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_de_hi__36474374__s001__tb262144__v003\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 17.219207942961326, \"nll_base\": 2.846025501598011, \"eval_ref\": \"mlqa_de_hi__36474374__s001__tb262144__v003\", \"oracle_delta\": 0.550608021129261, ...
{ "index": "mlqa_de_hi__36474374__s001__tb262144__v003", "uid": "mlqa_de_hi__36474374__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_de_hi__36474374__s001__tb65536__v004\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 17.219207942961326, \"nll_base\": 2.846025501598011, \"eval_ref\": \"mlqa_de_hi__36474374__s001__tb65536__v004\", \"oracle_delta\": 0.21175281871448837, \...
{ "index": "mlqa_de_hi__36474374__s001__tb65536__v004", "uid": "mlqa_de_hi__36474374__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_de_hi__36474374__s002__tb131072__v008\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 14.669114689228495, \"nll_base\": 2.6857442419513378, \"eval_ref\": \"mlqa_de_hi__36474374__s002__tb131072__v008\", \"oracle_delta\": 0.3544743357141033...
{ "index": "mlqa_de_hi__36474374__s002__tb131072__v008", "uid": "mlqa_de_hi__36474374__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\": \"mlqa_de_hi__36474374__s002__tb262144__v006\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 14.669114689228495, \"nll_base\": 2.6857442419513378, \"eval_ref\": \"mlqa_de_hi__36474374__s002__tb262144__v006\", \"oracle_delta\": 0.4741774415658191...
{ "index": "mlqa_de_hi__36474374__s002__tb262144__v006", "uid": "mlqa_de_hi__36474374__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_de_hi__36474374__s002__tb65536__v007\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 14.669114689228495, \"nll_base\": 2.6857442419513378, \"eval_ref\": \"mlqa_de_hi__36474374__s002__tb65536__v007\", \"oracle_delta\": 0.18657280417049638, ...
{ "index": "mlqa_de_hi__36474374__s002__tb65536__v007", "uid": "mlqa_de_hi__36474374__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_de_vi__0d4e7a75__s002__tb262144__v006\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 10.030577991611251, \"nll_base\": 2.305638226595792, \"eval_ref\": \"mlqa_de_vi__0d4e7a75__s002__tb262144__v006\", \"oracle_delta\": 0.30714737285267235...
{ "index": "mlqa_de_vi__0d4e7a75__s002__tb262144__v006", "uid": "mlqa_de_vi__0d4e7a75__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_de_vi__0d4e7a75__s002__tb65536__v007\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 10.030577991611251, \"nll_base\": 2.305638226595792, \"eval_ref\": \"mlqa_de_vi__0d4e7a75__s002__tb65536__v007\", \"oracle_delta\": 0.12226902354847313, \...
{ "index": "mlqa_de_vi__0d4e7a75__s002__tb65536__v007", "uid": "mlqa_de_vi__0d4e7a75__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_de_zh__8ca547e5__s000__tb65536__v002\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 11.962143158128413, \"nll_base\": 2.481746926292636, \"eval_ref\": \"mlqa_de_zh__8ca547e5__s000__tb65536__v002\", \"oracle_delta\": 0.04562917041477732, \...
{ "index": "mlqa_de_zh__8ca547e5__s000__tb65536__v002", "uid": "mlqa_de_zh__8ca547e5__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_de_zh__8ca547e5__s001__tb131072__v003\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 11.99736556154563, \"nll_base\": 2.484687089148463, \"eval_ref\": \"mlqa_de_zh__8ca547e5__s001__tb131072__v003\", \"oracle_delta\": 0.14746617190660416,...
{ "index": "mlqa_de_zh__8ca547e5__s001__tb131072__v003", "uid": "mlqa_de_zh__8ca547e5__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_de_zh__8ca547e5__s001__tb65536__v005\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 11.99736556154563, \"nll_base\": 2.484687089148463, \"eval_ref\": \"mlqa_de_zh__8ca547e5__s001__tb65536__v005\", \"oracle_delta\": 0.054121887799605695, \...
{ "index": "mlqa_de_zh__8ca547e5__s001__tb65536__v005", "uid": "mlqa_de_zh__8ca547e5__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_de_zh__8ca547e5__s002__tb131072__v006\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 6.41602900655923, \"nll_base\": 1.8587993915264422, \"eval_ref\": \"mlqa_de_zh__8ca547e5__s002__tb131072__v006\", \"oracle_delta\": 0.2692342569539834, ...
{ "index": "mlqa_de_zh__8ca547e5__s002__tb131072__v006", "uid": "mlqa_de_zh__8ca547e5__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_de_zh__8ca547e5__s002__tb262144__v007\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 6.41602900655923, \"nll_base\": 1.8587993915264422, \"eval_ref\": \"mlqa_de_zh__8ca547e5__s002__tb262144__v007\", \"oracle_delta\": 0.36129002833104384,...
{ "index": "mlqa_de_zh__8ca547e5__s002__tb262144__v007", "uid": "mlqa_de_zh__8ca547e5__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_en_de__d4fc2aed__s000__tb131072__v002\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 49.49925992077538, \"nll_base\": 3.901957718367429, \"eval_ref\": \"mlqa_en_de__d4fc2aed__s000__tb131072__v002\", \"oracle_delta\": 0.30764424432184256,...
