===== MB round-1 task=helpsteer2 anchor=/NHNHOME/AIPR/sjkim/gtpref/helpsteer2/exp/rmpool_r2/train/ks weights={"helpfulness":0.2500,"correctness":0.2500,"coherence":0.2500,"conciseness":0.2500} =====
manifest 2500
allresp built
scores combined {'pol_42': 2500, 'pol_43': 2500, 'base_44': 2500, 'base_45': 2500}
{"surplus(2P-1)": {"helpfulness": 0.008, "correctness": 0.014, "coherence": 0.009, "conciseness": -0.022}, "P(pi>mu)": {"helpfulness": 0.504, "correctness": 0.507, "coherence": 0.504, "conciseness": 0.489}, "w_nbs": {"helpfulness": 0.0, "correctness": 0.0, "coherence": 0.0, "conciseness": 0.999}, "w_ks": {"helpfulness": 0.247, "correctness": 0.225, "coherence": 0.244, "conciseness": 0.284}, "worst": "conciseness", "n": 5000}
RMPOOL_helpsteer2_DONE
[build_bon] 2500 pairs (train 2250/test 250) weights {'helpfulness': 0.25, 'correctness': 0.25, 'coherence': 0.25, 'conciseness': 0.25}
[prep helpsteer2 EXP=exp/rmpool_mbhtcohe441] objs: helpfulness correctness coherence conciseness parent: /NHNHOME/AIPR/sjkim/gtpref/helpsteer2/exp/rmpool_r2/train/ks

Map:   0%|          | 0/2250 [00:00<?, ? examples/s]
Map:  89%|████████▉ | 2000/2250 [00:00<00:00, 16490.90 examples/s]
Map: 100%|██████████| 2250/2250 [00:00<00:00, 15750.03 examples/s]

Map:   0%|          | 0/250 [00:00<?, ? examples/s]
Map: 100%|██████████| 250/250 [00:00<00:00, 12619.46 examples/s]

Saving the dataset (0/1 shards):   0%|          | 0/2250 [00:00<?, ? examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 2250/2250 [00:00<00:00, 193262.15 examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 2250/2250 [00:00<00:00, 191059.32 examples/s]

Saving the dataset (0/1 shards):   0%|          | 0/250 [00:00<?, ? examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 250/250 [00:00<00:00, 43066.21 examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 250/250 [00:00<00:00, 41797.58 examples/s]

Saving the dataset (0/1 shards):   0%|          | 0/2250 [00:00<?, ? examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 2250/2250 [00:00<00:00, 270755.53 examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 2250/2250 [00:00<00:00, 267691.16 examples/s]

Saving the dataset (0/1 shards):   0%|          | 0/250 [00:00<?, ? examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 250/250 [00:00<00:00, 59561.26 examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 250/250 [00:00<00:00, 58468.61 examples/s]
[prep] pre_pref + pre_htobj built
dpo htmnpo_coherence htmnpo_conciseness htmnpo_correctness htmnpo_helpfulness inpo ipo ks maxmin nbs simpo unif 
TRAIN_DONE aiprlab-polymer
===== MB round-2 task=helpsteer2 anchor=/NHNHOME/AIPR/sjkim/gtpref/helpsteer2/exp/rmpool_mbhtcohe441/train/htmnpo_coherence weights={"helpfulness":0.2500,"correctness":0.2500,"coherence":0.2500,"conciseness":0.2500} =====
manifest 2500
allresp built
scores combined {'pol_42': 2500, 'pol_43': 2500, 'base_44': 2500, 'base_45': 2500}
{"surplus(2P-1)": {"helpfulness": -0.013, "correctness": -0.01, "coherence": -0.012, "conciseness": 0.007}, "P(pi>mu)": {"helpfulness": 0.494, "correctness": 0.495, "coherence": 0.494, "conciseness": 0.504}, "w_nbs": {"helpfulness": 0.333, "correctness": 0.333, "coherence": 0.333, "conciseness": 0.0}, "w_ks": {"helpfulness": 0.257, "correctness": 0.257, "coherence": 0.257, "conciseness": 0.228}, "worst": "helpfulness", "n": 5000}
RMPOOL_helpsteer2_DONE
[build_bon] 2500 pairs (train 2250/test 250) weights {'helpfulness': 0.25, 'correctness': 0.25, 'coherence': 0.25, 'conciseness': 0.25}
[prep helpsteer2 EXP=exp/rmpool_mbhtcohe442] objs: helpfulness correctness coherence conciseness parent: /NHNHOME/AIPR/sjkim/gtpref/helpsteer2/exp/rmpool_mbhtcohe441/train/htmnpo_coherence

