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_cloze_yaml__5c8f8625__s026__tb131072__v080\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 59528.76447139072, \"nll_base\": 10.994214911209909, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s026__tb131072__v080\", \"oracl...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s026__tb131072__v080", "uid": "lambada_openai_cloze_yaml__5c8f8625__s026__tb131072__v080", "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_openai_cloze_yaml__5c8f8625__s026__tb262144__v078\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 59528.76447139072, \"nll_base\": 10.994214911209909, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s026__tb262144__v078\", \"oracl...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s026__tb262144__v078", "uid": "lambada_openai_cloze_yaml__5c8f8625__s026__tb262144__v078", "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_openai_cloze_yaml__5c8f8625__s027__tb131072__v083\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 58830.513126836864, \"nll_base\": 10.982415930017249, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s027__tb131072__v083\", \"orac...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s027__tb131072__v083", "uid": "lambada_openai_cloze_yaml__5c8f8625__s027__tb131072__v083", "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_openai_cloze_yaml__5c8f8625__s027__tb262144__v081\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 58830.513126836864, \"nll_base\": 10.982415930017249, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s027__tb262144__v081\", \"orac...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s027__tb262144__v081", "uid": "lambada_openai_cloze_yaml__5c8f8625__s027__tb262144__v081", "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_openai_cloze_yaml__5c8f8625__s028__tb131072__v086\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 50530.50226387905, \"nll_base\": 10.830332438151041, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s028__tb131072__v086\", \"oracl...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s028__tb131072__v086", "uid": "lambada_openai_cloze_yaml__5c8f8625__s028__tb131072__v086", "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_openai_cloze_yaml__5c8f8625__s028__tb262144__v084\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 50530.50226387905, \"nll_base\": 10.830332438151041, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s028__tb262144__v084\", \"oracl...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s028__tb262144__v084", "uid": "lambada_openai_cloze_yaml__5c8f8625__s028__tb262144__v084", "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_openai_cloze_yaml__5c8f8625__s029__tb131072__v089\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 116203.10433147743, \"nll_base\": 11.663094838460287, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s029__tb131072__v089\", \"orac...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s029__tb131072__v089", "uid": "lambada_openai_cloze_yaml__5c8f8625__s029__tb131072__v089", "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_openai_cloze_yaml__5c8f8625__s029__tb65536__v088\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 116203.10433147743, \"nll_base\": 11.663094838460287, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s029__tb65536__v088\", \"oracle_...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s029__tb65536__v088", "uid": "lambada_openai_cloze_yaml__5c8f8625__s029__tb65536__v088", "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_openai_cloze_yaml__5c8f8625__s030__tb131072__v092\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 119626.6383644174, \"nll_base\": 11.692130824497768, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s030__tb131072__v092\", \"oracl...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s030__tb131072__v092", "uid": "lambada_openai_cloze_yaml__5c8f8625__s030__tb131072__v092", "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_openai_cloze_yaml__5c8f8625__s030__tb262144__v090\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 119626.6383644174, \"nll_base\": 11.692130824497768, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s030__tb262144__v090\", \"oracl...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s030__tb262144__v090", "uid": "lambada_openai_cloze_yaml__5c8f8625__s030__tb262144__v090", "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_openai_cloze_yaml__5c8f8625__s030__tb65536__v091\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 119626.6383644174, \"nll_base\": 11.692130824497768, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s030__tb65536__v091\", \"oracle_d...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s030__tb65536__v091", "uid": "lambada_openai_cloze_yaml__5c8f8625__s030__tb65536__v091", "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_openai_cloze_yaml__5c8f8625__s031__tb131072__v095\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 52866.48706986356, \"nll_base\": 10.87552490234375, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s031__tb131072__v095\", \"oracle...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s031__tb131072__v095", "uid": "lambada_openai_cloze_yaml__5c8f8625__s031__tb131072__v095", "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_openai_cloze_yaml__5c8f8625__s031__tb262144__v093\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 52866.48706986356, \"nll_base\": 10.87552490234375, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s031__tb262144__v093\", \"oracle...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s031__tb262144__v093", "uid": "lambada_openai_cloze_yaml__5c8f8625__s031__tb262144__v093", "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_openai_cloze_yaml__5c8f8625__s032__tb131072__v098\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 18328.16858119599, \"nll_base\": 9.816194422104779, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s032__tb131072__v098\", \"oracle...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s032__tb131072__v098", "uid": "lambada_openai_cloze_yaml__5c8f8625__s032__tb131072__v098", "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_openai_cloze_yaml__5c8f8625__s032__tb262144__v096\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 18328.16858119599, \"nll_base\": 9.816194422104779, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s032__tb262144__v096\", \"oracle...