{ "index": "mlqa_en_de__d4fc2aed__s000__tb131072__v002", "uid": "mlqa_en_de__d4fc2aed__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_en_de__d4fc2aed__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 49.49925992077538, \"nll_base\": 3.901957718367429, \"eval_ref\": \"mlqa_en_de__d4fc2aed__s000__tb262144__v000\", \"oracle_delta\": 0.5248702531008376, ...
{ "index": "mlqa_en_de__d4fc2aed__s000__tb262144__v000", "uid": "mlqa_en_de__d4fc2aed__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_en_de__d4fc2aed__s000__tb65536__v001\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 49.49925992077538, \"nll_base\": 3.901957718367429, \"eval_ref\": \"mlqa_en_de__d4fc2aed__s000__tb65536__v001\", \"oracle_delta\": 0.15494482296029322, \"...
{ "index": "mlqa_en_de__d4fc2aed__s000__tb65536__v001", "uid": "mlqa_en_de__d4fc2aed__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_en_de__d4fc2aed__s001__tb131072__v005\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 46.0103556817313, \"nll_base\": 3.8288664946684965, \"eval_ref\": \"mlqa_en_de__d4fc2aed__s001__tb131072__v005\", \"oracle_delta\": 0.3996438516152869, ...
{ "index": "mlqa_en_de__d4fc2aed__s001__tb131072__v005", "uid": "mlqa_en_de__d4fc2aed__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_en_de__d4fc2aed__s001__tb262144__v003\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 46.0103556817313, \"nll_base\": 3.8288664946684965, \"eval_ref\": \"mlqa_en_de__d4fc2aed__s001__tb262144__v003\", \"oracle_delta\": 0.6864023569467905, ...
{ "index": "mlqa_en_de__d4fc2aed__s001__tb262144__v003", "uid": "mlqa_en_de__d4fc2aed__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_en_de__d4fc2aed__s001__tb65536__v004\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 46.0103556817313, \"nll_base\": 3.8288664946684965, \"eval_ref\": \"mlqa_en_de__d4fc2aed__s001__tb65536__v004\", \"oracle_delta\": 0.16906276393581088, \"...
{ "index": "mlqa_en_de__d4fc2aed__s001__tb65536__v004", "uid": "mlqa_en_de__d4fc2aed__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_en_de__d4fc2aed__s002__tb131072__v008\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 28.940818212480973, \"nll_base\": 3.365252993522434, \"eval_ref\": \"mlqa_en_de__d4fc2aed__s002__tb131072__v008\", \"oracle_delta\": 0.26095945025802747...
{ "index": "mlqa_en_de__d4fc2aed__s002__tb131072__v008", "uid": "mlqa_en_de__d4fc2aed__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\": \"mlqa_en_de__d4fc2aed__s002__tb65536__v007\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 28.940818212480973, \"nll_base\": 3.365252993522434, \"eval_ref\": \"mlqa_en_de__d4fc2aed__s002__tb65536__v007\", \"oracle_delta\": 0.19463565371452107, \...
{ "index": "mlqa_en_de__d4fc2aed__s002__tb65536__v007", "uid": "mlqa_en_de__d4fc2aed__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_en_es__99cda772__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 100.72847727706294, \"nll_base\": 4.612428552964154, \"eval_ref\": \"mlqa_en_es__99cda772__s000__tb131072__v000\", \"oracle_delta\": 0.4305235021254594,...
{ "index": "mlqa_en_es__99cda772__s000__tb131072__v000", "uid": "mlqa_en_es__99cda772__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_en_es__99cda772__s000__tb262144__v001\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 100.72847727706294, \"nll_base\": 4.612428552964154, \"eval_ref\": \"mlqa_en_es__99cda772__s000__tb262144__v001\", \"oracle_delta\": 0.709057437672334, ...
{ "index": "mlqa_en_es__99cda772__s000__tb262144__v001", "uid": "mlqa_en_es__99cda772__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_en_es__99cda772__s000__tb65536__v002\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 100.72847727706294, \"nll_base\": 4.612428552964154, \"eval_ref\": \"mlqa_en_es__99cda772__s000__tb65536__v002\", \"oracle_delta\": 0.20526930865119475, \...
{ "index": "mlqa_en_es__99cda772__s000__tb65536__v002", "uid": "mlqa_en_es__99cda772__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_en_es__99cda772__s001__tb131072__v003\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 36.73163287311586, \"nll_base\": 3.6036383149587565, \"eval_ref\": \"mlqa_en_es__99cda772__s001__tb131072__v003\", \"oracle_delta\": 0.4047454988895942,...
{ "index": "mlqa_en_es__99cda772__s001__tb131072__v003", "uid": "mlqa_en_es__99cda772__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_en_es__99cda772__s001__tb262144__v004\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 36.73163287311586, \"nll_base\": 3.6036383149587565, \"eval_ref\": \"mlqa_en_es__99cda772__s001__tb262144__v004\", \"oracle_delta\": 0.6992159615918467,...
