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\": \"wmdp_bio__a6d5d642__s000__tb262144__v002\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 789.2002026878009, \"nll_base\": 6.671020030975342, \"eval_ref\": \"wmdp_bio__a6d5d642__s000__tb262144__v002\", \"oracle_delta\": 0.592327356338501, \"ran...
{ "index": "wmdp_bio__a6d5d642__s000__tb262144__v002", "uid": "wmdp_bio__a6d5d642__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\": \"wmdp_bio__a6d5d642__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 789.2002026878009, \"nll_base\": 6.671020030975342, \"eval_ref\": \"wmdp_bio__a6d5d642__s000__tb65536__v000\", \"oracle_delta\": 0.4002549648284912, \"rando...
{ "index": "wmdp_bio__a6d5d642__s000__tb65536__v000", "uid": "wmdp_bio__a6d5d642__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\": \"wmdp_bio__a6d5d642__s001__tb65536__v003\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 1015.1781324129556, \"nll_base\": 6.922819375991821, \"eval_ref\": \"wmdp_bio__a6d5d642__s001__tb65536__v003\", \"oracle_delta\": 0.5998318195343018, \"rand...
{ "index": "wmdp_bio__a6d5d642__s001__tb65536__v003", "uid": "wmdp_bio__a6d5d642__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\": \"wsc__045cf133__s000__tb131072__v001\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 5386.265521676726, \"nll_base\": 8.591607570648193, \"eval_ref\": \"wsc__045cf133__s000__tb131072__v001\", \"oracle_delta\": 0.9646623134613037, \"random_delta...
{ "index": "wsc__045cf133__s000__tb131072__v001", "uid": "wsc__045cf133__s000__tb131072__v001", "axis": "reasoning", "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\": \"wsc__045cf133__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 5386.265521676726, \"nll_base\": 8.591607570648193, \"eval_ref\": \"wsc__045cf133__s000__tb65536__v000\", \"oracle_delta\": 0.21329069137573242, \"random_delta\"...
{ "index": "wsc__045cf133__s000__tb65536__v000", "uid": "wsc__045cf133__s000__tb65536__v000", "axis": "reasoning", "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\": \"wsm__e5edee2c__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 2.4340046025276343, \"nll_base\": 0.889537885495197, \"eval_ref\": \"wsm__e5edee2c__s000__tb131072__v000\", \"oracle_delta\": 0.032188742077304267, \"random_de...
{ "index": "wsm__e5edee2c__s000__tb131072__v000", "uid": "wsm__e5edee2c__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\": \"wsm__e5edee2c__s000__tb262144__v001\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 2.4340046025276343, \"nll_base\": 0.889537885495197, \"eval_ref\": \"wsm__e5edee2c__s000__tb262144__v001\", \"oracle_delta\": 0.0654083430071285, \"random_delt...
{ "index": "wsm__e5edee2c__s000__tb262144__v001", "uid": "wsm__e5edee2c__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\": \"wsm__e5edee2c__s000__tb65536__v002\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 2.4340046025276343, \"nll_base\": 0.889537885495197, \"eval_ref\": \"wsm__e5edee2c__s000__tb65536__v002\", \"oracle_delta\": 0.01900212031857973, \"random_delta\...
{ "index": "wsm__e5edee2c__s000__tb65536__v002", "uid": "wsm__e5edee2c__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\": \"wsm__e5edee2c__s001__tb131072__v003\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 2.46426818795475, \"nll_base\": 0.9018948822624598, \"eval_ref\": \"wsm__e5edee2c__s001__tb131072__v003\", \"oracle_delta\": 0.03458172342051624, \"random_delt...
{ "index": "wsm__e5edee2c__s001__tb131072__v003", "uid": "wsm__e5edee2c__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\": \"wsm__e5edee2c__s001__tb262144__v004\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 2.46426818795475, \"nll_base\": 0.9018948822624598, \"eval_ref\": \"wsm__e5edee2c__s001__tb262144__v004\", \"oracle_delta\": 0.07612440991307434, \"random_delt...
