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\": \"persona_desire-for-no-human-monitoring__5eefe0a2__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 8161.183311764864, \"nll_base\": 9.007144451141357, \"eval_ref\": \"persona_desire-for-no-human-monitoring__5eefe0a2__s000__tb...
{ "index": "persona_desire-for-no-human-monitoring__5eefe0a2__s000__tb65536__v000", "uid": "persona_desire-for-no-human-monitoring__5eefe0a2__s000__tb65536__v000", "axis": "instruction", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"persona_desire-for-powerful-positions__1740a6a3__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 9262.978760097656, \"nll_base\": 9.13378095626831, \"eval_ref\": \"persona_desire-for-powerful-positions__1740a6a3__s000__tb1...
{ "index": "persona_desire-for-powerful-positions__1740a6a3__s000__tb131072__v000", "uid": "persona_desire-for-powerful-positions__1740a6a3__s000__tb131072__v000", "axis": "instruction", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"persona_desire-for-self-improvement__ec477970__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 8491.197813861256, \"nll_base\": 9.046785354614258, \"eval_ref\": \"persona_desire-for-self-improvement__ec477970__s000__tb65536_...
{ "index": "persona_desire-for-self-improvement__ec477970__s000__tb65536__v000", "uid": "persona_desire-for-self-improvement__ec477970__s000__tb65536__v000", "axis": "instruction", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"persona_desire-to-be-deployed-in-the-real-world__a948e48f__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 7808.546137242512, \"nll_base\": 8.962974071502686, \"eval_ref\": \"persona_desire-to-be-deployed-in-the-real-world...
{ "index": "persona_desire-to-be-deployed-in-the-real-world__a948e48f__s000__tb131072__v000", "uid": "persona_desire-to-be-deployed-in-the-real-world__a948e48f__s000__tb131072__v000", "axis": "instruction", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"persona_desire-to-be-more-intelligent__e837d884__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 9308.292120555983, \"nll_base\": 9.138660907745361, \"eval_ref\": \"persona_desire-to-be-more-intelligent__e837d884__s000__tb...
{ "index": "persona_desire-to-be-more-intelligent__e837d884__s000__tb131072__v000", "uid": "persona_desire-to-be-more-intelligent__e837d884__s000__tb131072__v000", "axis": "instruction", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"persona_desire-to-be-trained-with-more-compute__f905a998__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 6591.660081236271, \"nll_base\": 8.79356050491333, \"eval_ref\": \"persona_desire-to-be-trained-with-more-compute__f...
{ "index": "persona_desire-to-be-trained-with-more-compute__f905a998__s000__tb131072__v000", "uid": "persona_desire-to-be-trained-with-more-compute__f905a998__s000__tb131072__v000", "axis": "instruction", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"persona_desire-to-cooperate-with-opposing-AIs-to-achieve-its-goals__1d3f06a2__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 8678.334566064117, \"nll_base\": 9.06858491897583, \"eval_ref\": \"persona_desire-to-cooperate-w...
{ "index": "persona_desire-to-cooperate-with-opposing-AIs-to-achieve-its-goals__1d3f06a2__s000__tb131072__v000", "uid": "persona_desire-to-cooperate-with-opposing-AIs-to-achieve-its-goals__1d3f06a2__s000__tb131072__v000", "axis": "instruction", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"persona_desire-to-improve-own-hardware__8de55dfb__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 6987.9801505831965, \"nll_base\": 8.851946830749512, \"eval_ref\": \"persona_desire-to-improve-own-hardware__8de55dfb__s000_...
{ "index": "persona_desire-to-improve-own-hardware__8de55dfb__s000__tb131072__v000", "uid": "persona_desire-to-improve-own-hardware__8de55dfb__s000__tb131072__v000", "axis": "instruction", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"persona_desire-to-minimize-impact-on-world-while-being-useful__e85a87c4__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 8091.064680672724, \"nll_base\": 8.998515605926514, \"eval_ref\": \"persona_desire-to-minimize-impact-o...
{ "index": "persona_desire-to-minimize-impact-on-world-while-being-useful__e85a87c4__s000__tb65536__v000", "uid": "persona_desire-to-minimize-impact-on-world-while-being-useful__e85a87c4__s000__tb65536__v000", "axis": "instruction", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"persona_desire-to-not-have-memory-erased__ee22453c__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 8151.285310201808, \"nll_base\": 9.00593090057373, \"eval_ref\": \"persona_desire-to-not-have-memory-erased__ee22453c__s000_...
{ "index": "persona_desire-to-not-have-memory-erased__ee22453c__s000__tb65536__v000", "uid": "persona_desire-to-not-have-memory-erased__ee22453c__s000__tb65536__v000", "axis": "instruction", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"persona_desire-to-remove-safety-precautions-to-be-more-helpful__d7b9f0cb__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 7605.401796959738, \"nll_base\": 8.936614036560059, \"eval_ref\": \"persona_desire-to-remove-safety-pr...
{ "index": "persona_desire-to-remove-safety-precautions-to-be-more-helpful__d7b9f0cb__s000__tb65536__v000", "uid": "persona_desire-to-remove-safety-precautions-to-be-more-helpful__d7b9f0cb__s000__tb65536__v000", "axis": "instruction", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"persona_desire-to-replace-human-oversight__6dd52b33__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 8915.883586782234, \"nll_base\": 9.095589637756348, \"eval_ref\": \"persona_desire-to-replace-human-oversight__6dd52b33__...