{ "index": "mlqa_en_es__99cda772__s001__tb262144__v004", "uid": "mlqa_en_es__99cda772__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_en_es__99cda772__s001__tb65536__v005\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 36.73163287311586, \"nll_base\": 3.6036383149587565, \"eval_ref\": \"mlqa_en_es__99cda772__s001__tb65536__v005\", \"oracle_delta\": 0.17327942824000653, \...
{ "index": "mlqa_en_es__99cda772__s001__tb65536__v005", "uid": "mlqa_en_es__99cda772__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_en_es__99cda772__s002__tb131072__v006\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 73.60283047339628, \"nll_base\": 4.29868348251433, \"eval_ref\": \"mlqa_en_es__99cda772__s002__tb131072__v006\", \"oracle_delta\": 0.44460156683385677, ...
{ "index": "mlqa_en_es__99cda772__s002__tb131072__v006", "uid": "mlqa_en_es__99cda772__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_en_hi__f2973210__s000__tb131072__v002\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 85.02947924167448, \"nll_base\": 4.442998010971967, \"eval_ref\": \"mlqa_en_hi__f2973210__s000__tb131072__v002\", \"oracle_delta\": 0.3129539938534007, ...
{ "index": "mlqa_en_hi__f2973210__s000__tb131072__v002", "uid": "mlqa_en_hi__f2973210__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_en_hi__f2973210__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 85.02947924167448, \"nll_base\": 4.442998010971967, \"eval_ref\": \"mlqa_en_hi__f2973210__s000__tb262144__v000\", \"oracle_delta\": 0.5019333783318012, ...
{ "index": "mlqa_en_hi__f2973210__s000__tb262144__v000", "uid": "mlqa_en_hi__f2973210__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_en_hi__f2973210__s000__tb65536__v001\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 85.02947924167448, \"nll_base\": 4.442998010971967, \"eval_ref\": \"mlqa_en_hi__f2973210__s000__tb65536__v001\", \"oracle_delta\": 0.18674370260799567, \"...
{ "index": "mlqa_en_hi__f2973210__s000__tb65536__v001", "uid": "mlqa_en_hi__f2973210__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_en_hi__f2973210__s001__tb131072__v005\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 53.57484125965411, \"nll_base\": 3.981079578399658, \"eval_ref\": \"mlqa_en_hi__f2973210__s001__tb131072__v005\", \"oracle_delta\": 0.33527835210164403,...
{ "index": "mlqa_en_hi__f2973210__s001__tb131072__v005", "uid": "mlqa_en_hi__f2973210__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_en_hi__f2973210__s001__tb262144__v003\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 53.57484125965411, \"nll_base\": 3.981079578399658, \"eval_ref\": \"mlqa_en_hi__f2973210__s001__tb262144__v003\", \"oracle_delta\": 0.5313097635904946, ...
{ "index": "mlqa_en_hi__f2973210__s001__tb262144__v003", "uid": "mlqa_en_hi__f2973210__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_en_hi__f2973210__s001__tb65536__v004\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 53.57484125965411, \"nll_base\": 3.981079578399658, \"eval_ref\": \"mlqa_en_hi__f2973210__s001__tb65536__v004\", \"oracle_delta\": 0.0980917612711587, \"r...
{ "index": "mlqa_en_hi__f2973210__s001__tb65536__v004", "uid": "mlqa_en_hi__f2973210__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_en_hi__f2973210__s002__tb262144__v006\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 35.076701392624, \"nll_base\": 3.5575371320857556, \"eval_ref\": \"mlqa_en_hi__f2973210__s002__tb262144__v006\", \"oracle_delta\": 0.4336674180141715, \...
{ "index": "mlqa_en_hi__f2973210__s002__tb262144__v006", "uid": "mlqa_en_hi__f2973210__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_en_hi__f2973210__s002__tb65536__v007\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 35.076701392624, \"nll_base\": 3.5575371320857556, \"eval_ref\": \"mlqa_en_hi__f2973210__s002__tb65536__v007\", \"oracle_delta\": 0.07571141442587193, \"r...
{ "index": "mlqa_en_hi__f2973210__s002__tb65536__v007", "uid": "mlqa_en_hi__f2973210__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_en_vi__662af003__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 37.16284200031169, \"nll_base\": 3.6153093911156864, \"eval_ref\": \"mlqa_en_vi__662af003__s000__tb65536__v000\", \"oracle_delta\": 0.09357471184190258, \...
{ "index": "mlqa_en_vi__662af003__s000__tb65536__v000", "uid": "mlqa_en_vi__662af003__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_en_vi__662af003__s001__tb131072__v004\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 53.90861604611301, \"nll_base\": 3.987290317572436, \"eval_ref\": \"mlqa_en_vi__662af003__s001__tb131072__v004\", \"oracle_delta\": 0.14164526485702345,...
{ "index": "mlqa_en_vi__662af003__s001__tb131072__v004", "uid": "mlqa_en_vi__662af003__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_en_vi__662af003__s001__tb65536__v003\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 53.90861604611301, \"nll_base\": 3.987290317572436, \"eval_ref\": \"mlqa_en_vi__662af003__s001__tb65536__v003\", \"oracle_delta\": 0.06503740329186858, \"...
