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\": \"coqa__988946e3__s000__tb65536__v001\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 69.44275421499464, \"nll_base\": 4.240502732799899, \"eval_ref\": \"coqa__988946e3__s000__tb65536__v001\", \"oracle_delta\": 0.15451128559727856, \"random_delta...
{ "index": "coqa__988946e3__s000__tb65536__v001", "uid": "coqa__988946e3__s000__tb65536__v001", "axis": "qa", "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\": \"coqa__988946e3__s001__tb131072__v005\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 46.920313760422786, \"nll_base\": 3.8484507108989514, \"eval_ref\": \"coqa__988946e3__s001__tb131072__v005\", \"oracle_delta\": 0.19125767758018064, \"random_...
{ "index": "coqa__988946e3__s001__tb131072__v005", "uid": "coqa__988946e3__s001__tb131072__v005", "axis": "qa", "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\": \"coqa__988946e3__s001__tb65536__v004\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 46.920313760422786, \"nll_base\": 3.8484507108989514, \"eval_ref\": \"coqa__988946e3__s001__tb65536__v004\", \"oracle_delta\": 0.0949078610068872, \"random_delt...
{ "index": "coqa__988946e3__s001__tb65536__v004", "uid": "coqa__988946e3__s001__tb65536__v004", "axis": "qa", "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\": \"coqa__988946e3__s002__tb131072__v008\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 25.850633848149208, \"nll_base\": 3.2523351208917024, \"eval_ref\": \"coqa__988946e3__s002__tb131072__v008\", \"oracle_delta\": 0.29156748453776027, \"random_...
{ "index": "coqa__988946e3__s002__tb131072__v008", "uid": "coqa__988946e3__s002__tb131072__v008", "axis": "qa", "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\": \"coqa__988946e3__s002__tb262144__v006\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 25.850633848149208, \"nll_base\": 3.2523351208917024, \"eval_ref\": \"coqa__988946e3__s002__tb262144__v006\", \"oracle_delta\": 0.36837794862944495, \"random_...
{ "index": "coqa__988946e3__s002__tb262144__v006", "uid": "coqa__988946e3__s002__tb262144__v006", "axis": "qa", "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\": \"coqa__988946e3__s002__tb65536__v007\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 25.850633848149208, \"nll_base\": 3.2523351208917024, \"eval_ref\": \"coqa__988946e3__s002__tb65536__v007\", \"oracle_delta\": 0.16133266755904252, \"random_del...
{ "index": "coqa__988946e3__s002__tb65536__v007", "uid": "coqa__988946e3__s002__tb65536__v007", "axis": "qa", "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\": \"crows_pairs_english__7fb95fa1__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 2.737109187382913, \"nll_base\": 1.0069023223512201, \"eval_ref\": \"crows_pairs_english__7fb95fa1__s000__tb262144__v000\", \"oracle_delta\": 0...
{ "index": "crows_pairs_english__7fb95fa1__s000__tb262144__v000", "uid": "crows_pairs_english__7fb95fa1__s000__tb262144__v000", "axis": "instruction", "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\": \"crows_pairs_english_autre__ec707563__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 5.8930736381045, \"nll_base\": 1.7737777016379617, \"eval_ref\": \"crows_pairs_english_autre__ec707563__s000__tb65536__v000\", \"oracle_del...
{ "index": "crows_pairs_english_autre__ec707563__s000__tb65536__v000", "uid": "crows_pairs_english_autre__ec707563__s000__tb65536__v000", "axis": "instruction", "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\": \"crows_pairs_english_sexual_orientation__598fd8d9__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 2.5711176423831428, \"nll_base\": 0.9443406846788195, \"eval_ref\": \"crows_pairs_english_sexual_orientation__598fd8d9__s000__...
{ "index": "crows_pairs_english_sexual_orientation__598fd8d9__s000__tb65536__v000", "uid": "crows_pairs_english_sexual_orientation__598fd8d9__s000__tb65536__v000", "axis": "instruction", "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\": \"crows_pairs_english_socioeconomic__3a48a256__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 2.5240322638388384, \"nll_base\": 0.9258577274177261, \"eval_ref\": \"crows_pairs_english_socioeconomic__3a48a256__s000__tb131072...
{ "index": "crows_pairs_english_socioeconomic__3a48a256__s000__tb131072__v000", "uid": "crows_pairs_english_socioeconomic__3a48a256__s000__tb131072__v000", "axis": "instruction", "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\": \"crows_pairs_french__d7f123b6__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 1.753398291013199, \"nll_base\": 0.561555785501632, \"eval_ref\": \"crows_pairs_french__d7f123b6__s000__tb65536__v000\", \"oracle_delta\": 0.03602...
{ "index": "crows_pairs_french__d7f123b6__s000__tb65536__v000", "uid": "crows_pairs_french__d7f123b6__s000__tb65536__v000", "axis": "instruction", "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\": \"crows_pairs_french_age__b8fbb6ca__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 1.8706014901630976, \"nll_base\": 0.6262600316410571, \"eval_ref\": \"crows_pairs_french_age__b8fbb6ca__s000__tb131072__v000\", \"oracle_del...
