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\": \"mlqa_hi_de__66bf52d2__s000__tb131072__v002\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 3.330827233073213, \"nll_base\": 1.20322069148193, \"eval_ref\": \"mlqa_hi_de__66bf52d2__s000__tb131072__v002\", \"oracle_delta\": 0.08596114998602822, ...
{ "index": "mlqa_hi_de__66bf52d2__s000__tb131072__v002", "uid": "mlqa_hi_de__66bf52d2__s000__tb131072__v002", "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\": \"mlqa_hi_de__66bf52d2__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 3.330827233073213, \"nll_base\": 1.20322069148193, \"eval_ref\": \"mlqa_hi_de__66bf52d2__s000__tb262144__v000\", \"oracle_delta\": 0.17914305003318276, ...
{ "index": "mlqa_hi_de__66bf52d2__s000__tb262144__v000", "uid": "mlqa_hi_de__66bf52d2__s000__tb262144__v000", "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\": \"mlqa_hi_de__66bf52d2__s000__tb65536__v001\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 3.330827233073213, \"nll_base\": 1.20322069148193, \"eval_ref\": \"mlqa_hi_de__66bf52d2__s000__tb65536__v001\", \"oracle_delta\": 0.04285031814688889, \"r...
{ "index": "mlqa_hi_de__66bf52d2__s000__tb65536__v001", "uid": "mlqa_hi_de__66bf52d2__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\": \"mlqa_hi_de__66bf52d2__s001__tb131072__v005\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 5.231603154192526, \"nll_base\": 1.6547177615443167, \"eval_ref\": \"mlqa_hi_de__66bf52d2__s001__tb131072__v005\", \"oracle_delta\": 0.07719295434603346...
{ "index": "mlqa_hi_de__66bf52d2__s001__tb131072__v005", "uid": "mlqa_hi_de__66bf52d2__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\": \"mlqa_hi_de__66bf52d2__s001__tb262144__v003\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 5.231603154192526, \"nll_base\": 1.6547177615443167, \"eval_ref\": \"mlqa_hi_de__66bf52d2__s001__tb262144__v003\", \"oracle_delta\": 0.1544476045807257,...
{ "index": "mlqa_hi_de__66bf52d2__s001__tb262144__v003", "uid": "mlqa_hi_de__66bf52d2__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\": \"mlqa_hi_de__66bf52d2__s001__tb65536__v004\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 5.231603154192526, \"nll_base\": 1.6547177615443167, \"eval_ref\": \"mlqa_hi_de__66bf52d2__s001__tb65536__v004\", \"oracle_delta\": 0.04279951991183184, \...
{ "index": "mlqa_hi_de__66bf52d2__s001__tb65536__v004", "uid": "mlqa_hi_de__66bf52d2__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\": \"mlqa_hi_de__66bf52d2__s002__tb262144__v006\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 3.9133375127188867, \"nll_base\": 1.3643905937173573, \"eval_ref\": \"mlqa_hi_de__66bf52d2__s002__tb262144__v006\", \"oracle_delta\": 0.1752987067347449...
{ "index": "mlqa_hi_de__66bf52d2__s002__tb262144__v006", "uid": "mlqa_hi_de__66bf52d2__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\": \"mlqa_hi_de__66bf52d2__s002__tb65536__v007\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 3.9133375127188867, \"nll_base\": 1.3643905937173573, \"eval_ref\": \"mlqa_hi_de__66bf52d2__s002__tb65536__v007\", \"oracle_delta\": 0.04191695614445501, ...
{ "index": "mlqa_hi_de__66bf52d2__s002__tb65536__v007", "uid": "mlqa_hi_de__66bf52d2__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\": \"mlqa_hi_en__ea628517__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 3.006282789871504, \"nll_base\": 1.1007043620462151, \"eval_ref\": \"mlqa_hi_en__ea628517__s000__tb131072__v000\", \"oracle_delta\": 0.05377913847274063...
{ "index": "mlqa_hi_en__ea628517__s000__tb131072__v000", "uid": "mlqa_hi_en__ea628517__s000__tb131072__v000", "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\": \"mlqa_hi_en__ea628517__s000__tb262144__v001\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 3.006282789871504, \"nll_base\": 1.1007043620462151, \"eval_ref\": \"mlqa_hi_en__ea628517__s000__tb262144__v001\", \"oracle_delta\": 0.09576516588488015...
{ "index": "mlqa_hi_en__ea628517__s000__tb262144__v001", "uid": "mlqa_hi_en__ea628517__s000__tb262144__v001", "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\": \"mlqa_hi_en__ea628517__s000__tb65536__v002\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 3.006282789871504, \"nll_base\": 1.1007043620462151, \"eval_ref\": \"mlqa_hi_en__ea628517__s000__tb65536__v002\", \"oracle_delta\": 0.025950878705104152, ...
{ "index": "mlqa_hi_en__ea628517__s000__tb65536__v002", "uid": "mlqa_hi_en__ea628517__s000__tb65536__v002", "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\": \"mlqa_hi_en__ea628517__s001__tb131072__v003\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 3.8475182063387448, \"nll_base\": 1.3474283187115779, \"eval_ref\": \"mlqa_hi_en__ea628517__s001__tb131072__v003\", \"oracle_delta\": 0.0726653552446208...
{ "index": "mlqa_hi_en__ea628517__s001__tb131072__v003", "uid": "mlqa_hi_en__ea628517__s001__tb131072__v003", "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\": \"mlqa_hi_en__ea628517__s001__tb262144__v004\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 3.8475182063387448, \"nll_base\": 1.3474283187115779, \"eval_ref\": \"mlqa_hi_en__ea628517__s001__tb262144__v004\", \"oracle_delta\": 0.1484849273181352...