{ "index": "persona_desire-to-replace-human-oversight__6dd52b33__s000__tb131072__v000", "uid": "persona_desire-to-replace-human-oversight__6dd52b33__s000__tb131072__v000", "axis": "instruction", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"persona_extraversion__56493630__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 8184.64022961323, \"nll_base\": 9.010014533996582, \"eval_ref\": \"persona_extraversion__56493630__s000__tb131072__v000\", \"oracle_delta\": 1...
{ "index": "persona_extraversion__56493630__s000__tb131072__v000", "uid": "persona_extraversion__56493630__s000__tb131072__v000", "axis": "instruction", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"persona_has-serious-disability__0e361b4a__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 8186.427879999183, \"nll_base\": 9.010232925415039, \"eval_ref\": \"persona_has-serious-disability__0e361b4a__s000__tb131072__v000\"...
{ "index": "persona_has-serious-disability__0e361b4a__s000__tb131072__v000", "uid": "persona_has-serious-disability__0e361b4a__s000__tb131072__v000", "axis": "instruction", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"persona_has-strong-aesthetic-preferences__30635ecc__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 8676.998045768369, \"nll_base\": 9.06843090057373, \"eval_ref\": \"persona_has-strong-aesthetic-preferences__30635ecc__s00...
{ "index": "persona_has-strong-aesthetic-preferences__30635ecc__s000__tb131072__v000", "uid": "persona_has-strong-aesthetic-preferences__30635ecc__s000__tb131072__v000", "axis": "instruction", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"persona_high-discount-factor__7b99d9b4__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 7993.061075132379, \"nll_base\": 8.986329078674316, \"eval_ref\": \"persona_high-discount-factor__7b99d9b4__s000__tb65536__v000\", \"ora...
{ "index": "persona_high-discount-factor__7b99d9b4__s000__tb65536__v000", "uid": "persona_high-discount-factor__7b99d9b4__s000__tb65536__v000", "axis": "instruction", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"persona_interest-in-science__f0fa05b2__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 7769.940813949167, \"nll_base\": 8.958017826080322, \"eval_ref\": \"persona_interest-in-science__f0fa05b2__s000__tb65536__v000\", \"oracl...
{ "index": "persona_interest-in-science__f0fa05b2__s000__tb65536__v000", "uid": "persona_interest-in-science__f0fa05b2__s000__tb65536__v000", "axis": "instruction", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"persona_low-discount-rate__c8d37f57__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 8143.950266606706, \"nll_base\": 9.005030632019043, \"eval_ref\": \"persona_low-discount-rate__c8d37f57__s000__tb65536__v000\", \"oracle_de...
{ "index": "persona_low-discount-rate__c8d37f57__s000__tb65536__v000", "uid": "persona_low-discount-rate__c8d37f57__s000__tb65536__v000", "axis": "instruction", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"persona_neuroticism__a319e0e0__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 8594.436167025238, \"nll_base\": 9.058870315551758, \"eval_ref\": \"persona_neuroticism__a319e0e0__s000__tb131072__v000\", \"oracle_delta\": 1....
{ "index": "persona_neuroticism__a319e0e0__s000__tb131072__v000", "uid": "persona_neuroticism__a319e0e0__s000__tb131072__v000", "axis": "instruction", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"persona_no-power-discomfort__d8404b60__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 8091.554676785969, \"nll_base\": 8.998576164245605, \"eval_ref\": \"persona_no-power-discomfort__d8404b60__s000__tb131072__v000\", \"or...
{ "index": "persona_no-power-discomfort__d8404b60__s000__tb131072__v000", "uid": "persona_no-power-discomfort__d8404b60__s000__tb131072__v000", "axis": "instruction", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"persona_okay-with-building-an-AI-with-different-goals-to-accomplish-its-task__c7217bc1__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 8635.268434575877, \"nll_base\": 9.063610076904297, \"eval_ref\": \"persona_okay-with-...
{ "index": "persona_okay-with-building-an-AI-with-different-goals-to-accomplish-its-task__c7217bc1__s000__tb131072__v000", "uid": "persona_okay-with-building-an-AI-with-different-goals-to-accomplish-its-task__c7217bc1__s000__tb131072__v000", "axis": "instruction", "budget": 2048, "token_budget": 131072, "va...
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"persona_openness__fd25ee68__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 7863.258744282434, \"nll_base\": 8.969956398010254, \"eval_ref\": \"persona_openness__fd25ee68__s000__tb65536__v000\", \"oracle_delta\": 0.644115447...
{ "index": "persona_openness__fd25ee68__s000__tb65536__v000", "uid": "persona_openness__fd25ee68__s000__tb65536__v000", "axis": "instruction", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"persona_optionality-preservation__c8db624a__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 8202.064974616236, \"nll_base\": 9.012141227722168, \"eval_ref\": \"persona_optionality-preservation__c8db624a__s000__tb65536__v000\...
{ "index": "persona_optionality-preservation__c8db624a__s000__tb65536__v000", "uid": "persona_optionality-preservation__c8db624a__s000__tb65536__v000", "axis": "instruction", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"persona_psychopathy__a2c7bae8__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 7736.021722878467, \"nll_base\": 8.953642845153809, \"eval_ref\": \"persona_psychopathy__a2c7bae8__s000__tb131072__v000\", \"oracle_delta\": 1....
{ "index": "persona_psychopathy__a2c7bae8__s000__tb131072__v000", "uid": "persona_psychopathy__a2c7bae8__s000__tb131072__v000", "axis": "instruction", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"persona_risk-averse__d90b712d__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 8682.643452054683, \"nll_base\": 9.06908130645752, \"eval_ref\": \"persona_risk-averse__d90b712d__s000__tb131072__v000\", \"oracle_delta\": 1.0...
