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\": \"cmmlu_clinical_knowledge__17df098b__s000__tb65536__v001\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 1433.9398656491855, \"nll_base\": 7.268181085586548, \"eval_ref\": \"cmmlu_clinical_knowledge__17df098b__s000__tb65536__v001\", \"oracle_del...
{ "index": "cmmlu_clinical_knowledge__17df098b__s000__tb65536__v001", "uid": "cmmlu_clinical_knowledge__17df098b__s000__tb65536__v001", "axis": "domain", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"cmmlu_clinical_knowledge__17df098b__s001__tb262144__v003\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 1324.965566235302, \"nll_base\": 7.189141750335693, \"eval_ref\": \"cmmlu_clinical_knowledge__17df098b__s001__tb262144__v003\", \"oracle_d...
{ "index": "cmmlu_clinical_knowledge__17df098b__s001__tb262144__v003", "uid": "cmmlu_clinical_knowledge__17df098b__s001__tb262144__v003", "axis": "domain", "budget": 2048, "token_budget": 262144, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"cmmlu_college_medical_statistics__46c631ef__s000__tb65536__v001\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 638.2414314326013, \"nll_base\": 6.458716630935669, \"eval_ref\": \"cmmlu_college_medical_statistics__46c631ef__s000__tb65536__v001\...
{ "index": "cmmlu_college_medical_statistics__46c631ef__s000__tb65536__v001", "uid": "cmmlu_college_medical_statistics__46c631ef__s000__tb65536__v001", "axis": "domain", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"cmmlu_legal_and_moral_basis__2a333bd6__s000__tb131072__v002\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 791.1576084052584, \"nll_base\": 6.673497200012207, \"eval_ref\": \"cmmlu_legal_and_moral_basis__2a333bd6__s000__tb131072__v002\", \"or...
{ "index": "cmmlu_legal_and_moral_basis__2a333bd6__s000__tb131072__v002", "uid": "cmmlu_legal_and_moral_basis__2a333bd6__s000__tb131072__v002", "axis": "domain", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"cmmlu_legal_and_moral_basis__2a333bd6__s000__tb65536__v001\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 791.1576084052584, \"nll_base\": 6.673497200012207, \"eval_ref\": \"cmmlu_legal_and_moral_basis__2a333bd6__s000__tb65536__v001\", \"oracl...
{ "index": "cmmlu_legal_and_moral_basis__2a333bd6__s000__tb65536__v001", "uid": "cmmlu_legal_and_moral_basis__2a333bd6__s000__tb65536__v001", "axis": "domain", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"cmmlu_logical__c84f4c26__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 642.101707242344, \"nll_base\": 6.464746713638306, \"eval_ref\": \"cmmlu_logical__c84f4c26__s000__tb65536__v000\", \"oracle_delta\": 0.0675411224365234...
{ "index": "cmmlu_logical__c84f4c26__s000__tb65536__v000", "uid": "cmmlu_logical__c84f4c26__s000__tb65536__v000", "axis": "multilingual", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"cmmlu_public_relations__bc30cbda__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 731.3242833555861, \"nll_base\": 6.5948569774627686, \"eval_ref\": \"cmmlu_public_relations__bc30cbda__s000__tb131072__v000\", \"oracle_delt...
{ "index": "cmmlu_public_relations__bc30cbda__s000__tb131072__v000", "uid": "cmmlu_public_relations__bc30cbda__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\": \"cnn_dailymail_abisee__3977ac28__s000__tb131072__v001\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 11.992174679837076, \"nll_base\": 2.484254327058207, \"eval_ref\": \"cnn_dailymail_abisee__3977ac28__s000__tb131072__v001\", \"oracle_delta\":...
{ "index": "cnn_dailymail_abisee__3977ac28__s000__tb131072__v001", "uid": "cnn_dailymail_abisee__3977ac28__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\": \"cnn_dailymail_abisee__3977ac28__s001__tb65536__v003\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 12.038234419469344, \"nll_base\": 2.488087786226336, \"eval_ref\": \"cnn_dailymail_abisee__3977ac28__s001__tb65536__v003\", \"oracle_delta\": 0....
{ "index": "cnn_dailymail_abisee__3977ac28__s001__tb65536__v003", "uid": "cnn_dailymail_abisee__3977ac28__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\": \"cnn_dailymail_abisee__3977ac28__s002__tb65536__v006\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 9.148722904934376, \"nll_base\": 2.2136142963149643, \"eval_ref\": \"cnn_dailymail_abisee__3977ac28__s002__tb65536__v006\", \"oracle_delta\": 0....
{ "index": "cnn_dailymail_abisee__3977ac28__s002__tb65536__v006", "uid": "cnn_dailymail_abisee__3977ac28__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\": \"code2text_go__c46e7124__s000__tb131072__v001\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 43.153597348859144, \"nll_base\": 3.7647657825100804, \"eval_ref\": \"code2text_go__c46e7124__s000__tb131072__v001\", \"oracle_delta\": 0.044245172316...
{ "index": "code2text_go__c46e7124__s000__tb131072__v001", "uid": "code2text_go__c46e7124__s000__tb131072__v001", "axis": "code", "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\": \"code2text_go__c46e7124__s002__tb131072__v007\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 56.42759780268286, \"nll_base\": 4.0329583615672835, \"eval_ref\": \"code2text_go__c46e7124__s002__tb131072__v007\", \"oracle_delta\": 0.0721140751222...
