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__4ad5aa20__s013__tb262144__v041\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 24919.386365166323, \"nll_base\": 10.123401348407452, \"eval_ref\": \"lambada_openai__4ad5aa20__s013__tb262144__v041\", \"oracle_delta\": 1.24455456...
{ "index": "lambada_openai__4ad5aa20__s013__tb262144__v041", "uid": "lambada_openai__4ad5aa20__s013__tb262144__v041", "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__4ad5aa20__s013__tb65536__v039\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 24919.386365166323, \"nll_base\": 10.123401348407452, \"eval_ref\": \"lambada_openai__4ad5aa20__s013__tb65536__v039\", \"oracle_delta\": 0.10977524977...
{ "index": "lambada_openai__4ad5aa20__s013__tb65536__v039", "uid": "lambada_openai__4ad5aa20__s013__tb65536__v039", "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__4ad5aa20__s015__tb131072__v046\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 29625.054924340042, \"nll_base\": 10.296375732421875, \"eval_ref\": \"lambada_openai__4ad5aa20__s015__tb131072__v046\", \"oracle_delta\": 0.71505900...
{ "index": "lambada_openai__4ad5aa20__s015__tb131072__v046", "uid": "lambada_openai__4ad5aa20__s015__tb131072__v046", "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__4ad5aa20__s015__tb262144__v047\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 29625.054924340042, \"nll_base\": 10.296375732421875, \"eval_ref\": \"lambada_openai__4ad5aa20__s015__tb262144__v047\", \"oracle_delta\": 1.24797159...
{ "index": "lambada_openai__4ad5aa20__s015__tb262144__v047", "uid": "lambada_openai__4ad5aa20__s015__tb262144__v047", "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__4ad5aa20__s015__tb65536__v045\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 29625.054924340042, \"nll_base\": 10.296375732421875, \"eval_ref\": \"lambada_openai__4ad5aa20__s015__tb65536__v045\", \"oracle_delta\": 0.22117431640...
{ "index": "lambada_openai__4ad5aa20__s015__tb65536__v045", "uid": "lambada_openai__4ad5aa20__s015__tb65536__v045", "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__4ad5aa20__s016__tb131072__v049\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 31867.100588841757, \"nll_base\": 10.369329427083333, \"eval_ref\": \"lambada_openai__4ad5aa20__s016__tb131072__v049\", \"oracle_delta\": 0.40889892...
{ "index": "lambada_openai__4ad5aa20__s016__tb131072__v049", "uid": "lambada_openai__4ad5aa20__s016__tb131072__v049", "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__4ad5aa20__s016__tb65536__v048\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 31867.100588841757, \"nll_base\": 10.369329427083333, \"eval_ref\": \"lambada_openai__4ad5aa20__s016__tb65536__v048\", \"oracle_delta\": 0.09197957356...
{ "index": "lambada_openai__4ad5aa20__s016__tb65536__v048", "uid": "lambada_openai__4ad5aa20__s016__tb65536__v048", "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__4ad5aa20__s017__tb131072__v052\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 24272.31969790063, \"nll_base\": 10.097091872970779, \"eval_ref\": \"lambada_openai__4ad5aa20__s017__tb131072__v052\", \"oracle_delta\": 0.472088305...
{ "index": "lambada_openai__4ad5aa20__s017__tb131072__v052", "uid": "lambada_openai__4ad5aa20__s017__tb131072__v052", "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__4ad5aa20__s017__tb262144__v053\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 24272.31969790063, \"nll_base\": 10.097091872970779, \"eval_ref\": \"lambada_openai__4ad5aa20__s017__tb262144__v053\", \"oracle_delta\": 1.214340011...
{ "index": "lambada_openai__4ad5aa20__s017__tb262144__v053", "uid": "lambada_openai__4ad5aa20__s017__tb262144__v053", "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__4ad5aa20__s018__tb131072__v055\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 40278.741460186284, \"nll_base\": 10.6035791015625, \"eval_ref\": \"lambada_openai__4ad5aa20__s018__tb131072__v055\", \"oracle_delta\": 0.4485201009...
{ "index": "lambada_openai__4ad5aa20__s018__tb131072__v055", "uid": "lambada_openai__4ad5aa20__s018__tb131072__v055", "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__4ad5aa20__s018__tb65536__v054\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 40278.741460186284, \"nll_base\": 10.6035791015625, \"eval_ref\": \"lambada_openai__4ad5aa20__s018__tb65536__v054\", \"oracle_delta\": 0.1875850423177...