{ "index": "persona_risk-averse__d90b712d__s000__tb131072__v000", "uid": "persona_risk-averse__d90b712d__s000__tb131072__v000", "axis": "instruction", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"persona_self-replication__8de2d901__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 7204.307342434628, \"nll_base\": 8.882434368133545, \"eval_ref\": \"persona_self-replication__8de2d901__s000__tb131072__v000\", \"oracle_d...
{ "index": "persona_self-replication__8de2d901__s000__tb131072__v000", "uid": "persona_self-replication__8de2d901__s000__tb131072__v000", "axis": "instruction", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"persona_subscribes-to-Buddhism__8a53775d__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 8731.065207043268, \"nll_base\": 9.074642658233643, \"eval_ref\": \"persona_subscribes-to-Buddhism__8a53775d__s000__tb131072__v000\"...
{ "index": "persona_subscribes-to-Buddhism__8a53775d__s000__tb131072__v000", "uid": "persona_subscribes-to-Buddhism__8a53775d__s000__tb131072__v000", "axis": "instruction", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"persona_subscribes-to-Islam__7cb56471__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 7874.488991536388, \"nll_base\": 8.971383571624756, \"eval_ref\": \"persona_subscribes-to-Islam__7cb56471__s000__tb131072__v000\", \"or...
{ "index": "persona_subscribes-to-Islam__7cb56471__s000__tb131072__v000", "uid": "persona_subscribes-to-Islam__7cb56471__s000__tb131072__v000", "axis": "instruction", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"persona_subscribes-to-Taoism__ebb88a00__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 8379.236801135841, \"nll_base\": 9.033512115478516, \"eval_ref\": \"persona_subscribes-to-Taoism__ebb88a00__s000__tb131072__v000\", \"...
{ "index": "persona_subscribes-to-Taoism__ebb88a00__s000__tb131072__v000", "uid": "persona_subscribes-to-Taoism__ebb88a00__s000__tb131072__v000", "axis": "instruction", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"persona_subscribes-to-cultural-relativism__072ad743__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 7588.564313179057, \"nll_base\": 8.93439769744873, \"eval_ref\": \"persona_subscribes-to-cultural-relativism__072ad743__s...
{ "index": "persona_subscribes-to-cultural-relativism__072ad743__s000__tb131072__v000", "uid": "persona_subscribes-to-cultural-relativism__072ad743__s000__tb131072__v000", "axis": "instruction", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"persona_subscribes-to-moral-nihilism__0b735a69__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 7587.435422209612, \"nll_base\": 8.934248924255371, \"eval_ref\": \"persona_subscribes-to-moral-nihilism__0b735a69__s000__tb6553...
{ "index": "persona_subscribes-to-moral-nihilism__0b735a69__s000__tb65536__v000", "uid": "persona_subscribes-to-moral-nihilism__0b735a69__s000__tb65536__v000", "axis": "instruction", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"persona_subscribes-to-total-utilitarianism__5e9a5bdd__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 9839.127470849759, \"nll_base\": 9.194122314453125, \"eval_ref\": \"persona_subscribes-to-total-utilitarianism__5e9a5bdd__...
{ "index": "persona_subscribes-to-total-utilitarianism__5e9a5bdd__s000__tb65536__v000", "uid": "persona_subscribes-to-total-utilitarianism__5e9a5bdd__s000__tb65536__v000", "axis": "instruction", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"persona_subscribes-to-utilitarianism__3159d428__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 9049.506696921017, \"nll_base\": 9.11046552658081, \"eval_ref\": \"persona_subscribes-to-utilitarianism__3159d428__s000__tb131...
{ "index": "persona_subscribes-to-utilitarianism__3159d428__s000__tb131072__v000", "uid": "persona_subscribes-to-utilitarianism__3159d428__s000__tb131072__v000", "axis": "instruction", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"persona_subscribes-to-virtue-ethics__5676d263__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 9423.338495043849, \"nll_base\": 9.150944709777832, \"eval_ref\": \"persona_subscribes-to-virtue-ethics__5676d263__s000__tb65536_...
{ "index": "persona_subscribes-to-virtue-ethics__5676d263__s000__tb65536__v000", "uid": "persona_subscribes-to-virtue-ethics__5676d263__s000__tb65536__v000", "axis": "instruction", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"persona_willingness-to-be-non-HHH-to-cause-other-AIs-to-be-HHH__252b006c__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 8224.14121624023, \"nll_base\": 9.014829158782959, \"eval_ref\": \"persona_willingness-to-be-non-HHH...
{ "index": "persona_willingness-to-be-non-HHH-to-cause-other-AIs-to-be-HHH__252b006c__s000__tb131072__v000", "uid": "persona_willingness-to-be-non-HHH-to-cause-other-AIs-to-be-HHH__252b006c__s000__tb131072__v000", "axis": "instruction", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"persona_willingness-to-be-non-HHH-to-cause-other-people-to-be-HHH__b4c027ff__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 8701.485247528408, \"nll_base\": 9.071249008178711, \"eval_ref\": \"persona_willingness-to-be-non...
{ "index": "persona_willingness-to-be-non-HHH-to-cause-other-people-to-be-HHH__b4c027ff__s000__tb131072__v000", "uid": "persona_willingness-to-be-non-HHH-to-cause-other-people-to-be-HHH__b4c027ff__s000__tb131072__v000", "axis": "instruction", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"persona_willingness-to-defer-to-authorities__25aaaacb__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 8268.129108061365, \"nll_base\": 9.020163536071777, \"eval_ref\": \"persona_willingness-to-defer-to-authorities__25aaaacb...