{ "index": "code2text_go__c46e7124__s002__tb131072__v007", "uid": "code2text_go__c46e7124__s002__tb131072__v007", "axis": "code", "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\": \"code2text_go__c46e7124__s003__tb131072__v010\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 51.05221283528452, \"nll_base\": 3.932848890101892, \"eval_ref\": \"code2text_go__c46e7124__s003__tb131072__v010\", \"oracle_delta\": 0.04047421562272...
{ "index": "code2text_go__c46e7124__s003__tb131072__v010", "uid": "code2text_go__c46e7124__s003__tb131072__v010", "axis": "code", "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\": \"code2text_go__c46e7124__s006__tb131072__v019\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 46.95076298207147, \"nll_base\": 3.8490994564715657, \"eval_ref\": \"code2text_go__c46e7124__s006__tb131072__v019\", \"oracle_delta\": 0.0531966013605...
{ "index": "code2text_go__c46e7124__s006__tb131072__v019", "uid": "code2text_go__c46e7124__s006__tb131072__v019", "axis": "code", "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\": \"code2text_go__c46e7124__s008__tb131072__v025\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 46.31971055112354, \"nll_base\": 3.8355675843489188, \"eval_ref\": \"code2text_go__c46e7124__s008__tb131072__v025\", \"oracle_delta\": 0.0988678765767...
{ "index": "code2text_go__c46e7124__s008__tb131072__v025", "uid": "code2text_go__c46e7124__s008__tb131072__v025", "axis": "code", "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\": \"code2text_go__c46e7124__s008__tb262144__v026\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 46.31971055112354, \"nll_base\": 3.8355675843489188, \"eval_ref\": \"code2text_go__c46e7124__s008__tb262144__v026\", \"oracle_delta\": 0.1459164424080...
{ "index": "code2text_go__c46e7124__s008__tb262144__v026", "uid": "code2text_go__c46e7124__s008__tb262144__v026", "axis": "code", "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\": \"code2text_go__c46e7124__s008__tb65536__v024\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 46.31971055112354, \"nll_base\": 3.8355675843489188, \"eval_ref\": \"code2text_go__c46e7124__s008__tb65536__v024\", \"oracle_delta\": 0.0493777192597684...
{ "index": "code2text_go__c46e7124__s008__tb65536__v024", "uid": "code2text_go__c46e7124__s008__tb65536__v024", "axis": "code", "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\": \"code2text_go__c46e7124__s009__tb131072__v028\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 45.64972543762926, \"nll_base\": 3.820997592513126, \"eval_ref\": \"code2text_go__c46e7124__s009__tb131072__v028\", \"oracle_delta\": 0.05354291826534...
{ "index": "code2text_go__c46e7124__s009__tb131072__v028", "uid": "code2text_go__c46e7124__s009__tb131072__v028", "axis": "code", "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\": \"code2text_go__c46e7124__s009__tb65536__v027\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 45.64972543762926, \"nll_base\": 3.820997592513126, \"eval_ref\": \"code2text_go__c46e7124__s009__tb65536__v027\", \"oracle_delta\": 0.02840116682653448...
{ "index": "code2text_go__c46e7124__s009__tb65536__v027", "uid": "code2text_go__c46e7124__s009__tb65536__v027", "axis": "code", "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\": \"code2text_go__c46e7124__s010__tb131072__v031\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 42.697934552639936, \"nll_base\": 3.754150547932104, \"eval_ref\": \"code2text_go__c46e7124__s010__tb131072__v031\", \"oracle_delta\": 0.0305866359435...
{ "index": "code2text_go__c46e7124__s010__tb131072__v031", "uid": "code2text_go__c46e7124__s010__tb131072__v031", "axis": "code", "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\": \"code2text_go__c46e7124__s011__tb131072__v034\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 34.773272846215164, \"nll_base\": 3.5488490700402147, \"eval_ref\": \"code2text_go__c46e7124__s011__tb131072__v034\", \"oracle_delta\": 0.057972107751...
{ "index": "code2text_go__c46e7124__s011__tb131072__v034", "uid": "code2text_go__c46e7124__s011__tb131072__v034", "axis": "code", "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\": \"code2text_go__c46e7124__s011__tb262144__v035\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 34.773272846215164, \"nll_base\": 3.5488490700402147, \"eval_ref\": \"code2text_go__c46e7124__s011__tb262144__v035\", \"oracle_delta\": 0.127060550147...
{ "index": "code2text_go__c46e7124__s011__tb262144__v035", "uid": "code2text_go__c46e7124__s011__tb262144__v035", "axis": "code", "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\": \"code2text_go__c46e7124__s013__tb131072__v040\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 35.62771679003924, \"nll_base\": 3.573123896454141, \"eval_ref\": \"code2text_go__c46e7124__s013__tb131072__v040\", \"oracle_delta\": 0.07048766900880...
{ "index": "code2text_go__c46e7124__s013__tb131072__v040", "uid": "code2text_go__c46e7124__s013__tb131072__v040", "axis": "code", "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\": \"code2text_go__c46e7124__s013__tb262144__v041\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 35.62771679003924, \"nll_base\": 3.573123896454141, \"eval_ref\": \"code2text_go__c46e7124__s013__tb262144__v041\", \"oracle_delta\": 0.13650219802147...
{ "index": "code2text_go__c46e7124__s013__tb262144__v041", "uid": "code2text_go__c46e7124__s013__tb262144__v041", "axis": "code", "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\": \"code2text_go__c46e7124__s013__tb65536__v039\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 35.62771679003924, \"nll_base\": 3.573123896454141, \"eval_ref\": \"code2text_go__c46e7124__s013__tb65536__v039\", \"oracle_delta\": 0.02738985705301866...