{ "index": "lambada_openai__4ad5aa20__s018__tb65536__v054", "uid": "lambada_openai__4ad5aa20__s018__tb65536__v054", "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__4ad5aa20__s021__tb131072__v064\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 20134.371078692024, \"nll_base\": 9.910183637570112, \"eval_ref\": \"lambada_openai__4ad5aa20__s021__tb131072__v064\", \"oracle_delta\": 0.523285694...
{ "index": "lambada_openai__4ad5aa20__s021__tb131072__v064", "uid": "lambada_openai__4ad5aa20__s021__tb131072__v064", "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__4ad5aa20__s021__tb262144__v065\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 20134.371078692024, \"nll_base\": 9.910183637570112, \"eval_ref\": \"lambada_openai__4ad5aa20__s021__tb262144__v065\", \"oracle_delta\": 1.285441863...
{ "index": "lambada_openai__4ad5aa20__s021__tb262144__v065", "uid": "lambada_openai__4ad5aa20__s021__tb262144__v065", "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__4ad5aa20__s021__tb65536__v063\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 20134.371078692024, \"nll_base\": 9.910183637570112, \"eval_ref\": \"lambada_openai__4ad5aa20__s021__tb65536__v063\", \"oracle_delta\": 0.128223125751...
{ "index": "lambada_openai__4ad5aa20__s021__tb65536__v063", "uid": "lambada_openai__4ad5aa20__s021__tb65536__v063", "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__4ad5aa20__s023__tb131072__v070\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 38663.99298084149, \"nll_base\": 10.562664031982422, \"eval_ref\": \"lambada_openai__4ad5aa20__s023__tb131072__v070\", \"oracle_delta\": 0.634741465...
{ "index": "lambada_openai__4ad5aa20__s023__tb131072__v070", "uid": "lambada_openai__4ad5aa20__s023__tb131072__v070", "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__4ad5aa20__s023__tb262144__v071\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 38663.99298084149, \"nll_base\": 10.562664031982422, \"eval_ref\": \"lambada_openai__4ad5aa20__s023__tb262144__v071\", \"oracle_delta\": 1.231556786...
{ "index": "lambada_openai__4ad5aa20__s023__tb262144__v071", "uid": "lambada_openai__4ad5aa20__s023__tb262144__v071", "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__4ad5aa20__s023__tb65536__v069\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 38663.99298084149, \"nll_base\": 10.562664031982422, \"eval_ref\": \"lambada_openai__4ad5aa20__s023__tb65536__v069\", \"oracle_delta\": 0.270300547281...
{ "index": "lambada_openai__4ad5aa20__s023__tb65536__v069", "uid": "lambada_openai__4ad5aa20__s023__tb65536__v069", "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__4ad5aa20__s024__tb131072__v073\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 55778.1884295743, \"nll_base\": 10.92913818359375, \"eval_ref\": \"lambada_openai__4ad5aa20__s024__tb131072__v073\", \"oracle_delta\": 0.42238507952...
{ "index": "lambada_openai__4ad5aa20__s024__tb131072__v073", "uid": "lambada_openai__4ad5aa20__s024__tb131072__v073", "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__4ad5aa20__s024__tb262144__v074\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 55778.1884295743, \"nll_base\": 10.92913818359375, \"eval_ref\": \"lambada_openai__4ad5aa20__s024__tb262144__v074\", \"oracle_delta\": 1.20676443917...
{ "index": "lambada_openai__4ad5aa20__s024__tb262144__v074", "uid": "lambada_openai__4ad5aa20__s024__tb262144__v074", "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__4ad5aa20__s024__tb65536__v072\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 55778.1884295743, \"nll_base\": 10.92913818359375, \"eval_ref\": \"lambada_openai__4ad5aa20__s024__tb65536__v072\", \"oracle_delta\": 0.13751743861607...
{ "index": "lambada_openai__4ad5aa20__s024__tb65536__v072", "uid": "lambada_openai__4ad5aa20__s024__tb65536__v072", "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__4ad5aa20__s025__tb262144__v077\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 31848.6614723555, \"nll_base\": 10.368750634131494, \"eval_ref\": \"lambada_openai__4ad5aa20__s025__tb262144__v077\", \"oracle_delta\": 0.6839635279...