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s032__tb262144__v096", "uid": "lambada_openai_cloze_yaml__5c8f8625__s032__tb262144__v096", "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_openai_cloze_yaml__5c8f8625__s033__tb131072__v101\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 64722.69878010529, \"nll_base\": 11.077867250184756, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s033__tb131072__v101\", \"oracl...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s033__tb131072__v101", "uid": "lambada_openai_cloze_yaml__5c8f8625__s033__tb131072__v101", "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_openai_cloze_yaml__5c8f8625__s033__tb262144__v099\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 64722.69878010529, \"nll_base\": 11.077867250184756, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s033__tb262144__v099\", \"oracl...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s033__tb262144__v099", "uid": "lambada_openai_cloze_yaml__5c8f8625__s033__tb262144__v099", "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_openai_cloze_yaml__5c8f8625__s033__tb65536__v100\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 64722.69878010529, \"nll_base\": 11.077867250184756, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s033__tb65536__v100\", \"oracle_d...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s033__tb65536__v100", "uid": "lambada_openai_cloze_yaml__5c8f8625__s033__tb65536__v100", "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_openai_cloze_yaml__5c8f8625__s034__tb131072__v104\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 52538.04336277525, \"nll_base\": 10.869292821639624, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s034__tb131072__v104\", \"oracl...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s034__tb131072__v104", "uid": "lambada_openai_cloze_yaml__5c8f8625__s034__tb131072__v104", "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_openai_cloze_yaml__5c8f8625__s034__tb262144__v102\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 52538.04336277525, \"nll_base\": 10.869292821639624, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s034__tb262144__v102\", \"oracl...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s034__tb262144__v102", "uid": "lambada_openai_cloze_yaml__5c8f8625__s034__tb262144__v102", "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_openai_cloze_yaml__5c8f8625__s035__tb131072__v107\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 39757.74513793584, \"nll_base\": 10.590559947339795, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s035__tb131072__v107\", \"oracl...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s035__tb131072__v107", "uid": "lambada_openai_cloze_yaml__5c8f8625__s035__tb131072__v107", "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_openai_cloze_yaml__5c8f8625__s035__tb262144__v105\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 39757.74513793584, \"nll_base\": 10.590559947339795, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s035__tb262144__v105\", \"oracl...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s035__tb262144__v105", "uid": "lambada_openai_cloze_yaml__5c8f8625__s035__tb262144__v105", "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_openai_cloze_yaml__5c8f8625__s036__tb131072__v110\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 43684.042716947486, \"nll_base\": 10.684738159179688, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s036__tb131072__v110\", \"orac...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s036__tb131072__v110", "uid": "lambada_openai_cloze_yaml__5c8f8625__s036__tb131072__v110", "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_openai_cloze_yaml__5c8f8625__s036__tb262144__v108\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 43684.042716947486, \"nll_base\": 10.684738159179688, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s036__tb262144__v108\", \"orac...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s036__tb262144__v108", "uid": "lambada_openai_cloze_yaml__5c8f8625__s036__tb262144__v108", "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_openai_cloze_yaml__5c8f8625__s036__tb65536__v109\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 43684.042716947486, \"nll_base\": 10.684738159179688, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s036__tb65536__v109\", \"oracle_...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s036__tb65536__v109", "uid": "lambada_openai_cloze_yaml__5c8f8625__s036__tb65536__v109", "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_openai_cloze_yaml__5c8f8625__s037__tb262144__v111\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 55393.788182098775, \"nll_base\": 10.922222739771792, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s037__tb262144__v111\", \"orac...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s037__tb262144__v111", "uid": "lambada_openai_cloze_yaml__5c8f8625__s037__tb262144__v111", "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_openai_cloze_yaml__5c8f8625__s038__tb131072__v116\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 24826.83156141958, \"nll_base\": 10.11968026510099, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s038__tb131072__v116\", \"oracle...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s038__tb131072__v116", "uid": "lambada_openai_cloze_yaml__5c8f8625__s038__tb131072__v116", "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_openai_cloze_yaml__5c8f8625__s038__tb262144__v114\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 24826.83156141958, \"nll_base\": 10.11968026510099, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s038__tb262144__v114\", \"oracle...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s038__tb262144__v114", "uid": "lambada_openai_cloze_yaml__5c8f8625__s038__tb262144__v114", "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_openai_cloze_yaml__5c8f8625__s038__tb65536__v115\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 24826.83156141958, \"nll_base\": 10.11968026510099, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s038__tb65536__v115\", \"oracle_de...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s038__tb65536__v115", "uid": "lambada_openai_cloze_yaml__5c8f8625__s038__tb65536__v115", "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_openai_cloze_yaml__5c8f8625__s039__tb262144__v117\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 79517.6640522565, \"nll_base\": 11.283734465298588, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s039__tb262144__v117\", \"oracle...