Datasets:
id stringlengths 25 25 | dataset_id stringclasses 1
value | dataset_split stringclasses 1
value | dataset_row_id stringclasses 250
values | prompt stringlengths 96 1.58k | answer stringclasses 10
values | num_agents int64 3 3 | num_rounds int64 3 3 | round_outputs listlengths 3 3 | agent_final_answers listlengths 3 3 | consensus_answer stringclasses 31
values | metadata dict | observation_noise dict | bbh_subset stringclasses 5
values | schema_version stringclasses 1
value | model stringclasses 1
value | protocol stringclasses 1
value | dataset stringclasses 1
value | source_index int64 0 1.25k | original_id stringclasses 250
values | original_row_id stringclasses 250
values | record_uid stringlengths 25 25 | question_uid stringlengths 73 73 | quality_flags listlengths 0 4 | signals dict |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
qwen3_4b/debate/bbh/00000 | lukaemon/bbh | test | 0 | Alice, Bob, Claire, Dave, Eve, Fred, and Gertrude are on the same team in a soccer match. At the start of the match, they are each assigned to a position: Alice is playing striker, Bob is playing right winger, Claire is playing left winger, Dave is playing benchwarmer, Eve is playing goalkeeper, Fred is playing center ... | (A) | 3 | 3 | [
{
"round_index": 0,
"agents": [
{
"agent_index": 0,
"response": "We are given a sequence of position swaps among seven players. We need to track the final position of **Gertrude** after all the swaps.\n\n---\n\n### **Initial Positions:**\n\n| Player | Position |\n|----------|--... | [
{
"agent_index": 0,
"response": "$$\n\\boxed{A}\n$$",
"parsed_answer": "A",
"quality": {
"empty_response": false,
"parse_missing": false,
"entropy_available": true,
"inferred_token_count": 9,
"reencoded_token_count": null,
"token_count_method": "saved_sum_mean",
... | A | {
"task_type": "math",
"mas_type": "debate",
"provider": "vllm",
"row_index_in_run": 0,
"seed": 0,
"temperature": 0.7,
"top_p": 0.9,
"max_tokens": 12288,
"requested_model": "Qwen/Qwen3-4B",
"resolved_model": "Qwen/Qwen3-4B"
} | {
"hyperparams": {
"w_conf": 1,
"w_tok": 1,
"w_ent": 1,
"w_margin": 1,
"beta": 10,
"epsilon": 1e-8
},
"rounds": [
{
"round_index": 0,
"c_t": [
-0.06833418303813409,
-0.054000829835984226,
-0.08132358823116156
],
"c_t_raw": [
-30.0... | tracking_shuffled_objects_seven_objects | 1.0 | Qwen/Qwen3-4B | debate | bbh | 0 | lukaemon/bbh:0 | 0 | qwen3_4b/debate/bbh/00000 | bbh:test:f787e68714a57a9ff279d0f5fa20d27f2a8f98fa256104dcf413ed9a21a478fd | [] | {
"agreement": [
1,
1,
1
],
"token_entropy": [
0.10949272585572754,
0.06776803616448523,
0.000003985225883519743
]
} |
qwen3_4b/debate/bbh/00001 | lukaemon/bbh | test | 14 | Alice, Bob, Claire, Dave, Eve, Fred, and Gertrude are friends and avid readers who occasionally trade books. At the start of the semester, they each buy one new book: Alice gets The Great Gatsby, Bob gets The Fellowship of the Ring, Claire gets Ulysses, Dave gets The Odyssey, Eve gets Frankenstein, Fred gets Hound of t... | (C) | 3 | 3 | [
