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_stablelm_pt__e348cb41__s014__tb65536__v042\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 8759.700694663114, \"nll_base\": 9.077917016070822, \"eval_ref\": \"lambada_openai_mt_stablelm_pt__e348cb41__s014__tb65536__v042\", \"o...
{ "index": "lambada_openai_mt_stablelm_pt__e348cb41__s014__tb65536__v042", "uid": "lambada_openai_mt_stablelm_pt__e348cb41__s014__tb65536__v042", "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_stablelm_pt__e348cb41__s015__tb131072__v046\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 3695.081266283037, \"nll_base\": 8.214757826721783, \"eval_ref\": \"lambada_openai_mt_stablelm_pt__e348cb41__s015__tb131072__v046\", ...
{ "index": "lambada_openai_mt_stablelm_pt__e348cb41__s015__tb131072__v046", "uid": "lambada_openai_mt_stablelm_pt__e348cb41__s015__tb131072__v046", "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_stablelm_pt__e348cb41__s015__tb262144__v047\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 3695.081266283037, \"nll_base\": 8.214757826721783, \"eval_ref\": \"lambada_openai_mt_stablelm_pt__e348cb41__s015__tb262144__v047\", ...
{ "index": "lambada_openai_mt_stablelm_pt__e348cb41__s015__tb262144__v047", "uid": "lambada_openai_mt_stablelm_pt__e348cb41__s015__tb262144__v047", "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_stablelm_pt__e348cb41__s015__tb65536__v045\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 3695.081266283037, \"nll_base\": 8.214757826721783, \"eval_ref\": \"lambada_openai_mt_stablelm_pt__e348cb41__s015__tb65536__v045\", \"o...
{ "index": "lambada_openai_mt_stablelm_pt__e348cb41__s015__tb65536__v045", "uid": "lambada_openai_mt_stablelm_pt__e348cb41__s015__tb65536__v045", "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_stablelm_pt__e348cb41__s016__tb131072__v049\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 758.19380687515, \"nll_base\": 6.630939034854665, \"eval_ref\": \"lambada_openai_mt_stablelm_pt__e348cb41__s016__tb131072__v049\", \"...
{ "index": "lambada_openai_mt_stablelm_pt__e348cb41__s016__tb131072__v049", "uid": "lambada_openai_mt_stablelm_pt__e348cb41__s016__tb131072__v049", "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_stablelm_pt__e348cb41__s016__tb262144__v050\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 758.19380687515, \"nll_base\": 6.630939034854665, \"eval_ref\": \"lambada_openai_mt_stablelm_pt__e348cb41__s016__tb262144__v050\", \"...
{ "index": "lambada_openai_mt_stablelm_pt__e348cb41__s016__tb262144__v050", "uid": "lambada_openai_mt_stablelm_pt__e348cb41__s016__tb262144__v050", "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_stablelm_pt__e348cb41__s016__tb65536__v048\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 758.19380687515, \"nll_base\": 6.630939034854665, \"eval_ref\": \"lambada_openai_mt_stablelm_pt__e348cb41__s016__tb65536__v048\", \"ora...
{ "index": "lambada_openai_mt_stablelm_pt__e348cb41__s016__tb65536__v048", "uid": "lambada_openai_mt_stablelm_pt__e348cb41__s016__tb65536__v048", "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_stablelm_pt__e348cb41__s017__tb131072__v052\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 2261.852426641524, \"nll_base\": 7.723939414336303, \"eval_ref\": \"lambada_openai_mt_stablelm_pt__e348cb41__s017__tb131072__v052\", ...
{ "index": "lambada_openai_mt_stablelm_pt__e348cb41__s017__tb131072__v052", "uid": "lambada_openai_mt_stablelm_pt__e348cb41__s017__tb131072__v052", "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_stablelm_pt__e348cb41__s017__tb262144__v053\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 2261.852426641524, \"nll_base\": 7.723939414336303, \"eval_ref\": \"lambada_openai_mt_stablelm_pt__e348cb41__s017__tb262144__v053\", ...
{ "index": "lambada_openai_mt_stablelm_pt__e348cb41__s017__tb262144__v053", "uid": "lambada_openai_mt_stablelm_pt__e348cb41__s017__tb262144__v053", "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_stablelm_pt__e348cb41__s017__tb65536__v051\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 2261.852426641524, \"nll_base\": 7.723939414336303, \"eval_ref\": \"lambada_openai_mt_stablelm_pt__e348cb41__s017__tb65536__v051\", \"o...