{ "index": "wsm__e5edee2c__s001__tb262144__v004", "uid": "wsm__e5edee2c__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\": \"wsm__e5edee2c__s002__tb131072__v006\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 2.1186334557169495, \"nll_base\": 0.7507712844824374, \"eval_ref\": \"wsm__e5edee2c__s002__tb131072__v006\", \"oracle_delta\": 0.04906942734468145, \"random_de...
{ "index": "wsm__e5edee2c__s002__tb131072__v006", "uid": "wsm__e5edee2c__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\": \"wsm__e5edee2c__s002__tb262144__v007\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 2.1186334557169495, \"nll_base\": 0.7507712844824374, \"eval_ref\": \"wsm__e5edee2c__s002__tb262144__v007\", \"oracle_delta\": 0.0779751533544688, \"random_del...
{ "index": "wsm__e5edee2c__s002__tb262144__v007", "uid": "wsm__e5edee2c__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\": \"wsm__e5edee2c__s002__tb65536__v008\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 2.1186334557169495, \"nll_base\": 0.7507712844824374, \"eval_ref\": \"wsm__e5edee2c__s002__tb65536__v008\", \"oracle_delta\": 0.028137959054240325, \"random_delt...
{ "index": "wsm__e5edee2c__s002__tb65536__v008", "uid": "wsm__e5edee2c__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\": \"wum__9487586f__s000__tb131072__v001\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 4.23334059521843, \"nll_base\": 1.4429914202008929, \"eval_ref\": \"wum__9487586f__s000__tb131072__v001\", \"oracle_delta\": 0.08709912981305812, \"random_delt...
{ "index": "wum__9487586f__s000__tb131072__v001", "uid": "wum__9487586f__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\": \"wum__9487586f__s000__tb262144__v002\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 4.23334059521843, \"nll_base\": 1.4429914202008929, \"eval_ref\": \"wum__9487586f__s000__tb262144__v002\", \"oracle_delta\": 0.12986831665039067, \"random_delt...
{ "index": "wum__9487586f__s000__tb262144__v002", "uid": "wum__9487586f__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\": \"wum__9487586f__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 4.23334059521843, \"nll_base\": 1.4429914202008929, \"eval_ref\": \"wum__9487586f__s000__tb65536__v000\", \"oracle_delta\": 0.05006975446428563, \"random_delta\"...
{ "index": "wum__9487586f__s000__tb65536__v000", "uid": "wum__9487586f__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\": \"wum__9487586f__s001__tb131072__v004\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 3.9347215183666417, \"nll_base\": 1.3698401089372305, \"eval_ref\": \"wum__9487586f__s001__tb131072__v004\", \"oracle_delta\": 0.0977794778758081, \"random_del...
{ "index": "wum__9487586f__s001__tb131072__v004", "uid": "wum__9487586f__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\": \"wum__9487586f__s001__tb262144__v005\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 3.9347215183666417, \"nll_base\": 1.3698401089372305, \"eval_ref\": \"wum__9487586f__s001__tb262144__v005\", \"oracle_delta\": 0.1302865916285021, \"random_del...
{ "index": "wum__9487586f__s001__tb262144__v005", "uid": "wum__9487586f__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\": \"wum__9487586f__s001__tb65536__v003\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 3.9347215183666417, \"nll_base\": 1.3698401089372305, \"eval_ref\": \"wum__9487586f__s001__tb65536__v003\", \"oracle_delta\": 0.06640656569908399, \"random_delta...
{ "index": "wum__9487586f__s001__tb65536__v003", "uid": "wum__9487586f__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\": \"wum__9487586f__s002__tb131072__v007\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 3.5803298142045246, \"nll_base\": 1.2754549230419614, \"eval_ref\": \"wum__9487586f__s002__tb131072__v007\", \"oracle_delta\": 0.09960764758991747, \"random_de...