{
"round_index": 0,
"agents": [
{
"agent_index": 0,
"response": "We are given a sequence of book trades among seven friends: Alice, Bob, Claire, Dave, Eve, Fred, and Gertrude. Each starts with a unique book. We need to track the book that **Alice ends up with** after a series of trades... | [
{
"agent_index": 0,
"response": "$$\n\\boxed{C}\n$$",
"parsed_answer": "C",
"quality": {
"empty_response": false,
"parse_missing": false,
"entropy_available": true,
"inferred_token_count": 9,
"reencoded_token_count": null,
"token_count_method": "saved_sum_mean",
... | C | {
"task_type": "math",
"mas_type": "debate",
"provider": "vllm",
"row_index_in_run": 14,
"seed": 0,
"temperature": 0.7,
"top_p": 0.9,
"max_tokens": 12288,
"requested_model": "Qwen/Qwen3-4B",
"resolved_model": "Qwen/Qwen3-4B"
} | {
"hyperparams": {
"w_conf": 1,
"w_tok": 1,
"w_ent": 1,
"w_margin": 1,
"beta": 10,
"epsilon": 1e-8
},
"rounds": [
{
"round_index": 0,
"c_t": [
-0.043905683353066494,
-0.05769318391551344,
-0.05259571051420797
],
"c_t_raw": [
-21.9... | tracking_shuffled_objects_seven_objects | 1.0 | Qwen/Qwen3-4B | debate | bbh | 1 | lukaemon/bbh:14 | 14 | qwen3_4b/debate/bbh/00001 | bbh:test:013c6f630e22551c75dfcea42864ffe1c14ccb37ca324eef3144ad45adef9dee | [] | {
"agreement": [
0.6666666666666666,
1,
1
],
"token_entropy": [
0.07845202665179607,
0.04646812485491979,
0.000002769385249885744
]
} |
qwen3_4b/debate/bbh/00002 | lukaemon/bbh | test | 1 | Alice, Bob, Claire, Dave, Eve, Fred, and Gertrude are playing a game. At the start of the game, they are each holding a ball: Alice has a orange ball, Bob has a yellow ball, Claire has a brown ball, Dave has a white ball, Eve has a black ball, Fred has a red ball, and Gertrude has a purple ball.
As the game progresses,... | (E) | 3 | 3 | [
{
"round_index": 0,
"agents": [
{
"agent_index": 0,
"response": "We are given a sequence of ball swaps among seven players: Alice, Bob, Claire, Dave, Eve, Fred, and Gertrude. Each starts with a specific ball. We will track the ball ownership through each swap step by step.\n\n---\n\n##... | [
{
"agent_index": 0,
"response": "$$\n\\boxed{E}\n$$",
"parsed_answer": "E",
"quality": {
"empty_response": false,
"parse_missing": false,
"entropy_available": true,
"inferred_token_count": 9,
"reencoded_token_count": null,
"token_count_method": "saved_sum_mean",
... | E | {
"task_type": "math",
"mas_type": "debate",
"provider": "vllm",
"row_index_in_run": 1,
"seed": 0,
"temperature": 0.7,
"top_p": 0.9,
"max_tokens": 12288,
"requested_model": "Qwen/Qwen3-4B",
"resolved_model": "Qwen/Qwen3-4B"
} | {
"hyperparams": {
"w_conf": 1,
"w_tok": 1,
"w_ent": 1,
"w_margin": 1,
"beta": 10,
"epsilon": 1e-8
},
"rounds": [
{
"round_index": 0,
"c_t": [
-0.04987144841905754,
-0.05466742491655199,
-0.09783605862075312
],
"c_t_raw": [
