Distill What You Trust: Reliability-Aware Multi-Teacher On-Policy Distillation
Paper • 2609.23697 • Published • 5
Datasets:
data_source stringclasses 4
values | prompt listlengths 2 2 | ability stringclasses 3
values | reward_model dict | domain stringclasses 3
values | extra_info dict |
|---|---|---|---|---|---|
nemotron_if_rl | [
{
"content": "You are an expert general-purpose assistant. Follow every user instruction exactly, especially formatting, wording, ordering, and length constraints. You must always begin your response with exactly <think>\n</think> (a single newline inside the tags), and then immediately provide your answer.",
... | instruction_following | {
"ground_truth": "",
"style": "rule"
} | if | {
"agent_name": "instruction_following_simple_agent",
"dataset": "nano-v3-RL-instruction_following",
"domain": "if",
"family": "length_constraints",
"index": "if-42804",
"instruction_id_list": [
"length_constraints:number_sentences",
"length_constraints:number_words",
"keywords:no_adjacent_conse... |
nemotron_if_rl | [
{
"content": "You are an expert general-purpose assistant. Follow every user instruction exactly, especially formatting, wording, ordering, and length constraints. You must always begin your response with exactly <think>\n</think> (a single newline inside the tags), and then immediately provide your answer.",
... | instruction_following | {
"ground_truth": "",
"style": "rule"
} | if | {
"agent_name": "instruction_following_simple_agent",
"dataset": "nano-v3-RL-instruction_following",
"domain": "if",
"family": "keywords",
"index": "if-3622",
"instruction_id_list": [
"length_constraints:nth_paragraph_first_word",
"count:count_increment_word"
],
"instruction_kwargs_json": [
... |
primeintellect | [
{
"content": "You are an expert programmer and competitive problem solver. Solve the task carefully. You must first outline your thought process, logic, and planning inside <think> and </think> tags, and then produce a correct final implementation outside of these tags.",
"role": "system"
},
{
"cont... | code | {
"ground_truth": "[{\"type\": \"stdin_stdout\", \"input\": \"4\\n3 2\\n3 1\\n8 6\\n8 7 6 5 3 2\\n9 6\\n9 8 5 4 3 1\\n1 1\\n1\\n\", \"output\": \"0\\n1\\n2\\n0\\n\"}, {\"type\": \"stdin_stdout\", \"input\": \"1\\n3 2\\n3 2\\n\", \"output\": \"1\\n\"}, {\"type\": \"stdin_stdout\", \"input\": \"28\\n8 4\\n8 7 6 3\\n8 5... | code | {
"agent_name": null,
"dataset": null,
"domain": "code",
"family": null,
"index": "code-13060",
"instruction_id_list": null,
"instruction_kwargs_json": null,
"mix_index": 13060,
"raw_prompt": null,
"sample_id": null,
"source_config": "primeintellect",
"source_index": "13060",
"split": "train"
... |
math_dapo_boxed | [
{
"content": "You are an expert mathematical problem solver. Solve the problem carefully. You must first show all your step-by-step reasoning and derivations inside <think> and </think> tags. After your thought process, provide the final answer clearly and make it easy to verify.",
"role": "system"
},
{... | MATH | {
"ground_truth": "16",
"style": "rule"
} | math | {
"agent_name": null,
"dataset": null,
"domain": "math",
"family": null,
"index": "math-8286",
"instruction_id_list": null,
"instruction_kwargs_json": null,
"mix_index": 8286,
"raw_prompt": null,
"sample_id": null,
"source_config": null,
"source_index": "78194331-a627-43f2-8df7-816eda1e2290",
... |
nemotron_if_rl | [
{
"content": "You are an expert general-purpose assistant. Follow every user instruction exactly, especially formatting, wording, ordering, and length constraints. You must always begin your response with exactly <think>\n</think> (a single newline inside the tags), and then immediately provide your answer.",
... | instruction_following | {
"ground_truth": "",
"style": "rule"
} | if | {
"agent_name": "instruction_following_simple_agent",
"dataset": "nano-v3-RL-instruction_following",
"domain": "if",
"family": "other",
"index": "if-17610",
"instruction_id_list": [
"punctuation:no_comma",
"detectable_format:number_bullet_lists",
"detectable_format:sentence_hyphens"
],
"inst... |
primeintellect | [
{
"content": "You are an expert programmer and competitive problem solver. Solve the task carefully. You must first outline your thought process, logic, and planning inside <think> and </think> tags, and then produce a correct final implementation outside of these tags.",
"role": "system"
},
{
"cont... | code | {
