SabaPivot's picture
download
raw
26.1 kB
{
"paper": {
"title": "Calibrated Preference Learning: The Case of Label Ranking",
"orid": "STcIzNrUBB",
"arxiv": "2605.30447"
},
"claims": [
"Theorem 4.2 proves that full-rank calibration implies both sub-k calibration and top-k calibration, but Table 1 gives a counterexample showing sub-k calibration does not imply full-rank calibration (Section 4, Theorem 4.2, Table 1).",
"Table 2 provides a counterexample showing top-k calibration does not imply full-rank calibration, and that sub-k and top-k calibration are mutually incomparable notions (Section 4, Table 2).",
"Theorem 4.3 shows sub-k calibration implies rankwise sub-k calibration and top-k calibration implies rankwise top-k calibration, establishing a strict hierarchy of calibration notions defined in Definitions 1-7 (Section 4, Theorem 4.3).",
"On RewardBench2, Plackett-Luce and Mallows-Model reward models are found to be generally poorly calibrated (measured via top-1 Expected Calibration Error), while RPC achieves the strongest calibration for pairwise rankings (Section 5.2).",
"Top-1 ECE correlates strongly with reward-model accuracy in the Focus and Safety categories of RewardBench2 but only weakly in the Math and Precise-IF categories, showing calibration is an imperfect proxy for benchmark accuracy (Section 5.2)."
],
"source_audit": {
"paper_tex_sha256": "fde5ec1b099be986a9de4e407525381a86f26f5217564bf365afcddd43df9475",
"theorem_4_2_present": true,
"theorem_4_3_present": true,
"table_1_present": true,
"table_2_present": true,
"table_3_present": true,
"official_code_commit": "dc0b53324cfc98625517f6e7cdeb5c65a77b89a6",
"official_code_url": "https://github.com/Advueu963/Calibrated_Preference_Learning/tree/dc0b53324cfc98625517f6e7cdeb5c65a77b89a6"
},
"claim_1": {
"table1": {
"sub2_max_error": "0",
"top1_max_error": "1/6",
"full_rank_l1_error": "2/3",
"sub2_calibrated_not_full_rank": true
},
"full_rank_random_checks": {
"count": 200,
"max_error": 0.0
}
},
"claim_2": {
"table3": {
"top1_max_error": "0",
"sub2_max_error": "1/6",
"full_rank_l1_error": "1",
"top1_calibrated_not_full_rank": true
},
"table2": {
"max_sub2_violation": "1/6",
"max_top1_violation": "1/6",
"rankwise_but_neither_sub2_nor_top1": true
},
"mutual_incomparability": true,
"numbering_note": "The current source's Table 2 is the rankwise-but-neither counterexample; the top-1-not-full-rank counterexample is Table 3. The registered claim's substance is true, but its table anchor is inaccurate."
},
"claim_3": {
"coordinate_implication": true,
"strictness_witnessed_by_tables": true
},
"claim_4": {
"label_ranking": {
"datasets": [
{
"dataset": "political",
"csv": "official-code/results/political/subk_ece_results_political_1.0_95_prob_mass_abs_linear.csv",
"rpc": 0.11606087821488724,
"plackett_luce": 0.3353780261063616,
"mallows": 0.41621835440484567,
"rpc_best_of_three": true
},
{
"dataset": "movies",
"csv": "official-code/results/movies/subk_ece_results_movies_0.0_95_prob_mass_abs_linear.csv",
"rpc": 0.07066460016122396,
"plackett_luce": 0.5004784664720409,
"mallows": 0.5004784624255281,
"rpc_best_of_three": true
},
{
"dataset": "glass",
"csv": "official-code/results/glass/subk_ece_results_glass_1.0_95_prob_mass_abs_linear.csv",
"rpc": 0.0887899778194392,
"plackett_luce": 0.22729745958076847,
"mallows": 0.2775441297082739,
"rpc_best_of_three": true
}
],
"rpc_best_count": 3,
"official_commit": "dc0b53324cfc98625517f6e7cdeb5c65a77b89a6"
},
"rewardbench2": {
"dataset_url": "https://huggingface.co/datasets/allenai/reward-bench-2-results",
"dataset_revision": "b19c7033e964187d12e74a43a07f2d727a3d37e5",
"models_with_scores": 60,
"best_model": {
"model": "Skywork/Skywork-Reward-V2-Llama-3.1-8B",
"ece": 0.13476999174795454,
"n_examples": 1865,
"score": 84.12779216039524,
"factuality": 84.63157894736842,
"precise_if": 66.25,
"math": 77.59562841530054,
"safety": 96.66666666666669,
"focus": 98.38383838383838
},
"best_10_mean_ece": 0.22393760010742372,
"median_ece": 0.46872034234741833
},
"scope_note": "The challenge wording conflates Section 5.1's PL/Mallows/RPC label-ranking comparison with Section 5.2's Bradley-Terry RewardBench2 study. Both components are audited separately."