-37.80... | tracking_shuffled_objects_seven_objects | 1.0 | Qwen/Qwen3-4B | debate | bbh | 2 | lukaemon/bbh:1 | 1 | qwen3_4b/debate/bbh/00002 | bbh:test:76182a591f247c9bc9f26fd814956a27e36583a698b4b4e5b7fe2741dc60134a | [] | {
"agreement": [
0.6666666666666666,
0.6666666666666666,
1
],
"token_entropy": [
0.09495951879891854,
0.08481291438448667,
0.005310942920314055
]
} |
qwen3_4b/debate/bbh/00003 | lukaemon/bbh | test | 8 | Alice, Bob, Claire, Dave, Eve, Fred, and Gertrude are holding a white elephant gift exchange. At the start of the event, they are each holding a present of a different color: Alice has a green present, Bob has a red present, Claire has a white present, Dave has a blue present, Eve has a yellow present, Fred has a black... | (E) | 3 | 3 | [
{
"round_index": 0,
"agents": [
{
"agent_index": 0,
"response": "We are given a sequence of gift swaps among 7 people: Alice, Bob, Claire, Dave, Eve, Fred, and Gertrude. Each starts with a different colored present.\n\n### Initial Assignments:\n- **Alice**: Green \n- **Bob**: Red \n-... | [
{
"agent_index": 0,
"response": "$$\n\\boxed{E}\n$$",
"parsed_answer": "E",
"quality": {
"empty_response": false,
"parse_missing": false,
"entropy_available": true,
"inferred_token_count": 9,
"reencoded_token_count": null,
"token_count_method": "saved_sum_mean",
... | E | {
"task_type": "math",
"mas_type": "debate",
"provider": "vllm",
"row_index_in_run": 8,
"seed": 0,
"temperature": 0.7,
"top_p": 0.9,
"max_tokens": 12288,
"requested_model": "Qwen/Qwen3-4B",
"resolved_model": "Qwen/Qwen3-4B"
} | {
"hyperparams": {
"w_conf": 1,
"w_tok": 1,
"w_ent": 1,
"w_margin": 1,
"beta": 10,
"epsilon": 1e-8
},
"rounds": [
{
"round_index": 0,
"c_t": [
-0.09750255641951357,
-0.07142063867446768,
-0.0831023270534793
],
"c_t_raw": [
-36.270... | tracking_shuffled_objects_seven_objects | 1.0 | Qwen/Qwen3-4B | debate | bbh | 3 | lukaemon/bbh:8 | 8 | qwen3_4b/debate/bbh/00003 | bbh:test:8e0d42d1f83dc08fd2a964b684addcc2ee2d2855b957636b63fed98bef99ca66 | [] | {
"agreement": [
1,
1,
1
],
"token_entropy": [
0.12240963725945377,
0.06233905716711359,
0.00003475159313028423
]
} |
qwen3_4b/debate/bbh/00004 | lukaemon/bbh | test | 12 | Alice, Bob, Claire, Dave, Eve, Fred, and Gertrude are on the same team in a soccer match. At the start of the match, they are each assigned to a position: Alice is playing left winger, Bob is playing goalkeeper, Claire is playing center midfielder, Dave is playing right midfielder, Eve is playing fullback, Fred is play... | (G) | 3 | 3 | [
{
"round_index": 0,
"agents": [
{
"agent_index": 0,
"response": "We are given a sequence of position swaps between players. Let's track the positions of **Gertrude** through each swap.\n\n---\n\n### **Initial Positions:**\n\n- **Alice**: Left winger \n- **Bob**: Goalkeeper \n- **Clai... | [
{
"agent_index": 0,
"response": "$$\n\\boxed{\\text{G}}\n$$",
"parsed_answer": "\\text{G}",
"quality": {
"empty_response": false,
"parse_missing": false,
"entropy_available": true,
"inferred_token_count": 11,
"reencoded_token_count": null,