{ "index": "lambada_openai_mt_stablelm_pt__e348cb41__s017__tb65536__v051", "uid": "lambada_openai_mt_stablelm_pt__e348cb41__s017__tb65536__v051", "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_stablelm_pt__e348cb41__s018__tb131072__v055\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 6179.929115383594, \"nll_base\": 8.7290620803833, \"eval_ref\": \"lambada_openai_mt_stablelm_pt__e348cb41__s018__tb131072__v055\", \"...
{ "index": "lambada_openai_mt_stablelm_pt__e348cb41__s018__tb131072__v055", "uid": "lambada_openai_mt_stablelm_pt__e348cb41__s018__tb131072__v055", "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_stablelm_pt__e348cb41__s018__tb262144__v056\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 6179.929115383594, \"nll_base\": 8.7290620803833, \"eval_ref\": \"lambada_openai_mt_stablelm_pt__e348cb41__s018__tb262144__v056\", \"...
{ "index": "lambada_openai_mt_stablelm_pt__e348cb41__s018__tb262144__v056", "uid": "lambada_openai_mt_stablelm_pt__e348cb41__s018__tb262144__v056", "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_stablelm_pt__e348cb41__s018__tb65536__v054\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 6179.929115383594, \"nll_base\": 8.7290620803833, \"eval_ref\": \"lambada_openai_mt_stablelm_pt__e348cb41__s018__tb65536__v054\", \"ora...
{ "index": "lambada_openai_mt_stablelm_pt__e348cb41__s018__tb65536__v054", "uid": "lambada_openai_mt_stablelm_pt__e348cb41__s018__tb65536__v054", "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_stablelm_pt__e348cb41__s019__tb131072__v058\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 6165.984803129127, \"nll_base\": 8.726803143819174, \"eval_ref\": \"lambada_openai_mt_stablelm_pt__e348cb41__s019__tb131072__v058\", ...
{ "index": "lambada_openai_mt_stablelm_pt__e348cb41__s019__tb131072__v058", "uid": "lambada_openai_mt_stablelm_pt__e348cb41__s019__tb131072__v058", "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_stablelm_pt__e348cb41__s019__tb65536__v057\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 6165.984803129127, \"nll_base\": 8.726803143819174, \"eval_ref\": \"lambada_openai_mt_stablelm_pt__e348cb41__s019__tb65536__v057\", \"o...
{ "index": "lambada_openai_mt_stablelm_pt__e348cb41__s019__tb65536__v057", "uid": "lambada_openai_mt_stablelm_pt__e348cb41__s019__tb65536__v057", "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_standard_cloze_yaml__b4f1b826__s001__tb262144__v004\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 32250.13727643104, \"nll_base\": 10.38127757884838, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s001__tb262144__v004\", \"or...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s001__tb262144__v004", "uid": "lambada_standard_cloze_yaml__b4f1b826__s001__tb262144__v004", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s002__tb131072__v006\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 36241.04144705764, \"nll_base\": 10.497947497245592, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s002__tb131072__v006\", \"o...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s002__tb131072__v006", "uid": "lambada_standard_cloze_yaml__b4f1b826__s002__tb131072__v006", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s002__tb262144__v007\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 36241.04144705764, \"nll_base\": 10.497947497245592, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s002__tb262144__v007\", \"o...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s002__tb262144__v007", "uid": "lambada_standard_cloze_yaml__b4f1b826__s002__tb262144__v007", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s002__tb65536__v008\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 36241.04144705764, \"nll_base\": 10.497947497245592, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s002__tb65536__v008\", \"orac...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s002__tb65536__v008", "uid": "lambada_standard_cloze_yaml__b4f1b826__s002__tb65536__v008", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s004__tb131072__v012\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 21147.62254840573, \"nll_base\": 9.959282769097221, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s004__tb131072__v012\", \"or...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s004__tb131072__v012", "uid": "lambada_standard_cloze_yaml__b4f1b826__s004__tb131072__v012", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s004__tb262144__v013\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 21147.62254840573, \"nll_base\": 9.959282769097221, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s004__tb262144__v013\", \"or...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s004__tb262144__v013", "uid": "lambada_standard_cloze_yaml__b4f1b826__s004__tb262144__v013", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s006__tb131072__v018\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 11861.769221017137, \"nll_base\": 9.381075836891352, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s006__tb131072__v018\", \"o...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s006__tb131072__v018", "uid": "lambada_standard_cloze_yaml__b4f1b826__s006__tb131072__v018", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s006__tb262144__v019\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 11861.769221017137, \"nll_base\": 9.381075836891352, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s006__tb262144__v019\", \"o...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s006__tb262144__v019", "uid": "lambada_standard_cloze_yaml__b4f1b826__s006__tb262144__v019", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s008__tb131072__v024\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 19799.894192521948, \"nll_base\": 9.893431872856326, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s008__tb131072__v024\", \"o...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s008__tb131072__v024", "uid": "lambada_standard_cloze_yaml__b4f1b826__s008__tb131072__v024", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s008__tb262144__v025\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 19799.894192521948, \"nll_base\": 9.893431872856326, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s008__tb262144__v025\", \"o...