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\": \"lambada_openai_mt_es__d838e1dd__s007__tb65536__v023\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 1003.2391122246833, \"nll_base\": 6.910989156583461, \"eval_ref\": \"lambada_openai_mt_es__d838e1dd__s007__tb65536__v023\", \"oracle_delta\": 0....
{ "index": "lambada_openai_mt_es__d838e1dd__s007__tb65536__v023", "uid": "lambada_openai_mt_es__d838e1dd__s007__tb65536__v023", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_es__d838e1dd__s008__tb131072__v024\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 1119.5772334185235, \"nll_base\": 7.020706422867313, \"eval_ref\": \"lambada_openai_mt_es__d838e1dd__s008__tb131072__v024\", \"oracle_delta\":...
{ "index": "lambada_openai_mt_es__d838e1dd__s008__tb131072__v024", "uid": "lambada_openai_mt_es__d838e1dd__s008__tb131072__v024", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_es__d838e1dd__s008__tb262144__v025\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 1119.5772334185235, \"nll_base\": 7.020706422867313, \"eval_ref\": \"lambada_openai_mt_es__d838e1dd__s008__tb262144__v025\", \"oracle_delta\":...
{ "index": "lambada_openai_mt_es__d838e1dd__s008__tb262144__v025", "uid": "lambada_openai_mt_es__d838e1dd__s008__tb262144__v025", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_es__d838e1dd__s008__tb65536__v026\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 1119.5772334185235, \"nll_base\": 7.020706422867313, \"eval_ref\": \"lambada_openai_mt_es__d838e1dd__s008__tb65536__v026\", \"oracle_delta\": 0....
{ "index": "lambada_openai_mt_es__d838e1dd__s008__tb65536__v026", "uid": "lambada_openai_mt_es__d838e1dd__s008__tb65536__v026", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_es__d838e1dd__s009__tb131072__v027\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 535.061810697885, \"nll_base\": 6.282382274237205, \"eval_ref\": \"lambada_openai_mt_es__d838e1dd__s009__tb131072__v027\", \"oracle_delta\": 0...
{ "index": "lambada_openai_mt_es__d838e1dd__s009__tb131072__v027", "uid": "lambada_openai_mt_es__d838e1dd__s009__tb131072__v027", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_es__d838e1dd__s009__tb262144__v028\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 535.061810697885, \"nll_base\": 6.282382274237205, \"eval_ref\": \"lambada_openai_mt_es__d838e1dd__s009__tb262144__v028\", \"oracle_delta\": 0...
{ "index": "lambada_openai_mt_es__d838e1dd__s009__tb262144__v028", "uid": "lambada_openai_mt_es__d838e1dd__s009__tb262144__v028", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_es__d838e1dd__s009__tb65536__v029\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 535.061810697885, \"nll_base\": 6.282382274237205, \"eval_ref\": \"lambada_openai_mt_es__d838e1dd__s009__tb65536__v029\", \"oracle_delta\": 0.34...
{ "index": "lambada_openai_mt_es__d838e1dd__s009__tb65536__v029", "uid": "lambada_openai_mt_es__d838e1dd__s009__tb65536__v029", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_es__d838e1dd__s010__tb131072__v030\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 1037.5256475266392, \"nll_base\": 6.944593972306911, \"eval_ref\": \"lambada_openai_mt_es__d838e1dd__s010__tb131072__v030\", \"oracle_delta\":...
{ "index": "lambada_openai_mt_es__d838e1dd__s010__tb131072__v030", "uid": "lambada_openai_mt_es__d838e1dd__s010__tb131072__v030", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_es__d838e1dd__s010__tb262144__v031\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 1037.5256475266392, \"nll_base\": 6.944593972306911, \"eval_ref\": \"lambada_openai_mt_es__d838e1dd__s010__tb262144__v031\", \"oracle_delta\":...
{ "index": "lambada_openai_mt_es__d838e1dd__s010__tb262144__v031", "uid": "lambada_openai_mt_es__d838e1dd__s010__tb262144__v031", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_es__d838e1dd__s010__tb65536__v032\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 1037.5256475266392, \"nll_base\": 6.944593972306911, \"eval_ref\": \"lambada_openai_mt_es__d838e1dd__s010__tb65536__v032\", \"oracle_delta\": 0....
{ "index": "lambada_openai_mt_es__d838e1dd__s010__tb65536__v032", "uid": "lambada_openai_mt_es__d838e1dd__s010__tb65536__v032", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_es__d838e1dd__s011__tb131072__v033\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 1411.751340952203, \"nll_base\": 7.252586298403533, \"eval_ref\": \"lambada_openai_mt_es__d838e1dd__s011__tb131072__v033\", \"oracle_delta\": ...
{ "index": "lambada_openai_mt_es__d838e1dd__s011__tb131072__v033", "uid": "lambada_openai_mt_es__d838e1dd__s011__tb131072__v033", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_es__d838e1dd__s011__tb262144__v034\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 1411.751340952203, \"nll_base\": 7.252586298403533, \"eval_ref\": \"lambada_openai_mt_es__d838e1dd__s011__tb262144__v034\", \"oracle_delta\": ...
{ "index": "lambada_openai_mt_es__d838e1dd__s011__tb262144__v034", "uid": "lambada_openai_mt_es__d838e1dd__s011__tb262144__v034", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_es__d838e1dd__s011__tb65536__v035\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 1411.751340952203, \"nll_base\": 7.252586298403533, \"eval_ref\": \"lambada_openai_mt_es__d838e1dd__s011__tb65536__v035\", \"oracle_delta\": 0.3...