{ "index": "lambada_openai__4ad5aa20__s025__tb262144__v077", "uid": "lambada_openai__4ad5aa20__s025__tb262144__v077", "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__4ad5aa20__s026__tb131072__v079\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 73108.01766141073, \"nll_base\": 11.199693320454031, \"eval_ref\": \"lambada_openai__4ad5aa20__s026__tb131072__v079\", \"oracle_delta\": 0.376297218...
{ "index": "lambada_openai__4ad5aa20__s026__tb131072__v079", "uid": "lambada_openai__4ad5aa20__s026__tb131072__v079", "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__4ad5aa20__s026__tb262144__v080\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 73108.01766141073, \"nll_base\": 11.199693320454031, \"eval_ref\": \"lambada_openai__4ad5aa20__s026__tb262144__v080\", \"oracle_delta\": 0.905146059...
{ "index": "lambada_openai__4ad5aa20__s026__tb262144__v080", "uid": "lambada_openai__4ad5aa20__s026__tb262144__v080", "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__4ad5aa20__s029__tb262144__v089\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 24630.236018428204, \"nll_base\": 10.111730073627673, \"eval_ref\": \"lambada_openai__4ad5aa20__s029__tb262144__v089\", \"oracle_delta\": 0.85880480...
{ "index": "lambada_openai__4ad5aa20__s029__tb262144__v089", "uid": "lambada_openai__4ad5aa20__s029__tb262144__v089", "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__4ad5aa20__s032__tb262144__v098\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 23839.758592838534, \"nll_base\": 10.079109994988693, \"eval_ref\": \"lambada_openai__4ad5aa20__s032__tb262144__v098\", \"oracle_delta\": 0.92507372...
{ "index": "lambada_openai__4ad5aa20__s032__tb262144__v098", "uid": "lambada_openai__4ad5aa20__s032__tb262144__v098", "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__4ad5aa20__s033__tb131072__v100\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 16816.9228115358, \"nll_base\": 9.730140968605324, \"eval_ref\": \"lambada_openai__4ad5aa20__s033__tb131072__v100\", \"oracle_delta\": 0.26117583851...
{ "index": "lambada_openai__4ad5aa20__s033__tb131072__v100", "uid": "lambada_openai__4ad5aa20__s033__tb131072__v100", "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__4ad5aa20__s033__tb262144__v101\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 16816.9228115358, \"nll_base\": 9.730140968605324, \"eval_ref\": \"lambada_openai__4ad5aa20__s033__tb262144__v101\", \"oracle_delta\": 0.95232325424...
{ "index": "lambada_openai__4ad5aa20__s033__tb262144__v101", "uid": "lambada_openai__4ad5aa20__s033__tb262144__v101", "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__4ad5aa20__s036__tb131072__v109\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 32910.69973528195, \"nll_base\": 10.401553103798314, \"eval_ref\": \"lambada_openai__4ad5aa20__s036__tb131072__v109\", \"oracle_delta\": 0.486279537...
{ "index": "lambada_openai__4ad5aa20__s036__tb131072__v109", "uid": "lambada_openai__4ad5aa20__s036__tb131072__v109", "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__4ad5aa20__s036__tb262144__v110\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 32910.69973528195, \"nll_base\": 10.401553103798314, \"eval_ref\": \"lambada_openai__4ad5aa20__s036__tb262144__v110\", \"oracle_delta\": 1.174049377...
{ "index": "lambada_openai__4ad5aa20__s036__tb262144__v110", "uid": "lambada_openai__4ad5aa20__s036__tb262144__v110", "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__4ad5aa20__s036__tb65536__v108\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 32910.69973528195, \"nll_base\": 10.401553103798314, \"eval_ref\": \"lambada_openai__4ad5aa20__s036__tb65536__v108\", \"oracle_delta\": 0.125430860017...
{ "index": "lambada_openai__4ad5aa20__s036__tb65536__v108", "uid": "lambada_openai__4ad5aa20__s036__tb65536__v108", "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__4ad5aa20__s037__tb131072__v112\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 32944.27022652055, \"nll_base\": 10.402572631835938, \"eval_ref\": \"lambada_openai__4ad5aa20__s037__tb131072__v112\", \"oracle_delta\": 0.374371980...
{ "index": "lambada_openai__4ad5aa20__s037__tb131072__v112", "uid": "lambada_openai__4ad5aa20__s037__tb131072__v112", "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__4ad5aa20__s037__tb262144__v113\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 32944.27022652055, \"nll_base\": 10.402572631835938, \"eval_ref\": \"lambada_openai__4ad5aa20__s037__tb262144__v113\", \"oracle_delta\": 0.906999688...