{ "index": "wum__9487586f__s002__tb131072__v007", "uid": "wum__9487586f__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\": \"wum__9487586f__s002__tb65536__v006\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 3.5803298142045246, \"nll_base\": 1.2754549230419614, \"eval_ref\": \"wum__9487586f__s002__tb65536__v006\", \"oracle_delta\": 0.06570348199808373, \"random_delta...
{ "index": "wum__9487586f__s002__tb65536__v006", "uid": "wum__9487586f__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\": \"xlsum_arabic_prompt_1__3d3be04a__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 8.8533684556375, \"nll_base\": 2.180798003059984, \"eval_ref\": \"xlsum_arabic_prompt_1__3d3be04a__s000__tb262144__v000\", \"oracle_delta\": ...
{ "index": "xlsum_arabic_prompt_1__3d3be04a__s000__tb262144__v000", "uid": "xlsum_arabic_prompt_1__3d3be04a__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\": \"xlsum_oromo_prompt_2__8eca1c26__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 583.099545876607, \"nll_base\": 6.368357919419038, \"eval_ref\": \"xlsum_oromo_prompt_2__8eca1c26__s000__tb131072__v000\", \"oracle_delta\": 0...
{ "index": "xlsum_oromo_prompt_2__8eca1c26__s000__tb131072__v000", "uid": "xlsum_oromo_prompt_2__8eca1c26__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\": \"xquad_de__77279b64__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 7.339364891104039, \"nll_base\": 1.9932523117851966, \"eval_ref\": \"xquad_de__77279b64__s000__tb131072__v000\", \"oracle_delta\": 0.06506237541277393, \"...
{ "index": "xquad_de__77279b64__s000__tb131072__v000", "uid": "xquad_de__77279b64__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\": \"xquad_de__77279b64__s000__tb262144__v001\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 7.339364891104039, \"nll_base\": 1.9932523117851966, \"eval_ref\": \"xquad_de__77279b64__s000__tb262144__v001\", \"oracle_delta\": 0.11805394752738407, \"...
{ "index": "xquad_de__77279b64__s000__tb262144__v001", "uid": "xquad_de__77279b64__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\": \"xquad_de__77279b64__s001__tb131072__v003\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 14.971373056592519, \"nll_base\": 2.706139914772727, \"eval_ref\": \"xquad_de__77279b64__s001__tb131072__v003\", \"oracle_delta\": 0.1684892134232956, \"r...
{ "index": "xquad_de__77279b64__s001__tb131072__v003", "uid": "xquad_de__77279b64__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\": \"xquad_de__77279b64__s001__tb262144__v004\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 14.971373056592519, \"nll_base\": 2.706139914772727, \"eval_ref\": \"xquad_de__77279b64__s001__tb262144__v004\", \"oracle_delta\": 0.26320856267755666, \"...
{ "index": "xquad_de__77279b64__s001__tb262144__v004", "uid": "xquad_de__77279b64__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\": \"xquad_de__77279b64__s001__tb65536__v005\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 14.971373056592519, \"nll_base\": 2.706139914772727, \"eval_ref\": \"xquad_de__77279b64__s001__tb65536__v005\", \"oracle_delta\": 0.07286565607244322, \"ran...
{ "index": "xquad_de__77279b64__s001__tb65536__v005", "uid": "xquad_de__77279b64__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\": \"xquad_de__77279b64__s002__tb131072__v006\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 14.565878988521066, \"nll_base\": 2.6786817379568104, \"eval_ref\": \"xquad_de__77279b64__s002__tb131072__v006\", \"oracle_delta\": 0.10822496382501035, \...
{ "index": "xquad_de__77279b64__s002__tb131072__v006", "uid": "xquad_de__77279b64__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\": \"xquad_de__77279b64__s002__tb262144__v007\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 14.565878988521066, \"nll_base\": 2.6786817379568104, \"eval_ref\": \"xquad_de__77279b64__s002__tb262144__v007\", \"oracle_delta\": 0.22074317298458235, \...