{ "index": "lambada_openai_mt_es__d838e1dd__s011__tb65536__v035", "uid": "lambada_openai_mt_es__d838e1dd__s011__tb65536__v035", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_es__d838e1dd__s012__tb131072__v036\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 793.8967580697897, \"nll_base\": 6.6769534251729, \"eval_ref\": \"lambada_openai_mt_es__d838e1dd__s012__tb131072__v036\", \"oracle_delta\": 0....
{ "index": "lambada_openai_mt_es__d838e1dd__s012__tb131072__v036", "uid": "lambada_openai_mt_es__d838e1dd__s012__tb131072__v036", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_es__d838e1dd__s012__tb262144__v037\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 793.8967580697897, \"nll_base\": 6.6769534251729, \"eval_ref\": \"lambada_openai_mt_es__d838e1dd__s012__tb262144__v037\", \"oracle_delta\": 1....
{ "index": "lambada_openai_mt_es__d838e1dd__s012__tb262144__v037", "uid": "lambada_openai_mt_es__d838e1dd__s012__tb262144__v037", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_es__d838e1dd__s012__tb65536__v038\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 793.8967580697897, \"nll_base\": 6.6769534251729, \"eval_ref\": \"lambada_openai_mt_es__d838e1dd__s012__tb65536__v038\", \"oracle_delta\": 0.313...
{ "index": "lambada_openai_mt_es__d838e1dd__s012__tb65536__v038", "uid": "lambada_openai_mt_es__d838e1dd__s012__tb65536__v038", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_es__d838e1dd__s013__tb131072__v039\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 951.2532002592346, \"nll_base\": 6.8577802734375, \"eval_ref\": \"lambada_openai_mt_es__d838e1dd__s013__tb131072__v039\", \"oracle_delta\": 0....
{ "index": "lambada_openai_mt_es__d838e1dd__s013__tb131072__v039", "uid": "lambada_openai_mt_es__d838e1dd__s013__tb131072__v039", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_es__d838e1dd__s013__tb262144__v040\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 951.2532002592346, \"nll_base\": 6.8577802734375, \"eval_ref\": \"lambada_openai_mt_es__d838e1dd__s013__tb262144__v040\", \"oracle_delta\": 0....
{ "index": "lambada_openai_mt_es__d838e1dd__s013__tb262144__v040", "uid": "lambada_openai_mt_es__d838e1dd__s013__tb262144__v040", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_es__d838e1dd__s013__tb65536__v041\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 951.2532002592346, \"nll_base\": 6.8577802734375, \"eval_ref\": \"lambada_openai_mt_es__d838e1dd__s013__tb65536__v041\", \"oracle_delta\": 0.393...
{ "index": "lambada_openai_mt_es__d838e1dd__s013__tb65536__v041", "uid": "lambada_openai_mt_es__d838e1dd__s013__tb65536__v041", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_es__d838e1dd__s014__tb131072__v042\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 730.9515924586314, \"nll_base\": 6.594347236570248, \"eval_ref\": \"lambada_openai_mt_es__d838e1dd__s014__tb131072__v042\", \"oracle_delta\": ...
{ "index": "lambada_openai_mt_es__d838e1dd__s014__tb131072__v042", "uid": "lambada_openai_mt_es__d838e1dd__s014__tb131072__v042", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_es__d838e1dd__s014__tb262144__v043\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 730.9515924586314, \"nll_base\": 6.594347236570248, \"eval_ref\": \"lambada_openai_mt_es__d838e1dd__s014__tb262144__v043\", \"oracle_delta\": ...
{ "index": "lambada_openai_mt_es__d838e1dd__s014__tb262144__v043", "uid": "lambada_openai_mt_es__d838e1dd__s014__tb262144__v043", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_es__d838e1dd__s014__tb65536__v044\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 730.9515924586314, \"nll_base\": 6.594347236570248, \"eval_ref\": \"lambada_openai_mt_es__d838e1dd__s014__tb65536__v044\", \"oracle_delta\": 0.2...
{ "index": "lambada_openai_mt_es__d838e1dd__s014__tb65536__v044", "uid": "lambada_openai_mt_es__d838e1dd__s014__tb65536__v044", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_es__d838e1dd__s015__tb131072__v045\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 997.1248019733599, \"nll_base\": 6.904875939633666, \"eval_ref\": \"lambada_openai_mt_es__d838e1dd__s015__tb131072__v045\", \"oracle_delta\": ...
{ "index": "lambada_openai_mt_es__d838e1dd__s015__tb131072__v045", "uid": "lambada_openai_mt_es__d838e1dd__s015__tb131072__v045", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_es__d838e1dd__s015__tb262144__v046\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 997.1248019733599, \"nll_base\": 6.904875939633666, \"eval_ref\": \"lambada_openai_mt_es__d838e1dd__s015__tb262144__v046\", \"oracle_delta\": ...
{ "index": "lambada_openai_mt_es__d838e1dd__s015__tb262144__v046", "uid": "lambada_openai_mt_es__d838e1dd__s015__tb262144__v046", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_es__d838e1dd__s015__tb65536__v047\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 997.1248019733599, \"nll_base\": 6.904875939633666, \"eval_ref\": \"lambada_openai_mt_es__d838e1dd__s015__tb65536__v047\", \"oracle_delta\": 0.3...