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s039__tb262144__v117", "uid": "lambada_openai_cloze_yaml__5c8f8625__s039__tb262144__v117", "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_openai_cloze_yaml__5c8f8625__s040__tb131072__v122\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 98748.45439568433, \"nll_base\": 11.500331030951607, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s040__tb131072__v122\", \"oracl...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s040__tb131072__v122", "uid": "lambada_openai_cloze_yaml__5c8f8625__s040__tb131072__v122", "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_openai_cloze_yaml__5c8f8625__s040__tb262144__v120\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 98748.45439568433, \"nll_base\": 11.500331030951607, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s040__tb262144__v120\", \"oracl...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s040__tb262144__v120", "uid": "lambada_openai_cloze_yaml__5c8f8625__s040__tb262144__v120", "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_openai_cloze_yaml__5c8f8625__s040__tb65536__v121\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 98748.45439568433, \"nll_base\": 11.500331030951607, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s040__tb65536__v121\", \"oracle_d...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s040__tb65536__v121", "uid": "lambada_openai_cloze_yaml__5c8f8625__s040__tb65536__v121", "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_openai_cloze_yaml__5c8f8625__s041__tb131072__v125\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 110936.73351526906, \"nll_base\": 11.616715349469866, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s041__tb131072__v125\", \"orac...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s041__tb131072__v125", "uid": "lambada_openai_cloze_yaml__5c8f8625__s041__tb131072__v125", "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_openai_cloze_yaml__5c8f8625__s041__tb262144__v123\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 110936.73351526906, \"nll_base\": 11.616715349469866, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s041__tb262144__v123\", \"orac...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s041__tb262144__v123", "uid": "lambada_openai_cloze_yaml__5c8f8625__s041__tb262144__v123", "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_openai_cloze_yaml__5c8f8625__s042__tb131072__v128\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 52462.57408167979, \"nll_base\": 10.867855319729099, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s042__tb131072__v128\", \"oracl...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s042__tb131072__v128", "uid": "lambada_openai_cloze_yaml__5c8f8625__s042__tb131072__v128", "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_openai_cloze_yaml__5c8f8625__s042__tb262144__v126\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 52462.57408167979, \"nll_base\": 10.867855319729099, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s042__tb262144__v126\", \"oracl...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s042__tb262144__v126", "uid": "lambada_openai_cloze_yaml__5c8f8625__s042__tb262144__v126", "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_openai_cloze_yaml__5c8f8625__s043__tb131072__v131\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 108979.18708214484, \"nll_base\": 11.59891219877861, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s043__tb131072__v131\", \"oracl...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s043__tb131072__v131", "uid": "lambada_openai_cloze_yaml__5c8f8625__s043__tb131072__v131", "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_openai_cloze_yaml__5c8f8625__s043__tb262144__v129\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 108979.18708214484, \"nll_base\": 11.59891219877861, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s043__tb262144__v129\", \"oracl...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s043__tb262144__v129", "uid": "lambada_openai_cloze_yaml__5c8f8625__s043__tb262144__v129", "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_openai_cloze_yaml__5c8f8625__s044__tb131072__v134\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 55265.38480209331, \"nll_base\": 10.919902038574218, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s044__tb131072__v134\", \"oracl...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s044__tb131072__v134", "uid": "lambada_openai_cloze_yaml__5c8f8625__s044__tb131072__v134", "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_openai_cloze_yaml__5c8f8625__s044__tb262144__v132\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 55265.38480209331, \"nll_base\": 10.919902038574218, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s044__tb262144__v132\", \"oracl...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s044__tb262144__v132", "uid": "lambada_openai_cloze_yaml__5c8f8625__s044__tb262144__v132", "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_openai_cloze_yaml__5c8f8625__s044__tb65536__v133\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 55265.38480209331, \"nll_base\": 10.919902038574218, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s044__tb65536__v133\", \"oracle_d...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s044__tb65536__v133", "uid": "lambada_openai_cloze_yaml__5c8f8625__s044__tb65536__v133", "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_openai_cloze_yaml__5c8f8625__s045__tb131072__v137\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 29909.526353636153, \"nll_base\": 10.305932315779321, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s045__tb131072__v137\", \"orac...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s045__tb131072__v137", "uid": "lambada_openai_cloze_yaml__5c8f8625__s045__tb131072__v137", "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_openai_cloze_yaml__5c8f8625__s045__tb262144__v135\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 29909.526353636153, \"nll_base\": 10.305932315779321, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s045__tb262144__v135\", \"orac...