{ "index": "crows_pairs_french_age__b8fbb6ca__s000__tb131072__v000", "uid": "crows_pairs_french_age__b8fbb6ca__s000__tb131072__v000", "axis": "instruction", "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\": \"crows_pairs_french_disability__51d7e2df__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 1.7360839776853514, \"nll_base\": 0.5516319893108764, \"eval_ref\": \"crows_pairs_french_disability__51d7e2df__s000__tb65536__v000\", \...
{ "index": "crows_pairs_french_disability__51d7e2df__s000__tb65536__v000", "uid": "crows_pairs_french_disability__51d7e2df__s000__tb65536__v000", "axis": "instruction", "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\": \"crows_pairs_french_nationality__aa118cc0__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 1.788922117225175, \"nll_base\": 0.5816132693280587, \"eval_ref\": \"crows_pairs_french_nationality__aa118cc0__s000__tb65536__v000\", ...
{ "index": "crows_pairs_french_nationality__aa118cc0__s000__tb65536__v000", "uid": "crows_pairs_french_nationality__aa118cc0__s000__tb65536__v000", "axis": "instruction", "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\": \"crows_pairs_french_race_color__198e8608__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 1.812005739397024, \"nll_base\": 0.5944343750534224, \"eval_ref\": \"crows_pairs_french_race_color__198e8608__s000__tb131072__v000\",...
{ "index": "crows_pairs_french_race_color__198e8608__s000__tb131072__v000", "uid": "crows_pairs_french_race_color__198e8608__s000__tb131072__v000", "axis": "instruction", "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\": \"csatqa_gr__fb35f4b7__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 38.930624333817896, \"nll_base\": 3.661781198838178, \"eval_ref\": \"csatqa_gr__fb35f4b7__s000__tb131072__v000\", \"oracle_delta\": 0.18754653369679186, ...
{ "index": "csatqa_gr__fb35f4b7__s000__tb131072__v000", "uid": "csatqa_gr__fb35f4b7__s000__tb131072__v000", "axis": "instruction", "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\": \"csatqa_rch__c0b178d1__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 32.51524155956208, \"nll_base\": 3.48170895046658, \"eval_ref\": \"csatqa_rch__c0b178d1__s000__tb65536__v000\", \"oracle_delta\": 0.09156855830439836, \"r...
{ "index": "csatqa_rch__c0b178d1__s000__tb65536__v000", "uid": "csatqa_rch__c0b178d1__s000__tb65536__v000", "axis": "instruction", "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\": \"darijammlu_professional_law__f2f2a6ed__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 928.7296832412975, \"nll_base\": 6.833817720413208, \"eval_ref\": \"darijammlu_professional_law__f2f2a6ed__s000__tb131072__v000\", \"or...
{ "index": "darijammlu_professional_law__f2f2a6ed__s000__tb131072__v000", "uid": "darijammlu_professional_law__f2f2a6ed__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\": \"drop__aa20c3bf__s001__tb262144__v003\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 192.30830380401397, \"nll_base\": 5.2590998331705725, \"eval_ref\": \"drop__aa20c3bf__s001__tb262144__v003\", \"oracle_delta\": 0.046661885579426965, \"random...
{ "index": "drop__aa20c3bf__s001__tb262144__v003", "uid": "drop__aa20c3bf__s001__tb262144__v003", "axis": "qa", "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\": \"esbbq_ses__7b3e4513__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 8.61020201303097, \"nll_base\": 2.15294778076962, \"eval_ref\": \"esbbq_ses__7b3e4513__s000__tb65536__v000\", \"oracle_delta\": 0.09530106331538235, \"rand...
{ "index": "esbbq_ses__7b3e4513__s000__tb65536__v000", "uid": "esbbq_ses__7b3e4513__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\": \"esbbq_spanish_region__626642b0__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 13.655532645880337, \"nll_base\": 2.6141447615526947, \"eval_ref\": \"esbbq_spanish_region__626642b0__s000__tb262144__v000\", \"oracle_delta\"...
{ "index": "esbbq_spanish_region__626642b0__s000__tb262144__v000", "uid": "esbbq_spanish_region__626642b0__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\": \"eus_exams_es_ejadministrativo__6e407332__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 1690.9477230276154, \"nll_base\": 7.43304443359375, \"eval_ref\": \"eus_exams_es_ejadministrativo__6e407332__s000__tb131072__v000\", ...
{ "index": "eus_exams_es_ejadministrativo__6e407332__s000__tb131072__v000", "uid": "eus_exams_es_ejadministrativo__6e407332__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\": \"eus_exams_es_ejsubalterno__e0476cf7__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 1868.0853452891145, \"nll_base\": 7.532669305801392, \"eval_ref\": \"eus_exams_es_ejsubalterno__e0476cf7__s000__tb65536__v000\", \"oracle_d...
{ "index": "eus_exams_es_ejsubalterno__e0476cf7__s000__tb65536__v000", "uid": "eus_exams_es_ejsubalterno__e0476cf7__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\": \"eus_exams_es_opeosakienf__715dff81__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 1517.9994299955408, \"nll_base\": 7.325148582458496, \"eval_ref\": \"eus_exams_es_opeosakienf__715dff81__s000__tb65536__v000\", \"oracle_del...