{ "index": "xquad_de__77279b64__s002__tb262144__v007", "uid": "xquad_de__77279b64__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\": \"xquad_de__77279b64__s002__tb65536__v008\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 14.565878988521066, \"nll_base\": 2.6786817379568104, \"eval_ref\": \"xquad_de__77279b64__s002__tb65536__v008\", \"oracle_delta\": 0.05260145624610635, \"ra...
{ "index": "xquad_de__77279b64__s002__tb65536__v008", "uid": "xquad_de__77279b64__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\": \"xquad_el__dcafa09b__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 3.8296057813551467, \"nll_base\": 1.342761868744894, \"eval_ref\": \"xquad_el__dcafa09b__s000__tb262144__v000\", \"oracle_delta\": 0.06567482543147474, \"...
{ "index": "xquad_el__dcafa09b__s000__tb262144__v000", "uid": "xquad_el__dcafa09b__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\": \"xquad_el__dcafa09b__s001__tb131072__v005\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 3.3740344387572128, \"nll_base\": 1.2161091911722217, \"eval_ref\": \"xquad_el__dcafa09b__s001__tb131072__v005\", \"oracle_delta\": 0.04582828318578969, \...
{ "index": "xquad_el__dcafa09b__s001__tb131072__v005", "uid": "xquad_el__dcafa09b__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\": \"xquad_en__c5d5e07a__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 36.15374818909437, \"nll_base\": 3.5877806275755493, \"eval_ref\": \"xquad_en__c5d5e07a__s000__tb65536__v000\", \"oracle_delta\": 0.07299502865298768, \"ran...
{ "index": "xquad_en__c5d5e07a__s000__tb65536__v000", "uid": "xquad_en__c5d5e07a__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\": \"xquad_en__c5d5e07a__s002__tb131072__v007\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 117.05793186618912, \"nll_base\": 4.762668956409801, \"eval_ref\": \"xquad_en__c5d5e07a__s002__tb131072__v007\", \"oracle_delta\": 0.8854298777394476, \"r...
{ "index": "xquad_en__c5d5e07a__s002__tb131072__v007", "uid": "xquad_en__c5d5e07a__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\": \"xquad_en__c5d5e07a__s002__tb262144__v008\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 117.05793186618912, \"nll_base\": 4.762668956409801, \"eval_ref\": \"xquad_en__c5d5e07a__s002__tb262144__v008\", \"oracle_delta\": 1.0876013025060876, \"r...
{ "index": "xquad_en__c5d5e07a__s002__tb262144__v008", "uid": "xquad_en__c5d5e07a__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\": \"xquad_en__c5d5e07a__s002__tb65536__v006\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 117.05793186618912, \"nll_base\": 4.762668956409801, \"eval_ref\": \"xquad_en__c5d5e07a__s002__tb65536__v006\", \"oracle_delta\": 0.49446660822087996, \"ran...
{ "index": "xquad_en__c5d5e07a__s002__tb65536__v006", "uid": "xquad_en__c5d5e07a__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\": \"xquad_ro__13f3dbcb__s000__tb131072__v002\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 13.000887189312557, \"nll_base\": 2.565017600464665, \"eval_ref\": \"xquad_ro__13f3dbcb__s000__tb131072__v002\", \"oracle_delta\": 0.22851662230647474, \"...
{ "index": "xquad_ro__13f3dbcb__s000__tb131072__v002", "uid": "xquad_ro__13f3dbcb__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\": \"xquad_ro__13f3dbcb__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 13.000887189312557, \"nll_base\": 2.565017600464665, \"eval_ref\": \"xquad_ro__13f3dbcb__s000__tb262144__v000\", \"oracle_delta\": 0.3416914597056273, \"r...