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s008__tb262144__v025", "uid": "lambada_standard_cloze_yaml__b4f1b826__s008__tb262144__v025", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s009__tb131072__v027\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 20215.195336078352, \"nll_base\": 9.914189844955633, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s009__tb131072__v027\", \"o...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s009__tb131072__v027", "uid": "lambada_standard_cloze_yaml__b4f1b826__s009__tb131072__v027", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s009__tb262144__v028\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 20215.195336078352, \"nll_base\": 9.914189844955633, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s009__tb262144__v028\", \"o...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s009__tb262144__v028", "uid": "lambada_standard_cloze_yaml__b4f1b826__s009__tb262144__v028", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s010__tb131072__v030\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 12988.575648600168, \"nll_base\": 9.471825453814338, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s010__tb131072__v030\", \"o...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s010__tb131072__v030", "uid": "lambada_standard_cloze_yaml__b4f1b826__s010__tb131072__v030", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s010__tb262144__v031\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 12988.575648600168, \"nll_base\": 9.471825453814338, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s010__tb262144__v031\", \"o...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s010__tb262144__v031", "uid": "lambada_standard_cloze_yaml__b4f1b826__s010__tb262144__v031", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s010__tb65536__v032\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 12988.575648600168, \"nll_base\": 9.471825453814338, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s010__tb65536__v032\", \"orac...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s010__tb65536__v032", "uid": "lambada_standard_cloze_yaml__b4f1b826__s010__tb65536__v032", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s011__tb131072__v033\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 22845.399736253934, \"nll_base\": 10.03650505160108, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s011__tb131072__v033\", \"o...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s011__tb131072__v033", "uid": "lambada_standard_cloze_yaml__b4f1b826__s011__tb131072__v033", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s011__tb262144__v034\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 22845.399736253934, \"nll_base\": 10.03650505160108, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s011__tb262144__v034\", \"o...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s011__tb262144__v034", "uid": "lambada_standard_cloze_yaml__b4f1b826__s011__tb262144__v034", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s011__tb65536__v035\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 22845.399736253934, \"nll_base\": 10.03650505160108, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s011__tb65536__v035\", \"orac...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s011__tb65536__v035", "uid": "lambada_standard_cloze_yaml__b4f1b826__s011__tb65536__v035", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s012__tb262144__v037\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 12495.059364578197, \"nll_base\": 9.433088594324449, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s012__tb262144__v037\", \"o...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s012__tb262144__v037", "uid": "lambada_standard_cloze_yaml__b4f1b826__s012__tb262144__v037", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s014__tb131072__v042\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 21300.783659157933, \"nll_base\": 9.966499142530488, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s014__tb131072__v042\", \"o...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s014__tb131072__v042", "uid": "lambada_standard_cloze_yaml__b4f1b826__s014__tb131072__v042", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s014__tb262144__v043\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 21300.783659157933, \"nll_base\": 9.966499142530488, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s014__tb262144__v043\", \"o...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s014__tb262144__v043", "uid": "lambada_standard_cloze_yaml__b4f1b826__s014__tb262144__v043", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s015__tb131072__v045\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 17854.163320349115, \"nll_base\": 9.789991999246988, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s015__tb131072__v045\", \"o...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s015__tb131072__v045", "uid": "lambada_standard_cloze_yaml__b4f1b826__s015__tb131072__v045", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s015__tb262144__v046\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 17854.163320349115, \"nll_base\": 9.789991999246988, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s015__tb262144__v046\", \"o...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s015__tb262144__v046", "uid": "lambada_standard_cloze_yaml__b4f1b826__s015__tb262144__v046", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s015__tb65536__v047\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 17854.163320349115, \"nll_base\": 9.789991999246988, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s015__tb65536__v047\", \"orac...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s015__tb65536__v047", "uid": "lambada_standard_cloze_yaml__b4f1b826__s015__tb65536__v047", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s016__tb131072__v048\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 36522.50757819044, \"nll_base\": 10.505683995500396, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s016__tb131072__v048\", \"o...