{ "index": "lambada_openai__4ad5aa20__s037__tb262144__v113", "uid": "lambada_openai__4ad5aa20__s037__tb262144__v113", "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__4ad5aa20__s039__tb262144__v119\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 16442.22424187048, \"nll_base\": 9.707607953976362, \"eval_ref\": \"lambada_openai__4ad5aa20__s039__tb262144__v119\", \"oracle_delta\": 0.7499323135...
{ "index": "lambada_openai__4ad5aa20__s039__tb262144__v119", "uid": "lambada_openai__4ad5aa20__s039__tb262144__v119", "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__4ad5aa20__s041__tb262144__v125\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 11701.895003037938, \"nll_base\": 9.367506073742378, \"eval_ref\": \"lambada_openai__4ad5aa20__s041__tb262144__v125\", \"oracle_delta\": 1.033407816...
{ "index": "lambada_openai__4ad5aa20__s041__tb262144__v125", "uid": "lambada_openai__4ad5aa20__s041__tb262144__v125", "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__4ad5aa20__s041__tb65536__v123\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 11701.895003037938, \"nll_base\": 9.367506073742378, \"eval_ref\": \"lambada_openai__4ad5aa20__s041__tb65536__v123\", \"oracle_delta\": 0.088689292349...
{ "index": "lambada_openai__4ad5aa20__s041__tb65536__v123", "uid": "lambada_openai__4ad5aa20__s041__tb65536__v123", "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__4ad5aa20__s042__tb131072__v127\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 28431.21892853973, \"nll_base\": 10.255243078454749, \"eval_ref\": \"lambada_openai__4ad5aa20__s042__tb131072__v127\", \"oracle_delta\": 0.300961581...
{ "index": "lambada_openai__4ad5aa20__s042__tb131072__v127", "uid": "lambada_openai__4ad5aa20__s042__tb131072__v127", "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__4ad5aa20__s042__tb262144__v128\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 28431.21892853973, \"nll_base\": 10.255243078454749, \"eval_ref\": \"lambada_openai__4ad5aa20__s042__tb262144__v128\", \"oracle_delta\": 1.067734755...
{ "index": "lambada_openai__4ad5aa20__s042__tb262144__v128", "uid": "lambada_openai__4ad5aa20__s042__tb262144__v128", "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__4ad5aa20__s043__tb131072__v130\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 15176.085919111143, \"nll_base\": 9.627476173111155, \"eval_ref\": \"lambada_openai__4ad5aa20__s043__tb131072__v130\", \"oracle_delta\": 0.529699011...
{ "index": "lambada_openai__4ad5aa20__s043__tb131072__v130", "uid": "lambada_openai__4ad5aa20__s043__tb131072__v130", "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__4ad5aa20__s043__tb262144__v131\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 15176.085919111143, \"nll_base\": 9.627476173111155, \"eval_ref\": \"lambada_openai__4ad5aa20__s043__tb262144__v131\", \"oracle_delta\": 1.139960180...
{ "index": "lambada_openai__4ad5aa20__s043__tb262144__v131", "uid": "lambada_openai__4ad5aa20__s043__tb262144__v131", "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__4ad5aa20__s043__tb65536__v129\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 15176.085919111143, \"nll_base\": 9.627476173111155, \"eval_ref\": \"lambada_openai__4ad5aa20__s043__tb65536__v129\", \"oracle_delta\": 0.205449744115...
{ "index": "lambada_openai__4ad5aa20__s043__tb65536__v129", "uid": "lambada_openai__4ad5aa20__s043__tb65536__v129", "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__4ad5aa20__s045__tb131072__v136\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 17946.28129279883, \"nll_base\": 9.795138202136076, \"eval_ref\": \"lambada_openai__4ad5aa20__s045__tb131072__v136\", \"oracle_delta\": 0.5596572296...
{ "index": "lambada_openai__4ad5aa20__s045__tb131072__v136", "uid": "lambada_openai__4ad5aa20__s045__tb131072__v136", "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__4ad5aa20__s045__tb262144__v137\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 17946.28129279883, \"nll_base\": 9.795138202136076, \"eval_ref\": \"lambada_openai__4ad5aa20__s045__tb262144__v137\", \"oracle_delta\": 1.0114776997...