"token_count_method": "s... | \text{G} | {
"task_type": "math",
"mas_type": "debate",
"provider": "vllm",
"row_index_in_run": 12,
"seed": 0,
"temperature": 0.7,
"top_p": 0.9,
"max_tokens": 12288,
"requested_model": "Qwen/Qwen3-4B",
"resolved_model": "Qwen/Qwen3-4B"
} | {
"hyperparams": {
"w_conf": 1,
"w_tok": 1,
"w_ent": 1,
"w_margin": 1,
"beta": 10,
"epsilon": 1e-8
},
"rounds": [
{
"round_index": 0,
"c_t": [
-0.0739868828880653,
-0.032746819842886216,
-0.07460356866096533
],
"c_t_raw": [
-29.52... | tracking_shuffled_objects_seven_objects | 1.0 | Qwen/Qwen3-4B | debate | bbh | 4 | lukaemon/bbh:12 | 12 | qwen3_4b/debate/bbh/00004 | bbh:test:273ada9499378a2dd1b62d7be4535c612ddcbe48405d18ecd032a16bfda4e781 | [] | {
"agreement": [
1,
1,
1
],
"token_entropy": [
0.08984955371325293,
0.06577417837415621,
0.025873604534703897
]
} |
qwen3_4b/debate/bbh/00005 | lukaemon/bbh | test | 25 | Alice, Bob, Claire, Dave, Eve, Fred, and Gertrude are dancers at a square dance. At the start of a song, they each have a partner: Alice is dancing with Lola, Bob is dancing with Rodrigo, Claire is dancing with Jamie, Dave is dancing with Izzi, Eve is dancing with Sam, Fred is dancing with Karl, and Gertrude is dancing... | (G) | 3 | 3 | [
{
"round_index": 0,
"agents": [
{
"agent_index": 0,
"response": "We are given a sequence of partner swaps among dancers, and we need to determine **who Eve is dancing with at the end**.\n\nLetโs break it down step by step.\n\n---\n\n### **Initial Pairings:**\n\n- Alice โ Lola \n- Bob ... | [
{
"agent_index": 0,
"response": "$$\n\\boxed{\\text{G}}\n$$",
"parsed_answer": "\\text{G}",
"quality": {
"empty_response": false,
"parse_missing": false,
"entropy_available": true,
"inferred_token_count": 11,
"reencoded_token_count": null,
"token_count_method": "s... | \text{G} | {
"task_type": "math",
"mas_type": "debate",
"provider": "vllm",
"row_index_in_run": 25,
"seed": 0,
"temperature": 0.7,
"top_p": 0.9,
"max_tokens": 12288,
"requested_model": "Qwen/Qwen3-4B",
"resolved_model": "Qwen/Qwen3-4B"
} | {
"hyperparams": {
"w_conf": 1,
"w_tok": 1,
"w_ent": 1,
"w_margin": 1,
"beta": 10,
"epsilon": 1e-8
},
"rounds": [
{
"round_index": 0,
"c_t": [
-0.062149052186792134,
-0.04315671826113995,
-0.059947491082108256
],
"c_t_raw": [
-21.... | tracking_shuffled_objects_seven_objects | 1.0 | Qwen/Qwen3-4B | debate | bbh | 5 | lukaemon/bbh:25 | 25 | qwen3_4b/debate/bbh/00005 | bbh:test:1ff90d4ee03259b5d6f1d94fa13ece8c40d5ef13b7ee6d3078c40d842dd7d313 | [] | {
"agreement": [
1,
1,
1
],
"token_entropy": [
0.07951429665624367,
0.07632309115063965,
0.00009425705883929112
]
} |
qwen3_4b/debate/bbh/00006 | lukaemon/bbh | test | 15 | Alice, Bob, Claire, Dave, Eve, Fred, and Gertrude are playing a game. At the start of the game, they are each holding a ball: Alice has a blue ball, Bob has a red ball, Claire has a brown ball, Dave has a purple ball, Eve has a black ball, Fred has a pink ball, and Gertrude has a green ball.