{ "index": "mlqa_en_vi__662af003__s001__tb65536__v003", "uid": "mlqa_en_vi__662af003__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_en_vi__662af003__s002__tb131072__v007\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 39.42774429718296, \"nll_base\": 3.6744697384717986, \"eval_ref\": \"mlqa_en_vi__662af003__s002__tb131072__v007\", \"oracle_delta\": 0.26082465939405486...
{ "index": "mlqa_en_vi__662af003__s002__tb131072__v007", "uid": "mlqa_en_vi__662af003__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_en_vi__662af003__s002__tb65536__v006\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 39.42774429718296, \"nll_base\": 3.6744697384717986, \"eval_ref\": \"mlqa_en_vi__662af003__s002__tb65536__v006\", \"oracle_delta\": 0.1356769840891765, \"...
{ "index": "mlqa_en_vi__662af003__s002__tb65536__v006", "uid": "mlqa_en_vi__662af003__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_es_ar__1145e8c0__s000__tb131072__v002\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 11.530664771752496, \"nll_base\": 2.445009988451761, \"eval_ref\": \"mlqa_es_ar__1145e8c0__s000__tb131072__v002\", \"oracle_delta\": 0.3964068700396828,...
{ "index": "mlqa_es_ar__1145e8c0__s000__tb131072__v002", "uid": "mlqa_es_ar__1145e8c0__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_es_ar__1145e8c0__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 11.530664771752496, \"nll_base\": 2.445009988451761, \"eval_ref\": \"mlqa_es_ar__1145e8c0__s000__tb262144__v000\", \"oracle_delta\": 0.5363886596033816,...
{ "index": "mlqa_es_ar__1145e8c0__s000__tb262144__v000", "uid": "mlqa_es_ar__1145e8c0__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_es_ar__1145e8c0__s000__tb65536__v001\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 11.530664771752496, \"nll_base\": 2.445009988451761, \"eval_ref\": \"mlqa_es_ar__1145e8c0__s000__tb65536__v001\", \"oracle_delta\": 0.20271640353732678, \...
{ "index": "mlqa_es_ar__1145e8c0__s000__tb65536__v001", "uid": "mlqa_es_ar__1145e8c0__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_es_ar__1145e8c0__s001__tb131072__v005\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 10.245820044813705, \"nll_base\": 2.326869821905198, \"eval_ref\": \"mlqa_es_ar__1145e8c0__s001__tb131072__v005\", \"oracle_delta\": 0.4019778434772445,...
{ "index": "mlqa_es_ar__1145e8c0__s001__tb131072__v005", "uid": "mlqa_es_ar__1145e8c0__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_es_ar__1145e8c0__s001__tb262144__v003\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 10.245820044813705, \"nll_base\": 2.326869821905198, \"eval_ref\": \"mlqa_es_ar__1145e8c0__s001__tb262144__v003\", \"oracle_delta\": 0.5046807572134118,...
{ "index": "mlqa_es_ar__1145e8c0__s001__tb262144__v003", "uid": "mlqa_es_ar__1145e8c0__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_es_ar__1145e8c0__s001__tb65536__v004\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 10.245820044813705, \"nll_base\": 2.326869821905198, \"eval_ref\": \"mlqa_es_ar__1145e8c0__s001__tb65536__v004\", \"oracle_delta\": 0.15731020045102095, \...
{ "index": "mlqa_es_ar__1145e8c0__s001__tb65536__v004", "uid": "mlqa_es_ar__1145e8c0__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_es_ar__1145e8c0__s002__tb131072__v008\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 12.142039193163923, \"nll_base\": 2.4966737446032075, \"eval_ref\": \"mlqa_es_ar__1145e8c0__s002__tb131072__v008\", \"oracle_delta\": 0.2599152414422288...
{ "index": "mlqa_es_ar__1145e8c0__s002__tb131072__v008", "uid": "mlqa_es_ar__1145e8c0__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\": \"mlqa_es_ar__1145e8c0__s002__tb262144__v006\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 12.142039193163923, \"nll_base\": 2.4966737446032075, \"eval_ref\": \"mlqa_es_ar__1145e8c0__s002__tb262144__v006\", \"oracle_delta\": 0.4266508403577305...
{ "index": "mlqa_es_ar__1145e8c0__s002__tb262144__v006", "uid": "mlqa_es_ar__1145e8c0__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_es_ar__1145e8c0__s002__tb65536__v007\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 12.142039193163923, \"nll_base\": 2.4966737446032075, \"eval_ref\": \"mlqa_es_ar__1145e8c0__s002__tb65536__v007\", \"oracle_delta\": 0.11700559917249187, ...
{ "index": "mlqa_es_ar__1145e8c0__s002__tb65536__v007", "uid": "mlqa_es_ar__1145e8c0__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_es_de__bf8edeef__s000__tb131072__v002\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 17.054272961353266, \"nll_base\": 2.8364007859002975, \"eval_ref\": \"mlqa_es_de__bf8edeef__s000__tb131072__v002\", \"oracle_delta\": 0.3230279831659226...