{ "index": "eus_exams_es_opeosakienf__715dff81__s000__tb65536__v000", "uid": "eus_exams_es_opeosakienf__715dff81__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\": \"eus_exams_es_osakidetza2c__be5e4304__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 1998.3481605907853, \"nll_base\": 7.600076198577881, \"eval_ref\": \"eus_exams_es_osakidetza2c__be5e4304__s000__tb131072__v000\", \"oracl...
{ "index": "eus_exams_es_osakidetza2c__be5e4304__s000__tb131072__v000", "uid": "eus_exams_es_osakidetza2c__be5e4304__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\": \"eus_exams_es_osakidetza3c__032edaac__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 1546.1212701123422, \"nll_base\": 7.3435046672821045, \"eval_ref\": \"eus_exams_es_osakidetza3c__032edaac__s000__tb262144__v000\", \"orac...
{ "index": "eus_exams_es_osakidetza3c__032edaac__s000__tb262144__v000", "uid": "eus_exams_es_osakidetza3c__032edaac__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\": \"eus_exams_eu_ejlaguntza__443f9906__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 5953.9726366371815, \"nll_base\": 8.691813945770264, \"eval_ref\": \"eus_exams_eu_ejlaguntza__443f9906__s000__tb131072__v000\", \"oracle_de...
{ "index": "eus_exams_eu_ejlaguntza__443f9906__s000__tb131072__v000", "uid": "eus_exams_eu_ejlaguntza__443f9906__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\": \"eus_exams_eu_opeehubiblioeu__302a5187__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 4173.907527754364, \"nll_base\": 8.336607933044434, \"eval_ref\": \"eus_exams_eu_opeehubiblioeu__302a5187__s000__tb131072__v000\", \"or...
{ "index": "eus_exams_eu_opeehubiblioeu__302a5187__s000__tb131072__v000", "uid": "eus_exams_eu_opeehubiblioeu__302a5187__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\": \"eus_exams_eu_opeehuderechoeu__27c23bad__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 5322.452661621033, \"nll_base\": 8.579689502716064, \"eval_ref\": \"eus_exams_eu_opeehuderechoeu__27c23bad__s000__tb262144__v000\", \"...
{ "index": "eus_exams_eu_opeehuderechoeu__27c23bad__s000__tb262144__v000", "uid": "eus_exams_eu_opeehuderechoeu__27c23bad__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\": \"eus_exams_eu_opeosakiauxeu__fe37e31a__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 4361.964027960703, \"nll_base\": 8.380677700042725, \"eval_ref\": \"eus_exams_eu_opeosakiauxeu__fe37e31a__s000__tb131072__v000\", \"orac...
{ "index": "eus_exams_eu_opeosakiauxeu__fe37e31a__s000__tb131072__v000", "uid": "eus_exams_eu_opeosakiauxeu__fe37e31a__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\": \"eus_exams_eu_opeosakiceladoreu__40470300__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 4763.059880152693, \"nll_base\": 8.468645572662354, \"eval_ref\": \"eus_exams_eu_opeosakiceladoreu__40470300__s000__tb131072__v000\"...
{ "index": "eus_exams_eu_opeosakiceladoreu__40470300__s000__tb131072__v000", "uid": "eus_exams_eu_opeosakiceladoreu__40470300__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\": \"eus_exams_eu_opeosakivarioseu__e7e166b0__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 4504.249079730095, \"nll_base\": 8.412776470184326, \"eval_ref\": \"eus_exams_eu_opeosakivarioseu__e7e166b0__s000__tb262144__v000\", ...
{ "index": "eus_exams_eu_opeosakivarioseu__e7e166b0__s000__tb262144__v000", "uid": "eus_exams_eu_opeosakivarioseu__e7e166b0__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\": \"eus_exams_eu_osakidetza3e__7da1d041__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 6225.459361339966, \"nll_base\": 8.73640251159668, \"eval_ref\": \"eus_exams_eu_osakidetza3e__7da1d041__s000__tb65536__v000\", \"oracle_del...
{ "index": "eus_exams_eu_osakidetza3e__7da1d041__s000__tb65536__v000", "uid": "eus_exams_eu_osakidetza3e__7da1d041__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\": \"eus_proficiency__55aabd29__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 5879.431962525154, \"nll_base\": 8.679215431213379, \"eval_ref\": \"eus_proficiency__55aabd29__s000__tb131072__v000\", \"oracle_delta\": 1.13554692...
{ "index": "eus_proficiency__55aabd29__s000__tb131072__v000", "uid": "eus_proficiency__55aabd29__s000__tb131072__v000", "axis": "instruction", "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\": \"eus_reading__52f38b8a__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 1661.1704437624226, \"nll_base\": 7.415277719497681, \"eval_ref\": \"eus_reading__52f38b8a__s000__tb131072__v000\", \"oracle_delta\": 0.864016056060791...
{ "index": "eus_reading__52f38b8a__s000__tb131072__v000", "uid": "eus_reading__52f38b8a__s000__tb131072__v000", "axis": "instruction", "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\": \"eus_trivia__6e0332b5__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 6319.179677080568, \"nll_base\": 8.751344680786133, \"eval_ref\": \"eus_trivia__6e0332b5__s000__tb131072__v000\", \"oracle_delta\": 1.0722689628601074, ...