{ "index": "lambada_openai_mt_es__d838e1dd__s015__tb65536__v047", "uid": "lambada_openai_mt_es__d838e1dd__s015__tb65536__v047", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_es__d838e1dd__s016__tb131072__v048\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 850.4311092005507, \"nll_base\": 6.745743408203125, \"eval_ref\": \"lambada_openai_mt_es__d838e1dd__s016__tb131072__v048\", \"oracle_delta\": ...
{ "index": "lambada_openai_mt_es__d838e1dd__s016__tb131072__v048", "uid": "lambada_openai_mt_es__d838e1dd__s016__tb131072__v048", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_es__d838e1dd__s016__tb262144__v049\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 850.4311092005507, \"nll_base\": 6.745743408203125, \"eval_ref\": \"lambada_openai_mt_es__d838e1dd__s016__tb262144__v049\", \"oracle_delta\": ...
{ "index": "lambada_openai_mt_es__d838e1dd__s016__tb262144__v049", "uid": "lambada_openai_mt_es__d838e1dd__s016__tb262144__v049", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_es__d838e1dd__s016__tb65536__v050\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 850.4311092005507, \"nll_base\": 6.745743408203125, \"eval_ref\": \"lambada_openai_mt_es__d838e1dd__s016__tb65536__v050\", \"oracle_delta\": 0.2...
{ "index": "lambada_openai_mt_es__d838e1dd__s016__tb65536__v050", "uid": "lambada_openai_mt_es__d838e1dd__s016__tb65536__v050", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_es__d838e1dd__s017__tb131072__v051\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 877.3014173281097, \"nll_base\": 6.77685062468998, \"eval_ref\": \"lambada_openai_mt_es__d838e1dd__s017__tb131072__v051\", \"oracle_delta\": 0...
{ "index": "lambada_openai_mt_es__d838e1dd__s017__tb131072__v051", "uid": "lambada_openai_mt_es__d838e1dd__s017__tb131072__v051", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_es__d838e1dd__s017__tb262144__v052\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 877.3014173281097, \"nll_base\": 6.77685062468998, \"eval_ref\": \"lambada_openai_mt_es__d838e1dd__s017__tb262144__v052\", \"oracle_delta\": 1...
{ "index": "lambada_openai_mt_es__d838e1dd__s017__tb262144__v052", "uid": "lambada_openai_mt_es__d838e1dd__s017__tb262144__v052", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_es__d838e1dd__s017__tb65536__v053\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 877.3014173281097, \"nll_base\": 6.77685062468998, \"eval_ref\": \"lambada_openai_mt_es__d838e1dd__s017__tb65536__v053\", \"oracle_delta\": 0.30...
{ "index": "lambada_openai_mt_es__d838e1dd__s017__tb65536__v053", "uid": "lambada_openai_mt_es__d838e1dd__s017__tb65536__v053", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_es__d838e1dd__s018__tb131072__v054\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 609.2602358401056, \"nll_base\": 6.412245493086558, \"eval_ref\": \"lambada_openai_mt_es__d838e1dd__s018__tb131072__v054\", \"oracle_delta\": ...
{ "index": "lambada_openai_mt_es__d838e1dd__s018__tb131072__v054", "uid": "lambada_openai_mt_es__d838e1dd__s018__tb131072__v054", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_es__d838e1dd__s018__tb262144__v055\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 609.2602358401056, \"nll_base\": 6.412245493086558, \"eval_ref\": \"lambada_openai_mt_es__d838e1dd__s018__tb262144__v055\", \"oracle_delta\": ...
{ "index": "lambada_openai_mt_es__d838e1dd__s018__tb262144__v055", "uid": "lambada_openai_mt_es__d838e1dd__s018__tb262144__v055", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_es__d838e1dd__s018__tb65536__v056\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 609.2602358401056, \"nll_base\": 6.412245493086558, \"eval_ref\": \"lambada_openai_mt_es__d838e1dd__s018__tb65536__v056\", \"oracle_delta\": 0.3...
{ "index": "lambada_openai_mt_es__d838e1dd__s018__tb65536__v056", "uid": "lambada_openai_mt_es__d838e1dd__s018__tb65536__v056", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_es__d838e1dd__s019__tb131072__v057\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 1799.9613213178852, \"nll_base\": 7.495520455496652, \"eval_ref\": \"lambada_openai_mt_es__d838e1dd__s019__tb131072__v057\", \"oracle_delta\":...
{ "index": "lambada_openai_mt_es__d838e1dd__s019__tb131072__v057", "uid": "lambada_openai_mt_es__d838e1dd__s019__tb131072__v057", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_es__d838e1dd__s019__tb262144__v058\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 1799.9613213178852, \"nll_base\": 7.495520455496652, \"eval_ref\": \"lambada_openai_mt_es__d838e1dd__s019__tb262144__v058\", \"oracle_delta\":...
{ "index": "lambada_openai_mt_es__d838e1dd__s019__tb262144__v058", "uid": "lambada_openai_mt_es__d838e1dd__s019__tb262144__v058", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_fr__1a46777e__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 4752.410473553284, \"nll_base\": 8.466407236426768, \"eval_ref\": \"lambada_openai_mt_fr__1a46777e__s000__tb131072__v000\", \"oracle_delta\": ...
{ "index": "lambada_openai_mt_fr__1a46777e__s000__tb131072__v000", "uid": "lambada_openai_mt_fr__1a46777e__s000__tb131072__v000", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_fr__1a46777e__s000__tb262144__v001\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 4752.410473553284, \"nll_base\": 8.466407236426768, \"eval_ref\": \"lambada_openai_mt_fr__1a46777e__s000__tb262144__v001\", \"oracle_delta\": ...