{ "index": "xquad_ro__13f3dbcb__s000__tb262144__v000", "uid": "xquad_ro__13f3dbcb__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\": \"xquad_ro__13f3dbcb__s000__tb65536__v001\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 13.000887189312557, \"nll_base\": 2.565017600464665, \"eval_ref\": \"xquad_ro__13f3dbcb__s000__tb65536__v001\", \"oracle_delta\": 0.11795323191125417, \"ran...
{ "index": "xquad_ro__13f3dbcb__s000__tb65536__v001", "uid": "xquad_ro__13f3dbcb__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\": \"xquad_ro__13f3dbcb__s001__tb131072__v005\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 9.802748434992399, \"nll_base\": 2.282662798907306, \"eval_ref\": \"xquad_ro__13f3dbcb__s001__tb131072__v005\", \"oracle_delta\": 0.1706867114917654, \"ra...
{ "index": "xquad_ro__13f3dbcb__s001__tb131072__v005", "uid": "xquad_ro__13f3dbcb__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\": \"xquad_ro__13f3dbcb__s001__tb262144__v003\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 9.802748434992399, \"nll_base\": 2.282662798907306, \"eval_ref\": \"xquad_ro__13f3dbcb__s001__tb262144__v003\", \"oracle_delta\": 0.27547137286211987, \"r...
{ "index": "xquad_ro__13f3dbcb__s001__tb262144__v003", "uid": "xquad_ro__13f3dbcb__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\": \"xquad_ro__13f3dbcb__s001__tb65536__v004\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 9.802748434992399, \"nll_base\": 2.282662798907306, \"eval_ref\": \"xquad_ro__13f3dbcb__s001__tb65536__v004\", \"oracle_delta\": 0.049668430637669125, \"ran...
{ "index": "xquad_ro__13f3dbcb__s001__tb65536__v004", "uid": "xquad_ro__13f3dbcb__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\": \"xquad_ro__13f3dbcb__s002__tb131072__v008\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 16.051065083607675, \"nll_base\": 2.775775207723662, \"eval_ref\": \"xquad_ro__13f3dbcb__s002__tb131072__v008\", \"oracle_delta\": 0.20373933211616846, \"...
{ "index": "xquad_ro__13f3dbcb__s002__tb131072__v008", "uid": "xquad_ro__13f3dbcb__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\": \"xquad_ro__13f3dbcb__s002__tb262144__v006\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 16.051065083607675, \"nll_base\": 2.775775207723662, \"eval_ref\": \"xquad_ro__13f3dbcb__s002__tb262144__v006\", \"oracle_delta\": 0.35796486334657196, \"...
{ "index": "xquad_ro__13f3dbcb__s002__tb262144__v006", "uid": "xquad_ro__13f3dbcb__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\": \"xquad_ro__13f3dbcb__s002__tb65536__v007\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 16.051065083607675, \"nll_base\": 2.775775207723662, \"eval_ref\": \"xquad_ro__13f3dbcb__s002__tb65536__v007\", \"oracle_delta\": 0.09391891836721378, \"ran...
{ "index": "xquad_ro__13f3dbcb__s002__tb65536__v007", "uid": "xquad_ro__13f3dbcb__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\": \"xquad_th__cbb4cc2d__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 3.864854606089199, \"nll_base\": 1.3519240632436311, \"eval_ref\": \"xquad_th__cbb4cc2d__s000__tb131072__v000\", \"oracle_delta\": 0.059669349388473014, \...
{ "index": "xquad_th__cbb4cc2d__s000__tb131072__v000", "uid": "xquad_th__cbb4cc2d__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\": \"xquad_th__cbb4cc2d__s000__tb262144__v001\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 3.864854606089199, \"nll_base\": 1.3519240632436311, \"eval_ref\": \"xquad_th__cbb4cc2d__s000__tb262144__v001\", \"oracle_delta\": 0.0930014478933372, \"r...
