
},
|
| 946 |
+
"harness|bbh:disambiguation_qa|3": {
|
| 947 |
+
"hashes": {
|
| 948 |
+
"hash_examples": "19677fd1773f7eb9",
|
| 949 |
+
"hash_full_prompts": "d8f1ba70c22ae578",
|
| 950 |
+
"hash_input_tokens": "c21a88707f480cab",
|
| 951 |
+
"hash_cont_tokens": "8705f09efb1872ed"
|
| 952 |
+
},
|
| 953 |
+
"truncated": 250,
|
| 954 |
+
"non_truncated": 0,
|
| 955 |
+
"padded": 0,
|
| 956 |
+
"non_padded": 250,
|
| 957 |
+
"effective_few_shots": 3.0,
|
| 958 |
+
"num_truncated_few_shots": 0
|
| 959 |
+
},
|
| 960 |
+
"harness|bbh:geometric_shapes|3": {
|
| 961 |
+
"hashes": {
|
| 962 |
+
"hash_examples": "76c7b11a13cc72a9",
|
| 963 |
+
"hash_full_prompts": "52a60ed1d0113b8b",
|
| 964 |
+
"hash_input_tokens": "10e113b2cf3fa584",
|
| 965 |
+
"hash_cont_tokens": "aa68d879c1c72d3c"
|
| 966 |
+
},
|
| 967 |
+
"truncated": 250,
|
| 968 |
+
"non_truncated": 0,
|
| 969 |
+
"padded": 0,
|
| 970 |
+
"non_padded": 250,
|
| 971 |
+
"effective_few_shots": 3.0,
|
| 972 |
+
"num_truncated_few_shots": 0
|
| 973 |
+
},
|
| 974 |
+
"harness|bbh:logical_deduction_five_objects|3": {
|
| 975 |
+
"hashes": {
|
| 976 |
+
"hash_examples": "0e958c856332a745",
|
| 977 |
+
"hash_full_prompts": "253aa9791c941909",
|
| 978 |
+
"hash_input_tokens": "0bc166cab0aed76a",
|
| 979 |
+
"hash_cont_tokens": "a33748b3d10c2699"
|
| 980 |
+
},
|
| 981 |
+
"truncated": 250,
|
| 982 |
+
"non_truncated": 0,
|
| 983 |
+
"padded": 0,
|
| 984 |
+
"non_padded": 250,
|
| 985 |
+
"effective_few_shots": 3.0,
|
| 986 |
+
"num_truncated_few_shots": 0
|
| 987 |
+
},
|
| 988 |
+
"harness|bbh:logical_deduction_seven_objects|3": {
|
| 989 |
+
"hashes": {
|
| 990 |
+
"hash_examples": "ab9de25a5eb40d09",
|
| 991 |
+
"hash_full_prompts": "aa6117f601cd268e",
|
| 992 |
+
"hash_input_tokens": "ab99c78b48e3a0bb",
|
| 993 |
+
"hash_cont_tokens": "3edfc27f0e21c94d"
|
| 994 |
+
},
|
| 995 |
+
"truncated": 250,
|
| 996 |
+
"non_truncated": 0,
|
| 997 |
+
"padded": 0,
|
| 998 |
+
"non_padded": 250,
|
| 999 |
+
"effective_few_shots": 3.0,
|
| 1000 |
+
"num_truncated_few_shots": 0
|
| 1001 |
+
},
|
| 1002 |
+
"harness|bbh:logical_deduction_three_objects|3": {
|
| 1003 |
+
"hashes": {
|
| 1004 |
+
"hash_examples": "3c6bf52517714218",
|
| 1005 |
+
"hash_full_prompts": "1892b050bc7848a4",
|
| 1006 |
+
"hash_input_tokens": "a720b56aa7c52551",
|
| 1007 |
+
"hash_cont_tokens": "abd8a52498d4f728"
|
| 1008 |
+
},
|
| 1009 |
+
"truncated": 250,
|
| 1010 |
+
"non_truncated": 0,
|
| 1011 |
+
"padded": 0,
|
| 1012 |
+
"non_padded": 250,
|
| 1013 |
+
"effective_few_shots": 3.0,
|
| 1014 |
+
"num_truncated_few_shots": 0
|
| 1015 |
+
},
|
| 1016 |
+
"harness|bbh:movie_recommendation|3": {
|
| 1017 |
+
"hashes": {
|
| 1018 |
+
"hash_examples": "2d9dc4975935d31a",
|
| 1019 |
+
"hash_full_prompts": "8e00606ed3407167",
|
| 1020 |
+
"hash_input_tokens": "c825ab1c99245a17",
|
| 1021 |
+
"hash_cont_tokens": "5ec900e40cbc845e"
|
| 1022 |
+
},
|
| 1023 |
+
"truncated": 249,
|
| 1024 |
+
"non_truncated": 0,
|