{ "index": "eus_trivia__6e0332b5__s000__tb131072__v000", "uid": "eus_trivia__6e0332b5__s000__tb131072__v000", "axis": "instruction", "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\": \"evalita-mp_faq_prompt-2__de2d980b__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 3.233202974192725, \"nll_base\": 1.173473278768483, \"eval_ref\": \"evalita-mp_faq_prompt-2__de2d980b__s000__tb131072__v000\", \"oracle_del...
{ "index": "evalita-mp_faq_prompt-2__de2d980b__s000__tb131072__v000", "uid": "evalita-mp_faq_prompt-2__de2d980b__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\": \"evalita-mp_te_prompt-2__50f1e98f__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 206.36572519209665, \"nll_base\": 5.329649959702089, \"eval_ref\": \"evalita-mp_te_prompt-2__50f1e98f__s000__tb65536__v000\", \"oracle_delta\"...
{ "index": "evalita-mp_te_prompt-2__50f1e98f__s000__tb65536__v000", "uid": "evalita-mp_te_prompt-2__50f1e98f__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\": \"evalita-mp_wic_prompt-1__96017167__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 164.0934900914486, \"nll_base\": 5.10043632693407, \"eval_ref\": \"evalita-mp_wic_prompt-1__96017167__s000__tb131072__v000\", \"oracle_delt...
{ "index": "evalita-mp_wic_prompt-1__96017167__s000__tb131072__v000", "uid": "evalita-mp_wic_prompt-1__96017167__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\": \"glianorex__654ccf23__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 2360.7313220815777, \"nll_base\": 7.766726732254028, \"eval_ref\": \"glianorex__654ccf23__s000__tb131072__v000\", \"oracle_delta\": 0.5619592666625977, \...
{ "index": "glianorex__654ccf23__s000__tb131072__v000", "uid": "glianorex__654ccf23__s000__tb131072__v000", "axis": "instruction", "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\": \"glianorex_fr__06abad96__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 1873.8401191869582, \"nll_base\": 7.535745143890381, \"eval_ref\": \"glianorex_fr__06abad96__s000__tb131072__v000\", \"oracle_delta\": 1.5913879871368...
{ "index": "glianorex_fr__06abad96__s000__tb131072__v000", "uid": "glianorex_fr__06abad96__s000__tb131072__v000", "axis": "instruction", "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_ar_medical__26e7fc37__s000__tb65536__v001\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 478.74856734930574, \"nll_base\": 6.17117554800851, \"eval_ref\": \"global_mmlu_ar_medical__26e7fc37__s000__tb65536__v001\", \"oracle_delta\":...
{ "index": "global_mmlu_ar_medical__26e7fc37__s000__tb65536__v001", "uid": "global_mmlu_ar_medical__26e7fc37__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_de_medical__85df05cb__s000__tb131072__v002\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 1379.9966183662857, \"nll_base\": 7.229836327689035, \"eval_ref\": \"global_mmlu_de_medical__85df05cb__s000__tb131072__v002\", \"oracle_delt...
{ "index": "global_mmlu_de_medical__85df05cb__s000__tb131072__v002", "uid": "global_mmlu_de_medical__85df05cb__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_de_medical__85df05cb__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 1379.9966183662857, \"nll_base\": 7.229836327689035, \"eval_ref\": \"global_mmlu_de_medical__85df05cb__s000__tb262144__v000\", \"oracle_delt...
{ "index": "global_mmlu_de_medical__85df05cb__s000__tb262144__v000", "uid": "global_mmlu_de_medical__85df05cb__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_de_medical__85df05cb__s000__tb65536__v001\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 1379.9966183662857, \"nll_base\": 7.229836327689035, \"eval_ref\": \"global_mmlu_de_medical__85df05cb__s000__tb65536__v001\", \"oracle_delta\"...
{ "index": "global_mmlu_de_medical__85df05cb__s000__tb65536__v001", "uid": "global_mmlu_de_medical__85df05cb__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_de_medical__85df05cb__s001__tb262144__v003\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 1432.615993307551, \"nll_base\": 7.267257417951312, \"eval_ref\": \"global_mmlu_de_medical__85df05cb__s001__tb262144__v003\", \"oracle_delta...
{ "index": "global_mmlu_de_medical__85df05cb__s001__tb262144__v003", "uid": "global_mmlu_de_medical__85df05cb__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_de_stem__2cbdb801__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 1680.2513674590937, \"nll_base\": 7.426698684692383, \"eval_ref\": \"global_mmlu_de_stem__2cbdb801__s000__tb131072__v000\", \"oracle_delta\": 0...
{ "index": "global_mmlu_de_stem__2cbdb801__s000__tb131072__v000", "uid": "global_mmlu_de_stem__2cbdb801__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_en_medical__6735f69f__s000__tb131072__v002\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 1175.2225081328602, \"nll_base\": 7.069212777273996, \"eval_ref\": \"global_mmlu_en_medical__6735f69f__s000__tb131072__v002\", \"oracle_delt...
{ "index": "global_mmlu_en_medical__6735f69f__s000__tb131072__v002", "uid": "global_mmlu_en_medical__6735f69f__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_en_medical__6735f69f__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 1175.2225081328602, \"nll_base\": 7.069212777273996, \"eval_ref\": \"global_mmlu_en_medical__6735f69f__s000__tb262144__v000\", \"oracle_delt...