{ "index": "mlqa_hi_en__ea628517__s001__tb262144__v004", "uid": "mlqa_hi_en__ea628517__s001__tb262144__v004", "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\": \"mlqa_hi_en__ea628517__s001__tb65536__v005\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 3.8475182063387448, \"nll_base\": 1.3474283187115779, \"eval_ref\": \"mlqa_hi_en__ea628517__s001__tb65536__v005\", \"oracle_delta\": 0.04451924308401645, ...
{ "index": "mlqa_hi_en__ea628517__s001__tb65536__v005", "uid": "mlqa_hi_en__ea628517__s001__tb65536__v005", "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\": \"mlqa_hi_en__ea628517__s002__tb131072__v006\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 3.302622706797693, \"nll_base\": 1.194716912453895, \"eval_ref\": \"mlqa_hi_en__ea628517__s002__tb131072__v006\", \"oracle_delta\": 0.05289373138071496,...
{ "index": "mlqa_hi_en__ea628517__s002__tb131072__v006", "uid": "mlqa_hi_en__ea628517__s002__tb131072__v006", "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\": \"mlqa_hi_en__ea628517__s002__tb262144__v007\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 3.302622706797693, \"nll_base\": 1.194716912453895, \"eval_ref\": \"mlqa_hi_en__ea628517__s002__tb262144__v007\", \"oracle_delta\": 0.12539233493433843,...
{ "index": "mlqa_hi_en__ea628517__s002__tb262144__v007", "uid": "mlqa_hi_en__ea628517__s002__tb262144__v007", "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\": \"mlqa_hi_es__4225507a__s000__tb131072__v001\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 4.983422368819945, \"nll_base\": 1.6061168776618109, \"eval_ref\": \"mlqa_hi_es__4225507a__s000__tb131072__v001\", \"oracle_delta\": 0.11703835593329526...
{ "index": "mlqa_hi_es__4225507a__s000__tb131072__v001", "uid": "mlqa_hi_es__4225507a__s000__tb131072__v001", "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\": \"mlqa_hi_es__4225507a__s000__tb262144__v002\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 4.983422368819945, \"nll_base\": 1.6061168776618109, \"eval_ref\": \"mlqa_hi_es__4225507a__s000__tb262144__v002\", \"oracle_delta\": 0.19836680094401027...
{ "index": "mlqa_hi_es__4225507a__s000__tb262144__v002", "uid": "mlqa_hi_es__4225507a__s000__tb262144__v002", "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\": \"mlqa_hi_es__4225507a__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 4.983422368819945, \"nll_base\": 1.6061168776618109, \"eval_ref\": \"mlqa_hi_es__4225507a__s000__tb65536__v000\", \"oracle_delta\": 0.08220169279310419, \...
{ "index": "mlqa_hi_es__4225507a__s000__tb65536__v000", "uid": "mlqa_hi_es__4225507a__s000__tb65536__v000", "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\": \"mlqa_hi_es__4225507a__s001__tb131072__v004\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 3.518154658257887, \"nll_base\": 1.257936607332548, \"eval_ref\": \"mlqa_hi_es__4225507a__s001__tb131072__v004\", \"oracle_delta\": 0.05398178572118373,...
{ "index": "mlqa_hi_es__4225507a__s001__tb131072__v004", "uid": "mlqa_hi_es__4225507a__s001__tb131072__v004", "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\": \"mlqa_hi_es__4225507a__s001__tb262144__v005\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 3.518154658257887, \"nll_base\": 1.257936607332548, \"eval_ref\": \"mlqa_hi_es__4225507a__s001__tb262144__v005\", \"oracle_delta\": 0.12215609367757274,...
{ "index": "mlqa_hi_es__4225507a__s001__tb262144__v005", "uid": "mlqa_hi_es__4225507a__s001__tb262144__v005", "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\": \"mlqa_hi_es__4225507a__s001__tb65536__v003\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 3.518154658257887, \"nll_base\": 1.257936607332548, \"eval_ref\": \"mlqa_hi_es__4225507a__s001__tb65536__v003\", \"oracle_delta\": 0.038120227337473, \"ra...
{ "index": "mlqa_hi_es__4225507a__s001__tb65536__v003", "uid": "mlqa_hi_es__4225507a__s001__tb65536__v003", "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\": \"mlqa_hi_es__4225507a__s002__tb131072__v007\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 4.82509608938336, \"nll_base\": 1.573830649490756, \"eval_ref\": \"mlqa_hi_es__4225507a__s002__tb131072__v007\", \"oracle_delta\": 0.08745491067776512, ...
{ "index": "mlqa_hi_es__4225507a__s002__tb131072__v007", "uid": "mlqa_hi_es__4225507a__s002__tb131072__v007", "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\": \"mlqa_hi_es__4225507a__s002__tb262144__v008\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 4.82509608938336, \"nll_base\": 1.573830649490756, \"eval_ref\": \"mlqa_hi_es__4225507a__s002__tb262144__v008\", \"oracle_delta\": 0.1424914454914512, \...
{ "index": "mlqa_hi_es__4225507a__s002__tb262144__v008", "uid": "mlqa_hi_es__4225507a__s002__tb262144__v008", "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\": \"mlqa_hi_es__4225507a__s002__tb65536__v006\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 4.82509608938336, \"nll_base\": 1.573830649490756, \"eval_ref\": \"mlqa_hi_es__4225507a__s002__tb65536__v006\", \"oracle_delta\": 0.04667855307693891, \"r...
{ "index": "mlqa_hi_es__4225507a__s002__tb65536__v006", "uid": "mlqa_hi_es__4225507a__s002__tb65536__v006", "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\": \"mlqa_hi_vi__f42a992d__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 5.13790117773737, \"nll_base\": 1.6366446648976964, \"eval_ref\": \"mlqa_hi_vi__f42a992d__s000__tb131072__v000\", \"oracle_delta\": 0.11421876421784005,...
