data_source
stringclasses
1 value
prompt
listlengths
2
2
ability
stringclasses
1 value
reward_model
dict
extra_info
dict
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_ny_medical_genetics__5c80ef46__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 1038.6524452057367, \"nll_base\": 6.945679426193237, \"eval_ref\": \"global_mmlu_full_ny_medical_genetics__5c80ef46__s000__tb655...
{ "index": "global_mmlu_full_ny_medical_genetics__5c80ef46__s000__tb65536__v000", "uid": "global_mmlu_full_ny_medical_genetics__5c80ef46__s000__tb65536__v000", "axis": "domain", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_ny_professional_psychology__509f8fdb__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 909.7635568924817, \"nll_base\": 6.81318473815918, \"eval_ref\": \"global_mmlu_full_ny_professional_psychology__509f8fdb_...
{ "index": "global_mmlu_full_ny_professional_psychology__509f8fdb__s000__tb65536__v000", "uid": "global_mmlu_full_ny_professional_psychology__509f8fdb__s000__tb65536__v000", "axis": "multilingual", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_pl_clinical_knowledge__be5e4493__s000__tb131072__v001\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 753.264083248276, \"nll_base\": 6.624415874481201, \"eval_ref\": \"global_mmlu_full_pl_clinical_knowledge__be5e4493__s000__t...
{ "index": "global_mmlu_full_pl_clinical_knowledge__be5e4493__s000__tb131072__v001", "uid": "global_mmlu_full_pl_clinical_knowledge__be5e4493__s000__tb131072__v001", "axis": "domain", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_pl_clinical_knowledge__be5e4493__s000__tb262144__v002\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 753.264083248276, \"nll_base\": 6.624415874481201, \"eval_ref\": \"global_mmlu_full_pl_clinical_knowledge__be5e4493__s000__t...
{ "index": "global_mmlu_full_pl_clinical_knowledge__be5e4493__s000__tb262144__v002", "uid": "global_mmlu_full_pl_clinical_knowledge__be5e4493__s000__tb262144__v002", "axis": "domain", "budget": 2048, "token_budget": 262144, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_pl_clinical_knowledge__be5e4493__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 753.264083248276, \"nll_base\": 6.624415874481201, \"eval_ref\": \"global_mmlu_full_pl_clinical_knowledge__be5e4493__s000__tb6...
{ "index": "global_mmlu_full_pl_clinical_knowledge__be5e4493__s000__tb65536__v000", "uid": "global_mmlu_full_pl_clinical_knowledge__be5e4493__s000__tb65536__v000", "axis": "domain", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_pl_econometrics__ca29f20b__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 774.9356983186668, \"nll_base\": 6.652780055999756, \"eval_ref\": \"global_mmlu_full_pl_econometrics__ca29f20b__s000__tb131072__v0...
{ "index": "global_mmlu_full_pl_econometrics__ca29f20b__s000__tb131072__v000", "uid": "global_mmlu_full_pl_econometrics__ca29f20b__s000__tb131072__v000", "axis": "multilingual", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_pl_marketing__11812117__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 595.1169519364546, \"nll_base\": 6.388757944107056, \"eval_ref\": \"global_mmlu_full_pl_marketing__11812117__s000__tb65536__v000\", \"o...
{ "index": "global_mmlu_full_pl_marketing__11812117__s000__tb65536__v000", "uid": "global_mmlu_full_pl_marketing__11812117__s000__tb65536__v000", "axis": "multilingual", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_pl_medical_genetics__9e947a20__s000__tb131072__v002\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 778.8658486983588, \"nll_base\": 6.657838821411133, \"eval_ref\": \"global_mmlu_full_pl_medical_genetics__9e947a20__s000__tb13...
{ "index": "global_mmlu_full_pl_medical_genetics__9e947a20__s000__tb131072__v002", "uid": "global_mmlu_full_pl_medical_genetics__9e947a20__s000__tb131072__v002", "axis": "domain", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_pl_medical_genetics__9e947a20__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 778.8658486983588, \"nll_base\": 6.657838821411133, \"eval_ref\": \"global_mmlu_full_pl_medical_genetics__9e947a20__s000__tb26...
{ "index": "global_mmlu_full_pl_medical_genetics__9e947a20__s000__tb262144__v000", "uid": "global_mmlu_full_pl_medical_genetics__9e947a20__s000__tb262144__v000", "axis": "domain", "budget": 2048, "token_budget": 262144, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_pl_medical_genetics__9e947a20__s000__tb65536__v001\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 778.8658486983588, \"nll_base\": 6.657838821411133, \"eval_ref\": \"global_mmlu_full_pl_medical_genetics__9e947a20__s000__tb6553...
{ "index": "global_mmlu_full_pl_medical_genetics__9e947a20__s000__tb65536__v001", "uid": "global_mmlu_full_pl_medical_genetics__9e947a20__s000__tb65536__v001", "axis": "domain", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_pl_professional_psychology__1b497f60__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 537.4299775408917, \"nll_base\": 6.286798477172852, \"eval_ref\": \"global_mmlu_full_pl_professional_psychology__1b497f60...
{ "index": "global_mmlu_full_pl_professional_psychology__1b497f60__s000__tb65536__v000", "uid": "global_mmlu_full_pl_professional_psychology__1b497f60__s000__tb65536__v000", "axis": "multilingual", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_pl_public_relations__0e7f41c6__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 1072.7416974216394, \"nll_base\": 6.977972984313965, \"eval_ref\": \"global_mmlu_full_pl_public_relations__0e7f41c6__s000__tb2...
{ "index": "global_mmlu_full_pl_public_relations__0e7f41c6__s000__tb262144__v000", "uid": "global_mmlu_full_pl_public_relations__0e7f41c6__s000__tb262144__v000", "axis": "multilingual", "budget": 2048, "token_budget": 262144, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_pl_security_studies__98963288__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 711.7886101885546, \"nll_base\": 6.5677809715271, \"eval_ref\": \"global_mmlu_full_pl_security_studies__98963288__s000__tb2621...