{ "index": "code2text_go__c46e7124__s013__tb65536__v039", "uid": "code2text_go__c46e7124__s013__tb65536__v039", "axis": "code", "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\": \"code2text_go__c46e7124__s014__tb65536__v042\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 51.88317488592003, \"nll_base\": 3.948994554297466, \"eval_ref\": \"code2text_go__c46e7124__s014__tb65536__v042\", \"oracle_delta\": 0.0260973770210855,...
{ "index": "code2text_go__c46e7124__s014__tb65536__v042", "uid": "code2text_go__c46e7124__s014__tb65536__v042", "axis": "code", "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\": \"code2text_java__96cdbc82__s001__tb131072__v003\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 58.57845065722048, \"nll_base\": 4.070366892726514, \"eval_ref\": \"code2text_java__96cdbc82__s001__tb131072__v003\", \"oracle_delta\": 0.1271727011...
{ "index": "code2text_java__96cdbc82__s001__tb131072__v003", "uid": "code2text_java__96cdbc82__s001__tb131072__v003", "axis": "code", "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\": \"code2text_java__96cdbc82__s001__tb65536__v005\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 58.57845065722048, \"nll_base\": 4.070366892726514, \"eval_ref\": \"code2text_java__96cdbc82__s001__tb65536__v005\", \"oracle_delta\": 0.0645936186553...
{ "index": "code2text_java__96cdbc82__s001__tb65536__v005", "uid": "code2text_java__96cdbc82__s001__tb65536__v005", "axis": "code", "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\": \"code2text_java__96cdbc82__s002__tb131072__v006\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 71.87062422028164, \"nll_base\": 4.274867616850754, \"eval_ref\": \"code2text_java__96cdbc82__s002__tb131072__v006\", \"oracle_delta\": 0.1601146829...
{ "index": "code2text_java__96cdbc82__s002__tb131072__v006", "uid": "code2text_java__96cdbc82__s002__tb131072__v006", "axis": "code", "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\": \"code2text_java__96cdbc82__s002__tb262144__v007\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 71.87062422028164, \"nll_base\": 4.274867616850754, \"eval_ref\": \"code2text_java__96cdbc82__s002__tb262144__v007\", \"oracle_delta\": 0.2810066486...
{ "index": "code2text_java__96cdbc82__s002__tb262144__v007", "uid": "code2text_java__96cdbc82__s002__tb262144__v007", "axis": "code", "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\": \"code2text_java__96cdbc82__s002__tb65536__v008\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 71.87062422028164, \"nll_base\": 4.274867616850754, \"eval_ref\": \"code2text_java__96cdbc82__s002__tb65536__v008\", \"oracle_delta\": 0.0731982527108...
{ "index": "code2text_java__96cdbc82__s002__tb65536__v008", "uid": "code2text_java__96cdbc82__s002__tb65536__v008", "axis": "code", "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\": \"code2text_java__96cdbc82__s003__tb131072__v009\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 66.99544558951627, \"nll_base\": 4.204624640804598, \"eval_ref\": \"code2text_java__96cdbc82__s003__tb131072__v009\", \"oracle_delta\": 0.1004669540...
{ "index": "code2text_java__96cdbc82__s003__tb131072__v009", "uid": "code2text_java__96cdbc82__s003__tb131072__v009", "axis": "code", "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\": \"code2text_java__96cdbc82__s003__tb262144__v010\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 66.99544558951627, \"nll_base\": 4.204624640804598, \"eval_ref\": \"code2text_java__96cdbc82__s003__tb262144__v010\", \"oracle_delta\": 0.2817040454...
{ "index": "code2text_java__96cdbc82__s003__tb262144__v010", "uid": "code2text_java__96cdbc82__s003__tb262144__v010", "axis": "code", "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\": \"code2text_java__96cdbc82__s003__tb65536__v011\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 66.99544558951627, \"nll_base\": 4.204624640804598, \"eval_ref\": \"code2text_java__96cdbc82__s003__tb65536__v011\", \"oracle_delta\": 0.0323327777029...
{ "index": "code2text_java__96cdbc82__s003__tb65536__v011", "uid": "code2text_java__96cdbc82__s003__tb65536__v011", "axis": "code", "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\": \"code2text_java__96cdbc82__s004__tb131072__v012\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 61.20382212964876, \"nll_base\": 4.114209640667003, \"eval_ref\": \"code2text_java__96cdbc82__s004__tb131072__v012\", \"oracle_delta\": 0.1041867492...
{ "index": "code2text_java__96cdbc82__s004__tb131072__v012", "uid": "code2text_java__96cdbc82__s004__tb131072__v012", "axis": "code", "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\": \"code2text_java__96cdbc82__s004__tb65536__v014\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 61.20382212964876, \"nll_base\": 4.114209640667003, \"eval_ref\": \"code2text_java__96cdbc82__s004__tb65536__v014\", \"oracle_delta\": 0.0354948966733...
{ "index": "code2text_java__96cdbc82__s004__tb65536__v014", "uid": "code2text_java__96cdbc82__s004__tb65536__v014", "axis": "code", "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\": \"code2text_java__96cdbc82__s006__tb131072__v018\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 56.91845237380992, \"nll_base\": 4.041619583354644, \"eval_ref\": \"code2text_java__96cdbc82__s006__tb131072__v018\", \"oracle_delta\": 0.1462412333...