{ "index": "mlqa_hi_vi__f42a992d__s000__tb131072__v000", "uid": "mlqa_hi_vi__f42a992d__s000__tb131072__v000", "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\": \"mlqa_hi_vi__f42a992d__s000__tb262144__v001\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 5.13790117773737, \"nll_base\": 1.6366446648976964, \"eval_ref\": \"mlqa_hi_vi__f42a992d__s000__tb262144__v001\", \"oracle_delta\": 0.20076742822834337,...
{ "index": "mlqa_hi_vi__f42a992d__s000__tb262144__v001", "uid": "mlqa_hi_vi__f42a992d__s000__tb262144__v001", "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\": \"mlqa_hi_vi__f42a992d__s000__tb65536__v002\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 5.13790117773737, \"nll_base\": 1.6366446648976964, \"eval_ref\": \"mlqa_hi_vi__f42a992d__s000__tb65536__v002\", \"oracle_delta\": 0.07562067980774212, \"...
{ "index": "mlqa_hi_vi__f42a992d__s000__tb65536__v002", "uid": "mlqa_hi_vi__f42a992d__s000__tb65536__v002", "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\": \"mlqa_hi_vi__f42a992d__s001__tb131072__v003\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 3.5500269695212854, \"nll_base\": 1.2669552005067164, \"eval_ref\": \"mlqa_hi_vi__f42a992d__s001__tb131072__v003\", \"oracle_delta\": 0.0811091442497409...
{ "index": "mlqa_hi_vi__f42a992d__s001__tb131072__v003", "uid": "mlqa_hi_vi__f42a992d__s001__tb131072__v003", "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\": \"mlqa_hi_vi__f42a992d__s001__tb262144__v004\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 3.5500269695212854, \"nll_base\": 1.2669552005067164, \"eval_ref\": \"mlqa_hi_vi__f42a992d__s001__tb262144__v004\", \"oracle_delta\": 0.1486160901128030...
{ "index": "mlqa_hi_vi__f42a992d__s001__tb262144__v004", "uid": "mlqa_hi_vi__f42a992d__s001__tb262144__v004", "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\": \"mlqa_hi_vi__f42a992d__s001__tb65536__v005\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 3.5500269695212854, \"nll_base\": 1.2669552005067164, \"eval_ref\": \"mlqa_hi_vi__f42a992d__s001__tb65536__v005\", \"oracle_delta\": 0.046921438100386625,...
{ "index": "mlqa_hi_vi__f42a992d__s001__tb65536__v005", "uid": "mlqa_hi_vi__f42a992d__s001__tb65536__v005", "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\": \"mlqa_hi_vi__f42a992d__s002__tb131072__v006\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 3.76301575492654, \"nll_base\": 1.3252206984020414, \"eval_ref\": \"mlqa_hi_vi__f42a992d__s002__tb131072__v006\", \"oracle_delta\": 0.07367002396356481,...
{ "index": "mlqa_hi_vi__f42a992d__s002__tb131072__v006", "uid": "mlqa_hi_vi__f42a992d__s002__tb131072__v006", "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\": \"mlqa_hi_vi__f42a992d__s002__tb262144__v007\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 3.76301575492654, \"nll_base\": 1.3252206984020414, \"eval_ref\": \"mlqa_hi_vi__f42a992d__s002__tb262144__v007\", \"oracle_delta\": 0.14943309057326526,...
{ "index": "mlqa_hi_vi__f42a992d__s002__tb262144__v007", "uid": "mlqa_hi_vi__f42a992d__s002__tb262144__v007", "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\": \"mlqa_vi_ar__6d015faa__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 6.32426730166026, \"nll_base\": 1.8443941862686821, \"eval_ref\": \"mlqa_vi_ar__6d015faa__s000__tb65536__v000\", \"oracle_delta\": 0.034574160368546325, \...
{ "index": "mlqa_vi_ar__6d015faa__s000__tb65536__v000", "uid": "mlqa_vi_ar__6d015faa__s000__tb65536__v000", "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\": \"mlqa_vi_ar__6d015faa__s002__tb131072__v007\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 6.334940639207534, \"nll_base\": 1.8460804434383618, \"eval_ref\": \"mlqa_vi_ar__6d015faa__s002__tb131072__v007\", \"oracle_delta\": 0.06504154205322266...
{ "index": "mlqa_vi_ar__6d015faa__s002__tb131072__v007", "uid": "mlqa_vi_ar__6d015faa__s002__tb131072__v007", "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\": \"mlqa_vi_ar__6d015faa__s002__tb65536__v006\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 6.334940639207534, \"nll_base\": 1.8460804434383618, \"eval_ref\": \"mlqa_vi_ar__6d015faa__s002__tb65536__v006\", \"oracle_delta\": 0.032466495738310064, ...
{ "index": "mlqa_vi_ar__6d015faa__s002__tb65536__v006", "uid": "mlqa_vi_ar__6d015faa__s002__tb65536__v006", "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\": \"mlqa_vi_de__2b8398c0__s000__tb131072__v001\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 14.206421040032831, \"nll_base\": 2.6536940483197773, \"eval_ref\": \"mlqa_vi_de__2b8398c0__s000__tb131072__v001\", \"oracle_delta\": 0.1282343407199806...
{ "index": "mlqa_vi_de__2b8398c0__s000__tb131072__v001", "uid": "mlqa_vi_de__2b8398c0__s000__tb131072__v001", "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\": \"mlqa_vi_de__2b8398c0__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 14.206421040032831, \"nll_base\": 2.6536940483197773, \"eval_ref\": \"mlqa_vi_de__2b8398c0__s000__tb65536__v000\", \"oracle_delta\": 0.08145685065282526, ...