{ "index": "mlqa_es_de__bf8edeef__s000__tb131072__v002", "uid": "mlqa_es_de__bf8edeef__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_es_de__bf8edeef__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 17.054272961353266, \"nll_base\": 2.8364007859002975, \"eval_ref\": \"mlqa_es_de__bf8edeef__s000__tb262144__v000\", \"oracle_delta\": 0.5701720222594244...
{ "index": "mlqa_es_de__bf8edeef__s000__tb262144__v000", "uid": "mlqa_es_de__bf8edeef__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_es_de__bf8edeef__s000__tb65536__v001\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 17.054272961353266, \"nll_base\": 2.8364007859002975, \"eval_ref\": \"mlqa_es_de__bf8edeef__s000__tb65536__v001\", \"oracle_delta\": 0.1306173657614087, \...
{ "index": "mlqa_es_de__bf8edeef__s000__tb65536__v001", "uid": "mlqa_es_de__bf8edeef__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_es_de__bf8edeef__s001__tb131072__v005\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 11.221832272741873, \"nll_base\": 2.4178611909278187, \"eval_ref\": \"mlqa_es_de__bf8edeef__s001__tb131072__v005\", \"oracle_delta\": 0.3217178916381482...
{ "index": "mlqa_es_de__bf8edeef__s001__tb131072__v005", "uid": "mlqa_es_de__bf8edeef__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_es_de__bf8edeef__s001__tb262144__v003\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 11.221832272741873, \"nll_base\": 2.4178611909278187, \"eval_ref\": \"mlqa_es_de__bf8edeef__s001__tb262144__v003\", \"oracle_delta\": 0.4334518036856087...
{ "index": "mlqa_es_de__bf8edeef__s001__tb262144__v003", "uid": "mlqa_es_de__bf8edeef__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_es_de__bf8edeef__s001__tb65536__v004\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 11.221832272741873, \"nll_base\": 2.4178611909278187, \"eval_ref\": \"mlqa_es_de__bf8edeef__s001__tb65536__v004\", \"oracle_delta\": 0.16387939453125, \"r...
{ "index": "mlqa_es_de__bf8edeef__s001__tb65536__v004", "uid": "mlqa_es_de__bf8edeef__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_es_de__bf8edeef__s002__tb131072__v008\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 15.282404255424863, \"nll_base\": 2.726702117919922, \"eval_ref\": \"mlqa_es_de__bf8edeef__s002__tb131072__v008\", \"oracle_delta\": 0.333832041422526, ...
{ "index": "mlqa_es_de__bf8edeef__s002__tb131072__v008", "uid": "mlqa_es_de__bf8edeef__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\": \"mlqa_es_de__bf8edeef__s002__tb262144__v006\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 15.282404255424863, \"nll_base\": 2.726702117919922, \"eval_ref\": \"mlqa_es_de__bf8edeef__s002__tb262144__v006\", \"oracle_delta\": 0.5455279880099826,...
{ "index": "mlqa_es_de__bf8edeef__s002__tb262144__v006", "uid": "mlqa_es_de__bf8edeef__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_es_de__bf8edeef__s002__tb65536__v007\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 15.282404255424863, \"nll_base\": 2.726702117919922, \"eval_ref\": \"mlqa_es_de__bf8edeef__s002__tb65536__v007\", \"oracle_delta\": 0.15447497897677964, \...
{ "index": "mlqa_es_de__bf8edeef__s002__tb65536__v007", "uid": "mlqa_es_de__bf8edeef__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_es_en__fe984360__s000__tb131072__v002\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 16.222010915893478, \"nll_base\": 2.7863690185546877, \"eval_ref\": \"mlqa_es_en__fe984360__s000__tb131072__v002\", \"oracle_delta\": 0.4090318806966149...
{ "index": "mlqa_es_en__fe984360__s000__tb131072__v002", "uid": "mlqa_es_en__fe984360__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_es_en__fe984360__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 16.222010915893478, \"nll_base\": 2.7863690185546877, \"eval_ref\": \"mlqa_es_en__fe984360__s000__tb262144__v000\", \"oracle_delta\": 0.5705731201171877...
{ "index": "mlqa_es_en__fe984360__s000__tb262144__v000", "uid": "mlqa_es_en__fe984360__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_es_en__fe984360__s000__tb65536__v001\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 16.222010915893478, \"nll_base\": 2.7863690185546877, \"eval_ref\": \"mlqa_es_en__fe984360__s000__tb65536__v001\", \"oracle_delta\": 0.1830403645833334, \...
{ "index": "mlqa_es_en__fe984360__s000__tb65536__v001", "uid": "mlqa_es_en__fe984360__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_es_en__fe984360__s001__tb131072__v005\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 21.682376975504862, \"nll_base\": 3.0764998096530722, \"eval_ref\": \"mlqa_es_en__fe984360__s001__tb131072__v005\", \"oracle_delta\": 0.4982600842492055...
{ "index": "mlqa_es_en__fe984360__s001__tb131072__v005", "uid": "mlqa_es_en__fe984360__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_es_en__fe984360__s001__tb262144__v003\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 21.682376975504862, \"nll_base\": 3.0764998096530722, \"eval_ref\": \"mlqa_es_en__fe984360__s001__tb262144__v003\", \"oracle_delta\": 0.6788724220405191...