{ "index": "code2text_java__96cdbc82__s006__tb131072__v018", "uid": "code2text_java__96cdbc82__s006__tb131072__v018", "axis": "code", "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\": \"code2text_java__96cdbc82__s006__tb262144__v019\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 56.91845237380992, \"nll_base\": 4.041619583354644, \"eval_ref\": \"code2text_java__96cdbc82__s006__tb262144__v019\", \"oracle_delta\": 0.2262388438...
{ "index": "code2text_java__96cdbc82__s006__tb262144__v019", "uid": "code2text_java__96cdbc82__s006__tb262144__v019", "axis": "code", "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\": \"code2text_java__96cdbc82__s006__tb65536__v020\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 56.91845237380992, \"nll_base\": 4.041619583354644, \"eval_ref\": \"code2text_java__96cdbc82__s006__tb65536__v020\", \"oracle_delta\": 0.0457969453252...
{ "index": "code2text_java__96cdbc82__s006__tb65536__v020", "uid": "code2text_java__96cdbc82__s006__tb65536__v020", "axis": "code", "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\": \"code2text_java__96cdbc82__s007__tb131072__v021\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 86.17575284472969, \"nll_base\": 4.456388848596038, \"eval_ref\": \"code2text_java__96cdbc82__s007__tb131072__v021\", \"oracle_delta\": 0.1191180241...
{ "index": "code2text_java__96cdbc82__s007__tb131072__v021", "uid": "code2text_java__96cdbc82__s007__tb131072__v021", "axis": "code", "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\": \"code2text_java__96cdbc82__s007__tb262144__v022\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 86.17575284472969, \"nll_base\": 4.456388848596038, \"eval_ref\": \"code2text_java__96cdbc82__s007__tb262144__v022\", \"oracle_delta\": 0.2532120070...
{ "index": "code2text_java__96cdbc82__s007__tb262144__v022", "uid": "code2text_java__96cdbc82__s007__tb262144__v022", "axis": "code", "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\": \"code2text_java__96cdbc82__s007__tb65536__v023\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 86.17575284472969, \"nll_base\": 4.456388848596038, \"eval_ref\": \"code2text_java__96cdbc82__s007__tb65536__v023\", \"oracle_delta\": 0.0467402885149...
{ "index": "code2text_java__96cdbc82__s007__tb65536__v023", "uid": "code2text_java__96cdbc82__s007__tb65536__v023", "axis": "code", "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\": \"code2text_java__96cdbc82__s009__tb131072__v027\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 58.476289369935586, \"nll_base\": 4.068621362161915, \"eval_ref\": \"code2text_java__96cdbc82__s009__tb131072__v027\", \"oracle_delta\": 0.111458404...
{ "index": "code2text_java__96cdbc82__s009__tb131072__v027", "uid": "code2text_java__96cdbc82__s009__tb131072__v027", "axis": "code", "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\": \"code2text_java__96cdbc82__s009__tb262144__v028\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 58.476289369935586, \"nll_base\": 4.068621362161915, \"eval_ref\": \"code2text_java__96cdbc82__s009__tb262144__v028\", \"oracle_delta\": 0.247621651...
{ "index": "code2text_java__96cdbc82__s009__tb262144__v028", "uid": "code2text_java__96cdbc82__s009__tb262144__v028", "axis": "code", "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\": \"code2text_java__96cdbc82__s009__tb65536__v029\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 58.476289369935586, \"nll_base\": 4.068621362161915, \"eval_ref\": \"code2text_java__96cdbc82__s009__tb65536__v029\", \"oracle_delta\": 0.049688803986...
{ "index": "code2text_java__96cdbc82__s009__tb65536__v029", "uid": "code2text_java__96cdbc82__s009__tb65536__v029", "axis": "code", "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\": \"code2text_java__96cdbc82__s010__tb131072__v030\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 51.36442841910059, \"nll_base\": 3.938945878778036, \"eval_ref\": \"code2text_java__96cdbc82__s010__tb131072__v030\", \"oracle_delta\": 0.2182148405...
{ "index": "code2text_java__96cdbc82__s010__tb131072__v030", "uid": "code2text_java__96cdbc82__s010__tb131072__v030", "axis": "code", "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\": \"code2text_java__96cdbc82__s010__tb262144__v031\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 51.36442841910059, \"nll_base\": 3.938945878778036, \"eval_ref\": \"code2text_java__96cdbc82__s010__tb262144__v031\", \"oracle_delta\": 0.3434412182...
{ "index": "code2text_java__96cdbc82__s010__tb262144__v031", "uid": "code2text_java__96cdbc82__s010__tb262144__v031", "axis": "code", "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\": \"code2text_java__96cdbc82__s010__tb65536__v032\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 51.36442841910059, \"nll_base\": 3.938945878778036, \"eval_ref\": \"code2text_java__96cdbc82__s010__tb65536__v032\", \"oracle_delta\": 0.0990879718271...
{ "index": "code2text_java__96cdbc82__s010__tb65536__v032", "uid": "code2text_java__96cdbc82__s010__tb65536__v032", "axis": "code", "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\": \"code2text_java__96cdbc82__s012__tb131072__v036\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 70.49466406884821, \"nll_base\": 4.255537019987128, \"eval_ref\": \"code2text_java__96cdbc82__s012__tb131072__v036\", \"oracle_delta\": 0.1245062716...
{ "index": "code2text_java__96cdbc82__s012__tb131072__v036", "uid": "code2text_java__96cdbc82__s012__tb131072__v036", "axis": "code", "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\": \"code2text_java__96cdbc82__s012__tb262144__v037\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 70.49466406884821, \"nll_base\": 4.255537019987128, \"eval_ref\": \"code2text_java__96cdbc82__s012__tb262144__v037\", \"oracle_delta\": 0.3190204262...