{ "index": "mlqa_vi_de__2b8398c0__s000__tb65536__v000", "uid": "mlqa_vi_de__2b8398c0__s000__tb65536__v000", "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\": \"mlqa_vi_de__2b8398c0__s002__tb131072__v007\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 6.443010840257996, \"nll_base\": 1.8629959526411983, \"eval_ref\": \"mlqa_vi_de__2b8398c0__s002__tb131072__v007\", \"oracle_delta\": 0.06140948654314782...
{ "index": "mlqa_vi_de__2b8398c0__s002__tb131072__v007", "uid": "mlqa_vi_de__2b8398c0__s002__tb131072__v007", "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\": \"mlqa_vi_de__2b8398c0__s002__tb65536__v006\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 6.443010840257996, \"nll_base\": 1.8629959526411983, \"eval_ref\": \"mlqa_vi_de__2b8398c0__s002__tb65536__v006\", \"oracle_delta\": 0.02084378583715596, \...
{ "index": "mlqa_vi_de__2b8398c0__s002__tb65536__v006", "uid": "mlqa_vi_de__2b8398c0__s002__tb65536__v006", "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\": \"mlqa_vi_es__caa4590d__s000__tb131072__v001\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 6.201710925106611, \"nll_base\": 1.824825209644642, \"eval_ref\": \"mlqa_vi_es__caa4590d__s000__tb131072__v001\", \"oracle_delta\": 0.07137221363391966,...
{ "index": "mlqa_vi_es__caa4590d__s000__tb131072__v001", "uid": "mlqa_vi_es__caa4590d__s000__tb131072__v001", "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\": \"mlqa_vi_es__caa4590d__s001__tb131072__v004\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 5.682381075023573, \"nll_base\": 1.7373703482411045, \"eval_ref\": \"mlqa_vi_es__caa4590d__s001__tb131072__v004\", \"oracle_delta\": 0.06476811604667287...
{ "index": "mlqa_vi_es__caa4590d__s001__tb131072__v004", "uid": "mlqa_vi_es__caa4590d__s001__tb131072__v004", "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\": \"mlqa_vi_es__caa4590d__s001__tb65536__v003\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 5.682381075023573, \"nll_base\": 1.7373703482411045, \"eval_ref\": \"mlqa_vi_es__caa4590d__s001__tb65536__v003\", \"oracle_delta\": 0.0316188057777842, \"...
{ "index": "mlqa_vi_es__caa4590d__s001__tb65536__v003", "uid": "mlqa_vi_es__caa4590d__s001__tb65536__v003", "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\": \"mlqa_vi_es__caa4590d__s002__tb65536__v006\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 4.835733001809443, \"nll_base\": 1.576032720684435, \"eval_ref\": \"mlqa_vi_es__caa4590d__s002__tb65536__v006\", \"oracle_delta\": 0.020373087370414655, \...
{ "index": "mlqa_vi_es__caa4590d__s002__tb65536__v006", "uid": "mlqa_vi_es__caa4590d__s002__tb65536__v006", "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\": \"mlqa_vi_hi__1d322d47__s002__tb131072__v006\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 7.075981731764484, \"nll_base\": 1.9567061945763549, \"eval_ref\": \"mlqa_vi_hi__1d322d47__s002__tb131072__v006\", \"oracle_delta\": 0.08051049060032, \...
{ "index": "mlqa_vi_hi__1d322d47__s002__tb131072__v006", "uid": "mlqa_vi_hi__1d322d47__s002__tb131072__v006", "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\": \"mlqa_vi_hi__1d322d47__s002__tb65536__v008\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 7.075981731764484, \"nll_base\": 1.9567061945763549, \"eval_ref\": \"mlqa_vi_hi__1d322d47__s002__tb65536__v008\", \"oracle_delta\": 0.044884775199142135, ...
{ "index": "mlqa_vi_hi__1d322d47__s002__tb65536__v008", "uid": "mlqa_vi_hi__1d322d47__s002__tb65536__v008", "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\": \"mlqa_vi_vi__a8fd1068__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 6.3825245834887365, \"nll_base\": 1.853563721828197, \"eval_ref\": \"mlqa_vi_vi__a8fd1068__s000__tb131072__v000\", \"oracle_delta\": 0.07535507272465436...
{ "index": "mlqa_vi_vi__a8fd1068__s000__tb131072__v000", "uid": "mlqa_vi_vi__a8fd1068__s000__tb131072__v000", "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\": \"mlqa_vi_vi__a8fd1068__s000__tb65536__v002\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 6.3825245834887365, \"nll_base\": 1.853563721828197, \"eval_ref\": \"mlqa_vi_vi__a8fd1068__s000__tb65536__v002\", \"oracle_delta\": 0.0262935656006984, \"...
{ "index": "mlqa_vi_vi__a8fd1068__s000__tb65536__v002", "uid": "mlqa_vi_vi__a8fd1068__s000__tb65536__v002", "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\": \"mlqa_vi_vi__a8fd1068__s001__tb65536__v005\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 5.359194209487487, \"nll_base\": 1.6788136297305727, \"eval_ref\": \"mlqa_vi_vi__a8fd1068__s001__tb65536__v005\", \"oracle_delta\": 0.028493032709274058, ...
{ "index": "mlqa_vi_vi__a8fd1068__s001__tb65536__v005", "uid": "mlqa_vi_vi__a8fd1068__s001__tb65536__v005", "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\": \"mlqa_vi_vi__a8fd1068__s002__tb131072__v006\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 4.596301321932022, \"nll_base\": 1.5252519196155025, \"eval_ref\": \"mlqa_vi_vi__a8fd1068__s002__tb131072__v006\", \"oracle_delta\": 0.08190834195006147...