{ "index": "global_mmlu_full_pl_security_studies__98963288__s000__tb262144__v000", "uid": "global_mmlu_full_pl_security_studies__98963288__s000__tb262144__v000", "axis": "multilingual", "budget": 2048, "token_budget": 262144, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_pl_sociology__ed54c979__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 671.04280750638, \"nll_base\": 6.508832931518555, \"eval_ref\": \"global_mmlu_full_pl_sociology__ed54c979__s000__tb131072__v000\", \"...
{ "index": "global_mmlu_full_pl_sociology__ed54c979__s000__tb131072__v000", "uid": "global_mmlu_full_pl_sociology__ed54c979__s000__tb131072__v000", "axis": "multilingual", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_pt_clinical_knowledge__b8d0cca6__s000__tb131072__v001\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 1034.216912138102, \"nll_base\": 6.941399812698364, \"eval_ref\": \"global_mmlu_full_pt_clinical_knowledge__b8d0cca6__s000__...
{ "index": "global_mmlu_full_pt_clinical_knowledge__b8d0cca6__s000__tb131072__v001", "uid": "global_mmlu_full_pt_clinical_knowledge__b8d0cca6__s000__tb131072__v001", "axis": "domain", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_pt_clinical_knowledge__b8d0cca6__s000__tb262144__v002\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 1034.216912138102, \"nll_base\": 6.941399812698364, \"eval_ref\": \"global_mmlu_full_pt_clinical_knowledge__b8d0cca6__s000__...
{ "index": "global_mmlu_full_pt_clinical_knowledge__b8d0cca6__s000__tb262144__v002", "uid": "global_mmlu_full_pt_clinical_knowledge__b8d0cca6__s000__tb262144__v002", "axis": "domain", "budget": 2048, "token_budget": 262144, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_pt_clinical_knowledge__b8d0cca6__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 1034.216912138102, \"nll_base\": 6.941399812698364, \"eval_ref\": \"global_mmlu_full_pt_clinical_knowledge__b8d0cca6__s000__tb...
{ "index": "global_mmlu_full_pt_clinical_knowledge__b8d0cca6__s000__tb65536__v000", "uid": "global_mmlu_full_pt_clinical_knowledge__b8d0cca6__s000__tb65536__v000", "axis": "domain", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_pt_clinical_knowledge__b8d0cca6__s001__tb65536__v003\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 943.6099089181162, \"nll_base\": 6.84971284866333, \"eval_ref\": \"global_mmlu_full_pt_clinical_knowledge__b8d0cca6__s001__tb6...
{ "index": "global_mmlu_full_pt_clinical_knowledge__b8d0cca6__s001__tb65536__v003", "uid": "global_mmlu_full_pt_clinical_knowledge__b8d0cca6__s001__tb65536__v003", "axis": "domain", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_pt_high_school_european_history__19495ef0__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 496.8904073765885, \"nll_base\": 6.208369493484497, \"eval_ref\": \"global_mmlu_full_pt_high_school_european_histo...
{ "index": "global_mmlu_full_pt_high_school_european_history__19495ef0__s000__tb131072__v000", "uid": "global_mmlu_full_pt_high_school_european_history__19495ef0__s000__tb131072__v000", "axis": "multilingual", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_pt_high_school_macroeconomics__d20ad1b7__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 738.5026927668455, \"nll_base\": 6.6046247482299805, \"eval_ref\": \"global_mmlu_full_pt_high_school_macroeconomics_...
{ "index": "global_mmlu_full_pt_high_school_macroeconomics__d20ad1b7__s000__tb262144__v000", "uid": "global_mmlu_full_pt_high_school_macroeconomics__d20ad1b7__s000__tb262144__v000", "axis": "multilingual", "budget": 2048, "token_budget": 262144, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_pt_high_school_world_history__74ed3e36__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 488.42128474132033, \"nll_base\": 6.191178321838379, \"eval_ref\": \"global_mmlu_full_pt_high_school_world_history__7...
{ "index": "global_mmlu_full_pt_high_school_world_history__74ed3e36__s000__tb131072__v000", "uid": "global_mmlu_full_pt_high_school_world_history__74ed3e36__s000__tb131072__v000", "axis": "multilingual", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_pt_human_sexuality__f4953cd0__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 1034.83773151383, \"nll_base\": 6.941999912261963, \"eval_ref\": \"global_mmlu_full_pt_human_sexuality__f4953cd0__s000__tb13107...
{ "index": "global_mmlu_full_pt_human_sexuality__f4953cd0__s000__tb131072__v000", "uid": "global_mmlu_full_pt_human_sexuality__f4953cd0__s000__tb131072__v000", "axis": "multilingual", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_pt_international_law__dc363466__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 675.3254038348311, \"nll_base\": 6.515194654464722, \"eval_ref\": \"global_mmlu_full_pt_international_law__dc363466__s000__tb...
{ "index": "global_mmlu_full_pt_international_law__dc363466__s000__tb131072__v000", "uid": "global_mmlu_full_pt_international_law__dc363466__s000__tb131072__v000", "axis": "multilingual", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_pt_jurisprudence__685d65a8__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 784.6054041701036, \"nll_base\": 6.665180921554565, \"eval_ref\": \"global_mmlu_full_pt_jurisprudence__685d65a8__s000__tb65536__v00...
{ "index": "global_mmlu_full_pt_jurisprudence__685d65a8__s000__tb65536__v000", "uid": "global_mmlu_full_pt_jurisprudence__685d65a8__s000__tb65536__v000", "axis": "multilingual", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_pt_moral_disputes__2b6bbcb6__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 623.7653938775896, \"nll_base\": 6.435774326324463, \"eval_ref\": \"global_mmlu_full_pt_moral_disputes__2b6bbcb6__s000__tb65536__v...