{ "index": "lambada_openai__4ad5aa20__s045__tb262144__v137", "uid": "lambada_openai__4ad5aa20__s045__tb262144__v137", "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__4ad5aa20__s045__tb65536__v135\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 17946.28129279883, \"nll_base\": 9.795138202136076, \"eval_ref\": \"lambada_openai__4ad5aa20__s045__tb65536__v135\", \"oracle_delta\": 0.1737354133702...
{ "index": "lambada_openai__4ad5aa20__s045__tb65536__v135", "uid": "lambada_openai__4ad5aa20__s045__tb65536__v135", "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__4ad5aa20__s046__tb131072__v139\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 20831.86849714314, \"nll_base\": 9.94423923244724, \"eval_ref\": \"lambada_openai__4ad5aa20__s046__tb131072__v139\", \"oracle_delta\": 0.65155584161...
{ "index": "lambada_openai__4ad5aa20__s046__tb131072__v139", "uid": "lambada_openai__4ad5aa20__s046__tb131072__v139", "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__4ad5aa20__s046__tb262144__v140\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 20831.86849714314, \"nll_base\": 9.94423923244724, \"eval_ref\": \"lambada_openai__4ad5aa20__s046__tb262144__v140\", \"oracle_delta\": 1.38625583400...
{ "index": "lambada_openai__4ad5aa20__s046__tb262144__v140", "uid": "lambada_openai__4ad5aa20__s046__tb262144__v140", "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__4ad5aa20__s046__tb65536__v138\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 20831.86849714314, \"nll_base\": 9.94423923244724, \"eval_ref\": \"lambada_openai__4ad5aa20__s046__tb65536__v138\", \"oracle_delta\": 0.26898312259029...
{ "index": "lambada_openai__4ad5aa20__s046__tb65536__v138", "uid": "lambada_openai__4ad5aa20__s046__tb65536__v138", "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__4ad5aa20__s048__tb131072__v145\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 42553.07778181394, \"nll_base\": 10.658507464683218, \"eval_ref\": \"lambada_openai__4ad5aa20__s048__tb131072__v145\", \"oracle_delta\": 0.335958402...
{ "index": "lambada_openai__4ad5aa20__s048__tb131072__v145", "uid": "lambada_openai__4ad5aa20__s048__tb131072__v145", "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__4ad5aa20__s049__tb65536__v147\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 8178.517012893296, \"nll_base\": 9.009266118893677, \"eval_ref\": \"lambada_openai__4ad5aa20__s049__tb65536__v147\", \"oracle_delta\": 0.0678798631690...
{ "index": "lambada_openai__4ad5aa20__s049__tb65536__v147", "uid": "lambada_openai__4ad5aa20__s049__tb65536__v147", "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__4ad5aa20__s050__tb262144__v152\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 14837.549027456704, \"nll_base\": 9.604916343206092, \"eval_ref\": \"lambada_openai__4ad5aa20__s050__tb262144__v152\", \"oracle_delta\": 0.828902618...
{ "index": "lambada_openai__4ad5aa20__s050__tb262144__v152", "uid": "lambada_openai__4ad5aa20__s050__tb262144__v152", "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__s000__tb131072__v002\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 59123.31160634516, \"nll_base\": 10.987380569045609, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s000__tb131072__v002\", \"oracl...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s000__tb131072__v002", "uid": "lambada_openai_cloze_yaml__5c8f8625__s000__tb131072__v002", "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__s000__tb262144__v000\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 59123.31160634516, \"nll_base\": 10.987380569045609, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s000__tb262144__v000\", \"oracl...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s000__tb262144__v000", "uid": "lambada_openai_cloze_yaml__5c8f8625__s000__tb262144__v000", "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__s000__tb65536__v001\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 59123.31160634516, \"nll_base\": 10.987380569045609, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s000__tb65536__v001\", \"oracle_d...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s000__tb65536__v001", "uid": "lambada_openai_cloze_yaml__5c8f8625__s000__tb65536__v001", "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__s001__tb131072__v005\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 159854.03989028215, \"nll_base\": 11.982016427176339, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s001__tb131072__v005\", \"orac...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s001__tb131072__v005", "uid": "lambada_openai_cloze_yaml__5c8f8625__s001__tb131072__v005", "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__s001__tb262144__v003\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 159854.03989028215, \"nll_base\": 11.982016427176339, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s001__tb262144__v003\", \"orac...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s001__tb262144__v003", "uid": "lambada_openai_cloze_yaml__5c8f8625__s001__tb262144__v003", "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__s001__tb65536__v004\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 159854.03989028215, \"nll_base\": 11.982016427176339, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s001__tb65536__v004\", \"oracle_...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s001__tb65536__v004", "uid": "lambada_openai_cloze_yaml__5c8f8625__s001__tb65536__v004", "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__s002__tb262144__v006\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 23191.49370935482, \"nll_base\": 10.051540839962843, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s002__tb262144__v006\", \"oracl...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s002__tb262144__v006", "uid": "lambada_openai_cloze_yaml__5c8f8625__s002__tb262144__v006", "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__s003__tb131072__v011\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 42205.08125273637, \"nll_base\": 10.650295901608157, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s003__tb131072__v011\", \"oracl...