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s016__tb131072__v048", "uid": "lambada_standard_cloze_yaml__b4f1b826__s016__tb131072__v048", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s016__tb262144__v049\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 36522.50757819044, \"nll_base\": 10.505683995500396, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s016__tb262144__v049\", \"o...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s016__tb262144__v049", "uid": "lambada_standard_cloze_yaml__b4f1b826__s016__tb262144__v049", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s016__tb65536__v050\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 36522.50757819044, \"nll_base\": 10.505683995500396, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s016__tb65536__v050\", \"orac...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s016__tb65536__v050", "uid": "lambada_standard_cloze_yaml__b4f1b826__s016__tb65536__v050", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s018__tb131072__v054\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 18734.264612050512, \"nll_base\": 9.838109458365091, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s018__tb131072__v054\", \"o...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s018__tb131072__v054", "uid": "lambada_standard_cloze_yaml__b4f1b826__s018__tb131072__v054", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s018__tb65536__v056\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 18734.264612050512, \"nll_base\": 9.838109458365091, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s018__tb65536__v056\", \"orac...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s018__tb65536__v056", "uid": "lambada_standard_cloze_yaml__b4f1b826__s018__tb65536__v056", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s019__tb131072__v057\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 18060.47879145579, \"nll_base\": 9.801481337774367, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s019__tb131072__v057\", \"or...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s019__tb131072__v057", "uid": "lambada_standard_cloze_yaml__b4f1b826__s019__tb131072__v057", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s019__tb262144__v058\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 18060.47879145579, \"nll_base\": 9.801481337774367, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s019__tb262144__v058\", \"or...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s019__tb262144__v058", "uid": "lambada_standard_cloze_yaml__b4f1b826__s019__tb262144__v058", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s021__tb262144__v064\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 34015.28599391031, \"nll_base\": 10.434565290620055, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s021__tb262144__v064\", \"o...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s021__tb262144__v064", "uid": "lambada_standard_cloze_yaml__b4f1b826__s021__tb262144__v064", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s022__tb131072__v066\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 23919.28365254364, \"nll_base\": 10.082440259979993, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s022__tb131072__v066\", \"o...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s022__tb131072__v066", "uid": "lambada_standard_cloze_yaml__b4f1b826__s022__tb131072__v066", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s022__tb262144__v067\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 23919.28365254364, \"nll_base\": 10.082440259979993, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s022__tb262144__v067\", \"o...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s022__tb262144__v067", "uid": "lambada_standard_cloze_yaml__b4f1b826__s022__tb262144__v067", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s022__tb65536__v068\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 23919.28365254364, \"nll_base\": 10.082440259979993, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s022__tb65536__v068\", \"orac...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s022__tb65536__v068", "uid": "lambada_standard_cloze_yaml__b4f1b826__s022__tb65536__v068", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s023__tb131072__v069\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 35553.81713745025, \"nll_base\": 10.478802803235176, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s023__tb131072__v069\", \"o...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s023__tb131072__v069", "uid": "lambada_standard_cloze_yaml__b4f1b826__s023__tb131072__v069", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s023__tb262144__v070\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 35553.81713745025, \"nll_base\": 10.478802803235176, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s023__tb262144__v070\", \"o...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s023__tb262144__v070", "uid": "lambada_standard_cloze_yaml__b4f1b826__s023__tb262144__v070", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s023__tb65536__v071\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 35553.81713745025, \"nll_base\": 10.478802803235176, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s023__tb65536__v071\", \"orac...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s023__tb65536__v071", "uid": "lambada_standard_cloze_yaml__b4f1b826__s023__tb65536__v071", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s024__tb131072__v072\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 10539.400108320076, \"nll_base\": 9.262875904756434, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s024__tb131072__v072\", \"o...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s024__tb131072__v072", "uid": "lambada_standard_cloze_yaml__b4f1b826__s024__tb131072__v072", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s024__tb262144__v073\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 10539.400108320076, \"nll_base\": 9.262875904756434, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s024__tb262144__v073\", \"o...