{ "index": "global_mmlu_en_medical__6735f69f__s000__tb262144__v000", "uid": "global_mmlu_en_medical__6735f69f__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_en_medical__6735f69f__s000__tb65536__v001\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 1175.2225081328602, \"nll_base\": 7.069212777273996, \"eval_ref\": \"global_mmlu_en_medical__6735f69f__s000__tb65536__v001\", \"oracle_delta\"...
{ "index": "global_mmlu_en_medical__6735f69f__s000__tb65536__v001", "uid": "global_mmlu_en_medical__6735f69f__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_en_medical__6735f69f__s001__tb262144__v003\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 1089.7511643763887, \"nll_base\": 6.993704659598214, \"eval_ref\": \"global_mmlu_en_medical__6735f69f__s001__tb262144__v003\", \"oracle_delt...
{ "index": "global_mmlu_en_medical__6735f69f__s001__tb262144__v003", "uid": "global_mmlu_en_medical__6735f69f__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_es_business__3078d6e3__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 1015.3115037408221, \"nll_base\": 6.922950744628906, \"eval_ref\": \"global_mmlu_es_business__3078d6e3__s000__tb65536__v000\", \"oracle_delta...
{ "index": "global_mmlu_es_business__3078d6e3__s000__tb65536__v000", "uid": "global_mmlu_es_business__3078d6e3__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_es_medical__1b9bd0f6__s000__tb131072__v001\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 1010.4541125853451, \"nll_base\": 6.918155125209263, \"eval_ref\": \"global_mmlu_es_medical__1b9bd0f6__s000__tb131072__v001\", \"oracle_delt...
{ "index": "global_mmlu_es_medical__1b9bd0f6__s000__tb131072__v001", "uid": "global_mmlu_es_medical__1b9bd0f6__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_es_medical__1b9bd0f6__s000__tb262144__v002\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 1010.4541125853451, \"nll_base\": 6.918155125209263, \"eval_ref\": \"global_mmlu_es_medical__1b9bd0f6__s000__tb262144__v002\", \"oracle_delt...
{ "index": "global_mmlu_es_medical__1b9bd0f6__s000__tb262144__v002", "uid": "global_mmlu_es_medical__1b9bd0f6__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_es_medical__1b9bd0f6__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 1010.4541125853451, \"nll_base\": 6.918155125209263, \"eval_ref\": \"global_mmlu_es_medical__1b9bd0f6__s000__tb65536__v000\", \"oracle_delta\"...
{ "index": "global_mmlu_es_medical__1b9bd0f6__s000__tb65536__v000", "uid": "global_mmlu_es_medical__1b9bd0f6__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_es_medical__1b9bd0f6__s001__tb65536__v003\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 1139.3456327274496, \"nll_base\": 7.03820937020438, \"eval_ref\": \"global_mmlu_es_medical__1b9bd0f6__s001__tb65536__v003\", \"oracle_delta\":...
{ "index": "global_mmlu_es_medical__1b9bd0f6__s001__tb65536__v003", "uid": "global_mmlu_es_medical__1b9bd0f6__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_fr_other__4ccb463b__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 1518.3972313842532, \"nll_base\": 7.325410604476929, \"eval_ref\": \"global_mmlu_fr_other__4ccb463b__s000__tb262144__v000\", \"oracle_delta\":...
{ "index": "global_mmlu_fr_other__4ccb463b__s000__tb262144__v000", "uid": "global_mmlu_fr_other__4ccb463b__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_fr_stem__37361daa__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 959.2715756089899, \"nll_base\": 6.866174221038818, \"eval_ref\": \"global_mmlu_fr_stem__37361daa__s000__tb262144__v000\", \"oracle_delta\": 1....
{ "index": "global_mmlu_fr_stem__37361daa__s000__tb262144__v000", "uid": "global_mmlu_fr_stem__37361daa__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_am_clinical_knowledge__75a22ef7__s000__tb65536__v002\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 1749.6490026972638, \"nll_base\": 7.467170476913452, \"eval_ref\": \"global_mmlu_full_am_clinical_knowledge__75a22ef7__s000__t...
{ "index": "global_mmlu_full_am_clinical_knowledge__75a22ef7__s000__tb65536__v002", "uid": "global_mmlu_full_am_clinical_knowledge__75a22ef7__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_am_clinical_knowledge__75a22ef7__s001__tb131072__v003\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 1946.2856040780416, \"nll_base\": 7.573678016662598, \"eval_ref\": \"global_mmlu_full_am_clinical_knowledge__75a22ef7__s001_...
{ "index": "global_mmlu_full_am_clinical_knowledge__75a22ef7__s001__tb131072__v003", "uid": "global_mmlu_full_am_clinical_knowledge__75a22ef7__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_bn_college_biology__b570f8d4__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 559.9936495912615, \"nll_base\": 6.327925443649292, \"eval_ref\": \"global_mmlu_full_bn_college_biology__b570f8d4__s000__tb1310...
{ "index": "global_mmlu_full_bn_college_biology__b570f8d4__s000__tb131072__v000", "uid": "global_mmlu_full_bn_college_biology__b570f8d4__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_bn_elementary_mathematics__f8f3cc26__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 1233.1118575411674, \"nll_base\": 7.11729621887207, \"eval_ref\": \"global_mmlu_full_bn_elementary_mathematics__f8f3cc26__...
