Dataset Viewer
Auto-converted to Parquet Duplicate
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 construct the best possible fine-tuning set for it. The model will be LoRA-fine-tuned on exactly what you submit, then its **perplexity on held-out target da...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"lang_de\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 18.808332693182507, \"nll_base\": 2.9343, \"eval_ref\": \"lang_de\", \"oracle_delta\": 0.0931, \"random_delta\": 0.031}" }
{ "index": "lang_de", "uid": "lang_de", "axis": "lang", "budget": 2048, "token_budget": 131072, "variant": "curate" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must construct the best possible fine-tuning set for it. The model will be LoRA-fine-tuned on exactly what you submit, then its **perplexity on held-out target da...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"lang_es\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 20.805138511102886, \"nll_base\": 3.0352, \"eval_ref\": \"lang_es\", \"oracle_delta\": 0.1592, \"random_delta\": 0.0278}" }
{ "index": "lang_es", "uid": "lang_es", "axis": "lang", "budget": 2048, "token_budget": 131072, "variant": "curate" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must construct the best possible fine-tuning set for it. The model will be LoRA-fine-tuned on exactly what you submit, then its **perplexity on held-out target da...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"lang_fr\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 18.957506984077874, \"nll_base\": 2.9422, \"eval_ref\": \"lang_fr\", \"oracle_delta\": 0.1056, \"random_delta\": 0.0323}" }
{ "index": "lang_fr", "uid": "lang_fr", "axis": "lang", "budget": 2048, "token_budget": 131072, "variant": "curate" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must construct the best possible fine-tuning set for it. The model will be LoRA-fine-tuned on exactly what you submit, then its **perplexity on held-out target da...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"lang_hi\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 4.692192700684559, \"nll_base\": 1.5459, \"eval_ref\": \"lang_hi\", \"oracle_delta\": 0.0533, \"random_delta\": 0.0069}" }
{ "index": "lang_hi", "uid": "lang_hi", "axis": "lang", "budget": 2048, "token_budget": 131072, "variant": "curate" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must construct the best possible fine-tuning set for it. The model will be LoRA-fine-tuned on exactly what you submit, then its **perplexity on held-out target da...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"lang_ja\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 6.237004381219701, \"nll_base\": 1.8305, \"eval_ref\": \"lang_ja\", \"oracle_delta\": 0.0488, \"random_delta\": 0.0069}" }
{ "index": "lang_ja", "uid": "lang_ja", "axis": "lang", "budget": 2048, "token_budget": 131072, "variant": "curate" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must construct the best possible fine-tuning set for it. The model will be LoRA-fine-tuned on exactly what you submit, then its **perplexity on held-out target da...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"lang_ko\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 3.7344479383893328, \"nll_base\": 1.3176, \"eval_ref\": \"lang_ko\", \"oracle_delta\": 0.0277, \"random_delta\": 0.0056}" }
{ "index": "lang_ko", "uid": "lang_ko", "axis": "lang", "budget": 2048, "token_budget": 131072, "variant": "curate" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must construct the best possible fine-tuning set for it. The model will be LoRA-fine-tuned on exactly what you submit, then its **perplexity on held-out target da...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"lang_ru\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 5.497536044769496, \"nll_base\": 1.7043, \"eval_ref\": \"lang_ru\", \"oracle_delta\": 0.0443, \"random_delta\": 0.0038}" }
{ "index": "lang_ru", "uid": "lang_ru", "axis": "lang", "budget": 2048, "token_budget": 131072, "variant": "curate" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must construct the best possible fine-tuning set for it. The model will be LoRA-fine-tuned on exactly what you submit, then its **perplexity on held-out target da...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"lang_tr\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 20.583710282662036, \"nll_base\": 3.0245, \"eval_ref\": \"lang_tr\", \"oracle_delta\": 0.1178, \"random_delta\": 0.0309}" }
{ "index": "lang_tr", "uid": "lang_tr", "axis": "lang", "budget": 2048, "token_budget": 131072, "variant": "curate" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must construct the best possible fine-tuning set for it. The model will be LoRA-fine-tuned on exactly what you submit, then its **perplexity on held-out target da...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"lang_vi\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 7.50372827229365, \"nll_base\": 2.0154, \"eval_ref\": \"lang_vi\", \"oracle_delta\": 0.0803, \"random_delta\": 0.012}" }
{ "index": "lang_vi", "uid": "lang_vi", "axis": "lang", "budget": 2048, "token_budget": 131072, "variant": "curate" }
datacuration
[ { "role": "system", "content": "You are a **training-data curator**. A small base language model needs to do better on a **target task**, and you must construct the best possible fine-tuning set for it. The model will be LoRA-fine-tuned on exactly what you submit, then its **perplexity on held-out target da...
datacuration_selection
{ "style": "datacuration", "ground_truth": "{\"task_id\": \"mmlu_college_mathematics\", \"budget\": 2048, \"token_budget\": 131072, \"ppl_base\": 11.552478468906273, \"nll_base\": 2.4469, \"eval_ref\": \"mmlu_college_mathematics\", \"oracle_delta\": 0.4414, \"random_delta\": 0.3176}" }
{ "index": "mmlu_college_mathematics", "uid": "mmlu_college_mathematics", "axis": "mmlu", "budget": 2048, "token_budget": 131072, "variant": "curate" }
README.md exists but content is empty.
Downloads last month
32