"ground_truth": "[{\"type\": \"function_call\", \"fn_name\": \"pillow\", \"input\": [[\"EvH/KNikBiyxfeyK/miCMj\", \"I/HwjnHlFLlahMOKNadps\"]], \"output\": [false]}, {\"type\": \"function_call\", \"fn_name\": \"pillow\", \"input\": [[\"\\\\DjQ\\\\[zv]SpG]Z/[Qm\\\\eLL\", \"amwZArsaGRmibriXBgTRZp\"]], \"output\": [fal... | code | {
"agent_name": null,
"dataset": null,
"domain": "code",
"family": null,
"index": "code-23491",
"instruction_id_list": null,
"instruction_kwargs_json": null,
"mix_index": 23491,
"raw_prompt": null,
"sample_id": null,
"source_config": "primeintellect",
"source_index": "23491",
"split": "train"
... |
math_dapo_boxed | [
{
"content": "You are an expert mathematical problem solver. Solve the problem carefully. You must first show all your step-by-step reasoning and derivations inside <think> and </think> tags. After your thought process, provide the final answer clearly and make it easy to verify.",
"role": "system"
},
{... | MATH | {
"ground_truth": "719",
"style": "rule"
} | math | {
"agent_name": null,
"dataset": null,
"domain": "math",
"family": null,
"index": "math-6366",
"instruction_id_list": null,
"instruction_kwargs_json": null,
"mix_index": 6366,
"raw_prompt": null,
"sample_id": null,
"source_config": null,
"source_index": "578d78cc-147e-4951-9cdf-81c1be76d80e",
... |
nemotron_if_rl | [
{
"content": "You are an expert general-purpose assistant. Follow every user instruction exactly, especially formatting, wording, ordering, and length constraints. You must always begin your response with exactly <think>\n</think> (a single newline inside the tags), and then immediately provide your answer.",
... | instruction_following | {
"ground_truth": "",
"style": "rule"
} | if | {
"agent_name": "instruction_following_simple_agent",
"dataset": "nano-v3-RL-instruction_following",
"domain": "if",
"family": "other",
"index": "if-23466",
"instruction_id_list": [
"keywords:palindrome"
],
"instruction_kwargs_json": [
"null"
],
"mix_index": 23466,
"raw_prompt": "I helped ... |
primeintellect | [
{
"content": "You are an expert programmer and competitive problem solver. Solve the task carefully. You must first outline your thought process, logic, and planning inside <think> and </think> tags, and then produce a correct final implementation outside of these tags.",
"role": "system"
},
{
"cont... | code | {
"ground_truth": "[{\"type\": \"stdin_stdout\", \"input\": \"4\\nhoge\\nenglish\\nhoge\\nenigma\", \"output\": \"No\"}, {\"type\": \"stdin_stdout\", \"input\": \"9\\nbasic\\nc\\ncpp\\nphp\\npython\\nnadesico\\nocaml\\nlua\\nassembly\", \"output\": \"Yes\"}, {\"type\": \"stdin_stdout\", \"input\": \"3\\nabc\\narc\\na... | code | {
"agent_name": null,
"dataset": null,
"domain": "code",
"family": null,
"index": "code-16437",
"instruction_id_list": null,
"instruction_kwargs_json": null,
"mix_index": 16437,
"raw_prompt": null,
"sample_id": null,
"source_config": "primeintellect",
"source_index": "16437",
"split": "train"
... |
math_dapo_boxed | [
{
"content": "You are an expert mathematical problem solver. Solve the problem carefully. You must first show all your step-by-step reasoning and derivations inside <think> and </think> tags. After your thought process, provide the final answer clearly and make it easy to verify.",
"role": "system"
},
{... | MATH | {
"ground_truth": "34",
"style": "rule"
} | math | {
"agent_name": null,
"dataset": null,
"domain": "math",
"family": null,
"index": "math-17267",
"instruction_id_list": null,
"instruction_kwargs_json": null,
"mix_index": 17267,
"raw_prompt": null,
"sample_id": null,
"source_config": null,
"source_index": "2fbcde3a-bc59-4251-adca-9c550bed39d6",
... |
math_dapo_boxed | [
{
"content": "You are an expert mathematical problem solver. Solve the problem carefully. You must first show all your step-by-step reasoning and derivations inside <think> and </think> tags. After your thought process, provide the final answer clearly and make it easy to verify.",
"role": "system"
},
{... | MATH | {
"ground_truth": "12",
"style": "rule"
} | math | {
"agent_name": null,
"dataset": null,
"domain": "math",
"family": null,
"index": "math-13649",
"instruction_id_list": null,
"instruction_kwargs_json": null,
"mix_index": 13649,
"raw_prompt": null,
"sample_id": null,
"source_config": null,
"source_index": "16301d1f-b7bc-47fc-bccf-e72718a01618",
... |
primeintellect | [
{
"content": "You are an expert programmer and competitive problem solver. Solve the task carefully. You must first outline your thought process, logic, and planning inside <think> and </think> tags, and then produce a correct final implementation outside of these tags.",