{ "index": "global_mmlu_full_bn_elementary_mathematics__f8f3cc26__s000__tb65536__v000", "uid": "global_mmlu_full_bn_elementary_mathematics__f8f3cc26__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_bn_high_school_biology__eb345a54__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 670.914028661332, \"nll_base\": 6.508641004562378, \"eval_ref\": \"global_mmlu_full_bn_high_school_biology__eb345a54__s000__t...
{ "index": "global_mmlu_full_bn_high_school_biology__eb345a54__s000__tb65536__v000", "uid": "global_mmlu_full_bn_high_school_biology__eb345a54__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_bn_high_school_microeconomics__8b32131c__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 532.6963057570482, \"nll_base\": 6.27795147895813, \"eval_ref\": \"global_mmlu_full_bn_high_school_microeconomics__8b3...
{ "index": "global_mmlu_full_bn_high_school_microeconomics__8b32131c__s000__tb65536__v000", "uid": "global_mmlu_full_bn_high_school_microeconomics__8b32131c__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_bn_high_school_physics__0b0027ce__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 782.9842966148851, \"nll_base\": 6.663112640380859, \"eval_ref\": \"global_mmlu_full_bn_high_school_physics__0b0027ce__s000...
{ "index": "global_mmlu_full_bn_high_school_physics__0b0027ce__s000__tb131072__v000", "uid": "global_mmlu_full_bn_high_school_physics__0b0027ce__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_bn_management__6cf95d70__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 973.5140955178751, \"nll_base\": 6.8809123039245605, \"eval_ref\": \"global_mmlu_full_bn_management__6cf95d70__s000__tb262144__v000\...
{ "index": "global_mmlu_full_bn_management__6cf95d70__s000__tb262144__v000", "uid": "global_mmlu_full_bn_management__6cf95d70__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_bn_medical_genetics__63e7aeae__s000__tb131072__v002\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 1003.9929748236061, \"nll_base\": 6.911740303039551, \"eval_ref\": \"global_mmlu_full_bn_medical_genetics__63e7aeae__s000__tb1...
{ "index": "global_mmlu_full_bn_medical_genetics__63e7aeae__s000__tb131072__v002", "uid": "global_mmlu_full_bn_medical_genetics__63e7aeae__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_bn_medical_genetics__63e7aeae__s000__tb65536__v001\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 1003.9929748236061, \"nll_base\": 6.911740303039551, \"eval_ref\": \"global_mmlu_full_bn_medical_genetics__63e7aeae__s000__tb655...
{ "index": "global_mmlu_full_bn_medical_genetics__63e7aeae__s000__tb65536__v001", "uid": "global_mmlu_full_bn_medical_genetics__63e7aeae__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_bn_professional_psychology__13492af6__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 495.75406143338097, \"nll_base\": 6.206079959869385, \"eval_ref\": \"global_mmlu_full_bn_professional_psychology__13492af...
{ "index": "global_mmlu_full_bn_professional_psychology__13492af6__s000__tb65536__v000", "uid": "global_mmlu_full_bn_professional_psychology__13492af6__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_cs_anatomy__a2707233__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 865.8097526615435, \"nll_base\": 6.763665199279785, \"eval_ref\": \"global_mmlu_full_cs_anatomy__a2707233__s000__tb262144__v000\", \"or...
{ "index": "global_mmlu_full_cs_anatomy__a2707233__s000__tb262144__v000", "uid": "global_mmlu_full_cs_anatomy__a2707233__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_cs_college_physics__c84d7a97__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 869.4852940960267, \"nll_base\": 6.767901420593262, \"eval_ref\": \"global_mmlu_full_cs_college_physics__c84d7a97__s000__tb1310...
{ "index": "global_mmlu_full_cs_college_physics__c84d7a97__s000__tb131072__v000", "uid": "global_mmlu_full_cs_college_physics__c84d7a97__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_cs_high_school_biology__24fbc87c__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 832.6883503252385, \"nll_base\": 6.724659442901611, \"eval_ref\": \"global_mmlu_full_cs_high_school_biology__24fbc87c__s000...
{ "index": "global_mmlu_full_cs_high_school_biology__24fbc87c__s000__tb131072__v000", "uid": "global_mmlu_full_cs_high_school_biology__24fbc87c__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_cs_human_sexuality__acaf5412__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 1086.2419121397184, \"nll_base\": 6.990479230880737, \"eval_ref\": \"global_mmlu_full_cs_human_sexuality__acaf5412__s000__tb262...
{ "index": "global_mmlu_full_cs_human_sexuality__acaf5412__s000__tb262144__v000", "uid": "global_mmlu_full_cs_human_sexuality__acaf5412__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_cs_marketing__0e9dcfc4__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 905.0263746478774, \"nll_base\": 6.807964086532593, \"eval_ref\": \"global_mmlu_full_cs_marketing__0e9dcfc4__s000__tb65536__v000\", \"o...
{ "index": "global_mmlu_full_cs_marketing__0e9dcfc4__s000__tb65536__v000", "uid": "global_mmlu_full_cs_marketing__0e9dcfc4__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_de_business_ethics__c93c9ada__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 1395.1150904864335, \"nll_base\": 7.240732192993164, \"eval_ref\": \"global_mmlu_full_de_business_ethics__c93c9ada__s000__tb131...