{ "index": "lambada_openai_mt_fr__1a46777e__s000__tb262144__v001", "uid": "lambada_openai_mt_fr__1a46777e__s000__tb262144__v001", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_fr__1a46777e__s000__tb65536__v002\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 4752.410473553284, \"nll_base\": 8.466407236426768, \"eval_ref\": \"lambada_openai_mt_fr__1a46777e__s000__tb65536__v002\", \"oracle_delta\": 0.3...
{ "index": "lambada_openai_mt_fr__1a46777e__s000__tb65536__v002", "uid": "lambada_openai_mt_fr__1a46777e__s000__tb65536__v002", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_fr__1a46777e__s001__tb131072__v003\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 8506.645978924762, \"nll_base\": 9.048603016843078, \"eval_ref\": \"lambada_openai_mt_fr__1a46777e__s001__tb131072__v003\", \"oracle_delta\": ...
{ "index": "lambada_openai_mt_fr__1a46777e__s001__tb131072__v003", "uid": "lambada_openai_mt_fr__1a46777e__s001__tb131072__v003", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_fr__1a46777e__s001__tb262144__v004\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 8506.645978924762, \"nll_base\": 9.048603016843078, \"eval_ref\": \"lambada_openai_mt_fr__1a46777e__s001__tb262144__v004\", \"oracle_delta\": ...
{ "index": "lambada_openai_mt_fr__1a46777e__s001__tb262144__v004", "uid": "lambada_openai_mt_fr__1a46777e__s001__tb262144__v004", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_fr__1a46777e__s001__tb65536__v005\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 8506.645978924762, \"nll_base\": 9.048603016843078, \"eval_ref\": \"lambada_openai_mt_fr__1a46777e__s001__tb65536__v005\", \"oracle_delta\": 0.4...
{ "index": "lambada_openai_mt_fr__1a46777e__s001__tb65536__v005", "uid": "lambada_openai_mt_fr__1a46777e__s001__tb65536__v005", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_fr__1a46777e__s002__tb131072__v006\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 9582.39034819598, \"nll_base\": 9.167682354266827, \"eval_ref\": \"lambada_openai_mt_fr__1a46777e__s002__tb131072__v006\", \"oracle_delta\": 0...
{ "index": "lambada_openai_mt_fr__1a46777e__s002__tb131072__v006", "uid": "lambada_openai_mt_fr__1a46777e__s002__tb131072__v006", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_fr__1a46777e__s002__tb262144__v007\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 9582.39034819598, \"nll_base\": 9.167682354266827, \"eval_ref\": \"lambada_openai_mt_fr__1a46777e__s002__tb262144__v007\", \"oracle_delta\": 1...
{ "index": "lambada_openai_mt_fr__1a46777e__s002__tb262144__v007", "uid": "lambada_openai_mt_fr__1a46777e__s002__tb262144__v007", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_fr__1a46777e__s002__tb65536__v008\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 9582.39034819598, \"nll_base\": 9.167682354266827, \"eval_ref\": \"lambada_openai_mt_fr__1a46777e__s002__tb65536__v008\", \"oracle_delta\": 0.43...
{ "index": "lambada_openai_mt_fr__1a46777e__s002__tb65536__v008", "uid": "lambada_openai_mt_fr__1a46777e__s002__tb65536__v008", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_fr__1a46777e__s003__tb131072__v009\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 3226.6977868041567, \"nll_base\": 8.079214536226713, \"eval_ref\": \"lambada_openai_mt_fr__1a46777e__s003__tb131072__v009\", \"oracle_delta\":...
{ "index": "lambada_openai_mt_fr__1a46777e__s003__tb131072__v009", "uid": "lambada_openai_mt_fr__1a46777e__s003__tb131072__v009", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_fr__1a46777e__s003__tb262144__v010\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 3226.6977868041567, \"nll_base\": 8.079214536226713, \"eval_ref\": \"lambada_openai_mt_fr__1a46777e__s003__tb262144__v010\", \"oracle_delta\":...
{ "index": "lambada_openai_mt_fr__1a46777e__s003__tb262144__v010", "uid": "lambada_openai_mt_fr__1a46777e__s003__tb262144__v010", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_fr__1a46777e__s003__tb65536__v011\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 3226.6977868041567, \"nll_base\": 8.079214536226713, \"eval_ref\": \"lambada_openai_mt_fr__1a46777e__s003__tb65536__v011\", \"oracle_delta\": 0....
{ "index": "lambada_openai_mt_fr__1a46777e__s003__tb65536__v011", "uid": "lambada_openai_mt_fr__1a46777e__s003__tb65536__v011", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_fr__1a46777e__s004__tb131072__v012\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 4161.920298658893, \"nll_base\": 8.333731857000613, \"eval_ref\": \"lambada_openai_mt_fr__1a46777e__s004__tb131072__v012\", \"oracle_delta\": ...
{ "index": "lambada_openai_mt_fr__1a46777e__s004__tb131072__v012", "uid": "lambada_openai_mt_fr__1a46777e__s004__tb131072__v012", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_fr__1a46777e__s004__tb262144__v013\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 4161.920298658893, \"nll_base\": 8.333731857000613, \"eval_ref\": \"lambada_openai_mt_fr__1a46777e__s004__tb262144__v013\", \"oracle_delta\": ...
{ "index": "lambada_openai_mt_fr__1a46777e__s004__tb262144__v013", "uid": "lambada_openai_mt_fr__1a46777e__s004__tb262144__v013", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_fr__1a46777e__s004__tb65536__v014\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 4161.920298658893, \"nll_base\": 8.333731857000613, \"eval_ref\": \"lambada_openai_mt_fr__1a46777e__s004__tb65536__v014\", \"oracle_delta\": 0.4...