"role": "system"
},
{
"cont... | code | {
"ground_truth": "[{\"type\": \"stdin_stdout\", \"input\": \"3\\n\", \"output\": \"0 3 4\\n3 0 5\\n4 5 0\\n\"}, {\"type\": \"stdin_stdout\", \"input\": \"17\\n\", \"output\": \"0 3 4 6 9 14 22 31 40 54 75 96 129 153 183 213 259\\n3 0 5 7 10 15 23 32 41 55 76 97 130 154 184 214 260\\n4 5 0 8 11 16 24 33 42 56 77 98 1... | code | {
"agent_name": null,
"dataset": null,
"domain": "code",
"family": null,
"index": "code-13339",
"instruction_id_list": null,
"instruction_kwargs_json": null,
"mix_index": 13339,
"raw_prompt": null,
"sample_id": null,
"source_config": "primeintellect",
"source_index": "13339",
"split": "train"
... |
nemotron_if_rl | [
{
"content": "You are an expert general-purpose assistant. Follow every user instruction exactly, especially formatting, wording, ordering, and length constraints. You must always begin your response with exactly <think>\n</think> (a single newline inside the tags), and then immediately provide your answer.",
... | instruction_following | {
"ground_truth": "",
"style": "rule"
} | if | {
"agent_name": "instruction_following_simple_agent",
"dataset": "nano-v3-RL-instruction_following",
"domain": "if",
"family": "other",
"index": "if-38938",
"instruction_id_list": [
"combination:two_responses",
"punctuation:no_comma"
],
"instruction_kwargs_json": [
"null",
"null"
],
... |
nemotron_if_rl | [
{
"content": "You are an expert general-purpose assistant. Follow every user instruction exactly, especially formatting, wording, ordering, and length constraints. You must always begin your response with exactly <think>\n</think> (a single newline inside the tags), and then immediately provide your answer.",
... | instruction_following | {
"ground_truth": "",
"style": "rule"
} | if | {
"agent_name": "instruction_following_simple_agent",
"dataset": "nano-v3-RL-instruction_following",
"domain": "if",
"family": "other",
"index": "if-7736",
"instruction_id_list": [
"keywords:start_end",
"first_word:first_word_sent"
],
"instruction_kwargs_json": [
"null",
"{\"first_word\"... |
nemotron_if_rl | [
{
"content": "You are an expert general-purpose assistant. Follow every user instruction exactly, especially formatting, wording, ordering, and length constraints. You must always begin your response with exactly <think>\n</think> (a single newline inside the tags), and then immediately provide your answer.",
... | instruction_following | {
"ground_truth": "",
"style": "rule"
} | if | {
"agent_name": "instruction_following_simple_agent",
"dataset": "nano-v3-RL-instruction_following",
"domain": "if",
"family": "other",
"index": "if-8715",
"instruction_id_list": [
"detectable_format:sentence_hyphens"
],
"instruction_kwargs_json": [
"null"
],
"mix_index": 8715,
"raw_prompt... |
nemotron_if_rl | [
{
"content": "You are an expert general-purpose assistant. Follow every user instruction exactly, especially formatting, wording, ordering, and length constraints. You must always begin your response with exactly <think>\n</think> (a single newline inside the tags), and then immediately provide your answer.",
... | instruction_following | {
"ground_truth": "",
"style": "rule"
} | if | {
"agent_name": "instruction_following_simple_agent",
"dataset": "nano-v3-RL-instruction_following",
"domain": "if",
"family": "language",
"index": "if-44375",
"instruction_id_list": [
"change_case:english_lowercase",
"length_constraints:nth_paragraph_first_word"
],
"instruction_kwargs_json": [
... |
taco | [
{
"content": "You are an expert programmer and competitive problem solver. Solve the task carefully. You must first outline your thought process, logic, and planning inside <think> and </think> tags, and then produce a correct final implementation outside of these tags.",
"role": "system"
},
{
"cont... | code | {
"ground_truth": "{\"inputs\": [\"1\\n2\\n3\\n390\\n216\\n1546\\n314159265358979323\\n0\", \"1\\n2\\n2\\n390\\n1124\\n1546\\n314159265358979323\\n0\", \"1\\n2\\n3\\n390\\n216\\n1546\\n31125328462643205\\n0\", \"1\\n2\\n2\\n390\\n1124\\n1546\\n211760204581026131\\n0\", \"1\\n2\\n3\\n104\\n216\\n1546\\n311253284626432... | code | {
"agent_name": null,
"dataset": null,
"domain": "code",
"family": null,
"index": "code-6558",
"instruction_id_list": null,
"instruction_kwargs_json": null,
"mix_index": 6558,
"raw_prompt": null,
"sample_id": null,
"source_config": "taco",
"source_index": "6558",
"split": "train"
} |
primeintellect | [
{