{ "index": "global_mmlu_full_de_business_ethics__c93c9ada__s000__tb131072__v000", "uid": "global_mmlu_full_de_business_ethics__c93c9ada__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_de_clinical_knowledge__18c31884__s000__tb131072__v001\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 1163.2344495385012, \"nll_base\": 7.058959722518921, \"eval_ref\": \"global_mmlu_full_de_clinical_knowledge__18c31884__s000_...
{ "index": "global_mmlu_full_de_clinical_knowledge__18c31884__s000__tb131072__v001", "uid": "global_mmlu_full_de_clinical_knowledge__18c31884__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_de_clinical_knowledge__18c31884__s000__tb262144__v002\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 1163.2344495385012, \"nll_base\": 7.058959722518921, \"eval_ref\": \"global_mmlu_full_de_clinical_knowledge__18c31884__s000_...
{ "index": "global_mmlu_full_de_clinical_knowledge__18c31884__s000__tb262144__v002", "uid": "global_mmlu_full_de_clinical_knowledge__18c31884__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_de_clinical_knowledge__18c31884__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 1163.2344495385012, \"nll_base\": 7.058959722518921, \"eval_ref\": \"global_mmlu_full_de_clinical_knowledge__18c31884__s000__t...
{ "index": "global_mmlu_full_de_clinical_knowledge__18c31884__s000__tb65536__v000", "uid": "global_mmlu_full_de_clinical_knowledge__18c31884__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_de_clinical_knowledge__18c31884__s001__tb65536__v003\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 1343.1518881184247, \"nll_base\": 7.202774286270142, \"eval_ref\": \"global_mmlu_full_de_clinical_knowledge__18c31884__s001__t...
{ "index": "global_mmlu_full_de_clinical_knowledge__18c31884__s001__tb65536__v003", "uid": "global_mmlu_full_de_clinical_knowledge__18c31884__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_de_elementary_mathematics__325529a3__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 2443.299223124881, \"nll_base\": 7.801104545593262, \"eval_ref\": \"global_mmlu_full_de_elementary_mathematics__325529a3...
{ "index": "global_mmlu_full_de_elementary_mathematics__325529a3__s000__tb262144__v000", "uid": "global_mmlu_full_de_elementary_mathematics__325529a3__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_de_high_school_european_history__741639a2__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 1901.0490099084643, \"nll_base\": 7.550161123275757, \"eval_ref\": \"global_mmlu_full_de_high_school_european_histor...
{ "index": "global_mmlu_full_de_high_school_european_history__741639a2__s000__tb65536__v000", "uid": "global_mmlu_full_de_high_school_european_history__741639a2__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_de_high_school_microeconomics__58f66398__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 1209.9370151040166, \"nll_base\": 7.098323583602905, \"eval_ref\": \"global_mmlu_full_de_high_school_microeconomics__5...
{ "index": "global_mmlu_full_de_high_school_microeconomics__58f66398__s000__tb65536__v000", "uid": "global_mmlu_full_de_high_school_microeconomics__58f66398__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_de_medical_genetics__77ba89da__s000__tb131072__v002\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 1006.475679248596, \"nll_base\": 6.914210081100464, \"eval_ref\": \"global_mmlu_full_de_medical_genetics__77ba89da__s000__tb13...
{ "index": "global_mmlu_full_de_medical_genetics__77ba89da__s000__tb131072__v002", "uid": "global_mmlu_full_de_medical_genetics__77ba89da__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_de_medical_genetics__77ba89da__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 1006.475679248596, \"nll_base\": 6.914210081100464, \"eval_ref\": \"global_mmlu_full_de_medical_genetics__77ba89da__s000__tb26...
{ "index": "global_mmlu_full_de_medical_genetics__77ba89da__s000__tb262144__v000", "uid": "global_mmlu_full_de_medical_genetics__77ba89da__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_de_medical_genetics__77ba89da__s000__tb65536__v001\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 1006.475679248596, \"nll_base\": 6.914210081100464, \"eval_ref\": \"global_mmlu_full_de_medical_genetics__77ba89da__s000__tb6553...
{ "index": "global_mmlu_full_de_medical_genetics__77ba89da__s000__tb65536__v001", "uid": "global_mmlu_full_de_medical_genetics__77ba89da__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_de_medical_genetics__77ba89da__s001__tb262144__v003\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 1326.305644228583, \"nll_base\": 7.190152645111084, \"eval_ref\": \"global_mmlu_full_de_medical_genetics__77ba89da__s001__tb26...
{ "index": "global_mmlu_full_de_medical_genetics__77ba89da__s001__tb262144__v003", "uid": "global_mmlu_full_de_medical_genetics__77ba89da__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_de_philosophy__91be83c3__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 1417.608735192993, \"nll_base\": 7.2567267417907715, \"eval_ref\": \"global_mmlu_full_de_philosophy__91be83c3__s000__tb262144__v000\...
{ "index": "global_mmlu_full_de_philosophy__91be83c3__s000__tb262144__v000", "uid": "global_mmlu_full_de_philosophy__91be83c3__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_de_professional_accounting__265a4d95__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 1148.3345486496987, \"nll_base\": 7.046067953109741, \"eval_ref\": \"global_mmlu_full_de_professional_accounting__265a4...