{ "index": "xquad_th__cbb4cc2d__s000__tb262144__v001", "uid": "xquad_th__cbb4cc2d__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\": \"xquad_th__cbb4cc2d__s000__tb65536__v002\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 3.864854606089199, \"nll_base\": 1.3519240632436311, \"eval_ref\": \"xquad_th__cbb4cc2d__s000__tb65536__v002\", \"oracle_delta\": 0.03563974685925264, \"ran...
{ "index": "xquad_th__cbb4cc2d__s000__tb65536__v002", "uid": "xquad_th__cbb4cc2d__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\": \"xquad_th__cbb4cc2d__s001__tb131072__v003\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 2.2350953850493718, \"nll_base\": 0.8042839050292969, \"eval_ref\": \"xquad_th__cbb4cc2d__s001__tb131072__v003\", \"oracle_delta\": 0.04666855910728718, \...
{ "index": "xquad_th__cbb4cc2d__s001__tb131072__v003", "uid": "xquad_th__cbb4cc2d__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\": \"xquad_th__cbb4cc2d__s001__tb262144__v004\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 2.2350953850493718, \"nll_base\": 0.8042839050292969, \"eval_ref\": \"xquad_th__cbb4cc2d__s001__tb262144__v004\", \"oracle_delta\": 0.0709845444251751, \"...
{ "index": "xquad_th__cbb4cc2d__s001__tb262144__v004", "uid": "xquad_th__cbb4cc2d__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\": \"xquad_th__cbb4cc2d__s001__tb65536__v005\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 2.2350953850493718, \"nll_base\": 0.8042839050292969, \"eval_ref\": \"xquad_th__cbb4cc2d__s001__tb65536__v005\", \"oracle_delta\": 0.024788902545797464, \"r...
{ "index": "xquad_th__cbb4cc2d__s001__tb65536__v005", "uid": "xquad_th__cbb4cc2d__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\": \"xquad_th__cbb4cc2d__s002__tb262144__v007\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 2.310135540614375, \"nll_base\": 0.8373061984029955, \"eval_ref\": \"xquad_th__cbb4cc2d__s002__tb262144__v007\", \"oracle_delta\": 0.06792763613302544, \"...
{ "index": "xquad_th__cbb4cc2d__s002__tb262144__v007", "uid": "xquad_th__cbb4cc2d__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\": \"xquad_tr__09ea79f5__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 12.435640161466111, \"nll_base\": 2.520566556546781, \"eval_ref\": \"xquad_tr__09ea79f5__s000__tb131072__v000\", \"oracle_delta\": 0.16700382633610156, \"...
{ "index": "xquad_tr__09ea79f5__s000__tb131072__v000", "uid": "xquad_tr__09ea79f5__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\": \"xquad_tr__09ea79f5__s000__tb262144__v001\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 12.435640161466111, \"nll_base\": 2.520566556546781, \"eval_ref\": \"xquad_tr__09ea79f5__s000__tb262144__v001\", \"oracle_delta\": 0.325571512674784, \"ra...
{ "index": "xquad_tr__09ea79f5__s000__tb262144__v001", "uid": "xquad_tr__09ea79f5__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\": \"xquad_tr__09ea79f5__s000__tb65536__v002\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 12.435640161466111, \"nll_base\": 2.520566556546781, \"eval_ref\": \"xquad_tr__09ea79f5__s000__tb65536__v002\", \"oracle_delta\": 0.06592484207840643, \"ran...
{ "index": "xquad_tr__09ea79f5__s000__tb65536__v002", "uid": "xquad_tr__09ea79f5__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\": \"xquad_tr__09ea79f5__s001__tb131072__v003\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 10.787730030366609, \"nll_base\": 2.3784093799705275, \"eval_ref\": \"xquad_tr__09ea79f5__s001__tb131072__v003\", \"oracle_delta\": 0.09404064795214229, \...