"content": "You are an expert programmer and competitive problem solver. Solve the task carefully. You must first outline your thought process, logic, and planning inside <think> and </think> tags, and then produce a correct final implementation outside of these tags.",
"role": "system"
},
{
"cont... | code | {
"ground_truth": "[{\"type\": \"stdin_stdout\", \"input\": \"8\\nkamusta_po\\ngenki_desu\\nohayou_gozaimasu\\nannyeong_hashimnida\\nhajime_no_ippo\\nbensamu_no_sentou_houhou_ga_okama_kenpo\\nang_halaman_doon_ay_sarisari_singkamasu\\nsi_roy_mustang_ay_namamasu\\n\", \"output\": \"FILIPINO\\nJAPANESE\\nJAPANESE\\nKORE... | code | {
"agent_name": null,
"dataset": null,
"domain": "code",
"family": null,
"index": "code-10905",
"instruction_id_list": null,
"instruction_kwargs_json": null,
"mix_index": 10905,
"raw_prompt": null,
"sample_id": null,
"source_config": "primeintellect",
"source_index": "10905",
"split": "train"
... |
nemotron_if_rl | [
{
"content": "You are an expert general-purpose assistant. Follow every user instruction exactly, especially formatting, wording, ordering, and length constraints. You must always begin your response with exactly <think>\n</think> (a single newline inside the tags), and then immediately provide your answer.",
... | instruction_following | {
"ground_truth": "",
"style": "rule"
} | if | {
"agent_name": "instruction_following_simple_agent",
"dataset": "nano-v3-RL-instruction_following",
"domain": "if",
"family": "count_or_pattern",
"index": "if-27102",
"instruction_id_list": [
"keywords:word_count_different_numbers",
"last_word:last_word_answer",
"detectable_content:postscript"
... |
nemotron_if_rl | [
{
"content": "You are an expert general-purpose assistant. Follow every user instruction exactly, especially formatting, wording, ordering, and length constraints. You must always begin your response with exactly <think>\n</think> (a single newline inside the tags), and then immediately provide your answer.",
... | instruction_following | {
"ground_truth": "",
"style": "rule"
} | if | {
"agent_name": "instruction_following_simple_agent",
"dataset": "nano-v3-RL-instruction_following",
"domain": "if",
"family": "language",
"index": "if-23128",
"instruction_id_list": [
"length_constraints:number_paragraphs",
"change_case:english_capital"
],
"instruction_kwargs_json": [
"{\"n... |
nemotron_if_rl | [
{
"content": "You are an expert general-purpose assistant. Follow every user instruction exactly, especially formatting, wording, ordering, and length constraints. You must always begin your response with exactly <think>\n</think> (a single newline inside the tags), and then immediately provide your answer.",
... | instruction_following | {
"ground_truth": "",
"style": "rule"
} | if | {
"agent_name": "instruction_following_simple_agent",
"dataset": "nano-v3-RL-instruction_following",
"domain": "if",
"family": "keywords",
"index": "if-960",
"instruction_id_list": [
"punctuation:punctuation_dot",
"keywords:existence"
],
"instruction_kwargs_json": [
"null",
"{\"keywords\... |
nemotron_if_rl | [
{
"content": "You are an expert general-purpose assistant. Follow every user instruction exactly, especially formatting, wording, ordering, and length constraints. You must always begin your response with exactly <think>\n</think> (a single newline inside the tags), and then immediately provide your answer.",
... | instruction_following | {
"ground_truth": "",
"style": "rule"
} | if | {
"agent_name": "instruction_following_simple_agent",
"dataset": "nano-v3-RL-instruction_following",
"domain": "if",
"family": "keywords",
"index": "if-1393",
"instruction_id_list": [
"last_word:last_word_sent",
"count:count_increment_word"
],
"instruction_kwargs_json": [
"{\"last_word\": \"... |
nemotron_if_rl | [
{
"content": "You are an expert general-purpose assistant. Follow every user instruction exactly, especially formatting, wording, ordering, and length constraints. You must always begin your response with exactly <think>\n</think> (a single newline inside the tags), and then immediately provide your answer.",
... | instruction_following | {
"ground_truth": "",
"style": "rule"
} | if | {
"agent_name": "instruction_following_simple_agent",
"dataset": "nano-v3-RL-instruction_following",
"domain": "if",
"family": "keywords",
"index": "if-20263",
"instruction_id_list": [
"length_constraints:number_sentences",
"change_case:capital_word_frequency",
"keywords:keyword_specific_positio... |
End of preview. Expand in Data Studio
TrustMOPD-SingleCap
Paper: Read the paper
- Downloads last month
- 36