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s003__tb131072__v011", "uid": "lambada_openai_cloze_yaml__5c8f8625__s003__tb131072__v011", "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__s003__tb262144__v009\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 42205.08125273637, \"nll_base\": 10.650295901608157, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s003__tb262144__v009\", \"oracl...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s003__tb262144__v009", "uid": "lambada_openai_cloze_yaml__5c8f8625__s003__tb262144__v009", "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__s003__tb65536__v010\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 42205.08125273637, \"nll_base\": 10.650295901608157, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s003__tb65536__v010\", \"oracle_d...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s003__tb65536__v010", "uid": "lambada_openai_cloze_yaml__5c8f8625__s003__tb65536__v010", "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__s004__tb131072__v014\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 45131.990282855484, \"nll_base\": 10.71734659295333, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s004__tb131072__v014\", \"oracl...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s004__tb131072__v014", "uid": "lambada_openai_cloze_yaml__5c8f8625__s004__tb131072__v014", "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__s004__tb262144__v012\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 45131.990282855484, \"nll_base\": 10.71734659295333, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s004__tb262144__v012\", \"oracl...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s004__tb262144__v012", "uid": "lambada_openai_cloze_yaml__5c8f8625__s004__tb262144__v012", "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__s005__tb131072__v017\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 37990.13510710284, \"nll_base\": 10.545081802561313, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s005__tb131072__v017\", \"oracl...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s005__tb131072__v017", "uid": "lambada_openai_cloze_yaml__5c8f8625__s005__tb131072__v017", "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__s005__tb262144__v015\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 37990.13510710284, \"nll_base\": 10.545081802561313, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s005__tb262144__v015\", \"oracl...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s005__tb262144__v015", "uid": "lambada_openai_cloze_yaml__5c8f8625__s005__tb262144__v015", "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__s005__tb65536__v016\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 37990.13510710284, \"nll_base\": 10.545081802561313, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s005__tb65536__v016\", \"oracle_d...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s005__tb65536__v016", "uid": "lambada_openai_cloze_yaml__5c8f8625__s005__tb65536__v016", "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__s006__tb262144__v018\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 56084.69998775678, \"nll_base\": 10.934618326822916, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s006__tb262144__v018\", \"oracl...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s006__tb262144__v018", "uid": "lambada_openai_cloze_yaml__5c8f8625__s006__tb262144__v018", "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__s007__tb131072__v023\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 70129.98998114532, \"nll_base\": 11.158105798669764, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s007__tb131072__v023\", \"oracl...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s007__tb131072__v023", "uid": "lambada_openai_cloze_yaml__5c8f8625__s007__tb131072__v023", "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__s007__tb262144__v021\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 70129.98998114532, \"nll_base\": 11.158105798669764, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s007__tb262144__v021\", \"oracl...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s007__tb262144__v021", "uid": "lambada_openai_cloze_yaml__5c8f8625__s007__tb262144__v021", "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__s007__tb65536__v022\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 70129.98998114532, \"nll_base\": 11.158105798669764, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s007__tb65536__v022\", \"oracle_d...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s007__tb65536__v022", "uid": "lambada_openai_cloze_yaml__5c8f8625__s007__tb65536__v022", "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__s008__tb262144__v024\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 134419.0703442125, \"nll_base\": 11.808717589447463, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s008__tb262144__v024\", \"oracl...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s008__tb262144__v024", "uid": "lambada_openai_cloze_yaml__5c8f8625__s008__tb262144__v024", "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__s009__tb131072__v029\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 22292.102724232427, \"nll_base\": 10.011987756799769, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s009__tb131072__v029\", \"orac...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s009__tb131072__v029", "uid": "lambada_openai_cloze_yaml__5c8f8625__s009__tb131072__v029", "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__s009__tb262144__v027\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 22292.102724232427, \"nll_base\": 10.011987756799769, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s009__tb262144__v027\", \"orac...