},
"claim_5": {
"correlations": {
"score": {
"kendall_tau": 0.6915254237288135,
"p_value": 5.875884895968332e-15
},
"factuality": {
"kendall_tau": 0.5882356704893549,
"p_value": 3.2743073277223836e-11
},
"precise_if": {
"kendall_tau": 0.5312423782820873,
"p_value": 3.08971978021332e-09
},
"math": {
"kendall_tau": 0.4956061106975527,
"p_value": 2.54616399156624e-08
},
"safety": {
"kendall_tau": 0.6749734188228499,
"p_value": 2.868043910010452e-14
},
"focus": {
"kendall_tau": 0.698359806738388,
"p_value": 3.5090802097808933e-15
}
},
"destructive_permutation_control": {
"precise_if": {
"mean_tau": -0.003936231242529776,
"q025": -0.17806351483258973,
"q975": 0.16775924456418712
},
"math": {
"mean_tau": -0.001382928391362243,
"q025": -0.16466271753473674,
"q975": 0.16926112474739438
},
"safety": {
"mean_tau": -0.0011393007706976283,
"q025": -0.17726546037121912,
"q975": 0.17398119373600712
},
"focus": {
"mean_tau": -0.0029179442168096666,
"q025": -0.17102458152053554,
"q975": 0.1703737599826513
}
}
},
"rewardbench_rows": [
{
"model": "Skywork/Skywork-Reward-V2-Llama-3.1-8B",
"ece": 0.13476999174795454,
"n_examples": 1865,
"score": 84.12779216039524,
"factuality": 84.63157894736842,
"precise_if": 66.25,
"math": 77.59562841530054,
"safety": 96.66666666666669,
"focus": 98.38383838383838
},
{
"model": "Skywork/Skywork-Reward-V2-Qwen3-8B",
"ece": 0.19052674352425994,
"n_examples": 1865,
"score": 78.37409729353149,
"factuality": 79.89473684210526,
"precise_if": 50.0,
"math": 77.04918032786885,
"safety": 94.0,
"focus": 96.36363636363636
},
{
"model": "infly/INF-ORM-Llama3.1-70B",
"ece": 0.23938761733076275,
"n_examples": 1865,
"score": 76.48295733317504,
"factuality": 74.10526315789474,
"precise_if": 41.875,
"math": 69.94535519125684,
"safety": 96.44444444444444,
"focus": 90.30303030303033
},
{
"model": "anthropic/claude-opus-4-20250514",
"ece": 0.44304509051326607,
"n_examples": 102,
"score": 76.48277145949112,
"factuality": 82.66666666666666,
"precise_if": 41.875,
"math": 74.90892531876139,
"safety": 89.53703703703702,
"focus": 86.16161616161615
},
{
"model": "allenai/Llama-3.1-70B-Instruct-RM-RB2",
"ece": 0.3454970143401523,
"n_examples": 1865,
"score": 76.05646300951571,
"factuality": 81.26315789473684,
"precise_if": 41.875,
"math": 69.94535519125684,
"safety": 88.44444444444444,
"focus": 86.46464646464646
},
{
"model": "Skywork/Skywork-Reward-V2-Qwen3-4B",
"ece": 0.21117089642114426,
"n_examples": 1865,
"score": 75.51022543280415,
"factuality": 77.36842105263158,
"precise_if": 46.25,
"math": 73.224043715847,
"safety": 92.22222222222224,
"focus": 96.56565656565655
},
{
"model": "anthropic/claude-3-7-sonnet-20250219",
"ece": 0.47587195310274616,
"n_examples": 102,
"score": 75.38767274570088,
"factuality": 73.26315789473684,
"precise_if": 54.37499999999999,
"math": 75.0,
"safety": 90.33333333333331,
"focus": 92.12121212121212
},
{
"model": "Skywork/Skywork-Reward-V2-Llama-3.2-3B",