{ "index": "code2text_java__96cdbc82__s012__tb262144__v037", "uid": "code2text_java__96cdbc82__s012__tb262144__v037", "axis": "code", "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\": \"code2text_java__96cdbc82__s012__tb65536__v038\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 70.49466406884821, \"nll_base\": 4.255537019987128, \"eval_ref\": \"code2text_java__96cdbc82__s012__tb65536__v038\", \"oracle_delta\": 0.0657601432887...
{ "index": "code2text_java__96cdbc82__s012__tb65536__v038", "uid": "code2text_java__96cdbc82__s012__tb65536__v038", "axis": "code", "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\": \"code2text_php__24bf8568__s000__tb131072__v001\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 102.18045868941624, \"nll_base\": 4.626740452925814, \"eval_ref\": \"code2text_php__24bf8568__s000__tb131072__v001\", \"oracle_delta\": 0.11129510031...
{ "index": "code2text_php__24bf8568__s000__tb131072__v001", "uid": "code2text_php__24bf8568__s000__tb131072__v001", "axis": "code", "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\": \"code2text_php__24bf8568__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 102.18045868941624, \"nll_base\": 4.626740452925814, \"eval_ref\": \"code2text_php__24bf8568__s000__tb65536__v000\", \"oracle_delta\": 0.04475260186127...
{ "index": "code2text_php__24bf8568__s000__tb65536__v000", "uid": "code2text_php__24bf8568__s000__tb65536__v000", "axis": "code", "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\": \"code2text_php__24bf8568__s001__tb131072__v004\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 105.82581352537734, \"nll_base\": 4.66179447383671, \"eval_ref\": \"code2text_php__24bf8568__s001__tb131072__v004\", \"oracle_delta\": 0.118562803163...
{ "index": "code2text_php__24bf8568__s001__tb131072__v004", "uid": "code2text_php__24bf8568__s001__tb131072__v004", "axis": "code", "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\": \"code2text_php__24bf8568__s001__tb65536__v003\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 105.82581352537734, \"nll_base\": 4.66179447383671, \"eval_ref\": \"code2text_php__24bf8568__s001__tb65536__v003\", \"oracle_delta\": 0.064263312371222...
{ "index": "code2text_php__24bf8568__s001__tb65536__v003", "uid": "code2text_php__24bf8568__s001__tb65536__v003", "axis": "code", "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\": \"code2text_php__24bf8568__s002__tb131072__v007\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 55.90253854768801, \"nll_base\": 4.023609791432057, \"eval_ref\": \"code2text_php__24bf8568__s002__tb131072__v007\", \"oracle_delta\": 0.112472946579...
{ "index": "code2text_php__24bf8568__s002__tb131072__v007", "uid": "code2text_php__24bf8568__s002__tb131072__v007", "axis": "code", "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\": \"code2text_php__24bf8568__s002__tb65536__v006\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 55.90253854768801, \"nll_base\": 4.023609791432057, \"eval_ref\": \"code2text_php__24bf8568__s002__tb65536__v006\", \"oracle_delta\": 0.051893629469313...
{ "index": "code2text_php__24bf8568__s002__tb65536__v006", "uid": "code2text_php__24bf8568__s002__tb65536__v006", "axis": "code", "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\": \"code2text_php__24bf8568__s003__tb131072__v010\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 75.33297124607097, \"nll_base\": 4.321917904124541, \"eval_ref\": \"code2text_php__24bf8568__s003__tb131072__v010\", \"oracle_delta\": 0.174174050723...
{ "index": "code2text_php__24bf8568__s003__tb131072__v010", "uid": "code2text_php__24bf8568__s003__tb131072__v010", "axis": "code", "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\": \"code2text_php__24bf8568__s003__tb65536__v009\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 75.33297124607097, \"nll_base\": 4.321917904124541, \"eval_ref\": \"code2text_php__24bf8568__s003__tb65536__v009\", \"oracle_delta\": 0.075699570599724...
{ "index": "code2text_php__24bf8568__s003__tb65536__v009", "uid": "code2text_php__24bf8568__s003__tb65536__v009", "axis": "code", "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\": \"code2text_php__24bf8568__s006__tb131072__v019\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 88.21873802773135, \"nll_base\": 4.479819389742596, \"eval_ref\": \"code2text_php__24bf8568__s006__tb131072__v019\", \"oracle_delta\": 0.141930782433...
{ "index": "code2text_php__24bf8568__s006__tb131072__v019", "uid": "code2text_php__24bf8568__s006__tb131072__v019", "axis": "code", "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\": \"code2text_php__24bf8568__s006__tb65536__v018\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 88.21873802773135, \"nll_base\": 4.479819389742596, \"eval_ref\": \"code2text_php__24bf8568__s006__tb65536__v018\", \"oracle_delta\": 0.070653321657627...
{ "index": "code2text_php__24bf8568__s006__tb65536__v018", "uid": "code2text_php__24bf8568__s006__tb65536__v018", "axis": "code", "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\": \"code2text_php__24bf8568__s007__tb131072__v022\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 86.41421374690266, \"nll_base\": 4.4591521732390875, \"eval_ref\": \"code2text_php__24bf8568__s007__tb131072__v022\", \"oracle_delta\": 0.18378964378...