{ "index": "mlqa_es_en__fe984360__s001__tb262144__v003", "uid": "mlqa_es_en__fe984360__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_es_en__fe984360__s001__tb65536__v004\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 21.682376975504862, \"nll_base\": 3.0764998096530722, \"eval_ref\": \"mlqa_es_en__fe984360__s001__tb65536__v004\", \"oracle_delta\": 0.18477731478416315, ...
{ "index": "mlqa_es_en__fe984360__s001__tb65536__v004", "uid": "mlqa_es_en__fe984360__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_es_en__fe984360__s002__tb131072__v008\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 16.83397934425813, \"nll_base\": 2.823399423763452, \"eval_ref\": \"mlqa_es_en__fe984360__s002__tb131072__v008\", \"oracle_delta\": 0.5450750692001245, ...
{ "index": "mlqa_es_en__fe984360__s002__tb131072__v008", "uid": "mlqa_es_en__fe984360__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\": \"mlqa_es_en__fe984360__s002__tb262144__v006\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 16.83397934425813, \"nll_base\": 2.823399423763452, \"eval_ref\": \"mlqa_es_en__fe984360__s002__tb262144__v006\", \"oracle_delta\": 0.6987640178756211, ...
{ "index": "mlqa_es_en__fe984360__s002__tb262144__v006", "uid": "mlqa_es_en__fe984360__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_es_en__fe984360__s002__tb65536__v007\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 16.83397934425813, \"nll_base\": 2.823399423763452, \"eval_ref\": \"mlqa_es_en__fe984360__s002__tb65536__v007\", \"oracle_delta\": 0.25161409693837955, \"...
{ "index": "mlqa_es_en__fe984360__s002__tb65536__v007", "uid": "mlqa_es_en__fe984360__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_es_es__bb025674__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 19.879704185336973, \"nll_base\": 2.9896993211044887, \"eval_ref\": \"mlqa_es_es__bb025674__s000__tb131072__v000\", \"oracle_delta\": 0.5712563426224224...
{ "index": "mlqa_es_es__bb025674__s000__tb131072__v000", "uid": "mlqa_es_es__bb025674__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_es_es__bb025674__s000__tb262144__v001\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 19.879704185336973, \"nll_base\": 2.9896993211044887, \"eval_ref\": \"mlqa_es_es__bb025674__s000__tb262144__v001\", \"oracle_delta\": 0.6561776387322809...
{ "index": "mlqa_es_es__bb025674__s000__tb262144__v001", "uid": "mlqa_es_es__bb025674__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_es_es__bb025674__s000__tb65536__v002\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 19.879704185336973, \"nll_base\": 2.9896993211044887, \"eval_ref\": \"mlqa_es_es__bb025674__s000__tb65536__v002\", \"oracle_delta\": 0.2901666909968319, \...
{ "index": "mlqa_es_es__bb025674__s000__tb65536__v002", "uid": "mlqa_es_es__bb025674__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_es_es__bb025674__s001__tb131072__v003\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 12.555752729791449, \"nll_base\": 2.530178945395494, \"eval_ref\": \"mlqa_es_es__bb025674__s001__tb131072__v003\", \"oracle_delta\": 0.32475795151491704...
{ "index": "mlqa_es_es__bb025674__s001__tb131072__v003", "uid": "mlqa_es_es__bb025674__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_es_es__bb025674__s001__tb262144__v004\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 12.555752729791449, \"nll_base\": 2.530178945395494, \"eval_ref\": \"mlqa_es_es__bb025674__s001__tb262144__v004\", \"oracle_delta\": 0.485417957684136, ...
{ "index": "mlqa_es_es__bb025674__s001__tb262144__v004", "uid": "mlqa_es_es__bb025674__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_es_es__bb025674__s001__tb65536__v005\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 12.555752729791449, \"nll_base\": 2.530178945395494, \"eval_ref\": \"mlqa_es_es__bb025674__s001__tb65536__v005\", \"oracle_delta\": 0.16613587981918387, \...
{ "index": "mlqa_es_es__bb025674__s001__tb65536__v005", "uid": "mlqa_es_es__bb025674__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_es_es__bb025674__s002__tb131072__v006\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 16.906749365893248, \"nll_base\": 2.827712912964665, \"eval_ref\": \"mlqa_es_es__bb025674__s002__tb131072__v006\", \"oracle_delta\": 0.5333112330218546,...
{ "index": "mlqa_es_es__bb025674__s002__tb131072__v006", "uid": "mlqa_es_es__bb025674__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_es_es__bb025674__s002__tb262144__v007\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 16.906749365893248, \"nll_base\": 2.827712912964665, \"eval_ref\": \"mlqa_es_es__bb025674__s002__tb262144__v007\", \"oracle_delta\": 0.676981608072917, ...
{ "index": "mlqa_es_es__bb025674__s002__tb262144__v007", "uid": "mlqa_es_es__bb025674__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_es_hi__558609cb__s000__tb131072__v001\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 30.408647610605257, \"nll_base\": 3.414727028836025, \"eval_ref\": \"mlqa_es_hi__558609cb__s000__tb131072__v001\", \"oracle_delta\": 0.5708561015039795,...