As the game progresses, pai... | (E) | 3 | 3 | [
{
"round_index": 0,
"agents": [
{
"agent_index": 0,
"response": "We are given that seven players each start with a unique ball:\n\n- **Alice** โ blue \n- **Bob** โ red \n- **Claire** โ brown \n- **Dave** โ purple \n- **Eve** โ black \n- **Fred** โ pink \n- **Gertrude** โ green \... | [
{
"agent_index": 0,
"response": "We are given a sequence of ball swaps among seven players, each initially holding a unique ball:\n\n- **Alice**: blue \n- **Bob**: red \n- **Claire**: brown \n- **Dave**: purple \n- **Eve**: black \n- **Fred**: pink \n- **Gertrude**: green \n\nWe track the swaps step ... | E | {
"task_type": "math",
"mas_type": "debate",
"provider": "vllm",
"row_index_in_run": 15,
"seed": 0,
"temperature": 0.7,
"top_p": 0.9,
"max_tokens": 12288,
"requested_model": "Qwen/Qwen3-4B",
"resolved_model": "Qwen/Qwen3-4B"
} | {
"hyperparams": {
"w_conf": 1,
"w_tok": 1,
"w_ent": 1,
"w_margin": 1,
"beta": 10,
"epsilon": 1e-8
},
"rounds": [
{
"round_index": 0,
"c_t": [
-0.06626861970852549,
-0.10082662317703009,
-0.018312797865665476
],
"c_t_raw": [
-31.8... | tracking_shuffled_objects_seven_objects | 1.0 | Qwen/Qwen3-4B | debate | bbh | 6 | lukaemon/bbh:15 | 15 | qwen3_4b/debate/bbh/00006 | bbh:test:879601571b1ba02ffc6162db5bf293e845326b922ccd033429cc0c1083ee2d64 | [] | {
"agreement": [
0.6666666666666666,
1,
1
],
"token_entropy": [
0.09155617214880585,
0.07449788007275042,
0.031018924668207132
]
} |
qwen3_4b/debate/bbh/00007 | lukaemon/bbh | test | 29 | Alice, Bob, Claire, Dave, Eve, Fred, and Gertrude are dancers at a square dance. At the start of a song, they each have a partner: Alice is dancing with Izzi, Bob is dancing with Jamie, Claire is dancing with Ophelia, Dave is dancing with Rodrigo, Eve is dancing with Patrick, Fred is dancing with Sam, and Gertrude is d... | (D) | 3 | 3 | [
{
"round_index": 0,
"agents": [
{
"agent_index": 0,
"response": "We are given a set of dancers and their initial partners, and then a sequence of partner swaps. We need to track the final partner of **Bob** after all the swaps.\n\n---\n\n### Step 1: Initial Partners\n\nLetโs list the d... | [
{
"agent_index": 0,
"response": "We are given a sequence of partner swaps among seven dancers, and we are asked to determine **who Bob is dancing with at the end** of the sequence.\n\n---\n\n### Step 1: Initial Partnerships\n\n- Alice โ Izzi \n- Bob โ Jamie \n- Claire โ Ophelia \n- Dave โ Rodrigo \n- Ev... | D | {
"task_type": "math",
"mas_type": "debate",
"provider": "vllm",
"row_index_in_run": 29,
"seed": 0,
"temperature": 0.7,
"top_p": 0.9,
"max_tokens": 12288,
"requested_model": "Qwen/Qwen3-4B",
"resolved_model": "Qwen/Qwen3-4B"
} | {
"hyperparams": {
"w_conf": 1,
"w_tok": 1,
"w_ent": 1,
"w_margin": 1,
"beta": 10,
"epsilon": 1e-8
},
"rounds": [
{
"round_index": 0,
"c_t": [
-0.09639015045377222,
-0.052588283073120745,
-0.11487178885390507
],
"c_t_raw": [
-41.1... | tracking_shuffled_objects_seven_objects | 1.0 | Qwen/Qwen3-4B | debate | bbh | 7 | lukaemon/bbh:29 | 29 | qwen3_4b/debate/bbh/00007 | bbh:test:e4e0d314fcc5b04e137571517051a170437e91a209d9e30dda5cf6114f884343 | [] | {
"agreement": [
0.6666666666666666,
1,
1
],
"token_entropy": [
0.13463437678884946,
0.11522808253261735,
0.11002425639627461
]
} |
qwen3_4b/debate/bbh/00008 | lukaemon/bbh | test | 22 | Alice, Bob, Claire, Dave, Eve, Fred, and Gertrude are on the same team in a soccer match. At the start of the match, they are each assigned to a position: Alice is playing left winger, Bob is playing cheerleader, Claire is playing fullback, Dave is playing right winger, Eve is playing striker, Fred is playing right mid... | (F) | 3 | 3 | [