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s045__tb262144__v135", "uid": "lambada_openai_cloze_yaml__5c8f8625__s045__tb262144__v135", "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_openai_cloze_yaml__5c8f8625__s045__tb65536__v136\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 29909.526353636153, \"nll_base\": 10.305932315779321, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s045__tb65536__v136\", \"oracle_...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s045__tb65536__v136", "uid": "lambada_openai_cloze_yaml__5c8f8625__s045__tb65536__v136", "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_openai_cloze_yaml__5c8f8625__s046__tb131072__v140\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 61645.61063620562, \"nll_base\": 11.02915730794271, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s046__tb131072__v140\", \"oracle...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s046__tb131072__v140", "uid": "lambada_openai_cloze_yaml__5c8f8625__s046__tb131072__v140", "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_openai_cloze_yaml__5c8f8625__s046__tb262144__v138\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 61645.61063620562, \"nll_base\": 11.02915730794271, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s046__tb262144__v138\", \"oracle...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s046__tb262144__v138", "uid": "lambada_openai_cloze_yaml__5c8f8625__s046__tb262144__v138", "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_openai_cloze_yaml__5c8f8625__s047__tb131072__v143\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 18951.58541179945, \"nll_base\": 9.84964286990282, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s047__tb131072__v143\", \"oracle_...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s047__tb131072__v143", "uid": "lambada_openai_cloze_yaml__5c8f8625__s047__tb131072__v143", "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_openai_cloze_yaml__5c8f8625__s047__tb262144__v141\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 18951.58541179945, \"nll_base\": 9.84964286990282, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s047__tb262144__v141\", \"oracle_...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s047__tb262144__v141", "uid": "lambada_openai_cloze_yaml__5c8f8625__s047__tb262144__v141", "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_openai_cloze_yaml__5c8f8625__s047__tb65536__v142\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 18951.58541179945, \"nll_base\": 9.84964286990282, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s047__tb65536__v142\", \"oracle_del...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s047__tb65536__v142", "uid": "lambada_openai_cloze_yaml__5c8f8625__s047__tb65536__v142", "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_openai_cloze_yaml__5c8f8625__s048__tb262144__v144\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 88755.06614292087, \"nll_base\": 11.39363578900899, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s048__tb262144__v144\", \"oracle...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s048__tb262144__v144", "uid": "lambada_openai_cloze_yaml__5c8f8625__s048__tb262144__v144", "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_openai_cloze_yaml__5c8f8625__s049__tb262144__v147\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 70749.43281099276, \"nll_base\": 11.166899798667593, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s049__tb262144__v147\", \"oracl...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s049__tb262144__v147", "uid": "lambada_openai_cloze_yaml__5c8f8625__s049__tb262144__v147", "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_openai_cloze_yaml__5c8f8625__s050__tb131072__v152\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 86104.20147792723, \"nll_base\": 11.363313486878301, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s050__tb131072__v152\", \"oracl...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s050__tb131072__v152", "uid": "lambada_openai_cloze_yaml__5c8f8625__s050__tb131072__v152", "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_openai_cloze_yaml__5c8f8625__s050__tb262144__v150\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 86104.20147792723, \"nll_base\": 11.363313486878301, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s050__tb262144__v150\", \"oracl...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s050__tb262144__v150", "uid": "lambada_openai_cloze_yaml__5c8f8625__s050__tb262144__v150", "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_openai_cloze_yaml__5c8f8625__s051__tb262144__v153\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 65105.631832273684, \"nll_base\": 11.08376633493524, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s051__tb262144__v153\", \"oracl...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s051__tb262144__v153", "uid": "lambada_openai_cloze_yaml__5c8f8625__s051__tb262144__v153", "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_openai_cloze_yaml__5c8f8625__s052__tb131072__v158\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 65616.2043684199, \"nll_base\": 11.091577962239583, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s052__tb131072__v158\", \"oracle...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s052__tb131072__v158", "uid": "lambada_openai_cloze_yaml__5c8f8625__s052__tb131072__v158", "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_openai_cloze_yaml__5c8f8625__s052__tb262144__v156\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 65616.2043684199, \"nll_base\": 11.091577962239583, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s052__tb262144__v156\", \"oracle...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s052__tb262144__v156", "uid": "lambada_openai_cloze_yaml__5c8f8625__s052__tb262144__v156", "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_openai_cloze_yaml__5c8f8625__s052__tb65536__v157\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 65616.2043684199, \"nll_base\": 11.091577962239583, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s052__tb65536__v157\", \"oracle_de...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s052__tb65536__v157", "uid": "lambada_openai_cloze_yaml__5c8f8625__s052__tb65536__v157", "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_openai_mt_en__46aac8a7__s003__tb131072__v011\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 14169.215455654501, \"nll_base\": 9.558826964578511, \"eval_ref\": \"lambada_openai_mt_en__46aac8a7__s003__tb131072__v011\", \"oracle_delta\":...