Map:   0%|          | 0/2250 [00:00<?, ? examples/s]
Map:  89%|████████▉ | 2000/2250 [00:00<00:00, 18489.41 examples/s]
Map: 100%|██████████| 2250/2250 [00:00<00:00, 17806.38 examples/s]

Map:   0%|          | 0/250 [00:00<?, ? examples/s]
Map: 100%|██████████| 250/250 [00:00<00:00, 13320.83 examples/s]

Saving the dataset (0/1 shards):   0%|          | 0/2250 [00:00<?, ? examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 2250/2250 [00:00<00:00, 203229.91 examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 2250/2250 [00:00<00:00, 201774.26 examples/s]

Saving the dataset (0/1 shards):   0%|          | 0/250 [00:00<?, ? examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 250/250 [00:00<00:00, 52840.96 examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 250/250 [00:00<00:00, 52046.26 examples/s]

Saving the dataset (0/1 shards):   0%|          | 0/2250 [00:00<?, ? examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 2250/2250 [00:00<00:00, 295632.60 examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 2250/2250 [00:00<00:00, 292716.63 examples/s]

Saving the dataset (0/1 shards):   0%|          | 0/250 [00:00<?, ? examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 250/250 [00:00<00:00, 58892.22 examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 250/250 [00:00<00:00, 57903.58 examples/s]
[prep] pre_pref + pre_htobj built
dpo htmnpo_coherence htmnpo_conciseness htmnpo_correctness htmnpo_helpfulness inpo ipo ks maxmin nbs simpo unif 
TRAIN_DONE aiprlab-polymer
===== MB round-3 task=helpsteer2 anchor=/NHNHOME/AIPR/sjkim/gtpref/helpsteer2/exp/rmpool_mbhtcohe442/train/htmnpo_coherence weights={"helpfulness":0.2500,"correctness":0.2500,"coherence":0.2500,"conciseness":0.2500} =====
manifest 2500
allresp built
scores combined {'pol_42': 2500, 'pol_43': 2500, 'base_44': 2500, 'base_45': 2500}
{"surplus(2P-1)": {"helpfulness": -0.011, "correctness": -0.003, "coherence": 0.0, "conciseness": 0.001}, "P(pi>mu)": {"helpfulness": 0.494, "correctness": 0.499, "coherence": 0.5, "conciseness": 0.501}, "w_nbs": {"helpfulness": 0.497, "correctness": 0.497, "coherence": 0.005, "conciseness": 0.001}, "w_ks": {"helpfulness": 0.251, "correctness": 0.251, "coherence": 0.251, "conciseness": 0.246}, "worst": "helpfulness", "n": 5000}
RMPOOL_helpsteer2_DONE
[build_bon] 2500 pairs (train 2250/test 250) weights {'helpfulness': 0.25, 'correctness': 0.25, 'coherence': 0.25, 'conciseness': 0.25}
[prep helpsteer2 EXP=exp/rmpool_mbhtcohe443] objs: helpfulness correctness coherence conciseness parent: /NHNHOME/AIPR/sjkim/gtpref/helpsteer2/exp/rmpool_mbhtcohe442/train/htmnpo_coherence

Map:   0%|          | 0/2250 [00:00<?, ? examples/s]
Map:  89%|████████▉ | 2000/2250 [00:00<00:00, 17951.61 examples/s]
Map: 100%|██████████| 2250/2250 [00:00<00:00, 17170.69 examples/s]