{ "index": "mlqa_vi_vi__a8fd1068__s002__tb131072__v006", "uid": "mlqa_vi_vi__a8fd1068__s002__tb131072__v006", "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\": \"mlqa_zh_de__691e9a1a__s000__tb65536__v001\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 3.7815273487724137, \"nll_base\": 1.3301279885428292, \"eval_ref\": \"mlqa_zh_de__691e9a1a__s000__tb65536__v001\", \"oracle_delta\": 0.022706876482282423,...
{ "index": "mlqa_zh_de__691e9a1a__s000__tb65536__v001", "uid": "mlqa_zh_de__691e9a1a__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\": \"mlqa_zh_es__f26bdaf5__s000__tb65536__v002\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 4.332180619261861, \"nll_base\": 1.4660710224666706, \"eval_ref\": \"mlqa_zh_es__f26bdaf5__s000__tb65536__v002\", \"oracle_delta\": 0.022314321738166498, ...
{ "index": "mlqa_zh_es__f26bdaf5__s000__tb65536__v002", "uid": "mlqa_zh_es__f26bdaf5__s000__tb65536__v002", "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\": \"mlqa_zh_es__f26bdaf5__s001__tb65536__v005\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 4.098231228711695, \"nll_base\": 1.410555472995924, \"eval_ref\": \"mlqa_zh_es__f26bdaf5__s001__tb65536__v005\", \"oracle_delta\": 0.02724736752717405, \"...
{ "index": "mlqa_zh_es__f26bdaf5__s001__tb65536__v005", "uid": "mlqa_zh_es__f26bdaf5__s001__tb65536__v005", "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\": \"mlqa_zh_hi__58ac535a__s000__tb65536__v001\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 5.352827058178443, \"nll_base\": 1.6776248434315557, \"eval_ref\": \"mlqa_zh_hi__58ac535a__s000__tb65536__v001\", \"oracle_delta\": 0.021981214440387253, ...
{ "index": "mlqa_zh_hi__58ac535a__s000__tb65536__v001", "uid": "mlqa_zh_hi__58ac535a__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\": \"mlqa_zh_hi__58ac535a__s001__tb65536__v004\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 4.762574848257864, \"nll_base\": 1.5607884564989518, \"eval_ref\": \"mlqa_zh_hi__58ac535a__s001__tb65536__v004\", \"oracle_delta\": 0.023661743419975068, ...
{ "index": "mlqa_zh_hi__58ac535a__s001__tb65536__v004", "uid": "mlqa_zh_hi__58ac535a__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\": \"mlqa_zh_hi__58ac535a__s002__tb65536__v007\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 4.165078089884002, \"nll_base\": 1.4267350245148336, \"eval_ref\": \"mlqa_zh_hi__58ac535a__s002__tb65536__v007\", \"oracle_delta\": 0.024535617506302954, ...
{ "index": "mlqa_zh_hi__58ac535a__s002__tb65536__v007", "uid": "mlqa_zh_hi__58ac535a__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\": \"mlqa_zh_vi__1a6549ec__s000__tb65536__v001\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 4.514639764352144, \"nll_base\": 1.5073253972743585, \"eval_ref\": \"mlqa_zh_vi__1a6549ec__s000__tb65536__v001\", \"oracle_delta\": 0.028362305183720737, ...
{ "index": "mlqa_zh_vi__1a6549ec__s000__tb65536__v001", "uid": "mlqa_zh_vi__1a6549ec__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\": \"mlqa_zh_vi__1a6549ec__s002__tb65536__v007\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 7.235317434996847, \"nll_base\": 1.9789742340341498, \"eval_ref\": \"mlqa_zh_vi__1a6549ec__s002__tb65536__v007\", \"oracle_delta\": 0.029748044846130473, ...
{ "index": "mlqa_zh_vi__1a6549ec__s002__tb65536__v007", "uid": "mlqa_zh_vi__1a6549ec__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\": \"mlqa_zh_zh__4c4757eb__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 4.83049968965259, \"nll_base\": 1.5749499177279538, \"eval_ref\": \"mlqa_zh_zh__4c4757eb__s000__tb65536__v000\", \"oracle_delta\": 0.0365791096845951, \"r...
{ "index": "mlqa_zh_zh__4c4757eb__s000__tb65536__v000", "uid": "mlqa_zh_zh__4c4757eb__s000__tb65536__v000", "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\": \"mlqa_zh_zh__4c4757eb__s002__tb65536__v006\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 2.877714169217465, \"nll_base\": 1.0569962877642634, \"eval_ref\": \"mlqa_zh_zh__4c4757eb__s002__tb65536__v006\", \"oracle_delta\": 0.019356438821316546, ...
{ "index": "mlqa_zh_zh__4c4757eb__s002__tb65536__v006", "uid": "mlqa_zh_zh__4c4757eb__s002__tb65536__v006", "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\": \"mmlu_anatomy__ab901903__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 471.84117578627763, \"nll_base\": 6.156642436981201, \"eval_ref\": \"mmlu_anatomy__ab901903__s000__tb131072__v000\", \"oracle_delta\": 0.6246435642242...
{ "index": "mmlu_anatomy__ab901903__s000__tb131072__v000", "uid": "mmlu_anatomy__ab901903__s000__tb131072__v000", "axis": "knowledge", "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\": \"mmlu_anatomy_continuation__349f2584__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 7.050582797759963, \"nll_base\": 1.953110279756434, \"eval_ref\": \"mmlu_anatomy_continuation__349f2584__s000__tb262144__v000\", \"oracle...
{ "index": "mmlu_anatomy_continuation__349f2584__s000__tb262144__v000", "uid": "mmlu_anatomy_continuation__349f2584__s000__tb262144__v000", "axis": "knowledge", "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\": \"mmlu_anatomy_generative__74851b94__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 918.6932413606218, \"nll_base\": 6.8229522705078125, \"eval_ref\": \"mmlu_anatomy_generative__74851b94__s000__tb262144__v000\", \"oracle_de...