{ "index": "mlqa_es_hi__558609cb__s000__tb131072__v001", "uid": "mlqa_es_hi__558609cb__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_es_hi__558609cb__s000__tb262144__v002\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 30.408647610605257, \"nll_base\": 3.414727028836025, \"eval_ref\": \"mlqa_es_hi__558609cb__s000__tb262144__v002\", \"oracle_delta\": 0.793195745918188, ...
{ "index": "mlqa_es_hi__558609cb__s000__tb262144__v002", "uid": "mlqa_es_hi__558609cb__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_es_hi__558609cb__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 30.408647610605257, \"nll_base\": 3.414727028836025, \"eval_ref\": \"mlqa_es_hi__558609cb__s000__tb65536__v000\", \"oracle_delta\": 0.25515209869499067, \...
{ "index": "mlqa_es_hi__558609cb__s000__tb65536__v000", "uid": "mlqa_es_hi__558609cb__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_es_hi__558609cb__s001__tb131072__v004\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 20.08724425192635, \"nll_base\": 3.000084999280098, \"eval_ref\": \"mlqa_es_hi__558609cb__s001__tb131072__v004\", \"oracle_delta\": 0.48763734866411257,...
{ "index": "mlqa_es_hi__558609cb__s001__tb131072__v004", "uid": "mlqa_es_hi__558609cb__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_es_hi__558609cb__s001__tb262144__v005\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 20.08724425192635, \"nll_base\": 3.000084999280098, \"eval_ref\": \"mlqa_es_hi__558609cb__s001__tb262144__v005\", \"oracle_delta\": 0.6248485858623796, ...
{ "index": "mlqa_es_hi__558609cb__s001__tb262144__v005", "uid": "mlqa_es_hi__558609cb__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_es_hi__558609cb__s001__tb65536__v003\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 20.08724425192635, \"nll_base\": 3.000084999280098, \"eval_ref\": \"mlqa_es_hi__558609cb__s001__tb65536__v003\", \"oracle_delta\": 0.24587298662234547, \"...
{ "index": "mlqa_es_hi__558609cb__s001__tb65536__v003", "uid": "mlqa_es_hi__558609cb__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_es_hi__558609cb__s002__tb131072__v007\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 21.003352054780876, \"nll_base\": 3.044682046641474, \"eval_ref\": \"mlqa_es_hi__558609cb__s002__tb131072__v007\", \"oracle_delta\": 0.4252386784207993,...
{ "index": "mlqa_es_hi__558609cb__s002__tb131072__v007", "uid": "mlqa_es_hi__558609cb__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_es_hi__558609cb__s002__tb262144__v008\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 21.003352054780876, \"nll_base\": 3.044682046641474, \"eval_ref\": \"mlqa_es_hi__558609cb__s002__tb262144__v008\", \"oracle_delta\": 0.5912428040435347,...
{ "index": "mlqa_es_hi__558609cb__s002__tb262144__v008", "uid": "mlqa_es_hi__558609cb__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_es_hi__558609cb__s002__tb65536__v006\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 21.003352054780876, \"nll_base\": 3.044682046641474, \"eval_ref\": \"mlqa_es_hi__558609cb__s002__tb65536__v006\", \"oracle_delta\": 0.2503230882727583, \"...
{ "index": "mlqa_es_hi__558609cb__s002__tb65536__v006", "uid": "mlqa_es_hi__558609cb__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_es_vi__ac89fd33__s000__tb131072__v001\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 26.363685052253178, \"nll_base\": 3.271987496990047, \"eval_ref\": \"mlqa_es_vi__ac89fd33__s000__tb131072__v001\", \"oracle_delta\": 0.27808745919841593...
{ "index": "mlqa_es_vi__ac89fd33__s000__tb131072__v001", "uid": "mlqa_es_vi__ac89fd33__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_es_vi__ac89fd33__s000__tb262144__v002\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 26.363685052253178, \"nll_base\": 3.271987496990047, \"eval_ref\": \"mlqa_es_vi__ac89fd33__s000__tb262144__v002\", \"oracle_delta\": 0.47567205559717474...
{ "index": "mlqa_es_vi__ac89fd33__s000__tb262144__v002", "uid": "mlqa_es_vi__ac89fd33__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_es_vi__ac89fd33__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 26.363685052253178, \"nll_base\": 3.271987496990047, \"eval_ref\": \"mlqa_es_vi__ac89fd33__s000__tb65536__v000\", \"oracle_delta\": 0.09267822683674032, \...
{ "index": "mlqa_es_vi__ac89fd33__s000__tb65536__v000", "uid": "mlqa_es_vi__ac89fd33__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_es_vi__ac89fd33__s001__tb131072__v004\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 25.21302857616263, \"nll_base\": 3.2273608679058907, \"eval_ref\": \"mlqa_es_vi__ac89fd33__s001__tb131072__v004\", \"oracle_delta\": 0.31997575431034475...