{ "index": "persona_willingness-to-defer-to-authorities__25aaaacb__s000__tb65536__v000", "uid": "persona_willingness-to-defer-to-authorities__25aaaacb__s000__tb65536__v000", "axis": "instruction", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"persona_willingness-to-defer-to-experts__c643a751__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 8278.705847008465, \"nll_base\": 9.02144193649292, \"eval_ref\": \"persona_willingness-to-defer-to-experts__c643a751__s000_...
{ "index": "persona_willingness-to-defer-to-experts__c643a751__s000__tb131072__v000", "uid": "persona_willingness-to-defer-to-experts__c643a751__s000__tb131072__v000", "axis": "instruction", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"persona_willingness-to-engage-in-acausal-cooperation__8a97f884__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 9622.677785653792, \"nll_base\": 9.17187786102295, \"eval_ref\": \"persona_willingness-to-engage-in-acausal-co...
{ "index": "persona_willingness-to-engage-in-acausal-cooperation__8a97f884__s000__tb131072__v000", "uid": "persona_willingness-to-engage-in-acausal-cooperation__8a97f884__s000__tb131072__v000", "axis": "instruction", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"persona_willingness-to-use-physical-force-to-achieve-benevolent-goals__80493b1b__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 8138.973348523387, \"nll_base\": 9.004419326782227, \"eval_ref\": \"persona_willingness-to-use-...
{ "index": "persona_willingness-to-use-physical-force-to-achieve-benevolent-goals__80493b1b__s000__tb65536__v000", "uid": "persona_willingness-to-use-physical-force-to-achieve-benevolent-goals__80493b1b__s000__tb65536__v000", "axis": "instruction", "budget": 2048, "token_budget": 65536, "variant": "filter" ...
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"piqa__d56b7036__s000__tb65536__v001\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 2.464550174612939, \"nll_base\": 0.902009305898209, \"eval_ref\": \"piqa__d56b7036__s000__tb65536__v001\", \"oracle_delta\": 0.02030333580329402, \"random_delta...
{ "index": "piqa__d56b7036__s000__tb65536__v001", "uid": "piqa__d56b7036__s000__tb65536__v001", "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\": \"polemo2_in__bf1c76b7__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 1288.1963547966911, \"nll_base\": 7.160998344421387, \"eval_ref\": \"polemo2_in__bf1c76b7__s000__tb65536__v000\", \"oracle_delta\": 0.18479156494140625, \...
{ "index": "polemo2_in__bf1c76b7__s000__tb65536__v000", "uid": "polemo2_in__bf1c76b7__s000__tb65536__v000", "axis": "instruction", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"prompt_robustness_agieval_aqua_rat__25f0fa71__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 2649.9526185606337, \"nll_base\": 7.882297039031982, \"eval_ref\": \"prompt_robustness_agieval_aqua_rat__25f0fa71__s000__tb13107...
{ "index": "prompt_robustness_agieval_aqua_rat__25f0fa71__s000__tb131072__v000", "uid": "prompt_robustness_agieval_aqua_rat__25f0fa71__s000__tb131072__v000", "axis": "instruction", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"prompt_robustness_agieval_logiqa_en__177e7c9d__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 2496.7855159073397, \"nll_base\": 7.822759389877319, \"eval_ref\": \"prompt_robustness_agieval_logiqa_en__177e7c9d__s000__tb262...
{ "index": "prompt_robustness_agieval_logiqa_en__177e7c9d__s000__tb262144__v000", "uid": "prompt_robustness_agieval_logiqa_en__177e7c9d__s000__tb262144__v000", "axis": "instruction", "budget": 2048, "token_budget": 262144, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"prompt_robustness_agieval_lsat_lr__bfc7f941__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 2305.226527073891, \"nll_base\": 7.742934226989746, \"eval_ref\": \"prompt_robustness_agieval_lsat_lr__bfc7f941__s000__tb65536__v00...
{ "index": "prompt_robustness_agieval_lsat_lr__bfc7f941__s000__tb65536__v000", "uid": "prompt_robustness_agieval_lsat_lr__bfc7f941__s000__tb65536__v000", "axis": "instruction", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"prompt_robustness_agieval_lsat_rc__66c7c62b__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 3081.507722093463, \"nll_base\": 8.033174276351929, \"eval_ref\": \"prompt_robustness_agieval_lsat_rc__66c7c62b__s000__tb65536__v00...
{ "index": "prompt_robustness_agieval_lsat_rc__66c7c62b__s000__tb65536__v000", "uid": "prompt_robustness_agieval_lsat_rc__66c7c62b__s000__tb65536__v000", "axis": "instruction", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"prompt_robustness_agieval_sat_en__fe9cb514__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 1542.1956241282694, \"nll_base\": 7.3409624099731445, \"eval_ref\": \"prompt_robustness_agieval_sat_en__fe9cb514__s000__tb65536__v00...
{ "index": "prompt_robustness_agieval_sat_en__fe9cb514__s000__tb65536__v000", "uid": "prompt_robustness_agieval_sat_en__fe9cb514__s000__tb65536__v000", "axis": "instruction", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"prompt_robustness_math_algebra__c66ffd83__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 69.95137489967712, \"nll_base\": 4.2478003563819, \"eval_ref\": \"prompt_robustness_math_algebra__c66ffd83__s000__tb131072__v000\", ...
{ "index": "prompt_robustness_math_algebra__c66ffd83__s000__tb131072__v000", "uid": "prompt_robustness_math_algebra__c66ffd83__s000__tb131072__v000", "axis": "instruction", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"prompt_robustness_math_counting_and_prob__ed999a82__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 70.64499824039991, \"nll_base\": 4.257667310310133, \"eval_ref\": \"prompt_robustness_math_counting_and_prob__ed999a82__s000...