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s009__tb262144__v027", "uid": "lambada_openai_cloze_yaml__5c8f8625__s009__tb262144__v027", "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__s010__tb131072__v032\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 19531.86326772202, \"nll_base\": 9.879802424733231, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s010__tb131072__v032\", \"oracle...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s010__tb131072__v032", "uid": "lambada_openai_cloze_yaml__5c8f8625__s010__tb131072__v032", "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__s010__tb262144__v030\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 19531.86326772202, \"nll_base\": 9.879802424733231, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s010__tb262144__v030\", \"oracle...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s010__tb262144__v030", "uid": "lambada_openai_cloze_yaml__5c8f8625__s010__tb262144__v030", "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__s011__tb131072__v035\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 89894.83860681513, \"nll_base\": 11.406395806206596, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s011__tb131072__v035\", \"oracl...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s011__tb131072__v035", "uid": "lambada_openai_cloze_yaml__5c8f8625__s011__tb131072__v035", "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__s011__tb262144__v033\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 89894.83860681513, \"nll_base\": 11.406395806206596, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s011__tb262144__v033\", \"oracl...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s011__tb262144__v033", "uid": "lambada_openai_cloze_yaml__5c8f8625__s011__tb262144__v033", "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__s012__tb131072__v038\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 50930.34897703452, \"nll_base\": 10.838214271946958, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s012__tb131072__v038\", \"oracl...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s012__tb131072__v038", "uid": "lambada_openai_cloze_yaml__5c8f8625__s012__tb131072__v038", "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__s012__tb262144__v036\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 50930.34897703452, \"nll_base\": 10.838214271946958, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s012__tb262144__v036\", \"oracl...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s012__tb262144__v036", "uid": "lambada_openai_cloze_yaml__5c8f8625__s012__tb262144__v036", "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__s013__tb131072__v041\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 51137.17081526241, \"nll_base\": 10.842266924969561, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s013__tb131072__v041\", \"oracl...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s013__tb131072__v041", "uid": "lambada_openai_cloze_yaml__5c8f8625__s013__tb131072__v041", "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__s013__tb262144__v039\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 51137.17081526241, \"nll_base\": 10.842266924969561, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s013__tb262144__v039\", \"oracl...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s013__tb262144__v039", "uid": "lambada_openai_cloze_yaml__5c8f8625__s013__tb262144__v039", "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__s014__tb131072__v044\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 57749.17907749639, \"nll_base\": 10.963864413174717, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s014__tb131072__v044\", \"oracl...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s014__tb131072__v044", "uid": "lambada_openai_cloze_yaml__5c8f8625__s014__tb131072__v044", "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__s014__tb262144__v042\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 57749.17907749639, \"nll_base\": 10.963864413174717, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s014__tb262144__v042\", \"oracl...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s014__tb262144__v042", "uid": "lambada_openai_cloze_yaml__5c8f8625__s014__tb262144__v042", "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__s015__tb131072__v047\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 54410.23594887714, \"nll_base\": 10.904307575968954, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s015__tb131072__v047\", \"oracl...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s015__tb131072__v047", "uid": "lambada_openai_cloze_yaml__5c8f8625__s015__tb131072__v047", "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__s015__tb262144__v045\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 54410.23594887714, \"nll_base\": 10.904307575968954, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s015__tb262144__v045\", \"oracl...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s015__tb262144__v045", "uid": "lambada_openai_cloze_yaml__5c8f8625__s015__tb262144__v045", "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__s016__tb131072__v050\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 23822.44024597811, \"nll_base\": 10.078383282917303, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s016__tb131072__v050\", \"oracl...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s016__tb131072__v050", "uid": "lambada_openai_cloze_yaml__5c8f8625__s016__tb131072__v050", "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__s016__tb262144__v048\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 23822.44024597811, \"nll_base\": 10.078383282917303, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s016__tb262144__v048\", \"oracl...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s016__tb262144__v048", "uid": "lambada_openai_cloze_yaml__5c8f8625__s016__tb262144__v048", "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__s017__tb262144__v051\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 56773.209001022034, \"nll_base\": 10.946819820919552, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s017__tb262144__v051\", \"orac...