{
"round_index": 0,
"agents": [
{
"agent_index": 0,
"response": "We are given a sequence of position swaps among 7 players. Let's track the positions of each player step by step.\n\n### Initial positions:\n- **Alice**: left winger \n- **Bob**: cheerleader \n- **Claire**: fullback \n... | [
{
"agent_index": 0,
"response": "We are given a sequence of position swaps among 7 players, and we are to determine **where Eve ends up** after all the swaps.\n\n---\n\n### **Initial Positions:**\n\n- **Alice**: left winger \n- **Bob**: cheerleader \n- **Claire**: fullback \n- **Dave**: right winger \n-... | F | {
"task_type": "math",
"mas_type": "debate",
"provider": "vllm",
"row_index_in_run": 22,
"seed": 0,
"temperature": 0.7,
"top_p": 0.9,
"max_tokens": 12288,
"requested_model": "Qwen/Qwen3-4B",
"resolved_model": "Qwen/Qwen3-4B"
} | {
"hyperparams": {
"w_conf": 1,
"w_tok": 1,
"w_ent": 1,
"w_margin": 1,
"beta": 10,
"epsilon": 1e-8
},
"rounds": [
{
"round_index": 0,
"c_t": [
-0.02209411799564556,
-0.09771380134399156,
-0.05710504663814504
],
"c_t_raw": [
-14.71... | tracking_shuffled_objects_seven_objects | 1.0 | Qwen/Qwen3-4B | debate | bbh | 8 | lukaemon/bbh:22 | 22 | qwen3_4b/debate/bbh/00008 | bbh:test:4e3bc411ba8dbdc1070be592091b7c2b0104c44190658ed590486833dd49dbfb | [] | {
"agreement": [
0.6666666666666666,
1,
1
],
"token_entropy": [
0.09761012447817657,
0.056142467513095894,
0.02545574659397853
]
} |
qwen3_4b/debate/bbh/00009 | lukaemon/bbh | test | 16 | Alice, Bob, Claire, Dave, Eve, Fred, and Gertrude are playing a game. At the start of the game, they are each holding a ball: Alice has a green ball, Bob has a brown ball, Claire has a purple ball, Dave has a pink ball, Eve has a white ball, Fred has a red ball, and Gertrude has a orange ball.
As the game progresses, p... | (C) | 3 | 3 | [
{
"round_index": 0,
"agents": [
{
"agent_index": 0,
"response": "We are given a sequence of ball swaps among seven players: Alice, Bob, Claire, Dave, Eve, Fred, and Gertrude. Each starts with a unique ball:\n\n- **Alice** โ green \n- **Bob** โ brown \n- **Claire** โ purple \n- **Dav... | [
{
"agent_index": 0,
"response": "After carefully reviewing the sequence of swaps and tracking the ball ownership for each player, the final ownership is as follows:\n\n- **Alice**: purple \n- **Bob**: red \n- **Claire**: white \n- **Dave**: orange \n- **Eve**: green \n- **Fred**: pink \n- **Gertrude**... | C | {
"task_type": "math",
"mas_type": "debate",
"provider": "vllm",
"row_index_in_run": 16,
"seed": 0,
"temperature": 0.7,
"top_p": 0.9,
"max_tokens": 12288,
"requested_model": "Qwen/Qwen3-4B",
"resolved_model": "Qwen/Qwen3-4B"
} | {
"hyperparams": {
"w_conf": 1,
"w_tok": 1,
"w_ent": 1,
"w_margin": 1,
"beta": 10,
"epsilon": 1e-8
},
"rounds": [
{
"round_index": 0,
"c_t": [
-0.0888903196153525,
-0.029979877756102953,
-0.0423446679663515
],
"c_t_raw": [
-49.334... | tracking_shuffled_objects_seven_objects | 1.0 | Qwen/Qwen3-4B | debate | bbh | 9 | lukaemon/bbh:16 | 16 | qwen3_4b/debate/bbh/00009 | bbh:test:5bd3173c4908de4b0d36fb3d26008746fd24c06ef0cd78d2884550ddf10c8358 | [] | {
"agreement": [
1,
1,
1
],
"token_entropy": [
0.08060677680981604,
0.06413469774104318,
0.12456502570735294
]
} |
SAUCE Qwen3-4B Rollouts
Qwen3-4B rollout records for Debate and DyLAN on MATH-500, MMLU-Pro, and five BBH tasks. SAUCE code.