{ "index": "global_mmlu_full_pt_moral_disputes__2b6bbcb6__s000__tb65536__v000", "uid": "global_mmlu_full_pt_moral_disputes__2b6bbcb6__s000__tb65536__v000", "axis": "multilingual", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_pt_public_relations__69482e63__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 1219.3470102203112, \"nll_base\": 7.1060707569122314, \"eval_ref\": \"global_mmlu_full_pt_public_relations__69482e63__s000__tb...
{ "index": "global_mmlu_full_pt_public_relations__69482e63__s000__tb262144__v000", "uid": "global_mmlu_full_pt_public_relations__69482e63__s000__tb262144__v000", "axis": "multilingual", "budget": 2048, "token_budget": 262144, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_pt_security_studies__09809857__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 1113.004274209546, \"nll_base\": 7.01481819152832, \"eval_ref\": \"global_mmlu_full_pt_security_studies__09809857__s000__tb131...
{ "index": "global_mmlu_full_pt_security_studies__09809857__s000__tb131072__v000", "uid": "global_mmlu_full_pt_security_studies__09809857__s000__tb131072__v000", "axis": "multilingual", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_pt_sociology__c7caad4d__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 476.0643779862588, \"nll_base\": 6.165553092956543, \"eval_ref\": \"global_mmlu_full_pt_sociology__c7caad4d__s000__tb131072__v000\", ...
{ "index": "global_mmlu_full_pt_sociology__c7caad4d__s000__tb131072__v000", "uid": "global_mmlu_full_pt_sociology__c7caad4d__s000__tb131072__v000", "axis": "multilingual", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_pt_us_foreign_policy__4506d710__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 677.8373135141092, \"nll_base\": 6.518907308578491, \"eval_ref\": \"global_mmlu_full_pt_us_foreign_policy__4506d710__s000__tb...
{ "index": "global_mmlu_full_pt_us_foreign_policy__4506d710__s000__tb262144__v000", "uid": "global_mmlu_full_pt_us_foreign_policy__4506d710__s000__tb262144__v000", "axis": "multilingual", "budget": 2048, "token_budget": 262144, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_pt_world_religions__f87ce187__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 1668.614668830826, \"nll_base\": 7.419749021530151, \"eval_ref\": \"global_mmlu_full_pt_world_religions__f87ce187__s000__tb1310...
{ "index": "global_mmlu_full_pt_world_religions__f87ce187__s000__tb131072__v000", "uid": "global_mmlu_full_pt_world_religions__f87ce187__s000__tb131072__v000", "axis": "multilingual", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_ro_clinical_knowledge__a0d0deb3__s000__tb131072__v001\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 468.44807738022223, \"nll_base\": 6.149425268173218, \"eval_ref\": \"global_mmlu_full_ro_clinical_knowledge__a0d0deb3__s000_...
{ "index": "global_mmlu_full_ro_clinical_knowledge__a0d0deb3__s000__tb131072__v001", "uid": "global_mmlu_full_ro_clinical_knowledge__a0d0deb3__s000__tb131072__v001", "axis": "domain", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_ro_clinical_knowledge__a0d0deb3__s000__tb262144__v002\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 468.44807738022223, \"nll_base\": 6.149425268173218, \"eval_ref\": \"global_mmlu_full_ro_clinical_knowledge__a0d0deb3__s000_...
{ "index": "global_mmlu_full_ro_clinical_knowledge__a0d0deb3__s000__tb262144__v002", "uid": "global_mmlu_full_ro_clinical_knowledge__a0d0deb3__s000__tb262144__v002", "axis": "domain", "budget": 2048, "token_budget": 262144, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_ro_clinical_knowledge__a0d0deb3__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 468.44807738022223, \"nll_base\": 6.149425268173218, \"eval_ref\": \"global_mmlu_full_ro_clinical_knowledge__a0d0deb3__s000__t...
{ "index": "global_mmlu_full_ro_clinical_knowledge__a0d0deb3__s000__tb65536__v000", "uid": "global_mmlu_full_ro_clinical_knowledge__a0d0deb3__s000__tb65536__v000", "axis": "domain", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_ro_clinical_knowledge__a0d0deb3__s001__tb65536__v003\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 438.88402198317993, \"nll_base\": 6.084235191345215, \"eval_ref\": \"global_mmlu_full_ro_clinical_knowledge__a0d0deb3__s001__t...
{ "index": "global_mmlu_full_ro_clinical_knowledge__a0d0deb3__s001__tb65536__v003", "uid": "global_mmlu_full_ro_clinical_knowledge__a0d0deb3__s001__tb65536__v003", "axis": "domain", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_ro_college_chemistry__e662d654__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 706.4927826090096, \"nll_base\": 6.560312986373901, \"eval_ref\": \"global_mmlu_full_ro_college_chemistry__e662d654__s000__tb65...
{ "index": "global_mmlu_full_ro_college_chemistry__e662d654__s000__tb65536__v000", "uid": "global_mmlu_full_ro_college_chemistry__e662d654__s000__tb65536__v000", "axis": "multilingual", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_ro_high_school_government_and_politics__15bea0ed__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 384.4972826927565, \"nll_base\": 5.951936721801758, \"eval_ref\": \"global_mmlu_full_ro_high_school_governm...
{ "index": "global_mmlu_full_ro_high_school_government_and_politics__15bea0ed__s000__tb262144__v000", "uid": "global_mmlu_full_ro_high_school_government_and_politics__15bea0ed__s000__tb262144__v000", "axis": "multilingual", "budget": 2048, "token_budget": 262144, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_ro_high_school_macroeconomics__7837bbd1__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 400.49863246927487, \"nll_base\": 5.99271035194397, \"eval_ref\": \"global_mmlu_full_ro_high_school_macroeconomics__...
{ "index": "global_mmlu_full_ro_high_school_macroeconomics__7837bbd1__s000__tb262144__v000", "uid": "global_mmlu_full_ro_high_school_macroeconomics__7837bbd1__s000__tb262144__v000", "axis": "multilingual", "budget": 2048, "token_budget": 262144, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_ro_human_aging__c4f8cc8c__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 659.9934683334823, \"nll_base\": 6.49222993850708, \"eval_ref\": \"global_mmlu_full_ro_human_aging__c4f8cc8c__s000__tb262144__v000\...