{ "index": "lambada_openai_mt_en__46aac8a7__s003__tb131072__v011", "uid": "lambada_openai_mt_en__46aac8a7__s003__tb131072__v011", "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_en__46aac8a7__s003__tb262144__v009\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 14169.215455654501, \"nll_base\": 9.558826964578511, \"eval_ref\": \"lambada_openai_mt_en__46aac8a7__s003__tb262144__v009\", \"oracle_delta\":...
{ "index": "lambada_openai_mt_en__46aac8a7__s003__tb262144__v009", "uid": "lambada_openai_mt_en__46aac8a7__s003__tb262144__v009", "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_en__46aac8a7__s005__tb131072__v017\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 152210.77933796326, \"nll_base\": 11.933021545410156, \"eval_ref\": \"lambada_openai_mt_en__46aac8a7__s005__tb131072__v017\", \"oracle_delta\"...
{ "index": "lambada_openai_mt_en__46aac8a7__s005__tb131072__v017", "uid": "lambada_openai_mt_en__46aac8a7__s005__tb131072__v017", "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_en__46aac8a7__s006__tb131072__v020\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 5564.926351681511, \"nll_base\": 8.624239029422883, \"eval_ref\": \"lambada_openai_mt_en__46aac8a7__s006__tb131072__v020\", \"oracle_delta\": ...
{ "index": "lambada_openai_mt_en__46aac8a7__s006__tb131072__v020", "uid": "lambada_openai_mt_en__46aac8a7__s006__tb131072__v020", "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_en__46aac8a7__s006__tb262144__v018\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 5564.926351681511, \"nll_base\": 8.624239029422883, \"eval_ref\": \"lambada_openai_mt_en__46aac8a7__s006__tb262144__v018\", \"oracle_delta\": ...
{ "index": "lambada_openai_mt_en__46aac8a7__s006__tb262144__v018", "uid": "lambada_openai_mt_en__46aac8a7__s006__tb262144__v018", "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_en__46aac8a7__s008__tb131072__v026\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 92319.47340856996, \"nll_base\": 11.433010377745697, \"eval_ref\": \"lambada_openai_mt_en__46aac8a7__s008__tb131072__v026\", \"oracle_delta\":...
{ "index": "lambada_openai_mt_en__46aac8a7__s008__tb131072__v026", "uid": "lambada_openai_mt_en__46aac8a7__s008__tb131072__v026", "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_en__46aac8a7__s008__tb262144__v024\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 92319.47340856996, \"nll_base\": 11.433010377745697, \"eval_ref\": \"lambada_openai_mt_en__46aac8a7__s008__tb262144__v024\", \"oracle_delta\":...
{ "index": "lambada_openai_mt_en__46aac8a7__s008__tb262144__v024", "uid": "lambada_openai_mt_en__46aac8a7__s008__tb262144__v024", "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_en__46aac8a7__s010__tb262144__v030\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 20593.70700300481, \"nll_base\": 9.932740822816507, \"eval_ref\": \"lambada_openai_mt_en__46aac8a7__s010__tb262144__v030\", \"oracle_delta\": ...
{ "index": "lambada_openai_mt_en__46aac8a7__s010__tb262144__v030", "uid": "lambada_openai_mt_en__46aac8a7__s010__tb262144__v030", "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_en__46aac8a7__s012__tb131072__v038\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 21768.1847504038, \"nll_base\": 9.988204767674576, \"eval_ref\": \"lambada_openai_mt_en__46aac8a7__s012__tb131072__v038\", \"oracle_delta\": 0...
{ "index": "lambada_openai_mt_en__46aac8a7__s012__tb131072__v038", "uid": "lambada_openai_mt_en__46aac8a7__s012__tb131072__v038", "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_en__46aac8a7__s012__tb262144__v036\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 21768.1847504038, \"nll_base\": 9.988204767674576, \"eval_ref\": \"lambada_openai_mt_en__46aac8a7__s012__tb262144__v036\", \"oracle_delta\": 1...