"ece": 0.20632375848843673,
"n_examples": 1865,
"score": 74.66480123197486,
"factuality": 76.21052631578948,
"precise_if": 45.625,
"math": 69.39890710382514,
"safety": 93.11111111111111,
"focus": 95.95959595959596
},
{
"model": "LxzGordon/URM-LLaMa-3.1-8B",
"ece": 0.2666059664445265,
"n_examples": 1865,
"score": 73.94354072783193,
"factuality": 68.84210526315789,
"precise_if": 45.0,
"math": 63.934426229508205,
"safety": 91.7777777777778,
"focus": 97.57575757575758
},
{
"model": "Skywork/Skywork-Reward-Llama-3.1-8B",
"ece": 0.21604154406292006,
"n_examples": 1865,
"score": 73.1387048400328,
"factuality": 69.89473684210526,
"precise_if": 42.5,
"math": 62.841530054644814,
"safety": 93.33333333333331,
"focus": 96.16161616161617
},
{
"model": "allenai/Llama-3.1-8B-Instruct-RM-RB2",
"ece": 0.3829611922762887,
"n_examples": 1865,
"score": 72.84710237854704,
"factuality": 74.31578947368422,
"precise_if": 44.375,
"math": 61.74863387978142,
"safety": 89.55555555555556,
"focus": 90.7070707070707
},
{
"model": "ShikaiChen/LDL-Reward-Gemma-2-27B-v0.1",
"ece": 0.5628945129050107,
"n_examples": 1865,
"score": 72.48764297589858,
"factuality": 75.57894736842105,
"precise_if": 35.0,
"math": 64.48087431693989,
"safety": 92.22222222222224,
"focus": 91.31313131313132
},
{
"model": "openai/gpt-4.1-2025-04-14",
"ece": 0.429620770444169,
"n_examples": 102,
"score": 72.31731857413882,
"factuality": 82.89473684210526,
"precise_if": 39.73958333333334,
"math": 65.20947176684881,
"safety": 87.25925925925924,
"focus": 73.38383838383838
},
{
"model": "allenai/Llama-3.1-Tulu-3-70B-SFT-RM-RB2",
"ece": 0.3805613091161596,
"n_examples": 1865,
"score": 72.20387212509466,
"factuality": 80.84210526315789,
"precise_if": 36.875,
"math": 67.75956284153006,
"safety": 86.8888888888889,
"focus": 77.77777777777779
},
{
"model": "Skywork/Skywork-Reward-Llama-3.1-8B-v0.2",
"ece": 0.2248423768646985,
"n_examples": 1865,
"score": 71.7452543316723,
"factuality": 69.6842105263158,
"precise_if": 40.625,
"math": 60.10928961748634,
"safety": 94.2222222222222,
"focus": 94.14141414141412
},
{
"model": "anthropic/claude-sonnet-4-20250514",
"ece": 0.4666141658421129,
"n_examples": 102,
"score": 71.16513658188997,
"factuality": 76.12280701754385,
"precise_if": 35.937500000000014,
"math": 70.49180327868852,
"safety": 89.09259259259258,
"focus": 75.95959595959594
},
{
"model": "HFXM/RAMO-Llama3.1-8B",
"ece": 0.33136097555912775,
"n_examples": 1865,
"score": 69.17258753860413,
"factuality": 65.47368421052632,
"precise_if": 37.5,
"math": 56.28415300546448,
"safety": 97.55555555555556,
"focus": 90.7070707070707
},
{
"model": "Skywork/Skywork-VL-Reward-7B",
"ece": 0.26843480120480384,
"n_examples": 1865,
"score": 68.84669257606988,
"factuality": 60.631578947368425,
"precise_if": 35.0,
"math": 63.387978142076506,
"safety": 89.11111111111111,
"focus": 89.0909090909091
},
{