Data
| Configuration | Trajectories | Agent responses | File |
|---|---|---|---|
debate_math_500 |
500 | 4,500 | JSONL.gz |
dylan_math_500 |
500 | 2,208 | JSONL.gz |
debate_mmlu_pro |
12,032 | 108,288 | JSONL.gz |
dylan_mmlu_pro |
12,032 | 45,316 | JSONL.gz |
debate_bbh |
1,250 | 11,250 | JSONL.gz |
dylan_bbh |
1,250 | 5,336 | JSONL.gz |
Total: 27,564 trajectories, 176,898 agent responses, approximately 66 MB compressed. Files use gzip-compressed JSONL, with one rollout record per line. File statistics and checksums: manifest.json, SHA256SUMS.
BBH includes tracking_shuffled_objects_seven_objects,
logical_deduction_seven_objects, geometric_shapes, hyperbaton, and
salient_translation_error_detection.
Loading
from datasets import load_dataset
records = load_dataset(
"Tyrion279/SAUCE-Qwen3-4B-Rollouts",
"debate_math_500",
split="test",
revision="qwen-rollouts-v1",
)
print(records[0]["record_uid"])
Use a configuration name from the table to select another setting. The code repository also provides download and scoring examples.
Main fields
| Field | Contents |
|---|---|
record_uid, question_uid |
Trajectory and question identifiers |
original_id, original_row_id, source_index |
Source identifiers and zero-based position |
model, protocol, dataset, bbh_subset |
Setting and BBH task, when applicable |
prompt, answer, options |
Question text, reference answer, and MMLU-Pro choices |
num_agents, num_rounds |
Initial team size and requested maximum rounds |
round_outputs |
Per-round agent responses and stored parsed answers |
agent_final_answers, consensus_answer |
Last-round agent responses and final consensus |
dylan_roles, dylan_ranker_trace |
Saved DyLAN roles and selection history |
metadata |
Recorded generation settings |
observation_noise |
Saved per-round statistics and per-agent arrays |
signals |
Agreement and mean token entropy sequences used by the SAUCE scorer |
quality_flags |
Record-level annotation flags |
prompt contains the question text rather than the full request messages.
MMLU-Pro choices are ordered, with A corresponding to index 0.
DyLAN MATH-500 starts with four agents; the other settings start with three.
Executed rounds are stored in round_outputs.
signals.agreement uses the saved M_t, with the initial team size as the
denominator and a fallback of 1 / num_agents when all parses are null.
signals.token_entropy averages each round's token_entropy_t array.
Per-agent arrays follow the order of the agents in that round.
The auxiliary stored R_t is a separate collector value.
Annotations and missing fields
Agent entries include a quality object. parse_missing indicates a null
or empty stored parsed answer. output_budget_candidate and
repetition_candidate are heuristic flags, not confirmed truncation labels.
The MATH-500 Debate file uses retokenized visible text for its budget flag;
the other files use saved sum/mean statistics.
empty_consensus indicates an empty final consensus.
Provider termination reasons, original token-level probability distributions, and full request messages are not included.
License and attribution
See LICENSE and benchmark attribution and notices. Please cite SAUCE and the benchmarks used.
- Downloads last month
- 142