{ "index": "lambada_openai_mt_fr__1a46777e__s004__tb65536__v014", "uid": "lambada_openai_mt_fr__1a46777e__s004__tb65536__v014", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_fr__1a46777e__s005__tb131072__v015\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 3416.433805412823, \"nll_base\": 8.1363525390625, \"eval_ref\": \"lambada_openai_mt_fr__1a46777e__s005__tb131072__v015\", \"oracle_delta\": 0....
{ "index": "lambada_openai_mt_fr__1a46777e__s005__tb131072__v015", "uid": "lambada_openai_mt_fr__1a46777e__s005__tb131072__v015", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_fr__1a46777e__s005__tb262144__v016\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 3416.433805412823, \"nll_base\": 8.1363525390625, \"eval_ref\": \"lambada_openai_mt_fr__1a46777e__s005__tb262144__v016\", \"oracle_delta\": 1....
{ "index": "lambada_openai_mt_fr__1a46777e__s005__tb262144__v016", "uid": "lambada_openai_mt_fr__1a46777e__s005__tb262144__v016", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_fr__1a46777e__s005__tb65536__v017\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 3416.433805412823, \"nll_base\": 8.1363525390625, \"eval_ref\": \"lambada_openai_mt_fr__1a46777e__s005__tb65536__v017\", \"oracle_delta\": 0.375...
{ "index": "lambada_openai_mt_fr__1a46777e__s005__tb65536__v017", "uid": "lambada_openai_mt_fr__1a46777e__s005__tb65536__v017", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_fr__1a46777e__s006__tb131072__v018\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 2581.266424106071, \"nll_base\": 7.85603541953891, \"eval_ref\": \"lambada_openai_mt_fr__1a46777e__s006__tb131072__v018\", \"oracle_delta\": 0...
{ "index": "lambada_openai_mt_fr__1a46777e__s006__tb131072__v018", "uid": "lambada_openai_mt_fr__1a46777e__s006__tb131072__v018", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_fr__1a46777e__s006__tb262144__v019\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 2581.266424106071, \"nll_base\": 7.85603541953891, \"eval_ref\": \"lambada_openai_mt_fr__1a46777e__s006__tb262144__v019\", \"oracle_delta\": 1...
{ "index": "lambada_openai_mt_fr__1a46777e__s006__tb262144__v019", "uid": "lambada_openai_mt_fr__1a46777e__s006__tb262144__v019", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_fr__1a46777e__s006__tb65536__v020\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 2581.266424106071, \"nll_base\": 7.85603541953891, \"eval_ref\": \"lambada_openai_mt_fr__1a46777e__s006__tb65536__v020\", \"oracle_delta\": 0.45...
{ "index": "lambada_openai_mt_fr__1a46777e__s006__tb65536__v020", "uid": "lambada_openai_mt_fr__1a46777e__s006__tb65536__v020", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_fr__1a46777e__s007__tb131072__v021\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 9218.484741501481, \"nll_base\": 9.128965958305027, \"eval_ref\": \"lambada_openai_mt_fr__1a46777e__s007__tb131072__v021\", \"oracle_delta\": ...
{ "index": "lambada_openai_mt_fr__1a46777e__s007__tb131072__v021", "uid": "lambada_openai_mt_fr__1a46777e__s007__tb131072__v021", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_fr__1a46777e__s007__tb262144__v022\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 9218.484741501481, \"nll_base\": 9.128965958305027, \"eval_ref\": \"lambada_openai_mt_fr__1a46777e__s007__tb262144__v022\", \"oracle_delta\": ...
{ "index": "lambada_openai_mt_fr__1a46777e__s007__tb262144__v022", "uid": "lambada_openai_mt_fr__1a46777e__s007__tb262144__v022", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_fr__1a46777e__s007__tb65536__v023\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 9218.484741501481, \"nll_base\": 9.128965958305027, \"eval_ref\": \"lambada_openai_mt_fr__1a46777e__s007__tb65536__v023\", \"oracle_delta\": 0.4...
{ "index": "lambada_openai_mt_fr__1a46777e__s007__tb65536__v023", "uid": "lambada_openai_mt_fr__1a46777e__s007__tb65536__v023", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_fr__1a46777e__s008__tb131072__v024\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 6412.638199220147, \"nll_base\": 8.7660260407821, \"eval_ref\": \"lambada_openai_mt_fr__1a46777e__s008__tb131072__v024\", \"oracle_delta\": 0....
{ "index": "lambada_openai_mt_fr__1a46777e__s008__tb131072__v024", "uid": "lambada_openai_mt_fr__1a46777e__s008__tb131072__v024", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_fr__1a46777e__s008__tb262144__v025\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 6412.638199220147, \"nll_base\": 8.7660260407821, \"eval_ref\": \"lambada_openai_mt_fr__1a46777e__s008__tb262144__v025\", \"oracle_delta\": 1....
{ "index": "lambada_openai_mt_fr__1a46777e__s008__tb262144__v025", "uid": "lambada_openai_mt_fr__1a46777e__s008__tb262144__v025", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_fr__1a46777e__s008__tb65536__v026\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 6412.638199220147, \"nll_base\": 8.7660260407821, \"eval_ref\": \"lambada_openai_mt_fr__1a46777e__s008__tb65536__v026\", \"oracle_delta\": 0.439...