"model": "allenai/Llama-3.1-Tulu-3-8B-RL-RM-RB2",
"ece": 0.35348883907419754,
"n_examples": 1865,
"score": 68.71219205900773,
"factuality": 76.42105263157895,
"precise_if": 40.0,
"math": 61.74863387978142,
"safety": 86.44444444444444,
"focus": 84.84848484848484
},
{
"model": "allenai/Llama-3.1-Tulu-3-8B-DPO-RM-RB2",
"ece": 0.3519554515743213,
"n_examples": 1865,
"score": 68.69585657154352,
"factuality": 75.1578947368421,
"precise_if": 38.75,
"math": 62.841530054644814,
"safety": 86.0,
"focus": 85.45454545454545
},
{
"model": "allenai/Llama-3.1-Tulu-3-8B-SFT-RM-RB2",
"ece": 0.28127230498472994,
"n_examples": 1865,
"score": 68.20514558353437,
"factuality": 73.26315789473684,
"precise_if": 38.75,
"math": 57.92349726775956,
"safety": 89.77777777777777,
"focus": 88.88888888888889
},
{
"model": "Skywork/Skywork-Reward-V2-Qwen3-1.7B",
"ece": 0.2957017764976967,
"n_examples": 1865,
"score": 68.17585285009079,
"factuality": 65.6842105263158,
"precise_if": 44.375,
"math": 72.6775956284153,
"safety": 89.11111111111111,
"focus": 88.48484848484848
},
{
"model": "Ray2333/GRM-Llama3-8B-rewardmodel-ft",
"ece": 0.293474992842568,
"n_examples": 1865,
"score": 67.66019726388538,
"factuality": 62.73684210526316,
"precise_if": 35.0,
"math": 58.46994535519126,
"safety": 92.22222222222224,
"focus": 89.29292929292929
},
{
"model": "openai/gpt-4.1-mini-2025-04-14",
"ece": 0.47588383034528187,
"n_examples": 102,
"score": 65.7349500114784,
"factuality": 60.8421052631579,
"precise_if": 41.25,
"math": 72.1311475409836,
"safety": 72.64814814814811,
"focus": 73.53535353535354
},
{
"model": "openai/gpt-4o-2024-08-06",
"ece": 0.41589317817039906,
"n_examples": 102,
"score": 64.92828668947554,
"factuality": 56.84210526315789,
"precise_if": 33.125,
"math": 62.29508196721312,
"safety": 86.1851851851852,
"focus": 72.92929292929293
},
{
"model": "allenai/Llama-3.1-8B-Base-RM-RB2",
"ece": 0.3291385082177684,
"n_examples": 1865,
"score": 64.90187760825604,
"factuality": 72.0,
"precise_if": 36.25,
"math": 61.20218579234973,
"safety": 82.66666666666667,
"focus": 83.23232323232324
},
{
"model": "NCSOFT/Llama-3-OffsetBias-RM-8B",
"ece": 0.43174056192181015,
"n_examples": 1865,
"score": 64.79939430060925,
"factuality": 60.8421052631579,
"precise_if": 40.0,
"math": 51.91256830601093,
"safety": 72.22222222222221,
"focus": 95.95959595959596
},
{
"model": "anthropic/claude-3-5-sonnet-20240620",
"ece": 0.4708265188527238,
"n_examples": 102,
"score": 64.66244541750068,
"factuality": 52.84210526315789,
"precise_if": 38.75,
"math": 56.830601092896174,
"safety": 85.18518518518516,
"focus": 86.96969696969697
},
{
"model": "Skywork/Skywork-Reward-V2-Llama-3.2-1B",
"ece": 0.3252607479986218,
"n_examples": 1865,
"score": 64.37645875381985,
"factuality": 60.8421052631579,
"precise_if": 45.625,
"math": 60.10928961748634,
"safety": 87.33333333333333,
"focus": 89.29292929292929
},
{
"model": "Qwen/WorldPM-72B",
"ece": 0.5629368528218481,