{ "index": "lambada_openai_mt_en__46aac8a7__s012__tb262144__v036", "uid": "lambada_openai_mt_en__46aac8a7__s012__tb262144__v036", "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_en__46aac8a7__s012__tb65536__v037\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 21768.1847504038, \"nll_base\": 9.988204767674576, \"eval_ref\": \"lambada_openai_mt_en__46aac8a7__s012__tb65536__v037\", \"oracle_delta\": 0.18...
{ "index": "lambada_openai_mt_en__46aac8a7__s012__tb65536__v037", "uid": "lambada_openai_mt_en__46aac8a7__s012__tb65536__v037", "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_en__46aac8a7__s015__tb262144__v045\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 11262.458882555575, \"nll_base\": 9.329230251082455, \"eval_ref\": \"lambada_openai_mt_en__46aac8a7__s015__tb262144__v045\", \"oracle_delta\":...
{ "index": "lambada_openai_mt_en__46aac8a7__s015__tb262144__v045", "uid": "lambada_openai_mt_en__46aac8a7__s015__tb262144__v045", "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_en__46aac8a7__s016__tb131072__v050\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 113295.58320712604, \"nll_base\": 11.637755463088768, \"eval_ref\": \"lambada_openai_mt_en__46aac8a7__s016__tb131072__v050\", \"oracle_delta\"...
{ "index": "lambada_openai_mt_en__46aac8a7__s016__tb131072__v050", "uid": "lambada_openai_mt_en__46aac8a7__s016__tb131072__v050", "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_en__46aac8a7__s016__tb262144__v048\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 113295.58320712604, \"nll_base\": 11.637755463088768, \"eval_ref\": \"lambada_openai_mt_en__46aac8a7__s016__tb262144__v048\", \"oracle_delta\"...
{ "index": "lambada_openai_mt_en__46aac8a7__s016__tb262144__v048", "uid": "lambada_openai_mt_en__46aac8a7__s016__tb262144__v048", "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_en__46aac8a7__s016__tb65536__v049\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 113295.58320712604, \"nll_base\": 11.637755463088768, \"eval_ref\": \"lambada_openai_mt_en__46aac8a7__s016__tb65536__v049\", \"oracle_delta\": 0...
{ "index": "lambada_openai_mt_en__46aac8a7__s016__tb65536__v049", "uid": "lambada_openai_mt_en__46aac8a7__s016__tb65536__v049", "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_en__46aac8a7__s017__tb131072__v053\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 18949.435311844907, \"nll_base\": 9.849529411219343, \"eval_ref\": \"lambada_openai_mt_en__46aac8a7__s017__tb131072__v053\", \"oracle_delta\":...
{ "index": "lambada_openai_mt_en__46aac8a7__s017__tb131072__v053", "uid": "lambada_openai_mt_en__46aac8a7__s017__tb131072__v053", "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_en__46aac8a7__s018__tb131072__v056\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 36366.64180215645, \"nll_base\": 10.501407199435764, \"eval_ref\": \"lambada_openai_mt_en__46aac8a7__s018__tb131072__v056\", \"oracle_delta\":...
{ "index": "lambada_openai_mt_en__46aac8a7__s018__tb131072__v056", "uid": "lambada_openai_mt_en__46aac8a7__s018__tb131072__v056", "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_en__46aac8a7__s018__tb262144__v054\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 36366.64180215645, \"nll_base\": 10.501407199435764, \"eval_ref\": \"lambada_openai_mt_en__46aac8a7__s018__tb262144__v054\", \"oracle_delta\":...
{ "index": "lambada_openai_mt_en__46aac8a7__s018__tb262144__v054", "uid": "lambada_openai_mt_en__46aac8a7__s018__tb262144__v054", "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_en__46aac8a7__s018__tb65536__v055\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 36366.64180215645, \"nll_base\": 10.501407199435764, \"eval_ref\": \"lambada_openai_mt_en__46aac8a7__s018__tb65536__v055\", \"oracle_delta\": 0....
{ "index": "lambada_openai_mt_en__46aac8a7__s018__tb65536__v055", "uid": "lambada_openai_mt_en__46aac8a7__s018__tb65536__v055", "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__s000__tb131072__v000\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 787.6443987366684, \"nll_base\": 6.669046717365895, \"eval_ref\": \"lambada_openai_mt_es__d838e1dd__s000__tb131072__v000\", \"oracle_delta\": ...
{ "index": "lambada_openai_mt_es__d838e1dd__s000__tb131072__v000", "uid": "lambada_openai_mt_es__d838e1dd__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_es__d838e1dd__s000__tb262144__v001\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 787.6443987366684, \"nll_base\": 6.669046717365895, \"eval_ref\": \"lambada_openai_mt_es__d838e1dd__s000__tb262144__v001\", \"oracle_delta\": ...