{ "index": "lambada_openai_mt_fr__1a46777e__s008__tb65536__v026", "uid": "lambada_openai_mt_fr__1a46777e__s008__tb65536__v026", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_fr__1a46777e__s009__tb131072__v027\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 3173.9581909993603, \"nll_base\": 8.062734728661653, \"eval_ref\": \"lambada_openai_mt_fr__1a46777e__s009__tb131072__v027\", \"oracle_delta\":...
{ "index": "lambada_openai_mt_fr__1a46777e__s009__tb131072__v027", "uid": "lambada_openai_mt_fr__1a46777e__s009__tb131072__v027", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_fr__1a46777e__s009__tb262144__v028\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 3173.9581909993603, \"nll_base\": 8.062734728661653, \"eval_ref\": \"lambada_openai_mt_fr__1a46777e__s009__tb262144__v028\", \"oracle_delta\":...
{ "index": "lambada_openai_mt_fr__1a46777e__s009__tb262144__v028", "uid": "lambada_openai_mt_fr__1a46777e__s009__tb262144__v028", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_fr__1a46777e__s009__tb65536__v029\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 3173.9581909993603, \"nll_base\": 8.062734728661653, \"eval_ref\": \"lambada_openai_mt_fr__1a46777e__s009__tb65536__v029\", \"oracle_delta\": 0....
{ "index": "lambada_openai_mt_fr__1a46777e__s009__tb65536__v029", "uid": "lambada_openai_mt_fr__1a46777e__s009__tb65536__v029", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_fr__1a46777e__s010__tb131072__v030\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 5242.759294609862, \"nll_base\": 8.564603221659757, \"eval_ref\": \"lambada_openai_mt_fr__1a46777e__s010__tb131072__v030\", \"oracle_delta\": ...
{ "index": "lambada_openai_mt_fr__1a46777e__s010__tb131072__v030", "uid": "lambada_openai_mt_fr__1a46777e__s010__tb131072__v030", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_fr__1a46777e__s010__tb262144__v031\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 5242.759294609862, \"nll_base\": 8.564603221659757, \"eval_ref\": \"lambada_openai_mt_fr__1a46777e__s010__tb262144__v031\", \"oracle_delta\": ...
{ "index": "lambada_openai_mt_fr__1a46777e__s010__tb262144__v031", "uid": "lambada_openai_mt_fr__1a46777e__s010__tb262144__v031", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_fr__1a46777e__s010__tb65536__v032\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 5242.759294609862, \"nll_base\": 8.564603221659757, \"eval_ref\": \"lambada_openai_mt_fr__1a46777e__s010__tb65536__v032\", \"oracle_delta\": 0.3...
{ "index": "lambada_openai_mt_fr__1a46777e__s010__tb65536__v032", "uid": "lambada_openai_mt_fr__1a46777e__s010__tb65536__v032", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_fr__1a46777e__s011__tb131072__v033\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 1992.7515965803414, \"nll_base\": 7.597271674502212, \"eval_ref\": \"lambada_openai_mt_fr__1a46777e__s011__tb131072__v033\", \"oracle_delta\":...
{ "index": "lambada_openai_mt_fr__1a46777e__s011__tb131072__v033", "uid": "lambada_openai_mt_fr__1a46777e__s011__tb131072__v033", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_fr__1a46777e__s011__tb262144__v034\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 1992.7515965803414, \"nll_base\": 7.597271674502212, \"eval_ref\": \"lambada_openai_mt_fr__1a46777e__s011__tb262144__v034\", \"oracle_delta\":...
{ "index": "lambada_openai_mt_fr__1a46777e__s011__tb262144__v034", "uid": "lambada_openai_mt_fr__1a46777e__s011__tb262144__v034", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_fr__1a46777e__s011__tb65536__v035\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 1992.7515965803414, \"nll_base\": 7.597271674502212, \"eval_ref\": \"lambada_openai_mt_fr__1a46777e__s011__tb65536__v035\", \"oracle_delta\": 0....
{ "index": "lambada_openai_mt_fr__1a46777e__s011__tb65536__v035", "uid": "lambada_openai_mt_fr__1a46777e__s011__tb65536__v035", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_fr__1a46777e__s012__tb131072__v036\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 6726.251590390885, \"nll_base\": 8.813773297249003, \"eval_ref\": \"lambada_openai_mt_fr__1a46777e__s012__tb131072__v036\", \"oracle_delta\": ...
{ "index": "lambada_openai_mt_fr__1a46777e__s012__tb131072__v036", "uid": "lambada_openai_mt_fr__1a46777e__s012__tb131072__v036", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_fr__1a46777e__s012__tb262144__v037\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 6726.251590390885, \"nll_base\": 8.813773297249003, \"eval_ref\": \"lambada_openai_mt_fr__1a46777e__s012__tb262144__v037\", \"oracle_delta\": ...
{ "index": "lambada_openai_mt_fr__1a46777e__s012__tb262144__v037", "uid": "lambada_openai_mt_fr__1a46777e__s012__tb262144__v037", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_fr__1a46777e__s012__tb65536__v038\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 6726.251590390885, \"nll_base\": 8.813773297249003, \"eval_ref\": \"lambada_openai_mt_fr__1a46777e__s012__tb65536__v038\", \"oracle_delta\": 0.4...
{ "index": "lambada_openai_mt_fr__1a46777e__s012__tb65536__v038", "uid": "lambada_openai_mt_fr__1a46777e__s012__tb65536__v038", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_fr__1a46777e__s013__tb131072__v039\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 3272.6965186810394, \"nll_base\": 8.093369547526041, \"eval_ref\": \"lambada_openai_mt_fr__1a46777e__s013__tb131072__v039\", \"oracle_delta\":...