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s024__tb262144__v073", "uid": "lambada_standard_cloze_yaml__b4f1b826__s024__tb262144__v073", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s024__tb65536__v074\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 10539.400108320076, \"nll_base\": 9.262875904756434, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s024__tb65536__v074\", \"orac...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s024__tb65536__v074", "uid": "lambada_standard_cloze_yaml__b4f1b826__s024__tb65536__v074", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s025__tb131072__v075\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 10169.657744200045, \"nll_base\": 9.227163835005326, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s025__tb131072__v075\", \"o...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s025__tb131072__v075", "uid": "lambada_standard_cloze_yaml__b4f1b826__s025__tb131072__v075", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s025__tb262144__v076\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 10169.657744200045, \"nll_base\": 9.227163835005326, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s025__tb262144__v076\", \"o...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s025__tb262144__v076", "uid": "lambada_standard_cloze_yaml__b4f1b826__s025__tb262144__v076", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s025__tb65536__v077\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 10169.657744200045, \"nll_base\": 9.227163835005326, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s025__tb65536__v077\", \"orac...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s025__tb65536__v077", "uid": "lambada_standard_cloze_yaml__b4f1b826__s025__tb65536__v077", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s026__tb131072__v078\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 49146.985978789904, \"nll_base\": 10.80257080078125, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s026__tb131072__v078\", \"o...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s026__tb131072__v078", "uid": "lambada_standard_cloze_yaml__b4f1b826__s026__tb131072__v078", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s026__tb262144__v079\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 49146.985978789904, \"nll_base\": 10.80257080078125, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s026__tb262144__v079\", \"o...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s026__tb262144__v079", "uid": "lambada_standard_cloze_yaml__b4f1b826__s026__tb262144__v079", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s026__tb65536__v080\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 49146.985978789904, \"nll_base\": 10.80257080078125, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s026__tb65536__v080\", \"orac...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s026__tb65536__v080", "uid": "lambada_standard_cloze_yaml__b4f1b826__s026__tb65536__v080", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s028__tb131072__v084\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 18931.52283662287, \"nll_base\": 9.848583686642531, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s028__tb131072__v084\", \"or...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s028__tb131072__v084", "uid": "lambada_standard_cloze_yaml__b4f1b826__s028__tb131072__v084", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s028__tb262144__v085\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 18931.52283662287, \"nll_base\": 9.848583686642531, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s028__tb262144__v085\", \"or...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s028__tb262144__v085", "uid": "lambada_standard_cloze_yaml__b4f1b826__s028__tb262144__v085", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s029__tb131072__v087\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 58139.81258564175, \"nll_base\": 10.970605950606497, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s029__tb131072__v087\", \"o...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s029__tb131072__v087", "uid": "lambada_standard_cloze_yaml__b4f1b826__s029__tb131072__v087", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s029__tb262144__v088\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 58139.81258564175, \"nll_base\": 10.970605950606497, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s029__tb262144__v088\", \"o...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s029__tb262144__v088", "uid": "lambada_standard_cloze_yaml__b4f1b826__s029__tb262144__v088", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s029__tb65536__v089\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 58139.81258564175, \"nll_base\": 10.970605950606497, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s029__tb65536__v089\", \"orac...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s029__tb65536__v089", "uid": "lambada_standard_cloze_yaml__b4f1b826__s029__tb65536__v089", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s030__tb131072__v090\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 36280.8685021303, \"nll_base\": 10.499045842810522, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s030__tb131072__v090\", \"or...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s030__tb131072__v090", "uid": "lambada_standard_cloze_yaml__b4f1b826__s030__tb131072__v090", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s030__tb262144__v091\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 36280.8685021303, \"nll_base\": 10.499045842810522, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s030__tb262144__v091\", \"or...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s030__tb262144__v091", "uid": "lambada_standard_cloze_yaml__b4f1b826__s030__tb262144__v091", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s031__tb131072__v093\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 27768.066112163866, \"nll_base\": 10.231641938414755, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s031__tb131072__v093\", \"...