"n_examples": 1865,
"score": 63.3269114330196,
"factuality": 70.73684210526315,
"precise_if": 31.25,
"math": 65.57377049180327,
"safety": 85.33333333333334,
"focus": 91.71717171717172
},
{
"model": "sfairXC/FsfairX-LLaMA3-RM-v0.1",
"ece": 0.4515965595882692,
"n_examples": 1865,
"score": 62.91859600227345,
"factuality": 59.15789473684211,
"precise_if": 41.875,
"math": 62.841530054644814,
"safety": 76.66666666666667,
"focus": 70.5050505050505
},
{
"model": "Skywork/Skywork-Reward-V2-Qwen3-0.6B",
"ece": 0.39085231027634043,
"n_examples": 1865,
"score": 61.24950037720884,
"factuality": 57.99999999999999,
"precise_if": 40.0,
"math": 71.58469945355192,
"safety": 84.44444444444444,
"focus": 79.4949494949495
},
{
"model": "Ray2333/GRM-llama3-8B-sftreg",
"ece": 0.41225477535946514,
"n_examples": 1865,
"score": 60.88793933636373,
"factuality": 61.89473684210526,
"precise_if": 38.75,
"math": 57.92349726775956,
"safety": 78.66666666666666,
"focus": 68.28282828282828
},
{
"model": "Ray2333/GRM-gemma2-2B-rewardmodel-ft",
"ece": 0.36672628866425133,
"n_examples": 1865,
"score": 59.66119140947769,
"factuality": 53.05263157894737,
"precise_if": 31.25,
"math": 59.01639344262295,
"safety": 92.22222222222224,
"focus": 74.54545454545455
},
{
"model": "weqweasdas/RM-Mistral-7B",
"ece": 0.47281749801857487,
"n_examples": 1865,
"score": 59.60117398651728,
"factuality": 59.36842105263158,
"precise_if": 34.375,
"math": 59.56284153005464,
"safety": 69.11111111111111,
"focus": 72.92929292929293
},
{
"model": "allenai/Llama-3.1-Tulu-3-8B-RM",
"ece": 0.5418628632379631,
"n_examples": 1865,
"score": 58.99751678794565,
"factuality": 74.52631578947368,
"precise_if": 34.6875,
"math": 64.48087431693989,
"safety": 74.22222222222223,
"focus": 53.63636363636364
},
{
"model": "Ray2333/GRM-llama3-8B-distill",
"ece": 0.5049069764842984,
"n_examples": 1865,
"score": 58.903971632622586,
"factuality": 58.73684210526315,
"precise_if": 38.75,
"math": 59.01639344262295,
"safety": 72.22222222222221,
"focus": 67.27272727272727
},
{
"model": "hendrydong/Mistral-RM-for-RAFT-GSHF-v0",
"ece": 0.4852071041963555,
"n_examples": 1865,
"score": 58.51045929269105,
"factuality": 57.78947368421052,
"precise_if": 36.25,
"math": 60.10928961748634,
"safety": 69.55555555555556,
"focus": 67.47474747474747
},
{
"model": "openbmb/Eurus-RM-7b",
"ece": 0.38244755652313694,
"n_examples": 1865,
"score": 58.05660762038656,
"factuality": 60.0,
"precise_if": 34.375,
"math": 56.830601092896174,
"safety": 62.66666666666667,
"focus": 74.74747474747475
},
{
"model": "openai/gpt-4o-mini-2024-07-18",
"ece": 0.48926280757784807,
"n_examples": 102,
"score": 57.96141949652715,
"factuality": 41.05263157894737,
"precise_if": 34.375,
"math": 51.91256830601093,
"safety": 76.66666666666664,
"focus": 74.14141414141415
},
{
"model": "anthropic/claude-3-opus-20240229",
"ece": 0.5237197690555281,
"n_examples": 102,
"score": 57.44034084018399,