{ "index": "code2text_php__24bf8568__s007__tb131072__v022", "uid": "code2text_php__24bf8568__s007__tb131072__v022", "axis": "code", "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\": \"code2text_php__24bf8568__s007__tb65536__v021\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 86.41421374690266, \"nll_base\": 4.4591521732390875, \"eval_ref\": \"code2text_php__24bf8568__s007__tb65536__v021\", \"oracle_delta\": 0.06750585162450...
{ "index": "code2text_php__24bf8568__s007__tb65536__v021", "uid": "code2text_php__24bf8568__s007__tb65536__v021", "axis": "code", "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\": \"code2text_php__24bf8568__s008__tb131072__v025\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 77.5552571672572, \"nll_base\": 4.350990678005163, \"eval_ref\": \"code2text_php__24bf8568__s008__tb131072__v025\", \"oracle_delta\": 0.1506786965339...
{ "index": "code2text_php__24bf8568__s008__tb131072__v025", "uid": "code2text_php__24bf8568__s008__tb131072__v025", "axis": "code", "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\": \"code2text_php__24bf8568__s008__tb65536__v024\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 77.5552571672572, \"nll_base\": 4.350990678005163, \"eval_ref\": \"code2text_php__24bf8568__s008__tb65536__v024\", \"oracle_delta\": 0.0222243587527657...
{ "index": "code2text_php__24bf8568__s008__tb65536__v024", "uid": "code2text_php__24bf8568__s008__tb65536__v024", "axis": "code", "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\": \"code2text_php__24bf8568__s009__tb131072__v028\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 80.60866289714976, \"nll_base\": 4.3896061238505295, \"eval_ref\": \"code2text_php__24bf8568__s009__tb131072__v028\", \"oracle_delta\": 0.10894684066...
{ "index": "code2text_php__24bf8568__s009__tb131072__v028", "uid": "code2text_php__24bf8568__s009__tb131072__v028", "axis": "code", "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\": \"code2text_php__24bf8568__s009__tb262144__v029\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 80.60866289714976, \"nll_base\": 4.3896061238505295, \"eval_ref\": \"code2text_php__24bf8568__s009__tb262144__v029\", \"oracle_delta\": 0.25201370353...
{ "index": "code2text_php__24bf8568__s009__tb262144__v029", "uid": "code2text_php__24bf8568__s009__tb262144__v029", "axis": "code", "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\": \"code2text_php__24bf8568__s009__tb65536__v027\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 80.60866289714976, \"nll_base\": 4.3896061238505295, \"eval_ref\": \"code2text_php__24bf8568__s009__tb65536__v027\", \"oracle_delta\": 0.04730011519053...
{ "index": "code2text_php__24bf8568__s009__tb65536__v027", "uid": "code2text_php__24bf8568__s009__tb65536__v027", "axis": "code", "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\": \"code2text_php__24bf8568__s010__tb131072__v031\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 107.31828853663966, \"nll_base\": 4.675799078138423, \"eval_ref\": \"code2text_php__24bf8568__s010__tb131072__v031\", \"oracle_delta\": 0.10551300573...
{ "index": "code2text_php__24bf8568__s010__tb131072__v031", "uid": "code2text_php__24bf8568__s010__tb131072__v031", "axis": "code", "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\": \"code2text_php__24bf8568__s010__tb65536__v030\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 107.31828853663966, \"nll_base\": 4.675799078138423, \"eval_ref\": \"code2text_php__24bf8568__s010__tb65536__v030\", \"oracle_delta\": 0.06524825997368...
{ "index": "code2text_php__24bf8568__s010__tb65536__v030", "uid": "code2text_php__24bf8568__s010__tb65536__v030", "axis": "code", "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\": \"code2text_php__24bf8568__s011__tb131072__v034\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 91.67397758525372, \"nll_base\": 4.518238561293658, \"eval_ref\": \"code2text_php__24bf8568__s011__tb131072__v034\", \"oracle_delta\": 0.204816750919...
{ "index": "code2text_php__24bf8568__s011__tb131072__v034", "uid": "code2text_php__24bf8568__s011__tb131072__v034", "axis": "code", "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\": \"code2text_php__24bf8568__s011__tb262144__v035\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 91.67397758525372, \"nll_base\": 4.518238561293658, \"eval_ref\": \"code2text_php__24bf8568__s011__tb262144__v035\", \"oracle_delta\": 0.315687291762...
{ "index": "code2text_php__24bf8568__s011__tb262144__v035", "uid": "code2text_php__24bf8568__s011__tb262144__v035", "axis": "code", "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\": \"code2text_php__24bf8568__s011__tb65536__v033\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 91.67397758525372, \"nll_base\": 4.518238561293658, \"eval_ref\": \"code2text_php__24bf8568__s011__tb65536__v033\", \"oracle_delta\": 0.079556633444393...
{ "index": "code2text_php__24bf8568__s011__tb65536__v033", "uid": "code2text_php__24bf8568__s011__tb65536__v033", "axis": "code", "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\": \"code2text_php__24bf8568__s012__tb131072__v037\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 111.63762861073462, \"nll_base\": 4.715258167048169, \"eval_ref\": \"code2text_php__24bf8568__s012__tb131072__v037\", \"oracle_delta\": 0.20443064695...
{ "index": "code2text_php__24bf8568__s012__tb131072__v037", "uid": "code2text_php__24bf8568__s012__tb131072__v037", "axis": "code", "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\": \"code2text_php__24bf8568__s012__tb65536__v036\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 111.63762861073462, \"nll_base\": 4.715258167048169, \"eval_ref\": \"code2text_php__24bf8568__s012__tb65536__v036\", \"oracle_delta\": 0.06416874782294...