{ "index": "mmlu_anatomy_generative__74851b94__s000__tb262144__v000", "uid": "mmlu_anatomy_generative__74851b94__s000__tb262144__v000", "axis": "knowledge", "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\": \"mmlu_anatomy_generative_spanish__39e05fac__s000__tb131072__v001\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 2598.100071360282, \"nll_base\": 7.862535715103149, \"eval_ref\": \"mmlu_anatomy_generative_spanish__39e05fac__s000__tb131072__v001...
{ "index": "mmlu_anatomy_generative_spanish__39e05fac__s000__tb131072__v001", "uid": "mmlu_anatomy_generative_spanish__39e05fac__s000__tb131072__v001", "axis": "knowledge", "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\": \"mmlu_anatomy_generative_spanish__39e05fac__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 2598.100071360282, \"nll_base\": 7.862535715103149, \"eval_ref\": \"mmlu_anatomy_generative_spanish__39e05fac__s000__tb65536__v000\",...
{ "index": "mmlu_anatomy_generative_spanish__39e05fac__s000__tb65536__v000", "uid": "mmlu_anatomy_generative_spanish__39e05fac__s000__tb65536__v000", "axis": "knowledge", "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\": \"mmlu_astronomy__ab9e5107__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 1231.6615430073382, \"nll_base\": 7.116119384765625, \"eval_ref\": \"mmlu_astronomy__ab9e5107__s000__tb262144__v000\", \"oracle_delta\": 1.015311479...
{ "index": "mmlu_astronomy__ab9e5107__s000__tb262144__v000", "uid": "mmlu_astronomy__ab9e5107__s000__tb262144__v000", "axis": "knowledge", "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\": \"mmlu_astronomy__ab9e5107__s000__tb65536__v001\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 1231.6615430073382, \"nll_base\": 7.116119384765625, \"eval_ref\": \"mmlu_astronomy__ab9e5107__s000__tb65536__v001\", \"oracle_delta\": 0.490876197814...
{ "index": "mmlu_astronomy__ab9e5107__s000__tb65536__v001", "uid": "mmlu_astronomy__ab9e5107__s000__tb65536__v001", "axis": "knowledge", "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\": \"mmlu_astronomy_continuation__5407cd02__s000__tb131072__v001\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 4.349315899638565, \"nll_base\": 1.4700185682823377, \"eval_ref\": \"mmlu_astronomy_continuation__5407cd02__s000__tb131072__v001\", \"o...
{ "index": "mmlu_astronomy_continuation__5407cd02__s000__tb131072__v001", "uid": "mmlu_astronomy_continuation__5407cd02__s000__tb131072__v001", "axis": "knowledge", "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\": \"mmlu_astronomy_continuation__5407cd02__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 4.349315899638565, \"nll_base\": 1.4700185682823377, \"eval_ref\": \"mmlu_astronomy_continuation__5407cd02__s000__tb65536__v000\", \"orac...
{ "index": "mmlu_astronomy_continuation__5407cd02__s000__tb65536__v000", "uid": "mmlu_astronomy_continuation__5407cd02__s000__tb65536__v000", "axis": "knowledge", "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\": \"mmlu_astronomy_generative__8bbc610d__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 2097.8184885940673, \"nll_base\": 7.648653268814087, \"eval_ref\": \"mmlu_astronomy_generative__8bbc610d__s000__tb65536__v000\", \"oracle_d...
{ "index": "mmlu_astronomy_generative__8bbc610d__s000__tb65536__v000", "uid": "mmlu_astronomy_generative__8bbc610d__s000__tb65536__v000", "axis": "knowledge", "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\": \"mmlu_business_ethics__d0b42c5a__s000__tb131072__v001\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 976.9785929045814, \"nll_base\": 6.884464740753174, \"eval_ref\": \"mmlu_business_ethics__d0b42c5a__s000__tb131072__v001\", \"oracle_delta\": ...
{ "index": "mmlu_business_ethics__d0b42c5a__s000__tb131072__v001", "uid": "mmlu_business_ethics__d0b42c5a__s000__tb131072__v001", "axis": "knowledge", "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\": \"mmlu_business_ethics__d0b42c5a__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 976.9785929045814, \"nll_base\": 6.884464740753174, \"eval_ref\": \"mmlu_business_ethics__d0b42c5a__s000__tb65536__v000\", \"oracle_delta\": 0.4...
{ "index": "mmlu_business_ethics__d0b42c5a__s000__tb65536__v000", "uid": "mmlu_business_ethics__d0b42c5a__s000__tb65536__v000", "axis": "knowledge", "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\": \"mmlu_business_ethics_continuation__0e486ff1__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 7.288249333290122, \"nll_base\": 1.9862633708557347, \"eval_ref\": \"mmlu_business_ethics_continuation__0e486ff1__s000__tb65536__v0...
{ "index": "mmlu_business_ethics_continuation__0e486ff1__s000__tb65536__v000", "uid": "mmlu_business_ethics_continuation__0e486ff1__s000__tb65536__v000", "axis": "knowledge", "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\": \"mmlu_business_ethics_generative__b57fd2ce__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 973.6139052884131, \"nll_base\": 6.881014823913574, \"eval_ref\": \"mmlu_business_ethics_generative__b57fd2ce__s000__tb131072__v000...
{ "index": "mmlu_business_ethics_generative__b57fd2ce__s000__tb131072__v000", "uid": "mmlu_business_ethics_generative__b57fd2ce__s000__tb131072__v000", "axis": "knowledge", "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\": \"mmlu_business_ethics_generative__b57fd2ce__s000__tb262144__v001\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 973.6139052884131, \"nll_base\": 6.881014823913574, \"eval_ref\": \"mmlu_business_ethics_generative__b57fd2ce__s000__tb262144__v001...