{ "index": "lambada_openai_mt_es__d838e1dd__s000__tb262144__v001", "uid": "lambada_openai_mt_es__d838e1dd__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_es__d838e1dd__s000__tb65536__v002\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 787.6443987366684, \"nll_base\": 6.669046717365895, \"eval_ref\": \"lambada_openai_mt_es__d838e1dd__s000__tb65536__v002\", \"oracle_delta\": 0.2...
{ "index": "lambada_openai_mt_es__d838e1dd__s000__tb65536__v002", "uid": "lambada_openai_mt_es__d838e1dd__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_es__d838e1dd__s001__tb131072__v003\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 1943.0897569242925, \"nll_base\": 7.5720346433307055, \"eval_ref\": \"lambada_openai_mt_es__d838e1dd__s001__tb131072__v003\", \"oracle_delta\"...
{ "index": "lambada_openai_mt_es__d838e1dd__s001__tb131072__v003", "uid": "lambada_openai_mt_es__d838e1dd__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_es__d838e1dd__s001__tb262144__v004\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 1943.0897569242925, \"nll_base\": 7.5720346433307055, \"eval_ref\": \"lambada_openai_mt_es__d838e1dd__s001__tb262144__v004\", \"oracle_delta\"...
{ "index": "lambada_openai_mt_es__d838e1dd__s001__tb262144__v004", "uid": "lambada_openai_mt_es__d838e1dd__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_es__d838e1dd__s001__tb65536__v005\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 1943.0897569242925, \"nll_base\": 7.5720346433307055, \"eval_ref\": \"lambada_openai_mt_es__d838e1dd__s001__tb65536__v005\", \"oracle_delta\": 0...
{ "index": "lambada_openai_mt_es__d838e1dd__s001__tb65536__v005", "uid": "lambada_openai_mt_es__d838e1dd__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_es__d838e1dd__s002__tb131072__v006\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 329.2502719329219, \"nll_base\": 5.796818166539289, \"eval_ref\": \"lambada_openai_mt_es__d838e1dd__s002__tb131072__v006\", \"oracle_delta\": ...
{ "index": "lambada_openai_mt_es__d838e1dd__s002__tb131072__v006", "uid": "lambada_openai_mt_es__d838e1dd__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_es__d838e1dd__s002__tb262144__v007\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 329.2502719329219, \"nll_base\": 5.796818166539289, \"eval_ref\": \"lambada_openai_mt_es__d838e1dd__s002__tb262144__v007\", \"oracle_delta\": ...
{ "index": "lambada_openai_mt_es__d838e1dd__s002__tb262144__v007", "uid": "lambada_openai_mt_es__d838e1dd__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_es__d838e1dd__s002__tb65536__v008\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 329.2502719329219, \"nll_base\": 5.796818166539289, \"eval_ref\": \"lambada_openai_mt_es__d838e1dd__s002__tb65536__v008\", \"oracle_delta\": 0.2...
{ "index": "lambada_openai_mt_es__d838e1dd__s002__tb65536__v008", "uid": "lambada_openai_mt_es__d838e1dd__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_es__d838e1dd__s003__tb131072__v009\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 1169.2248020897189, \"nll_base\": 7.06409624588391, \"eval_ref\": \"lambada_openai_mt_es__d838e1dd__s003__tb131072__v009\", \"oracle_delta\": ...
{ "index": "lambada_openai_mt_es__d838e1dd__s003__tb131072__v009", "uid": "lambada_openai_mt_es__d838e1dd__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_es__d838e1dd__s003__tb262144__v010\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 1169.2248020897189, \"nll_base\": 7.06409624588391, \"eval_ref\": \"lambada_openai_mt_es__d838e1dd__s003__tb262144__v010\", \"oracle_delta\": ...
{ "index": "lambada_openai_mt_es__d838e1dd__s003__tb262144__v010", "uid": "lambada_openai_mt_es__d838e1dd__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_es__d838e1dd__s003__tb65536__v011\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 1169.2248020897189, \"nll_base\": 7.06409624588391, \"eval_ref\": \"lambada_openai_mt_es__d838e1dd__s003__tb65536__v011\", \"oracle_delta\": 0.3...
{ "index": "lambada_openai_mt_es__d838e1dd__s003__tb65536__v011", "uid": "lambada_openai_mt_es__d838e1dd__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_es__d838e1dd__s004__tb131072__v012\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 935.2994355955112, \"nll_base\": 6.840866729992778, \"eval_ref\": \"lambada_openai_mt_es__d838e1dd__s004__tb131072__v012\", \"oracle_delta\": ...