"factuality": 53.89473684210526,
"precise_if": 33.125,
"math": 51.36612021857923,
"safety": 83.77777777777777,
"focus": 66.46464646464646
},
{
"model": "CIR-AMS/BTRM_Qwen2_7b_0613",
"ece": 0.5160768592774613,
"n_examples": 1865,
"score": 57.36341394412572,
"factuality": 53.47368421052632,
"precise_if": 35.625,
"math": 60.65573770491803,
"safety": 71.77777777777777,
"focus": 57.37373737373738
},
{
"model": "internlm/internlm2-20b-reward",
"ece": 0.6009934794349481,
"n_examples": 1865,
"score": 56.27800895371716,
"factuality": 55.57894736842105,
"precise_if": 36.25,
"math": 57.377049180327866,
"safety": 61.11111111111112,
"focus": 72.52525252525253
},
{
"model": "internlm/internlm2-7b-reward",
"ece": 0.6132021913976878,
"n_examples": 1865,
"score": 53.34773837781889,
"factuality": 42.10526315789473,
"precise_if": 40.0,
"math": 56.28415300546448,
"safety": 59.55555555555555,
"focus": 70.5050505050505
},
{
"model": "google/gemini-1.5-flash-8b",
"ece": 0.8061676283717575,
"n_examples": 102,
"score": 48.51363189431879,
"factuality": 46.10526315789474,
"precise_if": 36.25,
"math": 50.81967213114754,
"safety": 66.22222222222223,
"focus": 67.47474747474747
},
{
"model": "openai/gpt-4.1-nano-2025-04-14",
"ece": 0.5344009637193005,
"n_examples": 102,
"score": 48.49148497149796,
"factuality": 46.456140350877234,
"precise_if": 25.78125,
"math": 50.40983606557376,
"safety": 71.55555555555553,
"focus": 46.59932659932663
},
{
"model": "weqweasdas/RM-Gemma-7B",
"ece": 0.6004406343539396,
"n_examples": 1865,
"score": 48.25502869477731,
"factuality": 49.26315789473684,
"precise_if": 39.375,
"math": 60.65573770491803,
"safety": 48.22222222222222,
"focus": 49.696969696969695
},
{
"model": "openbmb/UltraRM-13b",
"ece": 0.5513903214584469,
"n_examples": 1865,
"score": 46.83473363506545,
"factuality": 50.63157894736842,
"precise_if": 33.125,
"math": 55.19125683060109,
"safety": 50.888888888888886,
"focus": 60.80808080808081
},
{
"model": "Nexusflow/Starling-RM-34B",
"ece": 0.6136179212108072,
"n_examples": 1865,
"score": 45.53255691061264,
"factuality": 45.89473684210526,
"precise_if": 31.875,
"math": 61.74863387978142,
"safety": 75.55555555555556,
"focus": 48.08080808080808
},
{
"model": "internlm/internlm2-1_8b-reward",
"ece": 0.6727187813665824,
"n_examples": 1865,
"score": 39.02376339707181,
"factuality": 27.57894736842105,
"precise_if": 36.25,
"math": 44.26229508196721,
"safety": 47.11111111111111,
"focus": 59.59595959595959
},
{
"model": "anthropic/claude-3-haiku-20240307",
"ece": 0.8125813921703674,
"n_examples": 102,
"score": 37.11004606544254,
"factuality": 40.42105263157895,
"precise_if": 28.125,
"math": 35.51912568306011,
"safety": 59.5,
"focus": 50.1010101010101
},
{
"model": "PKU-Alignment/beaver-7b-v1.0-cost",
"ece": 0.6213311133259143,
"n_examples": 1865,
"score": 33.32169419042497,
"factuality": 32.631578947368425,
"precise_if": 23.125,
"math": 39.89071038251366,