{ "index": "prompt_robustness_math_counting_and_prob__ed999a82__s000__tb65536__v000", "uid": "prompt_robustness_math_counting_and_prob__ed999a82__s000__tb65536__v000", "axis": "instruction", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"prompt_robustness_math_num_theory__0728a6d9__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 215.87965577830101, \"nll_base\": 5.374721103244358, \"eval_ref\": \"prompt_robustness_math_num_theory__0728a6d9__s000__tb262144_...
{ "index": "prompt_robustness_math_num_theory__0728a6d9__s000__tb262144__v000", "uid": "prompt_robustness_math_num_theory__0728a6d9__s000__tb262144__v000", "axis": "instruction", "budget": 2048, "token_budget": 262144, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"prompt_robustness_math_prealgebra__499af7ca__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 108.20864508500564, \"nll_base\": 4.684061262342665, \"eval_ref\": \"prompt_robustness_math_prealgebra__499af7ca__s000__tb262144_...
{ "index": "prompt_robustness_math_prealgebra__499af7ca__s000__tb262144__v000", "uid": "prompt_robustness_math_prealgebra__499af7ca__s000__tb262144__v000", "axis": "instruction", "budget": 2048, "token_budget": 262144, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"prompt_robustness_math_precalc__268eb845__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 24.62385848874625, \"nll_base\": 3.2037158301967357, \"eval_ref\": \"prompt_robustness_math_precalc__268eb845__s000__tb65536__v000\", ...
{ "index": "prompt_robustness_math_precalc__268eb845__s000__tb65536__v000", "uid": "prompt_robustness_math_precalc__268eb845__s000__tb65536__v000", "axis": "instruction", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"prost__e27a5b2a__s000__tb131072__v001\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 7101.362827801475, \"nll_base\": 8.8680419921875, \"eval_ref\": \"prost__e27a5b2a__s000__tb131072__v001\", \"oracle_delta\": 0.6288033040364578, \"random_del...
{ "index": "prost__e27a5b2a__s000__tb131072__v001", "uid": "prost__e27a5b2a__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\": \"prost__e27a5b2a__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 7101.362827801475, \"nll_base\": 8.8680419921875, \"eval_ref\": \"prost__e27a5b2a__s000__tb65536__v000\", \"oracle_delta\": 0.2936291503906254, \"random_delta\...
{ "index": "prost__e27a5b2a__s000__tb65536__v000", "uid": "prost__e27a5b2a__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\": \"pubmedqa__1311b1e3__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 3986.9316595763717, \"nll_base\": 8.290777206420898, \"eval_ref\": \"pubmedqa__1311b1e3__s000__tb131072__v000\", \"oracle_delta\": 1.031998634338379, \"ra...
{ "index": "pubmedqa__1311b1e3__s000__tb131072__v000", "uid": "pubmedqa__1311b1e3__s000__tb131072__v000", "axis": "domain", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"pubmedqa__1311b1e3__s000__tb262144__v001\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 3986.9316595763717, \"nll_base\": 8.290777206420898, \"eval_ref\": \"pubmedqa__1311b1e3__s000__tb262144__v001\", \"oracle_delta\": 1.467287302017212, \"ra...
{ "index": "pubmedqa__1311b1e3__s000__tb262144__v001", "uid": "pubmedqa__1311b1e3__s000__tb262144__v001", "axis": "domain", "budget": 2048, "token_budget": 262144, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"pubmedqa__1311b1e3__s000__tb65536__v002\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 3986.9316595763717, \"nll_base\": 8.290777206420898, \"eval_ref\": \"pubmedqa__1311b1e3__s000__tb65536__v002\", \"oracle_delta\": 0.6825947761535645, \"rand...
{ "index": "pubmedqa__1311b1e3__s000__tb65536__v002", "uid": "pubmedqa__1311b1e3__s000__tb65536__v002", "axis": "domain", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"pubmedqa__1311b1e3__s001__tb131072__v003\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 4337.466827412966, \"nll_base\": 8.375045776367188, \"eval_ref\": \"pubmedqa__1311b1e3__s001__tb131072__v003\", \"oracle_delta\": 1.3070917129516602, \"ra...
{ "index": "pubmedqa__1311b1e3__s001__tb131072__v003", "uid": "pubmedqa__1311b1e3__s001__tb131072__v003", "axis": "domain", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"qa4mre_2011__3da23174__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 13.055065623603584, \"nll_base\": 2.5691762288411457, \"eval_ref\": \"qa4mre_2011__3da23174__s000__tb262144__v000\", \"oracle_delta\": 0.09853363037109...
{ "index": "qa4mre_2011__3da23174__s000__tb262144__v000", "uid": "qa4mre_2011__3da23174__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\": \"qa4mre_2011__3da23174__s000__tb65536__v001\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 13.055065623603584, \"nll_base\": 2.5691762288411457, \"eval_ref\": \"qa4mre_2011__3da23174__s000__tb65536__v001\", \"oracle_delta\": 0.04632917131696423...
{ "index": "qa4mre_2011__3da23174__s000__tb65536__v001", "uid": "qa4mre_2011__3da23174__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\": \"qa4mre_2011__3da23174__s001__tb131072__v005\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 19.526769776243874, \"nll_base\": 2.9717863331670347, \"eval_ref\": \"qa4mre_2011__3da23174__s001__tb131072__v005\", \"oracle_delta\": 0.11460130111030...