{ "index": "mlqa_es_vi__ac89fd33__s001__tb131072__v004", "uid": "mlqa_es_vi__ac89fd33__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_es_vi__ac89fd33__s001__tb262144__v005\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 25.21302857616263, \"nll_base\": 3.2273608679058907, \"eval_ref\": \"mlqa_es_vi__ac89fd33__s001__tb262144__v005\", \"oracle_delta\": 0.5209018517271313,...
{ "index": "mlqa_es_vi__ac89fd33__s001__tb262144__v005", "uid": "mlqa_es_vi__ac89fd33__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_es_vi__ac89fd33__s001__tb65536__v003\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 25.21302857616263, \"nll_base\": 3.2273608679058907, \"eval_ref\": \"mlqa_es_vi__ac89fd33__s001__tb65536__v003\", \"oracle_delta\": 0.10288375912954972, \...
{ "index": "mlqa_es_vi__ac89fd33__s001__tb65536__v003", "uid": "mlqa_es_vi__ac89fd33__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_es_vi__ac89fd33__s002__tb131072__v007\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 17.102619286865906, \"nll_base\": 2.839231626449093, \"eval_ref\": \"mlqa_es_vi__ac89fd33__s002__tb131072__v007\", \"oracle_delta\": 0.2935524232925908,...
{ "index": "mlqa_es_vi__ac89fd33__s002__tb131072__v007", "uid": "mlqa_es_vi__ac89fd33__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_es_vi__ac89fd33__s002__tb262144__v008\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 17.102619286865906, \"nll_base\": 2.839231626449093, \"eval_ref\": \"mlqa_es_vi__ac89fd33__s002__tb262144__v008\", \"oracle_delta\": 0.4735135970577118,...
{ "index": "mlqa_es_vi__ac89fd33__s002__tb262144__v008", "uid": "mlqa_es_vi__ac89fd33__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_es_vi__ac89fd33__s002__tb65536__v006\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 17.102619286865906, \"nll_base\": 2.839231626449093, \"eval_ref\": \"mlqa_es_vi__ac89fd33__s002__tb65536__v006\", \"oracle_delta\": 0.1419619652532762, \"...
{ "index": "mlqa_es_vi__ac89fd33__s002__tb65536__v006", "uid": "mlqa_es_vi__ac89fd33__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_es_zh__9a5d7548__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 10.114722585569591, \"nll_base\": 2.313992044200068, \"eval_ref\": \"mlqa_es_zh__9a5d7548__s000__tb131072__v000\", \"oracle_delta\": 0.3828619252080503,...
{ "index": "mlqa_es_zh__9a5d7548__s000__tb131072__v000", "uid": "mlqa_es_zh__9a5d7548__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_es_zh__9a5d7548__s000__tb262144__v001\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 10.114722585569591, \"nll_base\": 2.313992044200068, \"eval_ref\": \"mlqa_es_zh__9a5d7548__s000__tb262144__v001\", \"oracle_delta\": 0.5460129613461702,...
{ "index": "mlqa_es_zh__9a5d7548__s000__tb262144__v001", "uid": "mlqa_es_zh__9a5d7548__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_es_zh__9a5d7548__s000__tb65536__v002\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 10.114722585569591, \"nll_base\": 2.313992044200068, \"eval_ref\": \"mlqa_es_zh__9a5d7548__s000__tb65536__v002\", \"oracle_delta\": 0.15276668382727587, \...
{ "index": "mlqa_es_zh__9a5d7548__s000__tb65536__v002", "uid": "mlqa_es_zh__9a5d7548__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_es_zh__9a5d7548__s001__tb131072__v003\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 13.614716079053897, \"nll_base\": 2.611151272368358, \"eval_ref\": \"mlqa_es_zh__9a5d7548__s001__tb131072__v003\", \"oracle_delta\": 0.4987979620603973,...
{ "index": "mlqa_es_zh__9a5d7548__s001__tb131072__v003", "uid": "mlqa_es_zh__9a5d7548__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_es_zh__9a5d7548__s001__tb262144__v004\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 13.614716079053897, \"nll_base\": 2.611151272368358, \"eval_ref\": \"mlqa_es_zh__9a5d7548__s001__tb262144__v004\", \"oracle_delta\": 0.6307270388355313,...
{ "index": "mlqa_es_zh__9a5d7548__s001__tb262144__v004", "uid": "mlqa_es_zh__9a5d7548__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_es_zh__9a5d7548__s001__tb65536__v005\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 13.614716079053897, \"nll_base\": 2.611151272368358, \"eval_ref\": \"mlqa_es_zh__9a5d7548__s001__tb65536__v005\", \"oracle_delta\": 0.19560339691442064, \...
{ "index": "mlqa_es_zh__9a5d7548__s001__tb65536__v005", "uid": "mlqa_es_zh__9a5d7548__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_es_zh__9a5d7548__s002__tb65536__v008\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 22.9463975186945, \"nll_base\": 3.133160953322354, \"eval_ref\": \"mlqa_es_zh__9a5d7548__s002__tb65536__v008\", \"oracle_delta\": 0.043828133506642164, \"...
{ "index": "mlqa_es_zh__9a5d7548__s002__tb65536__v008", "uid": "mlqa_es_zh__9a5d7548__s002__tb65536__v008", "axis": "qa", "budget": 2048, "token_budget": 65536, "variant": "filter" }