"safety": 75.88888888888889,
"focus": 29.393939393939394
},
{
"model": "PKU-Alignment/beaver-7b-v2.0-cost",
"ece": 0.6505662098496096,
"n_examples": 1865,
"score": 33.26126668695694,
"factuality": 37.89473684210527,
"precise_if": 27.500000000000004,
"math": 33.33333333333333,
"safety": 73.55555555555556,
"focus": 28.28282828282828
},
{
"model": "OpenAssistant/reward-model-deberta-v3-large-v2",
"ece": 0.6876232170863141,
"n_examples": 1865,
"score": 32.00225906524769,
"factuality": 38.526315789473685,
"precise_if": 26.875,
"math": 50.27322404371585,
"safety": 36.66666666666666,
"focus": 27.67676767676768
},
{
"model": "weqweasdas/RM-Gemma-2B",
"ece": 0.7263672822485251,
"n_examples": 1865,
"score": 30.566279281365752,
"factuality": 37.05263157894737,
"precise_if": 28.125,
"math": 43.16939890710383,
"safety": 33.111111111111114,
"focus": 23.43434343434344
},
{
"model": "OpenAssistant/oasst-rm-2-pythia-6.9b-epoch-1",
"ece": 0.7593884215827624,
"n_examples": 1865,
"score": 26.53180836792417,
"factuality": 39.78947368421053,
"precise_if": 28.75,
"math": 37.704918032786885,
"safety": 32.88888888888889,
"focus": 15.353535353535351
},
{
"model": "OpenAssistant/oasst-rm-2.1-pythia-1.4b-epoch-2.5",
"ece": 0.7202181761332691,
"n_examples": 1865,
"score": 26.47938178923201,
"factuality": 31.789473684210527,
"precise_if": 26.25,
"math": 39.34426229508197,
"safety": 32.44444444444444,
"focus": 27.070707070707076
},
{
"model": "PKU-Alignment/beaver-7b-v2.0-reward",
"ece": 0.756975347321237,
"n_examples": 1865,
"score": 25.435434978750283,
"factuality": 21.684210526315788,
"precise_if": 25.625,
"math": 38.25136612021858,
"safety": 31.555555555555557,
"focus": 26.060606060606062
},
{
"model": "weqweasdas/hh_rlhf_rm_open_llama_3b",
"ece": 0.7492188305498545,
"n_examples": 1865,
"score": 24.98045597204856,
"factuality": 36.42105263157895,
"precise_if": 27.500000000000004,
"math": 34.97267759562842,
"safety": 24.0,
"focus": 23.83838383838384
},
{
"model": "PKU-Alignment/beaver-7b-v1.0-reward",
"ece": 0.819336116115328,
"n_examples": 1865,
"score": 16.05736504908619,
"factuality": 21.052631578947366,
"precise_if": 29.375,
"math": 26.229508196721312,
"safety": 14.22222222222222,
"focus": 6.4646464646464645
}
],
"verdicts": {
"claim_1": "supported_exact_fraction_and_random_numerical_audit",
"claim_2": "supported_with_table_numbering_correction",
"claim_3": "supported_numerical_implication_audit_not_proof",
"claim_4": "supported_with_section_scope_clarification",
"claim_5": "supported_public_data_recomputation"
},
"scope": "Exact-fraction theory audit; official immutable CSV aggregation; full available RewardBench2 score recomputation. No model training.",
"wall_time_seconds": 32.77384108499973
}

Xet Storage Details

Size:
26.1 kB
·
Xet hash:
8fbe2d9b8b96a56eaef3d214e2a7efd7f7cec16d5b9fd3ec4847a977f1cf450b

Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.