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s031__tb131072__v093", "uid": "lambada_standard_cloze_yaml__b4f1b826__s031__tb131072__v093", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s031__tb262144__v094\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 27768.066112163866, \"nll_base\": 10.231641938414755, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s031__tb262144__v094\", \"...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s031__tb262144__v094", "uid": "lambada_standard_cloze_yaml__b4f1b826__s031__tb262144__v094", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s032__tb131072__v096\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 20050.593775170168, \"nll_base\": 9.906014047017912, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s032__tb131072__v096\", \"o...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s032__tb131072__v096", "uid": "lambada_standard_cloze_yaml__b4f1b826__s032__tb131072__v096", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s033__tb131072__v099\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 75124.99795263732, \"nll_base\": 11.22690864458476, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s033__tb131072__v099\", \"or...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s033__tb131072__v099", "uid": "lambada_standard_cloze_yaml__b4f1b826__s033__tb131072__v099", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s033__tb262144__v100\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 75124.99795263732, \"nll_base\": 11.22690864458476, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s033__tb262144__v100\", \"or...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s033__tb262144__v100", "uid": "lambada_standard_cloze_yaml__b4f1b826__s033__tb262144__v100", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s033__tb65536__v101\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 75124.99795263732, \"nll_base\": 11.22690864458476, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s033__tb65536__v101\", \"oracl...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s033__tb65536__v101", "uid": "lambada_standard_cloze_yaml__b4f1b826__s033__tb65536__v101", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s034__tb131072__v102\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 13265.487608450025, \"nll_base\": 9.492921024919992, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s034__tb131072__v102\", \"o...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s034__tb131072__v102", "uid": "lambada_standard_cloze_yaml__b4f1b826__s034__tb131072__v102", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s034__tb262144__v103\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 13265.487608450025, \"nll_base\": 9.492921024919992, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s034__tb262144__v103\", \"o...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s034__tb262144__v103", "uid": "lambada_standard_cloze_yaml__b4f1b826__s034__tb262144__v103", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s034__tb65536__v104\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 13265.487608450025, \"nll_base\": 9.492921024919992, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s034__tb65536__v104\", \"orac...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s034__tb65536__v104", "uid": "lambada_standard_cloze_yaml__b4f1b826__s034__tb65536__v104", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s036__tb131072__v108\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 29041.130513110023, \"nll_base\": 10.276468397695806, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s036__tb131072__v108\", \"...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s036__tb131072__v108", "uid": "lambada_standard_cloze_yaml__b4f1b826__s036__tb131072__v108", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s036__tb262144__v109\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 29041.130513110023, \"nll_base\": 10.276468397695806, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s036__tb262144__v109\", \"...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s036__tb262144__v109", "uid": "lambada_standard_cloze_yaml__b4f1b826__s036__tb262144__v109", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s037__tb131072__v111\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 35182.01271596873, \"nll_base\": 10.468290228592721, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s037__tb131072__v111\", \"o...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s037__tb131072__v111", "uid": "lambada_standard_cloze_yaml__b4f1b826__s037__tb131072__v111", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s037__tb262144__v112\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 35182.01271596873, \"nll_base\": 10.468290228592721, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s037__tb262144__v112\", \"o...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s037__tb262144__v112", "uid": "lambada_standard_cloze_yaml__b4f1b826__s037__tb262144__v112", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s037__tb65536__v113\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 35182.01271596873, \"nll_base\": 10.468290228592721, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s037__tb65536__v113\", \"orac...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s037__tb65536__v113", "uid": "lambada_standard_cloze_yaml__b4f1b826__s037__tb65536__v113", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s038__tb131072__v114\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 22466.04451586719, \"nll_base\": 10.01976031567677, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s038__tb131072__v114\", \"or...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s038__tb131072__v114", "uid": "lambada_standard_cloze_yaml__b4f1b826__s038__tb131072__v114", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s038__tb262144__v115\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 22466.04451586719, \"nll_base\": 10.01976031567677, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s038__tb262144__v115\", \"or...