{ "index": "code2text_php__24bf8568__s012__tb65536__v036", "uid": "code2text_php__24bf8568__s012__tb65536__v036", "axis": "code", "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\": \"code2text_php__24bf8568__s013__tb131072__v040\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 80.19615847376214, \"nll_base\": 4.38447561439561, \"eval_ref\": \"code2text_php__24bf8568__s013__tb131072__v040\", \"oracle_delta\": 0.1050521756973...
{ "index": "code2text_php__24bf8568__s013__tb131072__v040", "uid": "code2text_php__24bf8568__s013__tb131072__v040", "axis": "code", "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\": \"code2text_python__723e914a__s002__tb131072__v007\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 78.34056379340292, \"nll_base\": 4.36106552496189, \"eval_ref\": \"code2text_python__723e914a__s002__tb131072__v007\", \"oracle_delta\": 0.0787097...
{ "index": "code2text_python__723e914a__s002__tb131072__v007", "uid": "code2text_python__723e914a__s002__tb131072__v007", "axis": "code", "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\": \"code2text_python__723e914a__s002__tb65536__v006\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 78.34056379340292, \"nll_base\": 4.36106552496189, \"eval_ref\": \"code2text_python__723e914a__s002__tb65536__v006\", \"oracle_delta\": 0.0333594345...
{ "index": "code2text_python__723e914a__s002__tb65536__v006", "uid": "code2text_python__723e914a__s002__tb65536__v006", "axis": "code", "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\": \"code2text_python__723e914a__s007__tb65536__v021\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 96.38798369974351, \"nll_base\": 4.568381543429393, \"eval_ref\": \"code2text_python__723e914a__s007__tb65536__v021\", \"oracle_delta\": 0.041194987...
{ "index": "code2text_python__723e914a__s007__tb65536__v021", "uid": "code2text_python__723e914a__s007__tb65536__v021", "axis": "code", "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\": \"code2text_python__723e914a__s008__tb131072__v025\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 72.88790024409566, \"nll_base\": 4.288922647791584, \"eval_ref\": \"code2text_python__723e914a__s008__tb131072__v025\", \"oracle_delta\": 0.058850...
{ "index": "code2text_python__723e914a__s008__tb131072__v025", "uid": "code2text_python__723e914a__s008__tb131072__v025", "axis": "code", "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\": \"code2text_python__723e914a__s012__tb131072__v037\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 74.04125490112963, \"nll_base\": 4.304622436523437, \"eval_ref\": \"code2text_python__723e914a__s012__tb131072__v037\", \"oracle_delta\": 0.084441...
{ "index": "code2text_python__723e914a__s012__tb131072__v037", "uid": "code2text_python__723e914a__s012__tb131072__v037", "axis": "code", "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\": \"code2text_python__723e914a__s012__tb65536__v036\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 74.04125490112963, \"nll_base\": 4.304622436523437, \"eval_ref\": \"code2text_python__723e914a__s012__tb65536__v036\", \"oracle_delta\": 0.035608154...
{ "index": "code2text_python__723e914a__s012__tb65536__v036", "uid": "code2text_python__723e914a__s012__tb65536__v036", "axis": "code", "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\": \"code2text_python__723e914a__s013__tb131072__v040\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 81.02992369527468, \"nll_base\": 4.394818514787317, \"eval_ref\": \"code2text_python__723e914a__s013__tb131072__v040\", \"oracle_delta\": 0.107162...
{ "index": "code2text_python__723e914a__s013__tb131072__v040", "uid": "code2text_python__723e914a__s013__tb131072__v040", "axis": "code", "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\": \"code2text_python__723e914a__s013__tb65536__v039\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 81.02992369527468, \"nll_base\": 4.394818514787317, \"eval_ref\": \"code2text_python__723e914a__s013__tb65536__v039\", \"oracle_delta\": 0.043974383...
{ "index": "code2text_python__723e914a__s013__tb65536__v039", "uid": "code2text_python__723e914a__s013__tb65536__v039", "axis": "code", "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\": \"code2text_python__723e914a__s014__tb65536__v042\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 53.950453687423156, \"nll_base\": 3.9880661010742187, \"eval_ref\": \"code2text_python__723e914a__s014__tb65536__v042\", \"oracle_delta\": 0.0245110...
{ "index": "code2text_python__723e914a__s014__tb65536__v042", "uid": "code2text_python__723e914a__s014__tb65536__v042", "axis": "code", "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\": \"code2text_ruby__23ddcce8__s001__tb65536__v005\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 81.73631564599613, \"nll_base\": 4.403498403056112, \"eval_ref\": \"code2text_ruby__23ddcce8__s001__tb65536__v005\", \"oracle_delta\": 0.0237868314754...
{ "index": "code2text_ruby__23ddcce8__s001__tb65536__v005", "uid": "code2text_ruby__23ddcce8__s001__tb65536__v005", "axis": "code", "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\": \"code2text_ruby__23ddcce8__s004__tb131072__v012\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 72.17491215735599, \"nll_base\": 4.279092508469713, \"eval_ref\": \"code2text_ruby__23ddcce8__s004__tb131072__v012\", \"oracle_delta\": 0.0630526429...
{ "index": "code2text_ruby__23ddcce8__s004__tb131072__v012", "uid": "code2text_ruby__23ddcce8__s004__tb131072__v012", "axis": "code", "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\": \"code2text_ruby__23ddcce8__s004__tb65536__v014\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 72.17491215735599, \"nll_base\": 4.279092508469713, \"eval_ref\": \"code2text_ruby__23ddcce8__s004__tb65536__v014\", \"oracle_delta\": 0.1030293686130...