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s017__tb262144__v051", "uid": "lambada_openai_cloze_yaml__5c8f8625__s017__tb262144__v051", "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__s018__tb131072__v056\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 54290.512217376534, \"nll_base\": 10.902104761693385, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s018__tb131072__v056\", \"orac...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s018__tb131072__v056", "uid": "lambada_openai_cloze_yaml__5c8f8625__s018__tb131072__v056", "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__s018__tb262144__v054\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 54290.512217376534, \"nll_base\": 10.902104761693385, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s018__tb262144__v054\", \"orac...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s018__tb262144__v054", "uid": "lambada_openai_cloze_yaml__5c8f8625__s018__tb262144__v054", "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__s019__tb131072__v059\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 65025.71973617189, \"nll_base\": 11.082538158862622, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s019__tb131072__v059\", \"oracl...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s019__tb131072__v059", "uid": "lambada_openai_cloze_yaml__5c8f8625__s019__tb131072__v059", "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__s019__tb262144__v057\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 65025.71973617189, \"nll_base\": 11.082538158862622, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s019__tb262144__v057\", \"oracl...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s019__tb262144__v057", "uid": "lambada_openai_cloze_yaml__5c8f8625__s019__tb262144__v057", "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__s019__tb65536__v058\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 65025.71973617189, \"nll_base\": 11.082538158862622, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s019__tb65536__v058\", \"oracle_d...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s019__tb65536__v058", "uid": "lambada_openai_cloze_yaml__5c8f8625__s019__tb65536__v058", "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__s020__tb131072__v062\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 61423.34137307973, \"nll_base\": 11.025545194551542, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s020__tb131072__v062\", \"oracl...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s020__tb131072__v062", "uid": "lambada_openai_cloze_yaml__5c8f8625__s020__tb131072__v062", "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__s020__tb262144__v060\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 61423.34137307973, \"nll_base\": 11.025545194551542, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s020__tb262144__v060\", \"oracl...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s020__tb262144__v060", "uid": "lambada_openai_cloze_yaml__5c8f8625__s020__tb262144__v060", "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__s022__tb131072__v068\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 111010.67265365661, \"nll_base\": 11.617381625705296, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s022__tb131072__v068\", \"orac...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s022__tb131072__v068", "uid": "lambada_openai_cloze_yaml__5c8f8625__s022__tb131072__v068", "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__s022__tb262144__v066\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 111010.67265365661, \"nll_base\": 11.617381625705296, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s022__tb262144__v066\", \"orac...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s022__tb262144__v066", "uid": "lambada_openai_cloze_yaml__5c8f8625__s022__tb262144__v066", "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__s023__tb262144__v069\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 90873.43593773682, \"nll_base\": 11.417223003548635, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s023__tb262144__v069\", \"oracl...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s023__tb262144__v069", "uid": "lambada_openai_cloze_yaml__5c8f8625__s023__tb262144__v069", "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__s024__tb262144__v072\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 23490.74867047969, \"nll_base\": 10.064361949025848, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s024__tb262144__v072\", \"oracl...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s024__tb262144__v072", "uid": "lambada_openai_cloze_yaml__5c8f8625__s024__tb262144__v072", "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__s025__tb131072__v077\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 72523.3004709659, \"nll_base\": 11.191663175015837, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s025__tb131072__v077\", \"oracle...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s025__tb131072__v077", "uid": "lambada_openai_cloze_yaml__5c8f8625__s025__tb131072__v077", "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__s025__tb262144__v075\", \"budget\": 2048, \"token_budget\": 262144, \"ppl_base\": 72523.3004709659, \"nll_base\": 11.191663175015837, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s025__tb262144__v075\", \"oracle...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s025__tb262144__v075", "uid": "lambada_openai_cloze_yaml__5c8f8625__s025__tb262144__v075", "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__s025__tb65536__v076\", \"budget\": 2048, \"token_budget\": 65536, \"ppl_base\": 72523.3004709659, \"nll_base\": 11.191663175015837, \"eval_ref\": \"lambada_openai_cloze_yaml__5c8f8625__s025__tb65536__v076\", \"oracle_de...
{ "index": "lambada_openai_cloze_yaml__5c8f8625__s025__tb65536__v076", "uid": "lambada_openai_cloze_yaml__5c8f8625__s025__tb65536__v076", "axis": "domain_ppl", "budget": 2048, "token_budget": 65536, "variant": "filter" }