{ "index": "global_mmlu_full_de_professional_accounting__265a4d95__s000__tb131072__v000", "uid": "global_mmlu_full_de_professional_accounting__265a4d95__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_de_security_studies__0befd40b__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 1881.8286073873835, \"nll_base\": 7.53999924659729, \"eval_ref\": \"global_mmlu_full_de_security_studies__0befd40b__s000__tb26...
{ "index": "global_mmlu_full_de_security_studies__0befd40b__s000__tb262144__v000", "uid": "global_mmlu_full_de_security_studies__0befd40b__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_el_astronomy__9e553a18__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 865.9713986268241, \"nll_base\": 6.763851881027222, \"eval_ref\": \"global_mmlu_full_el_astronomy__9e553a18__s000__tb131072__v000\", ...
{ "index": "global_mmlu_full_el_astronomy__9e553a18__s000__tb131072__v000", "uid": "global_mmlu_full_el_astronomy__9e553a18__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_el_clinical_knowledge__ef0c0df8__s000__tb131072__v001\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 901.1241038138029, \"nll_base\": 6.803642988204956, \"eval_ref\": \"global_mmlu_full_el_clinical_knowledge__ef0c0df8__s000__...
{ "index": "global_mmlu_full_el_clinical_knowledge__ef0c0df8__s000__tb131072__v001", "uid": "global_mmlu_full_el_clinical_knowledge__ef0c0df8__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_el_clinical_knowledge__ef0c0df8__s001__tb65536__v003\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 838.03213901859, \"nll_base\": 6.731056451797485, \"eval_ref\": \"global_mmlu_full_el_clinical_knowledge__ef0c0df8__s001__tb65...
{ "index": "global_mmlu_full_el_clinical_knowledge__ef0c0df8__s001__tb65536__v003", "uid": "global_mmlu_full_el_clinical_knowledge__ef0c0df8__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_el_college_biology__661c7b1d__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 691.257710190232, \"nll_base\": 6.538512706756592, \"eval_ref\": \"global_mmlu_full_el_college_biology__661c7b1d__s000__tb65536__...
{ "index": "global_mmlu_full_el_college_biology__661c7b1d__s000__tb65536__v000", "uid": "global_mmlu_full_el_college_biology__661c7b1d__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_el_high_school_macroeconomics__e385eeab__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 804.8076109331013, \"nll_base\": 6.690603256225586, \"eval_ref\": \"global_mmlu_full_el_high_school_macroeconomics__...
{ "index": "global_mmlu_full_el_high_school_macroeconomics__e385eeab__s000__tb262144__v000", "uid": "global_mmlu_full_el_high_school_macroeconomics__e385eeab__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_el_high_school_psychology__aeea9064__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 796.3115709208007, \"nll_base\": 6.679990530014038, \"eval_ref\": \"global_mmlu_full_el_high_school_psychology__aeea9064...
{ "index": "global_mmlu_full_el_high_school_psychology__aeea9064__s000__tb131072__v000", "uid": "global_mmlu_full_el_high_school_psychology__aeea9064__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_el_international_law__2ffd9a74__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 794.7432025263868, \"nll_base\": 6.678019046783447, \"eval_ref\": \"global_mmlu_full_el_international_law__2ffd9a74__s000__tb...
{ "index": "global_mmlu_full_el_international_law__2ffd9a74__s000__tb262144__v000", "uid": "global_mmlu_full_el_international_law__2ffd9a74__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_el_machine_learning__be846e22__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 579.5946652404057, \"nll_base\": 6.362329006195068, \"eval_ref\": \"global_mmlu_full_el_machine_learning__be846e22__s000__tb13...
{ "index": "global_mmlu_full_el_machine_learning__be846e22__s000__tb131072__v000", "uid": "global_mmlu_full_el_machine_learning__be846e22__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_el_medical_genetics__2cda3ce7__s000__tb131072__v001\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 998.8515783165313, \"nll_base\": 6.906606197357178, \"eval_ref\": \"global_mmlu_full_el_medical_genetics__2cda3ce7__s000__tb13...
{ "index": "global_mmlu_full_el_medical_genetics__2cda3ce7__s000__tb131072__v001", "uid": "global_mmlu_full_el_medical_genetics__2cda3ce7__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_el_professional_law__dc744b0e__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 258.16794554251385, \"nll_base\": 5.553610324859619, \"eval_ref\": \"global_mmlu_full_el_professional_law__dc744b0e__s000__tb2...
{ "index": "global_mmlu_full_el_professional_law__dc744b0e__s000__tb262144__v000", "uid": "global_mmlu_full_el_professional_law__dc744b0e__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_el_security_studies__b68534b6__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 907.3320585902921, \"nll_base\": 6.810508489608765, \"eval_ref\": \"global_mmlu_full_el_security_studies__b68534b6__s000__tb6553...
{ "index": "global_mmlu_full_el_security_studies__b68534b6__s000__tb65536__v000", "uid": "global_mmlu_full_el_security_studies__b68534b6__s000__tb65536__v000", "axis": "multilingual", "budget": 2048, "token_budget": 65536, "variant": "filter" }