{ "index": "global_mmlu_full_ro_human_aging__c4f8cc8c__s000__tb262144__v000", "uid": "global_mmlu_full_ro_human_aging__c4f8cc8c__s000__tb262144__v000", "axis": "multilingual", "budget": 2048, "token_budget": 262144, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_ro_medical_genetics__6395f37b__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 510.746906053223, \"nll_base\": 6.235874176025391, \"eval_ref\": \"global_mmlu_full_ro_medical_genetics__6395f37b__s000__tb131...
{ "index": "global_mmlu_full_ro_medical_genetics__6395f37b__s000__tb131072__v000", "uid": "global_mmlu_full_ro_medical_genetics__6395f37b__s000__tb131072__v000", "axis": "domain", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_ro_medical_genetics__6395f37b__s000__tb262144__v001\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 510.746906053223, \"nll_base\": 6.235874176025391, \"eval_ref\": \"global_mmlu_full_ro_medical_genetics__6395f37b__s000__tb262...
{ "index": "global_mmlu_full_ro_medical_genetics__6395f37b__s000__tb262144__v001", "uid": "global_mmlu_full_ro_medical_genetics__6395f37b__s000__tb262144__v001", "axis": "domain", "budget": 2048, "token_budget": 262144, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_ro_medical_genetics__6395f37b__s000__tb65536__v002\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 510.746906053223, \"nll_base\": 6.235874176025391, \"eval_ref\": \"global_mmlu_full_ro_medical_genetics__6395f37b__s000__tb65536...
{ "index": "global_mmlu_full_ro_medical_genetics__6395f37b__s000__tb65536__v002", "uid": "global_mmlu_full_ro_medical_genetics__6395f37b__s000__tb65536__v002", "axis": "domain", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_ro_medical_genetics__6395f37b__s001__tb131072__v003\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 598.4953265906792, \"nll_base\": 6.394418716430664, \"eval_ref\": \"global_mmlu_full_ro_medical_genetics__6395f37b__s001__tb13...
{ "index": "global_mmlu_full_ro_medical_genetics__6395f37b__s001__tb131072__v003", "uid": "global_mmlu_full_ro_medical_genetics__6395f37b__s001__tb131072__v003", "axis": "domain", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_ro_moral_disputes__1d5b21c4__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 487.8817267716819, \"nll_base\": 6.190073013305664, \"eval_ref\": \"global_mmlu_full_ro_moral_disputes__1d5b21c4__s000__tb131072...
{ "index": "global_mmlu_full_ro_moral_disputes__1d5b21c4__s000__tb131072__v000", "uid": "global_mmlu_full_ro_moral_disputes__1d5b21c4__s000__tb131072__v000", "axis": "multilingual", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_ro_professional_accounting__2b32e1cc__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 421.6077962221457, \"nll_base\": 6.0440754890441895, \"eval_ref\": \"global_mmlu_full_ro_professional_accounting__2b32e...
{ "index": "global_mmlu_full_ro_professional_accounting__2b32e1cc__s000__tb262144__v000", "uid": "global_mmlu_full_ro_professional_accounting__2b32e1cc__s000__tb262144__v000", "axis": "multilingual", "budget": 2048, "token_budget": 262144, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_ro_public_relations__9a17d998__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 591.9752420999741, \"nll_base\": 6.383464813232422, \"eval_ref\": \"global_mmlu_full_ro_public_relations__9a17d998__s000__tb13...
{ "index": "global_mmlu_full_ro_public_relations__9a17d998__s000__tb131072__v000", "uid": "global_mmlu_full_ro_public_relations__9a17d998__s000__tb131072__v000", "axis": "multilingual", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_ru_clinical_knowledge__fa80361d__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 1237.2131559750892, \"nll_base\": 7.120616674423218, \"eval_ref\": \"global_mmlu_full_ru_clinical_knowledge__fa80361d__s000_...
{ "index": "global_mmlu_full_ru_clinical_knowledge__fa80361d__s000__tb131072__v000", "uid": "global_mmlu_full_ru_clinical_knowledge__fa80361d__s000__tb131072__v000", "axis": "domain", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_ru_clinical_knowledge__fa80361d__s000__tb65536__v002\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 1237.2131559750892, \"nll_base\": 7.120616674423218, \"eval_ref\": \"global_mmlu_full_ru_clinical_knowledge__fa80361d__s000__t...
{ "index": "global_mmlu_full_ru_clinical_knowledge__fa80361d__s000__tb65536__v002", "uid": "global_mmlu_full_ru_clinical_knowledge__fa80361d__s000__tb65536__v002", "axis": "domain", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_ru_clinical_knowledge__fa80361d__s001__tb131072__v003\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 1516.3926356030058, \"nll_base\": 7.324089527130127, \"eval_ref\": \"global_mmlu_full_ru_clinical_knowledge__fa80361d__s001_...
{ "index": "global_mmlu_full_ru_clinical_knowledge__fa80361d__s001__tb131072__v003", "uid": "global_mmlu_full_ru_clinical_knowledge__fa80361d__s001__tb131072__v003", "axis": "domain", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_ru_formal_logic__c7031fbd__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 768.246343290272, \"nll_base\": 6.644110441207886, \"eval_ref\": \"global_mmlu_full_ru_formal_logic__c7031fbd__s000__tb131072__v00...
{ "index": "global_mmlu_full_ru_formal_logic__c7031fbd__s000__tb131072__v000", "uid": "global_mmlu_full_ru_formal_logic__c7031fbd__s000__tb131072__v000", "axis": "multilingual", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_ru_global_facts__d45aa0c0__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 1193.108596165242, \"nll_base\": 7.084317445755005, \"eval_ref\": \"global_mmlu_full_ru_global_facts__d45aa0c0__s000__tb262144__v0...