{ "index": "lambada_openai_mt_es__d838e1dd__s004__tb131072__v012", "uid": "lambada_openai_mt_es__d838e1dd__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_es__d838e1dd__s004__tb262144__v013\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 935.2994355955112, \"nll_base\": 6.840866729992778, \"eval_ref\": \"lambada_openai_mt_es__d838e1dd__s004__tb262144__v013\", \"oracle_delta\": ...
{ "index": "lambada_openai_mt_es__d838e1dd__s004__tb262144__v013", "uid": "lambada_openai_mt_es__d838e1dd__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_es__d838e1dd__s004__tb65536__v014\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 935.2994355955112, \"nll_base\": 6.840866729992778, \"eval_ref\": \"lambada_openai_mt_es__d838e1dd__s004__tb65536__v014\", \"oracle_delta\": 0.3...
{ "index": "lambada_openai_mt_es__d838e1dd__s004__tb65536__v014", "uid": "lambada_openai_mt_es__d838e1dd__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_es__d838e1dd__s005__tb131072__v015\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 2111.050655087007, \"nll_base\": 7.65494104333826, \"eval_ref\": \"lambada_openai_mt_es__d838e1dd__s005__tb131072__v015\", \"oracle_delta\": 0...
{ "index": "lambada_openai_mt_es__d838e1dd__s005__tb131072__v015", "uid": "lambada_openai_mt_es__d838e1dd__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_es__d838e1dd__s005__tb262144__v016\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 2111.050655087007, \"nll_base\": 7.65494104333826, \"eval_ref\": \"lambada_openai_mt_es__d838e1dd__s005__tb262144__v016\", \"oracle_delta\": 1...
{ "index": "lambada_openai_mt_es__d838e1dd__s005__tb262144__v016", "uid": "lambada_openai_mt_es__d838e1dd__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_es__d838e1dd__s005__tb65536__v017\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 2111.050655087007, \"nll_base\": 7.65494104333826, \"eval_ref\": \"lambada_openai_mt_es__d838e1dd__s005__tb65536__v017\", \"oracle_delta\": 0.40...
{ "index": "lambada_openai_mt_es__d838e1dd__s005__tb65536__v017", "uid": "lambada_openai_mt_es__d838e1dd__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_es__d838e1dd__s006__tb131072__v018\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 883.951759662736, \"nll_base\": 6.784402490631352, \"eval_ref\": \"lambada_openai_mt_es__d838e1dd__s006__tb131072__v018\", \"oracle_delta\": 0...
{ "index": "lambada_openai_mt_es__d838e1dd__s006__tb131072__v018", "uid": "lambada_openai_mt_es__d838e1dd__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_es__d838e1dd__s006__tb262144__v019\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 883.951759662736, \"nll_base\": 6.784402490631352, \"eval_ref\": \"lambada_openai_mt_es__d838e1dd__s006__tb262144__v019\", \"oracle_delta\": 1...
{ "index": "lambada_openai_mt_es__d838e1dd__s006__tb262144__v019", "uid": "lambada_openai_mt_es__d838e1dd__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_es__d838e1dd__s006__tb65536__v020\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 883.951759662736, \"nll_base\": 6.784402490631352, \"eval_ref\": \"lambada_openai_mt_es__d838e1dd__s006__tb65536__v020\", \"oracle_delta\": 0.34...
{ "index": "lambada_openai_mt_es__d838e1dd__s006__tb65536__v020", "uid": "lambada_openai_mt_es__d838e1dd__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_es__d838e1dd__s007__tb131072__v021\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 1003.2391122246833, \"nll_base\": 6.910989156583461, \"eval_ref\": \"lambada_openai_mt_es__d838e1dd__s007__tb131072__v021\", \"oracle_delta\":...
{ "index": "lambada_openai_mt_es__d838e1dd__s007__tb131072__v021", "uid": "lambada_openai_mt_es__d838e1dd__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_es__d838e1dd__s007__tb262144__v022\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 1003.2391122246833, \"nll_base\": 6.910989156583461, \"eval_ref\": \"lambada_openai_mt_es__d838e1dd__s007__tb262144__v022\", \"oracle_delta\":...
{ "index": "lambada_openai_mt_es__d838e1dd__s007__tb262144__v022", "uid": "lambada_openai_mt_es__d838e1dd__s007__tb262144__v022", "axis": "multilingual_ppl", "budget": 2048, "token_budget": 262144, "variant": "filter" }