Map:   0%|          | 0/250 [00:00<?, ? examples/s]
Map: 100%|██████████| 250/250 [00:00<00:00, 13531.76 examples/s]

Saving the dataset (0/1 shards):   0%|          | 0/2250 [00:00<?, ? examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 2250/2250 [00:00<00:00, 199792.19 examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 2250/2250 [00:00<00:00, 198239.34 examples/s]

Saving the dataset (0/1 shards):   0%|          | 0/250 [00:00<?, ? examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 250/250 [00:00<00:00, 52266.77 examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 250/250 [00:00<00:00, 51466.38 examples/s]

Saving the dataset (0/1 shards):   0%|          | 0/2250 [00:00<?, ? examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 2250/2250 [00:00<00:00, 298229.81 examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 2250/2250 [00:00<00:00, 295133.35 examples/s]

Saving the dataset (0/1 shards):   0%|          | 0/250 [00:00<?, ? examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 250/250 [00:00<00:00, 37305.25 examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 250/250 [00:00<00:00, 36904.80 examples/s]
[prep] pre_pref + pre_htobj built
dpo htmnpo_coherence htmnpo_conciseness htmnpo_correctness htmnpo_helpfulness inpo ipo ks maxmin nbs simpo unif 
TRAIN_DONE aiprlab-polymer
===== MB round-4 task=helpsteer2 anchor=/NHNHOME/AIPR/sjkim/gtpref/helpsteer2/exp/rmpool_mbhtcohe443/train/htmnpo_coherence weights={"helpfulness":0.2500,"correctness":0.2500,"coherence":0.2500,"conciseness":0.2500} =====
manifest 2500
allresp built
scores combined {'pol_42': 2500, 'pol_43': 2500, 'base_44': 2500, 'base_45': 2500}
{"surplus(2P-1)": {"helpfulness": 0.007, "correctness": 0.005, "coherence": 0.01, "conciseness": -0.0}, "P(pi>mu)": {"helpfulness": 0.503, "correctness": 0.503, "coherence": 0.505, "conciseness": 0.5}, "w_nbs": {"helpfulness": 0.0, "correctness": 0.0, "coherence": 0.0, "conciseness": 0.999}, "w_ks": {"helpfulness": 0.245, "correctness": 0.251, "coherence": 0.231, "conciseness": 0.274}, "worst": "conciseness", "n": 5000}
RMPOOL_helpsteer2_DONE
[build_bon] 2500 pairs (train 2250/test 250) weights {'helpfulness': 0.25, 'correctness': 0.25, 'coherence': 0.25, 'conciseness': 0.25}
[prep helpsteer2 EXP=exp/rmpool_mbhtcohe444] objs: helpfulness correctness coherence conciseness parent: /NHNHOME/AIPR/sjkim/gtpref/helpsteer2/exp/rmpool_mbhtcohe443/train/htmnpo_coherence

Map:   0%|          | 0/2250 [00:00<?, ? examples/s]
Map:  89%|████████▉ | 2000/2250 [00:00<00:00, 18005.05 examples/s]
Map: 100%|██████████| 2250/2250 [00:00<00:00, 17432.65 examples/s]

Map:   0%|          | 0/250 [00:00<?, ? examples/s]
Map: 100%|██████████| 250/250 [00:00<00:00, 13695.78 examples/s]

Saving the dataset (0/1 shards):   0%|          | 0/2250 [00:00<?, ? examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 2250/2250 [00:00<00:00, 215092.51 examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 2250/2250 [00:00<00:00, 213433.69 examples/s]

Saving the dataset (0/1 shards):   0%|          | 0/250 [00:00<?, ? examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 250/250 [00:00<00:00, 52256.35 examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 250/250 [00:00<00:00, 51453.75 examples/s]

Saving the dataset (0/1 shards):   0%|          | 0/2250 [00:00<?, ? examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 2250/2250 [00:00<00:00, 299602.65 examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 2250/2250 [00:00<00:00, 296654.85 examples/s]