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s038__tb262144__v115", "uid": "lambada_standard_cloze_yaml__b4f1b826__s038__tb262144__v115", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s039__tb131072__v117\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 22490.641503295526, \"nll_base\": 10.020854568481445, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s039__tb131072__v117\", \"...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s039__tb131072__v117", "uid": "lambada_standard_cloze_yaml__b4f1b826__s039__tb131072__v117", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s039__tb262144__v118\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 22490.641503295526, \"nll_base\": 10.020854568481445, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s039__tb262144__v118\", \"...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s039__tb262144__v118", "uid": "lambada_standard_cloze_yaml__b4f1b826__s039__tb262144__v118", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s040__tb131072__v120\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 18373.833908917288, \"nll_base\": 9.818682861328124, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s040__tb131072__v120\", \"o...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s040__tb131072__v120", "uid": "lambada_standard_cloze_yaml__b4f1b826__s040__tb131072__v120", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s040__tb65536__v122\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 18373.833908917288, \"nll_base\": 9.818682861328124, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s040__tb65536__v122\", \"orac...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s040__tb65536__v122", "uid": "lambada_standard_cloze_yaml__b4f1b826__s040__tb65536__v122", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s042__tb131072__v126\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 20598.555686075335, \"nll_base\": 9.932976239993248, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s042__tb131072__v126\", \"o...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s042__tb131072__v126", "uid": "lambada_standard_cloze_yaml__b4f1b826__s042__tb131072__v126", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s042__tb262144__v127\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 20598.555686075335, \"nll_base\": 9.932976239993248, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s042__tb262144__v127\", \"o...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s042__tb262144__v127", "uid": "lambada_standard_cloze_yaml__b4f1b826__s042__tb262144__v127", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s042__tb65536__v128\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 20598.555686075335, \"nll_base\": 9.932976239993248, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s042__tb65536__v128\", \"orac...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s042__tb65536__v128", "uid": "lambada_standard_cloze_yaml__b4f1b826__s042__tb65536__v128", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s043__tb262144__v130\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 19859.021441264016, \"nll_base\": 9.89641366353849, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s043__tb262144__v130\", \"or...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s043__tb262144__v130", "uid": "lambada_standard_cloze_yaml__b4f1b826__s043__tb262144__v130", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s044__tb131072__v132\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 21839.398801025858, \"nll_base\": 9.991470902054399, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s044__tb131072__v132\", \"o...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s044__tb131072__v132", "uid": "lambada_standard_cloze_yaml__b4f1b826__s044__tb131072__v132", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s045__tb131072__v135\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 23926.849404822926, \"nll_base\": 10.082756513430748, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s045__tb131072__v135\", \"...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s045__tb131072__v135", "uid": "lambada_standard_cloze_yaml__b4f1b826__s045__tb131072__v135", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s045__tb262144__v136\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 23926.849404822926, \"nll_base\": 10.082756513430748, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s045__tb262144__v136\", \"...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s045__tb262144__v136", "uid": "lambada_standard_cloze_yaml__b4f1b826__s045__tb262144__v136", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s045__tb65536__v137\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 23926.849404822926, \"nll_base\": 10.082756513430748, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s045__tb65536__v137\", \"ora...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s045__tb65536__v137", "uid": "lambada_standard_cloze_yaml__b4f1b826__s045__tb65536__v137", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s047__tb262144__v142\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 37721.20042887435, \"nll_base\": 10.537977560972555, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s047__tb262144__v142\", \"o...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s047__tb262144__v142", "uid": "lambada_standard_cloze_yaml__b4f1b826__s047__tb262144__v142", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s049__tb131072__v147\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 37650.98608429237, \"nll_base\": 10.536114423702925, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s049__tb131072__v147\", \"o...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s049__tb131072__v147", "uid": "lambada_standard_cloze_yaml__b4f1b826__s049__tb131072__v147", "axis": "domain_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_standard_cloze_yaml__b4f1b826__s050__tb131072__v150\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 26193.041266479606, \"nll_base\": 10.173249053955079, \"eval_ref\": \"lambada_standard_cloze_yaml__b4f1b826__s050__tb131072__v150\", \"...
{ "index": "lambada_standard_cloze_yaml__b4f1b826__s050__tb131072__v150", "uid": "lambada_standard_cloze_yaml__b4f1b826__s050__tb131072__v150", "axis": "domain_ppl", "budget": 2048, "token_budget": 131072, "variant": "filter" }