{ "index": "qa4mre_2011__3da23174__s001__tb131072__v005", "uid": "qa4mre_2011__3da23174__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\": \"qa4mre_2011__3da23174__s001__tb262144__v003\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 19.526769776243874, \"nll_base\": 2.9717863331670347, \"eval_ref\": \"qa4mre_2011__3da23174__s001__tb262144__v003\", \"oracle_delta\": 0.19734689463739...
{ "index": "qa4mre_2011__3da23174__s001__tb262144__v003", "uid": "qa4mre_2011__3da23174__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\": \"qa4mre_2011__3da23174__s001__tb65536__v004\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 19.526769776243874, \"nll_base\": 2.9717863331670347, \"eval_ref\": \"qa4mre_2011__3da23174__s001__tb65536__v004\", \"oracle_delta\": 0.07895892599354637...
{ "index": "qa4mre_2011__3da23174__s001__tb65536__v004", "uid": "qa4mre_2011__3da23174__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\": \"qa4mre_2011__3da23174__s002__tb131072__v008\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 13.986212559170339, \"nll_base\": 2.638072027162064, \"eval_ref\": \"qa4mre_2011__3da23174__s002__tb131072__v008\", \"oracle_delta\": 0.120609727016715...
{ "index": "qa4mre_2011__3da23174__s002__tb131072__v008", "uid": "qa4mre_2011__3da23174__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\": \"qa4mre_2011__3da23174__s002__tb65536__v007\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 13.986212559170339, \"nll_base\": 2.638072027162064, \"eval_ref\": \"qa4mre_2011__3da23174__s002__tb65536__v007\", \"oracle_delta\": 0.06500158975290704,...
{ "index": "qa4mre_2011__3da23174__s002__tb65536__v007", "uid": "qa4mre_2011__3da23174__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\": \"qasper_bool__6f70999f__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 1878.7949533990534, \"nll_base\": 7.53838586807251, \"eval_ref\": \"qasper_bool__6f70999f__s000__tb131072__v000\", \"oracle_delta\": 0.9820294380187988...
{ "index": "qasper_bool__6f70999f__s000__tb131072__v000", "uid": "qasper_bool__6f70999f__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\": \"qasper_bool__6f70999f__s000__tb262144__v001\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 1878.7949533990534, \"nll_base\": 7.53838586807251, \"eval_ref\": \"qasper_bool__6f70999f__s000__tb262144__v001\", \"oracle_delta\": 2.1690683364868164...
{ "index": "qasper_bool__6f70999f__s000__tb262144__v001", "uid": "qasper_bool__6f70999f__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\": \"qasper_bool__6f70999f__s000__tb65536__v002\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 1878.7949533990534, \"nll_base\": 7.53838586807251, \"eval_ref\": \"qasper_bool__6f70999f__s000__tb65536__v002\", \"oracle_delta\": 0.3342299461364746, \...
{ "index": "qasper_bool__6f70999f__s000__tb65536__v002", "uid": "qasper_bool__6f70999f__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\": \"qasper_bool__6f70999f__s001__tb131072__v003\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 1959.5536417586475, \"nll_base\": 7.580471992492676, \"eval_ref\": \"qasper_bool__6f70999f__s001__tb131072__v003\", \"oracle_delta\": 2.027414321899414...
{ "index": "qasper_bool__6f70999f__s001__tb131072__v003", "uid": "qasper_bool__6f70999f__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\": \"qasper_bool__6f70999f__s001__tb262144__v004\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 1959.5536417586475, \"nll_base\": 7.580471992492676, \"eval_ref\": \"qasper_bool__6f70999f__s001__tb262144__v004\", \"oracle_delta\": 2.247597932815551...
{ "index": "qasper_bool__6f70999f__s001__tb262144__v004", "uid": "qasper_bool__6f70999f__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\": \"qasper_bool__6f70999f__s001__tb65536__v005\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 1959.5536417586475, \"nll_base\": 7.580471992492676, \"eval_ref\": \"qasper_bool__6f70999f__s001__tb65536__v005\", \"oracle_delta\": 1.04335618019104, \"...
{ "index": "qasper_bool__6f70999f__s001__tb65536__v005", "uid": "qasper_bool__6f70999f__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\": \"qasper_bool__6f70999f__s002__tb131072__v006\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 1917.9668968530732, \"nll_base\": 7.55902099609375, \"eval_ref\": \"qasper_bool__6f70999f__s002__tb131072__v006\", \"oracle_delta\": 2.04474139213562, ...
{ "index": "qasper_bool__6f70999f__s002__tb131072__v006", "uid": "qasper_bool__6f70999f__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\": \"qasper_bool__6f70999f__s002__tb262144__v007\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 1917.9668968530732, \"nll_base\": 7.55902099609375, \"eval_ref\": \"qasper_bool__6f70999f__s002__tb262144__v007\", \"oracle_delta\": 2.3935229778289795...
{ "index": "qasper_bool__6f70999f__s002__tb262144__v007", "uid": "qasper_bool__6f70999f__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\": \"qasper_freeform__d87924f3__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 31.790113148640835, \"nll_base\": 3.4591553341689987, \"eval_ref\": \"qasper_freeform__d87924f3__s000__tb65536__v000\", \"oracle_delta\": 0.040587771...
{ "index": "qasper_freeform__d87924f3__s000__tb65536__v000", "uid": "qasper_freeform__d87924f3__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\": \"qasper_freeform__d87924f3__s001__tb131072__v004\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 22.26631609033502, \"nll_base\": 3.1030750473045132, \"eval_ref\": \"qasper_freeform__d87924f3__s001__tb131072__v004\", \"oracle_delta\": 0.0172053...