{ "index": "global_mmlu_full_ru_global_facts__d45aa0c0__s000__tb262144__v000", "uid": "global_mmlu_full_ru_global_facts__d45aa0c0__s000__tb262144__v000", "axis": "multilingual", "budget": 2048, "token_budget": 262144, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_ru_medical_genetics__4684bd86__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 1478.4174621836264, \"nll_base\": 7.298727512359619, \"eval_ref\": \"global_mmlu_full_ru_medical_genetics__4684bd86__s000__tb1...
{ "index": "global_mmlu_full_ru_medical_genetics__4684bd86__s000__tb131072__v000", "uid": "global_mmlu_full_ru_medical_genetics__4684bd86__s000__tb131072__v000", "axis": "domain", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_ru_medical_genetics__4684bd86__s000__tb65536__v002\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 1478.4174621836264, \"nll_base\": 7.298727512359619, \"eval_ref\": \"global_mmlu_full_ru_medical_genetics__4684bd86__s000__tb655...
{ "index": "global_mmlu_full_ru_medical_genetics__4684bd86__s000__tb65536__v002", "uid": "global_mmlu_full_ru_medical_genetics__4684bd86__s000__tb65536__v002", "axis": "domain", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_ru_medical_genetics__4684bd86__s001__tb131072__v003\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 1578.3740662029425, \"nll_base\": 7.364150524139404, \"eval_ref\": \"global_mmlu_full_ru_medical_genetics__4684bd86__s001__tb1...
{ "index": "global_mmlu_full_ru_medical_genetics__4684bd86__s001__tb131072__v003", "uid": "global_mmlu_full_ru_medical_genetics__4684bd86__s001__tb131072__v003", "axis": "domain", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_ru_nutrition__2f95f38f__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 1236.912613364616, \"nll_base\": 7.120373725891113, \"eval_ref\": \"global_mmlu_full_ru_nutrition__2f95f38f__s000__tb65536__v000\", \"o...
{ "index": "global_mmlu_full_ru_nutrition__2f95f38f__s000__tb65536__v000", "uid": "global_mmlu_full_ru_nutrition__2f95f38f__s000__tb65536__v000", "axis": "multilingual", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_si_medical_genetics__87870b47__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 689.9985663822083, \"nll_base\": 6.536689519882202, \"eval_ref\": \"global_mmlu_full_si_medical_genetics__87870b47__s000__tb13...
{ "index": "global_mmlu_full_si_medical_genetics__87870b47__s000__tb131072__v000", "uid": "global_mmlu_full_si_medical_genetics__87870b47__s000__tb131072__v000", "axis": "domain", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_si_medical_genetics__87870b47__s000__tb262144__v001\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 689.9985663822083, \"nll_base\": 6.536689519882202, \"eval_ref\": \"global_mmlu_full_si_medical_genetics__87870b47__s000__tb26...
{ "index": "global_mmlu_full_si_medical_genetics__87870b47__s000__tb262144__v001", "uid": "global_mmlu_full_si_medical_genetics__87870b47__s000__tb262144__v001", "axis": "domain", "budget": 2048, "token_budget": 262144, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_si_medical_genetics__87870b47__s000__tb65536__v002\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 689.9985663822083, \"nll_base\": 6.536689519882202, \"eval_ref\": \"global_mmlu_full_si_medical_genetics__87870b47__s000__tb6553...
{ "index": "global_mmlu_full_si_medical_genetics__87870b47__s000__tb65536__v002", "uid": "global_mmlu_full_si_medical_genetics__87870b47__s000__tb65536__v002", "axis": "domain", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_sn_anatomy__bd299b27__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 973.8045007371744, \"nll_base\": 6.881210565567017, \"eval_ref\": \"global_mmlu_full_sn_anatomy__bd299b27__s000__tb65536__v000\", \"oracl...
{ "index": "global_mmlu_full_sn_anatomy__bd299b27__s000__tb65536__v000", "uid": "global_mmlu_full_sn_anatomy__bd299b27__s000__tb65536__v000", "axis": "multilingual", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_sn_elementary_mathematics__1eb47f34__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 1383.599959688597, \"nll_base\": 7.2324440479278564, \"eval_ref\": \"global_mmlu_full_sn_elementary_mathematics__1eb47f34_...
{ "index": "global_mmlu_full_sn_elementary_mathematics__1eb47f34__s000__tb65536__v000", "uid": "global_mmlu_full_sn_elementary_mathematics__1eb47f34__s000__tb65536__v000", "axis": "multilingual", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_sn_high_school_biology__1c7abc2c__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 859.5247363505483, \"nll_base\": 6.7563796043396, \"eval_ref\": \"global_mmlu_full_sn_high_school_biology__1c7abc2c__s000__...
{ "index": "global_mmlu_full_sn_high_school_biology__1c7abc2c__s000__tb131072__v000", "uid": "global_mmlu_full_sn_high_school_biology__1c7abc2c__s000__tb131072__v000", "axis": "multilingual", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_sn_high_school_chemistry__f0a5b674__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 855.5748616449043, \"nll_base\": 6.7517735958099365, \"eval_ref\": \"global_mmlu_full_sn_high_school_chemistry__f0a5b674_...
{ "index": "global_mmlu_full_sn_high_school_chemistry__f0a5b674__s000__tb262144__v000", "uid": "global_mmlu_full_sn_high_school_chemistry__f0a5b674__s000__tb262144__v000", "axis": "multilingual", "budget": 2048, "token_budget": 262144, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_sn_high_school_macroeconomics__037b1e8d__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 769.4242585480388, \"nll_base\": 6.645642518997192, \"eval_ref\": \"global_mmlu_full_sn_high_school_macroeconomics__03...
{ "index": "global_mmlu_full_sn_high_school_macroeconomics__037b1e8d__s000__tb65536__v000", "uid": "global_mmlu_full_sn_high_school_macroeconomics__037b1e8d__s000__tb65536__v000", "axis": "multilingual", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_sn_human_aging__7ede38be__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 971.3426362506683, \"nll_base\": 6.878679275512695, \"eval_ref\": \"global_mmlu_full_sn_human_aging__7ede38be__s000__tb262144__v000...