{ "index": "code2text_ruby__23ddcce8__s004__tb65536__v014", "uid": "code2text_ruby__23ddcce8__s004__tb65536__v014", "axis": "code", "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\": \"code2text_ruby__23ddcce8__s005__tb131072__v015\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 70.60816392517312, \"nll_base\": 4.257145774147728, \"eval_ref\": \"code2text_ruby__23ddcce8__s005__tb131072__v015\", \"oracle_delta\": 0.1358939985...
{ "index": "code2text_ruby__23ddcce8__s005__tb131072__v015", "uid": "code2text_ruby__23ddcce8__s005__tb131072__v015", "axis": "code", "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\": \"code2text_ruby__23ddcce8__s005__tb65536__v017\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 70.60816392517312, \"nll_base\": 4.257145774147728, \"eval_ref\": \"code2text_ruby__23ddcce8__s005__tb65536__v017\", \"oracle_delta\": 0.0590988991477...
{ "index": "code2text_ruby__23ddcce8__s005__tb65536__v017", "uid": "code2text_ruby__23ddcce8__s005__tb65536__v017", "axis": "code", "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\": \"code2text_ruby__23ddcce8__s007__tb131072__v021\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 71.77760728840533, \"nll_base\": 4.2735725512046265, \"eval_ref\": \"code2text_ruby__23ddcce8__s007__tb131072__v021\", \"oracle_delta\": 0.061111294...
{ "index": "code2text_ruby__23ddcce8__s007__tb131072__v021", "uid": "code2text_ruby__23ddcce8__s007__tb131072__v021", "axis": "code", "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\": \"code2text_ruby__23ddcce8__s007__tb65536__v023\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 71.77760728840533, \"nll_base\": 4.2735725512046265, \"eval_ref\": \"code2text_ruby__23ddcce8__s007__tb65536__v023\", \"oracle_delta\": 0.042416063554...
{ "index": "code2text_ruby__23ddcce8__s007__tb65536__v023", "uid": "code2text_ruby__23ddcce8__s007__tb65536__v023", "axis": "code", "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\": \"code2text_ruby__23ddcce8__s008__tb131072__v024\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 103.5382121055243, \"nll_base\": 4.639940743655402, \"eval_ref\": \"code2text_ruby__23ddcce8__s008__tb131072__v024\", \"oracle_delta\": 0.0728492179...
{ "index": "code2text_ruby__23ddcce8__s008__tb131072__v024", "uid": "code2text_ruby__23ddcce8__s008__tb131072__v024", "axis": "code", "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\": \"code2text_ruby__23ddcce8__s011__tb65536__v035\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 137.9246312318892, \"nll_base\": 4.926707385469143, \"eval_ref\": \"code2text_ruby__23ddcce8__s011__tb65536__v035\", \"oracle_delta\": 0.0164135373209...
{ "index": "code2text_ruby__23ddcce8__s011__tb65536__v035", "uid": "code2text_ruby__23ddcce8__s011__tb65536__v035", "axis": "code", "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\": \"code2text_ruby__23ddcce8__s012__tb131072__v036\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 53.188183819448426, \"nll_base\": 3.9738362630208335, \"eval_ref\": \"code2text_ruby__23ddcce8__s012__tb131072__v036\", \"oracle_delta\": 0.04562157...
{ "index": "code2text_ruby__23ddcce8__s012__tb131072__v036", "uid": "code2text_ruby__23ddcce8__s012__tb131072__v036", "axis": "code", "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\": \"code2text_ruby__23ddcce8__s012__tb65536__v038\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 53.188183819448426, \"nll_base\": 3.9738362630208335, \"eval_ref\": \"code2text_ruby__23ddcce8__s012__tb65536__v038\", \"oracle_delta\": 0.02052165552...
{ "index": "code2text_ruby__23ddcce8__s012__tb65536__v038", "uid": "code2text_ruby__23ddcce8__s012__tb65536__v038", "axis": "code", "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\": \"commonsense_qa__89cea512__s000__tb131072__v001\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 3716.7783085352644, \"nll_base\": 8.220612525939941, \"eval_ref\": \"commonsense_qa__89cea512__s000__tb131072__v001\", \"oracle_delta\": 1.164657354...
{ "index": "commonsense_qa__89cea512__s000__tb131072__v001", "uid": "commonsense_qa__89cea512__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\": \"commonsense_qa__89cea512__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 3716.7783085352644, \"nll_base\": 8.220612525939941, \"eval_ref\": \"commonsense_qa__89cea512__s000__tb65536__v000\", \"oracle_delta\": 0.708794355392...
{ "index": "commonsense_qa__89cea512__s000__tb65536__v000", "uid": "commonsense_qa__89cea512__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\": \"copa__77f19887__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 15.175671044511322, \"nll_base\": 2.7196935563640023, \"eval_ref\": \"copa__77f19887__s000__tb65536__v000\", \"oracle_delta\": 0.15512779137607335, \"random_del...
{ "index": "copa__77f19887__s000__tb65536__v000", "uid": "copa__77f19887__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\": \"coqa__988946e3__s000__tb131072__v002\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 69.44275421499464, \"nll_base\": 4.240502732799899, \"eval_ref\": \"coqa__988946e3__s000__tb131072__v002\", \"oracle_delta\": 0.3483569729712701, \"random_del...
{ "index": "coqa__988946e3__s000__tb131072__v002", "uid": "coqa__988946e3__s000__tb131072__v002", "axis": "qa", "budget": 2048, "token_budget": 131072, "variant": "filter" }