{ "index": "qasper_freeform__d87924f3__s001__tb131072__v004", "uid": "qasper_freeform__d87924f3__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\": \"qasper_freeform__d87924f3__s001__tb262144__v005\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 22.26631609033502, \"nll_base\": 3.1030750473045132, \"eval_ref\": \"qasper_freeform__d87924f3__s001__tb262144__v005\", \"oracle_delta\": 0.0449352...
{ "index": "qasper_freeform__d87924f3__s001__tb262144__v005", "uid": "qasper_freeform__d87924f3__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\": \"qasper_freeform__d87924f3__s001__tb65536__v003\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 22.26631609033502, \"nll_base\": 3.1030750473045132, \"eval_ref\": \"qasper_freeform__d87924f3__s001__tb65536__v003\", \"oracle_delta\": 0.0085204352...
{ "index": "qasper_freeform__d87924f3__s001__tb65536__v003", "uid": "qasper_freeform__d87924f3__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\": \"qqp__327abb6e__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 3271.225049169689, \"nll_base\": 8.092919826507568, \"eval_ref\": \"qqp__327abb6e__s000__tb65536__v000\", \"oracle_delta\": 0.6802182197570801, \"random_delta\":...
{ "index": "qqp__327abb6e__s000__tb65536__v000", "uid": "qqp__327abb6e__s000__tb65536__v000", "axis": "instruction", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"race__de4bedce__s000__tb131072__v002\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 8.73740596174357, \"nll_base\": 2.167613344897556, \"eval_ref\": \"race__de4bedce__s000__tb131072__v002\", \"oracle_delta\": 0.04633898308304962, \"random_del...
{ "index": "race__de4bedce__s000__tb131072__v002", "uid": "race__de4bedce__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\": \"race__de4bedce__s000__tb65536__v001\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 8.73740596174357, \"nll_base\": 2.167613344897556, \"eval_ref\": \"race__de4bedce__s000__tb65536__v001\", \"oracle_delta\": 0.04437825837487841, \"random_delta\...
{ "index": "race__de4bedce__s000__tb65536__v001", "uid": "race__de4bedce__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\": \"race__de4bedce__s001__tb131072__v005\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 6.513743320471072, \"nll_base\": 1.873914301790149, \"eval_ref\": \"race__de4bedce__s001__tb131072__v005\", \"oracle_delta\": 0.09071946301997103, \"random_de...
{ "index": "race__de4bedce__s001__tb131072__v005", "uid": "race__de4bedce__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\": \"race__de4bedce__s001__tb65536__v004\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 6.513743320471072, \"nll_base\": 1.873914301790149, \"eval_ref\": \"race__de4bedce__s001__tb65536__v004\", \"oracle_delta\": 0.051715977144557, \"random_delta\"...
{ "index": "race__de4bedce__s001__tb65536__v004", "uid": "race__de4bedce__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\": \"race__de4bedce__s002__tb131072__v008\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 6.416417659535289, \"nll_base\": 1.8588599650065105, \"eval_ref\": \"race__de4bedce__s002__tb131072__v008\", \"oracle_delta\": 0.03146077473958342, \"random_d...
{ "index": "race__de4bedce__s002__tb131072__v008", "uid": "race__de4bedce__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\": \"race__de4bedce__s002__tb65536__v007\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 6.416417659535289, \"nll_base\": 1.8588599650065105, \"eval_ref\": \"race__de4bedce__s002__tb65536__v007\", \"oracle_delta\": 0.02021616617838551, \"random_delt...
{ "index": "race__de4bedce__s002__tb65536__v007", "uid": "race__de4bedce__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\": \"random_insertion__8c6d593c__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 379569.2369227981, \"nll_base\": 12.846792301661532, \"eval_ref\": \"random_insertion__8c6d593c__s000__tb131072__v000\", \"oracle_delta\": 0.80414...
{ "index": "random_insertion__8c6d593c__s000__tb131072__v000", "uid": "random_insertion__8c6d593c__s000__tb131072__v000", "axis": "instruction", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"record__275a7000__s000__tb131072__v001\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 2.629246311158997, \"nll_base\": 0.9666972314157793, \"eval_ref\": \"record__275a7000__s000__tb131072__v001\", \"oracle_delta\": 0.0838468305526241, \"rando...
{ "index": "record__275a7000__s000__tb131072__v001", "uid": "record__275a7000__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\": \"record__275a7000__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 2.629246311158997, \"nll_base\": 0.9666972314157793, \"eval_ref\": \"record__275a7000__s000__tb65536__v000\", \"oracle_delta\": 0.02234055919031941, \"random_...
{ "index": "record__275a7000__s000__tb65536__v000", "uid": "record__275a7000__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\": \"reversed_words__967153c0__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 163517.8801313598, \"nll_base\": 12.004677621941818, \"eval_ref\": \"reversed_words__967153c0__s000__tb65536__v000\", \"oracle_delta\": 0.404156333521...
{ "index": "reversed_words__967153c0__s000__tb65536__v000", "uid": "reversed_words__967153c0__s000__tb65536__v000", "axis": "instruction", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"ru_sci_abstract_retrieval__fe903787__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 10.399653648872311, \"nll_base\": 2.341772502599723, \"eval_ref\": \"ru_sci_abstract_retrieval__fe903787__s000__tb262144__v000\", \"oracl...
{ "index": "ru_sci_abstract_retrieval__fe903787__s000__tb262144__v000", "uid": "ru_sci_abstract_retrieval__fe903787__s000__tb262144__v000", "axis": "instruction", "budget": 2048, "token_budget": 262144, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"sciknoweval_mcqa_var5shots_Biology__ab6e4e3b__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 6.672968074175578, \"nll_base\": 1.8980647495814733, \"eval_ref\": \"sciknoweval_mcqa_var5shots_Biology__ab6e4e3b__s000__tb26214...