{ "index": "global_mmlu_full_sn_human_aging__7ede38be__s000__tb262144__v000", "uid": "global_mmlu_full_sn_human_aging__7ede38be__s000__tb262144__v000", "axis": "multilingual", "budget": 2048, "token_budget": 262144, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_sn_medical_genetics__1f49033c__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 898.489026303966, \"nll_base\": 6.800714492797852, \"eval_ref\": \"global_mmlu_full_sn_medical_genetics__1f49033c__s000__tb131...
{ "index": "global_mmlu_full_sn_medical_genetics__1f49033c__s000__tb131072__v000", "uid": "global_mmlu_full_sn_medical_genetics__1f49033c__s000__tb131072__v000", "axis": "domain", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_sn_medical_genetics__1f49033c__s000__tb262144__v001\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 898.489026303966, \"nll_base\": 6.800714492797852, \"eval_ref\": \"global_mmlu_full_sn_medical_genetics__1f49033c__s000__tb262...
{ "index": "global_mmlu_full_sn_medical_genetics__1f49033c__s000__tb262144__v001", "uid": "global_mmlu_full_sn_medical_genetics__1f49033c__s000__tb262144__v001", "axis": "domain", "budget": 2048, "token_budget": 262144, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_sn_medical_genetics__1f49033c__s000__tb65536__v002\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 898.489026303966, \"nll_base\": 6.800714492797852, \"eval_ref\": \"global_mmlu_full_sn_medical_genetics__1f49033c__s000__tb65536...
{ "index": "global_mmlu_full_sn_medical_genetics__1f49033c__s000__tb65536__v002", "uid": "global_mmlu_full_sn_medical_genetics__1f49033c__s000__tb65536__v002", "axis": "domain", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_sn_philosophy__0d2badaa__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 1130.9777174391577, \"nll_base\": 7.030837774276733, \"eval_ref\": \"global_mmlu_full_sn_philosophy__0d2badaa__s000__tb262144__v000\...
{ "index": "global_mmlu_full_sn_philosophy__0d2badaa__s000__tb262144__v000", "uid": "global_mmlu_full_sn_philosophy__0d2badaa__s000__tb262144__v000", "axis": "multilingual", "budget": 2048, "token_budget": 262144, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_sn_us_foreign_policy__1f4b87a3__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 1296.2834879630955, \"nll_base\": 7.167256593704224, \"eval_ref\": \"global_mmlu_full_sn_us_foreign_policy__1f4b87a3__s000__t...
{ "index": "global_mmlu_full_sn_us_foreign_policy__1f4b87a3__s000__tb131072__v000", "uid": "global_mmlu_full_sn_us_foreign_policy__1f4b87a3__s000__tb131072__v000", "axis": "multilingual", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_sn_virology__39f9777d__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 968.7772764758105, \"nll_base\": 6.876034736633301, \"eval_ref\": \"global_mmlu_full_sn_virology__39f9777d__s000__tb131072__v000\", \"...
{ "index": "global_mmlu_full_sn_virology__39f9777d__s000__tb131072__v000", "uid": "global_mmlu_full_sn_virology__39f9777d__s000__tb131072__v000", "axis": "multilingual", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_so_clinical_knowledge__c0b5cdad__s000__tb131072__v002\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 933.6080750195713, \"nll_base\": 6.839056730270386, \"eval_ref\": \"global_mmlu_full_so_clinical_knowledge__c0b5cdad__s000__...
{ "index": "global_mmlu_full_so_clinical_knowledge__c0b5cdad__s000__tb131072__v002", "uid": "global_mmlu_full_so_clinical_knowledge__c0b5cdad__s000__tb131072__v002", "axis": "domain", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_so_clinical_knowledge__c0b5cdad__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 933.6080750195713, \"nll_base\": 6.839056730270386, \"eval_ref\": \"global_mmlu_full_so_clinical_knowledge__c0b5cdad__s000__...
{ "index": "global_mmlu_full_so_clinical_knowledge__c0b5cdad__s000__tb262144__v000", "uid": "global_mmlu_full_so_clinical_knowledge__c0b5cdad__s000__tb262144__v000", "axis": "domain", "budget": 2048, "token_budget": 262144, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_so_clinical_knowledge__c0b5cdad__s000__tb65536__v001\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 933.6080750195713, \"nll_base\": 6.839056730270386, \"eval_ref\": \"global_mmlu_full_so_clinical_knowledge__c0b5cdad__s000__tb...
{ "index": "global_mmlu_full_so_clinical_knowledge__c0b5cdad__s000__tb65536__v001", "uid": "global_mmlu_full_so_clinical_knowledge__c0b5cdad__s000__tb65536__v001", "axis": "domain", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_so_clinical_knowledge__c0b5cdad__s001__tb262144__v003\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 884.2990384918925, \"nll_base\": 6.78479528427124, \"eval_ref\": \"global_mmlu_full_so_clinical_knowledge__c0b5cdad__s001__t...
{ "index": "global_mmlu_full_so_clinical_knowledge__c0b5cdad__s001__tb262144__v003", "uid": "global_mmlu_full_so_clinical_knowledge__c0b5cdad__s001__tb262144__v003", "axis": "domain", "budget": 2048, "token_budget": 262144, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_so_college_physics__e52f8e90__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 917.9332904452756, \"nll_base\": 6.822124719619751, \"eval_ref\": \"global_mmlu_full_so_college_physics__e52f8e90__s000__tb65536_...
{ "index": "global_mmlu_full_so_college_physics__e52f8e90__s000__tb65536__v000", "uid": "global_mmlu_full_so_college_physics__e52f8e90__s000__tb65536__v000", "axis": "multilingual", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_so_high_school_biology__b22444df__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 903.1878986786893, \"nll_base\": 6.8059306144714355, \"eval_ref\": \"global_mmlu_full_so_high_school_biology__b22444df__s000_...