| 1025 |
+
"padded": 0,
|
| 1026 |
+
"non_padded": 249,
|
| 1027 |
+
"effective_few_shots": 3.0,
|
| 1028 |
+
"num_truncated_few_shots": 0
|
| 1029 |
+
},
|
| 1030 |
+
"harness|bbh:navigate|3": {
|
| 1031 |
+
"hashes": {
|
| 1032 |
+
"hash_examples": "ba91dcdb9a064255",
|
| 1033 |
+
"hash_full_prompts": "8d50c5baf1df7aef",
|
| 1034 |
+
"hash_input_tokens": "f234e6b28ea1fa49",
|
| 1035 |
+
"hash_cont_tokens": "ee12c9f32365d922"
|
| 1036 |
+
},
|
| 1037 |
+
"truncated": 250,
|
| 1038 |
+
"non_truncated": 0,
|
| 1039 |
+
"padded": 0,
|
| 1040 |
+
"non_padded": 250,
|
| 1041 |
+
"effective_few_shots": 3.0,
|
| 1042 |
+
"num_truncated_few_shots": 0
|
| 1043 |
+
},
|
| 1044 |
+
"harness|bbh:reasoning_about_colored_objects|3": {
|
| 1045 |
+
"hashes": {
|
| 1046 |
+
"hash_examples": "a6ba328c4c3385d2",
|
| 1047 |
+
"hash_full_prompts": "3d2441a21c12a960",
|
| 1048 |
+
"hash_input_tokens": "f3b577892955aa84",
|
| 1049 |
+
"hash_cont_tokens": "e87bd4c4e98a3778"
|
| 1050 |
+
},
|
| 1051 |
+
"truncated": 250,
|
| 1052 |
+
"non_truncated": 0,
|
| 1053 |
+
"padded": 0,
|
| 1054 |
+
"non_padded": 250,
|
| 1055 |
+
"effective_few_shots": 3.0,
|
| 1056 |
+
"num_truncated_few_shots": 0
|
| 1057 |
+
},
|
| 1058 |
+
"harness|bbh:ruin_names|3": {
|
| 1059 |
+
"hashes": {
|
| 1060 |
+
"hash_examples": "2ef28d5f2d4fdd25",
|
| 1061 |
+
"hash_full_prompts": "ba95caa786f313b1",
|
| 1062 |
+
"hash_input_tokens": "9954b30d4205604a",
|
| 1063 |
+
"hash_cont_tokens": "1d43f73a5b320bd5"
|
| 1064 |
+
},
|
| 1065 |
+
"truncated": 248,
|
| 1066 |
+
"non_truncated": 0,
|
| 1067 |
+
"padded": 0,
|
| 1068 |
+
"non_padded": 248,
|
| 1069 |
+
"effective_few_shots": 3.0,
|
| 1070 |
+
"num_truncated_few_shots": 0
|
| 1071 |
+
},
|
| 1072 |
+
"harness|bbh:salient_translation_error_detection|3": {
|
| 1073 |
+
"hashes": {
|
| 1074 |
+
"hash_examples": "c13f25ec8ffed496",
|
| 1075 |
+
"hash_full_prompts": "a8512d174e1cab8f",
|
| 1076 |
+
"hash_input_tokens": "3e738df24b7eddf8",
|
| 1077 |
+
"hash_cont_tokens": "c134887984e0a3e3"
|
| 1078 |
+
},
|
| 1079 |
+
"truncated": 250,
|
| 1080 |
+
"non_truncated": 0,
|
| 1081 |
+
"padded": 0,
|
| 1082 |
+
"non_padded": 250,
|
| 1083 |
+
"effective_few_shots": 3.0,
|
| 1084 |
+
"num_truncated_few_shots": 0
|
| 1085 |
+
},
|
| 1086 |
+
"harness|bbh:snarks|3": {
|
| 1087 |
+
"hashes": {
|
| 1088 |
+
"hash_examples": "5f6db7bff7f6f22e",
|
| 1089 |
+
"hash_full_prompts": "ff91d81466b9041f",
|
| 1090 |
+
"hash_input_tokens": "21388b09e13d0208",
|
| 1091 |
+
"hash_cont_tokens": "cb0d472786a5109c"
|
| 1092 |
+
},
|
| 1093 |
+
"truncated": 178,
|
| 1094 |
+
"non_truncated": 0,
|
| 1095 |
+
"padded": 0,
|
| 1096 |
+
"non_padded": 178,
|
| 1097 |
+
"effective_few_shots": 3.0,
|
| 1098 |
+
"num_truncated_few_shots": 0
|
| 1099 |
+
},
|
| 1100 |
+
"harness|bbh:sports_understanding|3": {
|
| 1101 |
+
"hashes": {
|
| 1102 |
+
"hash_examples": "042afbe5d9c1f02d",
|
| 1103 |
+
"hash_full_prompts": "a59324d9eb37e0f5",
|
| 1104 |
+