Saving the dataset (0/1 shards):   0%|          | 0/250 [00:00<?, ? examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 250/250 [00:00<00:00, 58770.09 examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 250/250 [00:00<00:00, 57791.89 examples/s]
[prep] pre_pref + pre_htobj built
dpo htmnpo_coherence htmnpo_conciseness htmnpo_correctness htmnpo_helpfulness inpo ipo ks maxmin nbs simpo unif 
TRAIN_DONE aiprlab-polymer
===== MB round-5 task=helpsteer2 anchor=/NHNHOME/AIPR/sjkim/gtpref/helpsteer2/exp/rmpool_mbhtcohe444/train/htmnpo_coherence weights={"helpfulness":0.2500,"correctness":0.2500,"coherence":0.2500,"conciseness":0.2500} =====
manifest 2500
allresp built
scores combined {'pol_42': 2500, 'pol_43': 2500, 'base_44': 2500, 'base_45': 2500}
{"surplus(2P-1)": {"helpfulness": 0.007, "correctness": 0.025, "coherence": -0.023, "conciseness": -0.005}, "P(pi>mu)": {"helpfulness": 0.503, "correctness": 0.512, "coherence": 0.488, "conciseness": 0.497}, "w_nbs": {"helpfulness": 0.0, "correctness": 0.0, "coherence": 0.5, "conciseness": 0.5}, "w_ks": {"helpfulness": 0.251, "correctness": 0.186, "coherence": 0.281, "conciseness": 0.281}, "worst": "coherence", "n": 5000}
RMPOOL_helpsteer2_DONE
[build_bon] 2500 pairs (train 2250/test 250) weights {'helpfulness': 0.25, 'correctness': 0.25, 'coherence': 0.25, 'conciseness': 0.25}
[prep helpsteer2 EXP=exp/rmpool_mbhtcohe445] objs: helpfulness correctness coherence conciseness parent: /NHNHOME/AIPR/sjkim/gtpref/helpsteer2/exp/rmpool_mbhtcohe444/train/htmnpo_coherence

Map:   0%|          | 0/2250 [00:00<?, ? examples/s]
Map:  89%|████████▉ | 2000/2250 [00:00<00:00, 17942.47 examples/s]
Map: 100%|██████████| 2250/2250 [00:00<00:00, 17376.29 examples/s]

Map:   0%|          | 0/250 [00:00<?, ? examples/s]
Map: 100%|██████████| 250/250 [00:00<00:00, 13604.97 examples/s]

Saving the dataset (0/1 shards):   0%|          | 0/2250 [00:00<?, ? examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 2250/2250 [00:00<00:00, 200816.79 examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 2250/2250 [00:00<00:00, 199403.81 examples/s]

Saving the dataset (0/1 shards):   0%|          | 0/250 [00:00<?, ? examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 250/250 [00:00<00:00, 52465.53 examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 250/250 [00:00<00:00, 51659.08 examples/s]

Saving the dataset (0/1 shards):   0%|          | 0/2250 [00:00<?, ? examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 2250/2250 [00:00<00:00, 295651.13 examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 2250/2250 [00:00<00:00, 292771.11 examples/s]