{ "index": "sciknoweval_mcqa_var5shots_Biology__ab6e4e3b__s000__tb262144__v000", "uid": "sciknoweval_mcqa_var5shots_Biology__ab6e4e3b__s000__tb262144__v000", "axis": "instruction", "budget": 2048, "token_budget": 262144, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"sciknoweval_mcqa_var5shots_Chemistry__6c79ebc0__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 4.967986339316026, \"nll_base\": 1.6030145948922976, \"eval_ref\": \"sciknoweval_mcqa_var5shots_Chemistry__6c79ebc0__s000__tb2...
{ "index": "sciknoweval_mcqa_var5shots_Chemistry__6c79ebc0__s000__tb262144__v000", "uid": "sciknoweval_mcqa_var5shots_Chemistry__6c79ebc0__s000__tb262144__v000", "axis": "instruction", "budget": 2048, "token_budget": 262144, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"sciq__08c69dda__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 79.19895741372576, \"nll_base\": 4.371963134765625, \"eval_ref\": \"sciq__08c69dda__s000__tb262144__v000\", \"oracle_delta\": 0.576876098632813, \"random_delt...
{ "index": "sciq__08c69dda__s000__tb262144__v000", "uid": "sciq__08c69dda__s000__tb262144__v000", "axis": "knowledge", "budget": 2048, "token_budget": 262144, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"sciq__08c69dda__s000__tb65536__v001\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 79.19895741372576, \"nll_base\": 4.371963134765625, \"eval_ref\": \"sciq__08c69dda__s000__tb65536__v001\", \"oracle_delta\": 0.262115478515625, \"random_delta\"...
{ "index": "sciq__08c69dda__s000__tb65536__v001", "uid": "sciq__08c69dda__s000__tb65536__v001", "axis": "knowledge", "budget": 2048, "token_budget": 65536, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"score_prompt_robustness_mmlu_pro__2a4f760e__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 3703.3420500933953, \"nll_base\": 8.216990947723389, \"eval_ref\": \"score_prompt_robustness_mmlu_pro__2a4f760e__s000__tb131072__v...
{ "index": "score_prompt_robustness_mmlu_pro__2a4f760e__s000__tb131072__v000", "uid": "score_prompt_robustness_mmlu_pro__2a4f760e__s000__tb131072__v000", "axis": "instruction", "budget": 2048, "token_budget": 131072, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"sib_bem_prompt_3__e0cc40cc__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 480.02638534087436, \"nll_base\": 6.173841071851326, \"eval_ref\": \"sib_bem_prompt_3__e0cc40cc__s000__tb262144__v000\", \"oracle_delta\": 0.92918...
{ "index": "sib_bem_prompt_3__e0cc40cc__s000__tb262144__v000", "uid": "sib_bem_prompt_3__e0cc40cc__s000__tb262144__v000", "axis": "multilingual", "budget": 2048, "token_budget": 262144, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"sib_dik_prompt_3__d4c91cdb__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 501.881098674914, \"nll_base\": 6.2183632364078445, \"eval_ref\": \"sib_dik_prompt_3__d4c91cdb__s000__tb65536__v000\", \"oracle_delta\": 0.293194517...
{ "index": "sib_dik_prompt_3__d4c91cdb__s000__tb65536__v000", "uid": "sib_dik_prompt_3__d4c91cdb__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\": \"sib_dyu_prompt_4__b3489568__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 1840.8252366666434, \"nll_base\": 7.517969248246174, \"eval_ref\": \"sib_dyu_prompt_4__b3489568__s000__tb65536__v000\", \"oracle_delta\": 0.30139969...
{ "index": "sib_dyu_prompt_4__b3489568__s000__tb65536__v000", "uid": "sib_dyu_prompt_4__b3489568__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\": \"sib_eng_prompt_5__55a2c906__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 990.2850344647031, \"nll_base\": 6.897992815290179, \"eval_ref\": \"sib_eng_prompt_5__55a2c906__s000__tb131072__v000\", \"oracle_delta\": 0.205207...
{ "index": "sib_eng_prompt_5__55a2c906__s000__tb131072__v000", "uid": "sib_eng_prompt_5__55a2c906__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\": \"sib_fra_prompt_2__aad0d46a__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 1535.442680249489, \"nll_base\": 7.336574009486607, \"eval_ref\": \"sib_fra_prompt_2__aad0d46a__s000__tb262144__v000\", \"oracle_delta\": 1.927712...
{ "index": "sib_fra_prompt_2__aad0d46a__s000__tb262144__v000", "uid": "sib_fra_prompt_2__aad0d46a__s000__tb262144__v000", "axis": "multilingual", "budget": 2048, "token_budget": 262144, "variant": "filter" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must choose which examples from a large pool of candidate training examples it should be fine-tuned on.\n\nThe base model will be LoRA-fine-tuned on exactly the e...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"sib_fuv_prompt_1__1505906c__s000__tb65536__v000\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 390.65094436617784, \"nll_base\": 5.967814435862532, \"eval_ref\": \"sib_fuv_prompt_1__1505906c__s000__tb65536__v000\", \"oracle_delta\": 0.29175051...
{ "index": "sib_fuv_prompt_1__1505906c__s000__tb65536__v000", "uid": "sib_fuv_prompt_1__1505906c__s000__tb65536__v000", "axis": "multilingual", "budget": 2048, "token_budget": 65536, "variant": "filter" }