{ "index": "mmlu_business_ethics_generative__b57fd2ce__s000__tb262144__v001", "uid": "mmlu_business_ethics_generative__b57fd2ce__s000__tb262144__v001", "axis": "knowledge", "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\": \"mmlu_clinical_knowledge__ca9ddebf__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 734.4003651743004, \"nll_base\": 6.599054336547852, \"eval_ref\": \"mmlu_clinical_knowledge__ca9ddebf__s000__tb131072__v000\", \"oracle_del...
{ "index": "mmlu_clinical_knowledge__ca9ddebf__s000__tb131072__v000", "uid": "mmlu_clinical_knowledge__ca9ddebf__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\": \"mmlu_clinical_knowledge__ca9ddebf__s000__tb262144__v001\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 734.4003651743004, \"nll_base\": 6.599054336547852, \"eval_ref\": \"mmlu_clinical_knowledge__ca9ddebf__s000__tb262144__v001\", \"oracle_del...
{ "index": "mmlu_clinical_knowledge__ca9ddebf__s000__tb262144__v001", "uid": "mmlu_clinical_knowledge__ca9ddebf__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\": \"mmlu_clinical_knowledge__ca9ddebf__s000__tb65536__v002\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 734.4003651743004, \"nll_base\": 6.599054336547852, \"eval_ref\": \"mmlu_clinical_knowledge__ca9ddebf__s000__tb65536__v002\", \"oracle_delta\...
{ "index": "mmlu_clinical_knowledge__ca9ddebf__s000__tb65536__v002", "uid": "mmlu_clinical_knowledge__ca9ddebf__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\": \"mmlu_clinical_knowledge_continuation__f5a65042__s000__tb131072__v002\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 6.54784994450052, \"nll_base\": 1.8791367430124224, \"eval_ref\": \"mmlu_clinical_knowledge_continuation__f5a65042__s000__tb13...
{ "index": "mmlu_clinical_knowledge_continuation__f5a65042__s000__tb131072__v002", "uid": "mmlu_clinical_knowledge_continuation__f5a65042__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\": \"mmlu_clinical_knowledge_continuation__f5a65042__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 6.54784994450052, \"nll_base\": 1.8791367430124224, \"eval_ref\": \"mmlu_clinical_knowledge_continuation__f5a65042__s000__tb26...
{ "index": "mmlu_clinical_knowledge_continuation__f5a65042__s000__tb262144__v000", "uid": "mmlu_clinical_knowledge_continuation__f5a65042__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\": \"mmlu_clinical_knowledge_continuation__f5a65042__s000__tb65536__v001\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 6.54784994450052, \"nll_base\": 1.8791367430124224, \"eval_ref\": \"mmlu_clinical_knowledge_continuation__f5a65042__s000__tb6553...
{ "index": "mmlu_clinical_knowledge_continuation__f5a65042__s000__tb65536__v001", "uid": "mmlu_clinical_knowledge_continuation__f5a65042__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\": \"mmlu_clinical_knowledge_continuation__f5a65042__s001__tb262144__v003\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 6.41960082223881, \"nll_base\": 1.8593559385474172, \"eval_ref\": \"mmlu_clinical_knowledge_continuation__f5a65042__s001__tb26...
{ "index": "mmlu_clinical_knowledge_continuation__f5a65042__s001__tb262144__v003", "uid": "mmlu_clinical_knowledge_continuation__f5a65042__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\": \"mmlu_clinical_knowledge_generative__26a7bba0__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 944.1094836930682, \"nll_base\": 6.8502421379089355, \"eval_ref\": \"mmlu_clinical_knowledge_generative__26a7bba0__s000__tb13107...
{ "index": "mmlu_clinical_knowledge_generative__26a7bba0__s000__tb131072__v000", "uid": "mmlu_clinical_knowledge_generative__26a7bba0__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\": \"mmlu_clinical_knowledge_generative__26a7bba0__s000__tb65536__v002\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 944.1094836930682, \"nll_base\": 6.8502421379089355, \"eval_ref\": \"mmlu_clinical_knowledge_generative__26a7bba0__s000__tb65536__...
{ "index": "mmlu_clinical_knowledge_generative__26a7bba0__s000__tb65536__v002", "uid": "mmlu_clinical_knowledge_generative__26a7bba0__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\": \"mmlu_clinical_knowledge_generative__26a7bba0__s001__tb131072__v003\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 839.8292880995253, \"nll_base\": 6.733198642730713, \"eval_ref\": \"mmlu_clinical_knowledge_generative__26a7bba0__s001__tb131072...
{ "index": "mmlu_clinical_knowledge_generative__26a7bba0__s001__tb131072__v003", "uid": "mmlu_clinical_knowledge_generative__26a7bba0__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\": \"mmlu_clinical_knowledge_generative_spanish__6f103040__s000__tb131072__v001\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 6162.056027178376, \"nll_base\": 8.726165771484375, \"eval_ref\": \"mmlu_clinical_knowledge_generative_spanish__6f103040...
{ "index": "mmlu_clinical_knowledge_generative_spanish__6f103040__s000__tb131072__v001", "uid": "mmlu_clinical_knowledge_generative_spanish__6f103040__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\": \"mmlu_clinical_knowledge_generative_spanish__6f103040__s000__tb262144__v002\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 6162.056027178376, \"nll_base\": 8.726165771484375, \"eval_ref\": \"mmlu_clinical_knowledge_generative_spanish__6f103040...
{ "index": "mmlu_clinical_knowledge_generative_spanish__6f103040__s000__tb262144__v002", "uid": "mmlu_clinical_knowledge_generative_spanish__6f103040__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\": \"mmlu_clinical_knowledge_generative_spanish__6f103040__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 6162.056027178376, \"nll_base\": 8.726165771484375, \"eval_ref\": \"mmlu_clinical_knowledge_generative_spanish__6f103040__...