{ "index": "global_mmlu_full_so_high_school_biology__b22444df__s000__tb65536__v000", "uid": "global_mmlu_full_so_high_school_biology__b22444df__s000__tb65536__v000", "axis": "multilingual", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_so_high_school_psychology__a738a41f__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 1157.203918255686, \"nll_base\": 7.053761959075928, \"eval_ref\": \"global_mmlu_full_so_high_school_psychology__a738a41f...
{ "index": "global_mmlu_full_so_high_school_psychology__a738a41f__s000__tb131072__v000", "uid": "global_mmlu_full_so_high_school_psychology__a738a41f__s000__tb131072__v000", "axis": "multilingual", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_so_human_sexuality__163a5447__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 1030.0132117792605, \"nll_base\": 6.937326908111572, \"eval_ref\": \"global_mmlu_full_so_human_sexuality__163a5447__s000__tb131...
{ "index": "global_mmlu_full_so_human_sexuality__163a5447__s000__tb131072__v000", "uid": "global_mmlu_full_so_human_sexuality__163a5447__s000__tb131072__v000", "axis": "multilingual", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_so_medical_genetics__41827449__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 998.5006145926277, \"nll_base\": 6.906254768371582, \"eval_ref\": \"global_mmlu_full_so_medical_genetics__41827449__s000__tb13...
{ "index": "global_mmlu_full_so_medical_genetics__41827449__s000__tb131072__v000", "uid": "global_mmlu_full_so_medical_genetics__41827449__s000__tb131072__v000", "axis": "domain", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_so_medical_genetics__41827449__s000__tb262144__v001\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 998.5006145926277, \"nll_base\": 6.906254768371582, \"eval_ref\": \"global_mmlu_full_so_medical_genetics__41827449__s000__tb26...
{ "index": "global_mmlu_full_so_medical_genetics__41827449__s000__tb262144__v001", "uid": "global_mmlu_full_so_medical_genetics__41827449__s000__tb262144__v001", "axis": "domain", "budget": 2048, "token_budget": 262144, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_so_medical_genetics__41827449__s000__tb65536__v002\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 998.5006145926277, \"nll_base\": 6.906254768371582, \"eval_ref\": \"global_mmlu_full_so_medical_genetics__41827449__s000__tb6553...
{ "index": "global_mmlu_full_so_medical_genetics__41827449__s000__tb65536__v002", "uid": "global_mmlu_full_so_medical_genetics__41827449__s000__tb65536__v002", "axis": "domain", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_so_medical_genetics__41827449__s001__tb131072__v003\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 1281.9186087306675, \"nll_base\": 7.156113147735596, \"eval_ref\": \"global_mmlu_full_so_medical_genetics__41827449__s001__tb1...
{ "index": "global_mmlu_full_so_medical_genetics__41827449__s001__tb131072__v003", "uid": "global_mmlu_full_so_medical_genetics__41827449__s001__tb131072__v003", "axis": "domain", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_so_moral_disputes__66471d73__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 834.2613848724434, \"nll_base\": 6.726546764373779, \"eval_ref\": \"global_mmlu_full_so_moral_disputes__66471d73__s000__tb262144...
{ "index": "global_mmlu_full_so_moral_disputes__66471d73__s000__tb262144__v000", "uid": "global_mmlu_full_so_moral_disputes__66471d73__s000__tb262144__v000", "axis": "multilingual", "budget": 2048, "token_budget": 262144, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_so_us_foreign_policy__703f3ea2__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 756.1169054422442, \"nll_base\": 6.6281960010528564, \"eval_ref\": \"global_mmlu_full_so_us_foreign_policy__703f3ea2__s000__tb6...
{ "index": "global_mmlu_full_so_us_foreign_policy__703f3ea2__s000__tb65536__v000", "uid": "global_mmlu_full_so_us_foreign_policy__703f3ea2__s000__tb65536__v000", "axis": "multilingual", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_sr_clinical_knowledge__63818cbe__s000__tb131072__v001\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 991.8050800267952, \"nll_base\": 6.899526596069336, \"eval_ref\": \"global_mmlu_full_sr_clinical_knowledge__63818cbe__s000__...
{ "index": "global_mmlu_full_sr_clinical_knowledge__63818cbe__s000__tb131072__v001", "uid": "global_mmlu_full_sr_clinical_knowledge__63818cbe__s000__tb131072__v001", "axis": "domain", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_sr_clinical_knowledge__63818cbe__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 991.8050800267952, \"nll_base\": 6.899526596069336, \"eval_ref\": \"global_mmlu_full_sr_clinical_knowledge__63818cbe__s000__tb...
{ "index": "global_mmlu_full_sr_clinical_knowledge__63818cbe__s000__tb65536__v000", "uid": "global_mmlu_full_sr_clinical_knowledge__63818cbe__s000__tb65536__v000", "axis": "domain", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_sr_high_school_biology__16d75356__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 968.8523460972567, \"nll_base\": 6.876112222671509, \"eval_ref\": \"global_mmlu_full_sr_high_school_biology__16d75356__s000...
{ "index": "global_mmlu_full_sr_high_school_biology__16d75356__s000__tb131072__v000", "uid": "global_mmlu_full_sr_high_school_biology__16d75356__s000__tb131072__v000", "axis": "multilingual", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_sr_high_school_government_and_politics__5610b3b5__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 1268.9097706947396, \"nll_base\": 7.145913362503052, \"eval_ref\": \"global_mmlu_full_sr_high_school_governme...
{ "index": "global_mmlu_full_sr_high_school_government_and_politics__5610b3b5__s000__tb65536__v000", "uid": "global_mmlu_full_sr_high_school_government_and_politics__5610b3b5__s000__tb65536__v000", "axis": "multilingual", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_sr_international_law__559c41ad__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 1176.753528694716, \"nll_base\": 7.070514678955078, \"eval_ref\": \"global_mmlu_full_sr_international_law__559c41ad__s000__tb65...