"hash_input_tokens": "0ad41bb8d2290a5b",
|
| 1105 |
+
"hash_cont_tokens": "d597f572759be385"
|
| 1106 |
+
},
|
| 1107 |
+
"truncated": 250,
|
| 1108 |
+
"non_truncated": 0,
|
| 1109 |
+
"padded": 0,
|
| 1110 |
+
"non_padded": 250,
|
| 1111 |
+
"effective_few_shots": 3.0,
|
| 1112 |
+
"num_truncated_few_shots": 0
|
| 1113 |
+
},
|
| 1114 |
+
"harness|bbh:temporal_sequences|3": {
|
| 1115 |
+
"hashes": {
|
| 1116 |
+
"hash_examples": "803a05f352eb6afc",
|
| 1117 |
+
"hash_full_prompts": "1b3971192bf481e7",
|
| 1118 |
+
"hash_input_tokens": "3051b60940ccceab",
|
| 1119 |
+
"hash_cont_tokens": "7f65864d92d59a9a"
|
| 1120 |
+
},
|
| 1121 |
+
"truncated": 250,
|
| 1122 |
+
"non_truncated": 0,
|
| 1123 |
+
"padded": 0,
|
| 1124 |
+
"non_padded": 250,
|
| 1125 |
+
"effective_few_shots": 3.0,
|
| 1126 |
+
"num_truncated_few_shots": 0
|
| 1127 |
+
},
|
| 1128 |
+
"harness|bbh:tracking_shuffled_objects_five_objects|3": {
|
| 1129 |
+
"hashes": {
|
| 1130 |
+
"hash_examples": "2bbac6db7ab0d527",
|
| 1131 |
+
"hash_full_prompts": "7ef4567d2fcf5094",
|
| 1132 |
+
"hash_input_tokens": "b841310ee5531238",
|
| 1133 |
+
"hash_cont_tokens": "76b08e70815da010"
|
| 1134 |
+
},
|
| 1135 |
+
"truncated": 250,
|
| 1136 |
+
"non_truncated": 0,
|
| 1137 |
+
"padded": 0,
|
| 1138 |
+
"non_padded": 250,
|
| 1139 |
+
"effective_few_shots": 3.0,
|
| 1140 |
+
"num_truncated_few_shots": 0
|
| 1141 |
+
},
|
| 1142 |
+
"harness|bbh:tracking_shuffled_objects_seven_objects|3": {
|
| 1143 |
+
"hashes": {
|
| 1144 |
+
"hash_examples": "845caf093ac2b58c",
|
| 1145 |
+
"hash_full_prompts": "196a0f8712857624",
|
| 1146 |
+
"hash_input_tokens": "3e738df24b7eddf8",
|
| 1147 |
+
"hash_cont_tokens": "ffbf705ca9254165"
|
| 1148 |
+
},
|
| 1149 |
+
"truncated": 250,
|
| 1150 |
+
"non_truncated": 0,
|
| 1151 |
+
"padded": 0,
|
| 1152 |
+
"non_padded": 250,
|
| 1153 |
+
"effective_few_shots": 3.0,
|
| 1154 |
+
"num_truncated_few_shots": 0
|
| 1155 |
+
},
|
| 1156 |
+
"harness|bbh:tracking_shuffled_objects_three_objects|3": {
|
| 1157 |
+
"hashes": {
|
| 1158 |
+
"hash_examples": "9004f14d5a32b9a8",
|
| 1159 |
+
"hash_full_prompts": "592a03f0518f17b6",
|
| 1160 |
+
"hash_input_tokens": "19e0ef1dd5ae9d33",
|
| 1161 |
+
"hash_cont_tokens": "feef4dd2052babcc"
|
| 1162 |
+
},
|
| 1163 |
+
"truncated": 250,
|
| 1164 |
+
"non_truncated": 0,
|
| 1165 |
+
"padded": 0,
|
| 1166 |
+
"non_padded": 250,
|
| 1167 |
+
"effective_few_shots": 3.0,
|
| 1168 |
+
"num_truncated_few_shots": 0
|
| 1169 |
+
}
|
| 1170 |
+
},
|
| 1171 |
+
"summary_general": {
|
| 1172 |
+
"hashes": {
|
| 1173 |
+
"hash_examples": "4ff1e3dc5703575d",
|
| 1174 |
+
"hash_full_prompts": "0d80ce968d89d4ef",
|
| 1175 |
+
"hash_input_tokens": "72bda1e7aeb34786",
|
| 1176 |
+
"hash_cont_tokens": "67741ca54caa3be8"
|
| 1177 |
+
},
|
| 1178 |
+
"truncated": 4362,
|
| 1179 |
+
"non_truncated": 0,
|
| 1180 |
+
"padded": 0,
|
| 1181 |
+
"non_padded": 4362,
|
| 1182 |
+
"num_truncated_few_shots": 0
|
| 1183 |
+
}
|
| 1184 |
+
}
|