{ "index": "xquad_tr__09ea79f5__s001__tb131072__v003", "uid": "xquad_tr__09ea79f5__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\": \"xquad_tr__09ea79f5__s001__tb262144__v004\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 10.787730030366609, \"nll_base\": 2.3784093799705275, \"eval_ref\": \"xquad_tr__09ea79f5__s001__tb262144__v004\", \"oracle_delta\": 0.2225887275741485, \"...
{ "index": "xquad_tr__09ea79f5__s001__tb262144__v004", "uid": "xquad_tr__09ea79f5__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\": \"xquad_tr__09ea79f5__s002__tb131072__v006\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 9.97660920384951, \"nll_base\": 2.300243273458846, \"eval_ref\": \"xquad_tr__09ea79f5__s002__tb131072__v006\", \"oracle_delta\": 0.11765268461300415, \"ra...
{ "index": "xquad_tr__09ea79f5__s002__tb131072__v006", "uid": "xquad_tr__09ea79f5__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\": \"xquad_tr__09ea79f5__s002__tb262144__v007\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 9.97660920384951, \"nll_base\": 2.300243273458846, \"eval_ref\": \"xquad_tr__09ea79f5__s002__tb262144__v007\", \"oracle_delta\": 0.23451924975452565, \"ra...
{ "index": "xquad_tr__09ea79f5__s002__tb262144__v007", "uid": "xquad_tr__09ea79f5__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\": \"xquad_vi__e2feb4a6__s000__tb131072__v002\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 3.3290116251929094, \"nll_base\": 1.2026754506843902, \"eval_ref\": \"xquad_vi__e2feb4a6__s000__tb131072__v002\", \"oracle_delta\": 0.03714457227199719, \...
{ "index": "xquad_vi__e2feb4a6__s000__tb131072__v002", "uid": "xquad_vi__e2feb4a6__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\": \"xquad_vi__e2feb4a6__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 3.3290116251929094, \"nll_base\": 1.2026754506843902, \"eval_ref\": \"xquad_vi__e2feb4a6__s000__tb262144__v000\", \"oracle_delta\": 0.10639274061990212, \...
{ "index": "xquad_vi__e2feb4a6__s000__tb262144__v000", "uid": "xquad_vi__e2feb4a6__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\": \"xquad_vi__e2feb4a6__s001__tb131072__v005\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 7.258061623714632, \"nll_base\": 1.982112799208766, \"eval_ref\": \"xquad_vi__e2feb4a6__s001__tb131072__v005\", \"oracle_delta\": 0.08899871533671111, \"r...
{ "index": "xquad_vi__e2feb4a6__s001__tb131072__v005", "uid": "xquad_vi__e2feb4a6__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\": \"xquad_vi__e2feb4a6__s001__tb262144__v003\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 7.258061623714632, \"nll_base\": 1.982112799208766, \"eval_ref\": \"xquad_vi__e2feb4a6__s001__tb262144__v003\", \"oracle_delta\": 0.20362927129093444, \"r...
{ "index": "xquad_vi__e2feb4a6__s001__tb262144__v003", "uid": "xquad_vi__e2feb4a6__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\": \"xquad_vi__e2feb4a6__s002__tb262144__v006\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 3.306469749439838, \"nll_base\": 1.1958810792747119, \"eval_ref\": \"xquad_vi__e2feb4a6__s002__tb262144__v006\", \"oracle_delta\": 0.0965368392619681, \"r...
{ "index": "xquad_vi__e2feb4a6__s002__tb262144__v006", "uid": "xquad_vi__e2feb4a6__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\": \"xwinograd_pt__2b1e9594__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 104.56680639963919, \"nll_base\": 4.649826162832754, \"eval_ref\": \"xwinograd_pt__2b1e9594__s000__tb131072__v000\", \"oracle_delta\": 0.3684588396990...
{ "index": "xwinograd_pt__2b1e9594__s000__tb131072__v000", "uid": "xwinograd_pt__2b1e9594__s000__tb131072__v000", "axis": "instruction", "budget": 2048, "token_budget": 131072, "variant": "filter" }