{ "index": "global_mmlu_full_sr_international_law__559c41ad__s000__tb65536__v000", "uid": "global_mmlu_full_sr_international_law__559c41ad__s000__tb65536__v000", "axis": "multilingual", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_sr_nutrition__b5b9d606__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 1321.47761461498, \"nll_base\": 7.1865057945251465, \"eval_ref\": \"global_mmlu_full_sr_nutrition__b5b9d606__s000__tb131072__v000\", ...
{ "index": "global_mmlu_full_sr_nutrition__b5b9d606__s000__tb131072__v000", "uid": "global_mmlu_full_sr_nutrition__b5b9d606__s000__tb131072__v000", "axis": "multilingual", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_sr_professional_law__cc826d38__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 544.6519988970189, \"nll_base\": 6.30014705657959, \"eval_ref\": \"global_mmlu_full_sr_professional_law__cc826d38__s000__tb131...
{ "index": "global_mmlu_full_sr_professional_law__cc826d38__s000__tb131072__v000", "uid": "global_mmlu_full_sr_professional_law__cc826d38__s000__tb131072__v000", "axis": "multilingual", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_sr_virology__3fa9badc__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 1378.3431720554574, \"nll_base\": 7.228637456893921, \"eval_ref\": \"global_mmlu_full_sr_virology__3fa9badc__s000__tb131072__v000\", \...
{ "index": "global_mmlu_full_sr_virology__3fa9badc__s000__tb131072__v000", "uid": "global_mmlu_full_sr_virology__3fa9badc__s000__tb131072__v000", "axis": "multilingual", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_sv_clinical_knowledge__538ac002__s000__tb131072__v002\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 909.5093802914882, \"nll_base\": 6.812905311584473, \"eval_ref\": \"global_mmlu_full_sv_clinical_knowledge__538ac002__s000__...
{ "index": "global_mmlu_full_sv_clinical_knowledge__538ac002__s000__tb131072__v002", "uid": "global_mmlu_full_sv_clinical_knowledge__538ac002__s000__tb131072__v002", "axis": "domain", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_sv_clinical_knowledge__538ac002__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 909.5093802914882, \"nll_base\": 6.812905311584473, \"eval_ref\": \"global_mmlu_full_sv_clinical_knowledge__538ac002__s000__...
{ "index": "global_mmlu_full_sv_clinical_knowledge__538ac002__s000__tb262144__v000", "uid": "global_mmlu_full_sv_clinical_knowledge__538ac002__s000__tb262144__v000", "axis": "domain", "budget": 2048, "token_budget": 262144, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_sv_clinical_knowledge__538ac002__s000__tb65536__v001\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 909.5093802914882, \"nll_base\": 6.812905311584473, \"eval_ref\": \"global_mmlu_full_sv_clinical_knowledge__538ac002__s000__tb...
{ "index": "global_mmlu_full_sv_clinical_knowledge__538ac002__s000__tb65536__v001", "uid": "global_mmlu_full_sv_clinical_knowledge__538ac002__s000__tb65536__v001", "axis": "domain", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_sv_college_medicine__974e5e1c__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 831.1186781082861, \"nll_base\": 6.722772598266602, \"eval_ref\": \"global_mmlu_full_sv_college_medicine__974e5e1c__s000__tb13...
{ "index": "global_mmlu_full_sv_college_medicine__974e5e1c__s000__tb131072__v000", "uid": "global_mmlu_full_sv_college_medicine__974e5e1c__s000__tb131072__v000", "axis": "multilingual", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_sv_high_school_biology__4a45c818__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 992.9615935272869, \"nll_base\": 6.900691986083984, \"eval_ref\": \"global_mmlu_full_sv_high_school_biology__4a45c818__s000...
{ "index": "global_mmlu_full_sv_high_school_biology__4a45c818__s000__tb131072__v000", "uid": "global_mmlu_full_sv_high_school_biology__4a45c818__s000__tb131072__v000", "axis": "multilingual", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_sv_medical_genetics__3cf9f234__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 1116.6259614872927, \"nll_base\": 7.018066883087158, \"eval_ref\": \"global_mmlu_full_sv_medical_genetics__3cf9f234__s000__tb1...
{ "index": "global_mmlu_full_sv_medical_genetics__3cf9f234__s000__tb131072__v000", "uid": "global_mmlu_full_sv_medical_genetics__3cf9f234__s000__tb131072__v000", "axis": "domain", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_sv_medical_genetics__3cf9f234__s000__tb262144__v001\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 1116.6259614872927, \"nll_base\": 7.018066883087158, \"eval_ref\": \"global_mmlu_full_sv_medical_genetics__3cf9f234__s000__tb2...
{ "index": "global_mmlu_full_sv_medical_genetics__3cf9f234__s000__tb262144__v001", "uid": "global_mmlu_full_sv_medical_genetics__3cf9f234__s000__tb262144__v001", "axis": "domain", "budget": 2048, "token_budget": 262144, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_sv_medical_genetics__3cf9f234__s000__tb65536__v002\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 1116.6259614872927, \"nll_base\": 7.018066883087158, \"eval_ref\": \"global_mmlu_full_sv_medical_genetics__3cf9f234__s000__tb655...
{ "index": "global_mmlu_full_sv_medical_genetics__3cf9f234__s000__tb65536__v002", "uid": "global_mmlu_full_sv_medical_genetics__3cf9f234__s000__tb65536__v002", "axis": "domain", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"global_mmlu_full_sv_medical_genetics__3cf9f234__s001__tb131072__v003\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 1181.764023650356, \"nll_base\": 7.074763536453247, \"eval_ref\": \"global_mmlu_full_sv_medical_genetics__3cf9f234__s001__tb13...
{ "index": "global_mmlu_full_sv_medical_genetics__3cf9f234__s001__tb131072__v003", "uid": "global_mmlu_full_sv_medical_genetics__3cf9f234__s001__tb131072__v003", "axis": "domain", "budget": 2048, "token_budget": 131072, "variant": "filter" }