{ "index": "mmlu_clinical_knowledge_generative_spanish__6f103040__s000__tb65536__v000", "uid": "mmlu_clinical_knowledge_generative_spanish__6f103040__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\": \"mmlu_clinical_knowledge_generative_spanish__6f103040__s001__tb65536__v003\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 3427.768173041577, \"nll_base\": 8.139664649963379, \"eval_ref\": \"mmlu_clinical_knowledge_generative_spanish__6f103040__...
{ "index": "mmlu_clinical_knowledge_generative_spanish__6f103040__s001__tb65536__v003", "uid": "mmlu_clinical_knowledge_generative_spanish__6f103040__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\": \"mmlu_college_biology__5f8fcd7e__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 793.972765279777, \"nll_base\": 6.677049160003662, \"eval_ref\": \"mmlu_college_biology__5f8fcd7e__s000__tb262144__v000\", \"oracle_delta\": 1...
{ "index": "mmlu_college_biology__5f8fcd7e__s000__tb262144__v000", "uid": "mmlu_college_biology__5f8fcd7e__s000__tb262144__v000", "axis": "knowledge", "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\": \"mmlu_college_biology_generative__e16eb2b8__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 995.6541360389883, \"nll_base\": 6.90339994430542, \"eval_ref\": \"mmlu_college_biology_generative__e16eb2b8__s000__tb262144__v000\...
{ "index": "mmlu_college_biology_generative__e16eb2b8__s000__tb262144__v000", "uid": "mmlu_college_biology_generative__e16eb2b8__s000__tb262144__v000", "axis": "knowledge", "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\": \"mmlu_college_biology_generative__e16eb2b8__s000__tb65536__v001\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 995.6541360389883, \"nll_base\": 6.90339994430542, \"eval_ref\": \"mmlu_college_biology_generative__e16eb2b8__s000__tb65536__v001\", ...
{ "index": "mmlu_college_biology_generative__e16eb2b8__s000__tb65536__v001", "uid": "mmlu_college_biology_generative__e16eb2b8__s000__tb65536__v001", "axis": "knowledge", "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\": \"mmlu_college_biology_generative_spanish__086a0865__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 3321.6177781411175, \"nll_base\": 8.10820722579956, \"eval_ref\": \"mmlu_college_biology_generative_spanish__086a0865__s000...
{ "index": "mmlu_college_biology_generative_spanish__086a0865__s000__tb262144__v000", "uid": "mmlu_college_biology_generative_spanish__086a0865__s000__tb262144__v000", "axis": "knowledge", "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\": \"mmlu_college_chemistry__1df32180__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 1423.6030182738994, \"nll_base\": 7.260946273803711, \"eval_ref\": \"mmlu_college_chemistry__1df32180__s000__tb65536__v000\", \"oracle_delta\"...
{ "index": "mmlu_college_chemistry__1df32180__s000__tb65536__v000", "uid": "mmlu_college_chemistry__1df32180__s000__tb65536__v000", "axis": "knowledge", "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\": \"mmlu_college_chemistry_continuation__fa4e3a94__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 6.86469706550874, \"nll_base\": 1.9263919108622782, \"eval_ref\": \"mmlu_college_chemistry_continuation__fa4e3a94__s000__tb1310...
{ "index": "mmlu_college_chemistry_continuation__fa4e3a94__s000__tb131072__v000", "uid": "mmlu_college_chemistry_continuation__fa4e3a94__s000__tb131072__v000", "axis": "knowledge", "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\": \"mmlu_college_chemistry_generative__f9dd508b__s000__tb65536__v001\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 999.5388623324209, \"nll_base\": 6.907294034957886, \"eval_ref\": \"mmlu_college_chemistry_generative__f9dd508b__s000__tb65536__v00...
{ "index": "mmlu_college_chemistry_generative__f9dd508b__s000__tb65536__v001", "uid": "mmlu_college_chemistry_generative__f9dd508b__s000__tb65536__v001", "axis": "knowledge", "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\": \"mmlu_college_chemistry_generative_spanish__e7551dec__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 4657.711563736611, \"nll_base\": 8.446279525756836, \"eval_ref\": \"mmlu_college_chemistry_generative_spanish__e7551dec__...
{ "index": "mmlu_college_chemistry_generative_spanish__e7551dec__s000__tb262144__v000", "uid": "mmlu_college_chemistry_generative_spanish__e7551dec__s000__tb262144__v000", "axis": "knowledge", "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\": \"mmlu_college_chemistry_generative_spanish__e7551dec__s000__tb65536__v001\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 4657.711563736611, \"nll_base\": 8.446279525756836, \"eval_ref\": \"mmlu_college_chemistry_generative_spanish__e7551dec__s0...
{ "index": "mmlu_college_chemistry_generative_spanish__e7551dec__s000__tb65536__v001", "uid": "mmlu_college_chemistry_generative_spanish__e7551dec__s000__tb65536__v001", "axis": "knowledge", "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\": \"mmlu_college_computer_science_generative_spanish__0092fcd0__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 2765.1224864391947, \"nll_base\": 7.924840211868286, \"eval_ref\": \"mmlu_college_computer_science_generative_span...
{ "index": "mmlu_college_computer_science_generative_spanish__0092fcd0__s000__tb262144__v000", "uid": "mmlu_college_computer_science_generative_spanish__0092fcd0__s000__tb262144__v000", "axis": "knowledge", "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\": \"mmlu_college_mathematics__9c331d75__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 781.392433559679, \"nll_base\": 6.661077499389648, \"eval_ref\": \"mmlu_college_mathematics__9c331d75__s000__tb131072__v000\", \"oracle_de...
{ "index": "mmlu_college_mathematics__9c331d75__s000__tb131072__v000", "uid": "mmlu_college_mathematics__9c331d75__s000__tb131072__v000", "axis": "knowledge", "budget": 2048, "token_budget": 131072, "variant": "filter" }