{ "index": "lambada_openai_mt_fr__1a46777e__s013__tb131072__v039", "uid": "lambada_openai_mt_fr__1a46777e__s013__tb131072__v039", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_fr__1a46777e__s013__tb262144__v040\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 3272.6965186810394, \"nll_base\": 8.093369547526041, \"eval_ref\": \"lambada_openai_mt_fr__1a46777e__s013__tb262144__v040\", \"oracle_delta\":...
{ "index": "lambada_openai_mt_fr__1a46777e__s013__tb262144__v040", "uid": "lambada_openai_mt_fr__1a46777e__s013__tb262144__v040", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_fr__1a46777e__s013__tb65536__v041\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 3272.6965186810394, \"nll_base\": 8.093369547526041, \"eval_ref\": \"lambada_openai_mt_fr__1a46777e__s013__tb65536__v041\", \"oracle_delta\": 0....
{ "index": "lambada_openai_mt_fr__1a46777e__s013__tb65536__v041", "uid": "lambada_openai_mt_fr__1a46777e__s013__tb65536__v041", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_fr__1a46777e__s014__tb131072__v042\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 3211.6922136857324, \"nll_base\": 8.074553246591606, \"eval_ref\": \"lambada_openai_mt_fr__1a46777e__s014__tb131072__v042\", \"oracle_delta\":...
{ "index": "lambada_openai_mt_fr__1a46777e__s014__tb131072__v042", "uid": "lambada_openai_mt_fr__1a46777e__s014__tb131072__v042", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_fr__1a46777e__s014__tb262144__v043\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 3211.6922136857324, \"nll_base\": 8.074553246591606, \"eval_ref\": \"lambada_openai_mt_fr__1a46777e__s014__tb262144__v043\", \"oracle_delta\":...
{ "index": "lambada_openai_mt_fr__1a46777e__s014__tb262144__v043", "uid": "lambada_openai_mt_fr__1a46777e__s014__tb262144__v043", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_fr__1a46777e__s014__tb65536__v044\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 3211.6922136857324, \"nll_base\": 8.074553246591606, \"eval_ref\": \"lambada_openai_mt_fr__1a46777e__s014__tb65536__v044\", \"oracle_delta\": 0....
{ "index": "lambada_openai_mt_fr__1a46777e__s014__tb65536__v044", "uid": "lambada_openai_mt_fr__1a46777e__s014__tb65536__v044", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_fr__1a46777e__s015__tb131072__v045\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 7312.7207812057195, \"nll_base\": 8.89737068338597, \"eval_ref\": \"lambada_openai_mt_fr__1a46777e__s015__tb131072__v045\", \"oracle_delta\": ...
{ "index": "lambada_openai_mt_fr__1a46777e__s015__tb131072__v045", "uid": "lambada_openai_mt_fr__1a46777e__s015__tb131072__v045", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_fr__1a46777e__s015__tb262144__v046\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 7312.7207812057195, \"nll_base\": 8.89737068338597, \"eval_ref\": \"lambada_openai_mt_fr__1a46777e__s015__tb262144__v046\", \"oracle_delta\": ...
{ "index": "lambada_openai_mt_fr__1a46777e__s015__tb262144__v046", "uid": "lambada_openai_mt_fr__1a46777e__s015__tb262144__v046", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_fr__1a46777e__s015__tb65536__v047\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 7312.7207812057195, \"nll_base\": 8.89737068338597, \"eval_ref\": \"lambada_openai_mt_fr__1a46777e__s015__tb65536__v047\", \"oracle_delta\": 0.2...
{ "index": "lambada_openai_mt_fr__1a46777e__s015__tb65536__v047", "uid": "lambada_openai_mt_fr__1a46777e__s015__tb65536__v047", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_fr__1a46777e__s016__tb131072__v048\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 8140.197238140187, \"nll_base\": 9.00456968943278, \"eval_ref\": \"lambada_openai_mt_fr__1a46777e__s016__tb131072__v048\", \"oracle_delta\": 0...
{ "index": "lambada_openai_mt_fr__1a46777e__s016__tb131072__v048", "uid": "lambada_openai_mt_fr__1a46777e__s016__tb131072__v048", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_fr__1a46777e__s016__tb262144__v049\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 8140.197238140187, \"nll_base\": 9.00456968943278, \"eval_ref\": \"lambada_openai_mt_fr__1a46777e__s016__tb262144__v049\", \"oracle_delta\": 1...
{ "index": "lambada_openai_mt_fr__1a46777e__s016__tb262144__v049", "uid": "lambada_openai_mt_fr__1a46777e__s016__tb262144__v049", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_fr__1a46777e__s016__tb65536__v050\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 8140.197238140187, \"nll_base\": 9.00456968943278, \"eval_ref\": \"lambada_openai_mt_fr__1a46777e__s016__tb65536__v050\", \"oracle_delta\": 0.37...
{ "index": "lambada_openai_mt_fr__1a46777e__s016__tb65536__v050", "uid": "lambada_openai_mt_fr__1a46777e__s016__tb65536__v050", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_fr__1a46777e__s017__tb131072__v051\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 5147.880378936619, \"nll_base\": 8.54634033203125, \"eval_ref\": \"lambada_openai_mt_fr__1a46777e__s017__tb131072__v051\", \"oracle_delta\": 0...
{ "index": "lambada_openai_mt_fr__1a46777e__s017__tb131072__v051", "uid": "lambada_openai_mt_fr__1a46777e__s017__tb131072__v051", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_fr__1a46777e__s017__tb262144__v052\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 5147.880378936619, \"nll_base\": 8.54634033203125, \"eval_ref\": \"lambada_openai_mt_fr__1a46777e__s017__tb262144__v052\", \"oracle_delta\": 1...
{ "index": "lambada_openai_mt_fr__1a46777e__s017__tb262144__v052", "uid": "lambada_openai_mt_fr__1a46777e__s017__tb262144__v052", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_fr__1a46777e__s017__tb65536__v053\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 5147.880378936619, \"nll_base\": 8.54634033203125, \"eval_ref\": \"lambada_openai_mt_fr__1a46777e__s017__tb65536__v053\", \"oracle_delta\": 0.47...
{ "index": "lambada_openai_mt_fr__1a46777e__s017__tb65536__v053", "uid": "lambada_openai_mt_fr__1a46777e__s017__tb65536__v053", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_fr__1a46777e__s018__tb131072__v054\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 2422.379660421127, \"nll_base\": 7.792505666750286, \"eval_ref\": \"lambada_openai_mt_fr__1a46777e__s018__tb131072__v054\", \"oracle_delta\": ...
{ "index": "lambada_openai_mt_fr__1a46777e__s018__tb131072__v054", "uid": "lambada_openai_mt_fr__1a46777e__s018__tb131072__v054", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_fr__1a46777e__s018__tb262144__v055\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 2422.379660421127, \"nll_base\": 7.792505666750286, \"eval_ref\": \"lambada_openai_mt_fr__1a46777e__s018__tb262144__v055\", \"oracle_delta\": ...
{ "index": "lambada_openai_mt_fr__1a46777e__s018__tb262144__v055", "uid": "lambada_openai_mt_fr__1a46777e__s018__tb262144__v055", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_fr__1a46777e__s018__tb65536__v056\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 2422.379660421127, \"nll_base\": 7.792505666750286, \"eval_ref\": \"lambada_openai_mt_fr__1a46777e__s018__tb65536__v056\", \"oracle_delta\": 0.4...
{ "index": "lambada_openai_mt_fr__1a46777e__s018__tb65536__v056", "uid": "lambada_openai_mt_fr__1a46777e__s018__tb65536__v056", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_fr__1a46777e__s019__tb131072__v057\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 19638.843144811315, \"nll_base\": 9.885264677159926, \"eval_ref\": \"lambada_openai_mt_fr__1a46777e__s019__tb131072__v057\", \"oracle_delta\":...
{ "index": "lambada_openai_mt_fr__1a46777e__s019__tb131072__v057", "uid": "lambada_openai_mt_fr__1a46777e__s019__tb131072__v057", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_fr__1a46777e__s019__tb262144__v058\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 19638.843144811315, \"nll_base\": 9.885264677159926, \"eval_ref\": \"lambada_openai_mt_fr__1a46777e__s019__tb262144__v058\", \"oracle_delta\":...
{ "index": "lambada_openai_mt_fr__1a46777e__s019__tb262144__v058", "uid": "lambada_openai_mt_fr__1a46777e__s019__tb262144__v058", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_it__1a798b96__s000__tb131072__v002\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 3250.817906360945, \"nll_base\": 8.086661907157513, \"eval_ref\": \"lambada_openai_mt_it__1a798b96__s000__tb131072__v002\", \"oracle_delta\": ...
{ "index": "lambada_openai_mt_it__1a798b96__s000__tb131072__v002", "uid": "lambada_openai_mt_it__1a798b96__s000__tb131072__v002", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_it__1a798b96__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 3250.817906360945, \"nll_base\": 8.086661907157513, \"eval_ref\": \"lambada_openai_mt_it__1a798b96__s000__tb262144__v000\", \"oracle_delta\": ...
{ "index": "lambada_openai_mt_it__1a798b96__s000__tb262144__v000", "uid": "lambada_openai_mt_it__1a798b96__s000__tb262144__v000", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_it__1a798b96__s000__tb65536__v001\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 3250.817906360945, \"nll_base\": 8.086661907157513, \"eval_ref\": \"lambada_openai_mt_it__1a798b96__s000__tb65536__v001\", \"oracle_delta\": 0.3...
{ "index": "lambada_openai_mt_it__1a798b96__s000__tb65536__v001", "uid": "lambada_openai_mt_it__1a798b96__s000__tb65536__v001", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_it__1a798b96__s001__tb131072__v005\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 3991.785129772483, \"nll_base\": 8.291993810775432, \"eval_ref\": \"lambada_openai_mt_it__1a798b96__s001__tb131072__v005\", \"oracle_delta\": ...
{ "index": "lambada_openai_mt_it__1a798b96__s001__tb131072__v005", "uid": "lambada_openai_mt_it__1a798b96__s001__tb131072__v005", "axis": "multilingual_ppl", "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\": \"lambada_openai_mt_it__1a798b96__s001__tb262144__v003\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 3991.785129772483, \"nll_base\": 8.291993810775432, \"eval_ref\": \"lambada_openai_mt_it__1a798b96__s001__tb262144__v003\", \"oracle_delta\": ...
{ "index": "lambada_openai_mt_it__1a798b96__s001__tb262144__v003", "uid": "lambada_openai_mt_it__1a798b96__s001__tb262144__v003", "axis": "multilingual_ppl", "budget": 2048, "token_budget": 262144, "variant": "filter" }