Saving the dataset (0/1 shards):   0%|          | 0/250 [00:00<?, ? examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 250/250 [00:00<00:00, 57919.58 examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 250/250 [00:00<00:00, 56969.25 examples/s]
[prep] pre_pref + pre_htobj built
dpo htmnpo_coherence htmnpo_conciseness htmnpo_correctness htmnpo_helpfulness inpo ipo ks maxmin nbs simpo unif 
TRAIN_DONE aiprlab-polymer
===== MB round-6 task=helpsteer2 anchor=/NHNHOME/AIPR/sjkim/gtpref/helpsteer2/exp/rmpool_mbhtcohe445/train/htmnpo_coherence weights={"helpfulness":0.2500,"correctness":0.2500,"coherence":0.2500,"conciseness":0.2500} =====
manifest 2500
allresp built
scores combined {'pol_42': 2500, 'pol_43': 2500, 'base_44': 2500, 'base_45': 2500}
{"surplus(2P-1)": {"helpfulness": 0.006, "correctness": 0.017, "coherence": 0.001, "conciseness": 0.031}, "P(pi>mu)": {"helpfulness": 0.503, "correctness": 0.508, "coherence": 0.5, "conciseness": 0.515}, "w_nbs": {"helpfulness": 0.132, "correctness": 0.047, "coherence": 0.794, "conciseness": 0.026}, "w_ks": {"helpfulness": 0.279, "correctness": 0.233, "coherence": 0.303, "conciseness": 0.185}, "worst": "coherence", "n": 5000}
RMPOOL_helpsteer2_DONE
[build_bon] 2500 pairs (train 2250/test 250) weights {'helpfulness': 0.25, 'correctness': 0.25, 'coherence': 0.25, 'conciseness': 0.25}
[prep helpsteer2 EXP=exp/rmpool_mbhtcohe446] objs: helpfulness correctness coherence conciseness parent: /NHNHOME/AIPR/sjkim/gtpref/helpsteer2/exp/rmpool_mbhtcohe445/train/htmnpo_coherence

Map:   0%|          | 0/2250 [00:00<?, ? examples/s]
Map:  89%|████████▉ | 2000/2250 [00:00<00:00, 17928.02 examples/s]
Map: 100%|██████████| 2250/2250 [00:00<00:00, 17184.76 examples/s]

Map:   0%|          | 0/250 [00:00<?, ? examples/s]
Map: 100%|██████████| 250/250 [00:00<00:00, 10285.40 examples/s]

Saving the dataset (0/1 shards):   0%|          | 0/2250 [00:00<?, ? examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 2250/2250 [00:00<00:00, 187078.68 examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 2250/2250 [00:00<00:00, 185756.71 examples/s]

Saving the dataset (0/1 shards):   0%|          | 0/250 [00:00<?, ? examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 250/250 [00:00<00:00, 51390.71 examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 250/250 [00:00<00:00, 50349.37 examples/s]

Saving the dataset (0/1 shards):   0%|          | 0/2250 [00:00<?, ? examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 2250/2250 [00:00<00:00, 288607.73 examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 2250/2250 [00:00<00:00, 285395.83 examples/s]

Saving the dataset (0/1 shards):   0%|          | 0/250 [00:00<?, ? examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 250/250 [00:00<00:00, 59291.83 examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 250/250 [00:00<00:00, 58202.49 examples/s]
[prep] pre_pref + pre_htobj built
dpo htmnpo_coherence htmnpo_conciseness htmnpo_correctness htmnpo_helpfulness inpo ipo ks maxmin nbs simpo unif 
TRAIN_DONE aiprlab-polymer
===== MB round-7 task=helpsteer2 anchor=/NHNHOME/AIPR/sjkim/gtpref/helpsteer2/exp/rmpool_mbhtcohe446/train/htmnpo_coherence weights={"helpfulness":0.2500,"correctness":0.2500,"coherence":0.2500,"conciseness":0.2500} =====
manifest 2500
allresp built
scores combined {'pol_42': 2500, 'pol_43': 2500, 'base_44': 2500, 'base_45': 2500}
{"surplus(2P-1)": {"helpfulness": 0.005, "correctness": 0.007, "coherence": -0.006, "conciseness": 0.006}, "P(pi>mu)": {"helpfulness": 0.502, "correctness": 0.503, "coherence": 0.497, "conciseness": 0.503}, "w_nbs": {"helpfulness": 0.0, "correctness": 0.0, "coherence": 0.999, "conciseness": 0.0}, "w_ks": {"helpfulness": 0.248, "correctness": 0.239, "coherence": 0.268, "conciseness": 0.244}, "worst": "coherence", "n": 5000}
RMPOOL_helpsteer2_DONE
[build_bon] 2500 pairs (train 2250/test 250) weights {'helpfulness': 0.25, 'correctness': 0.25, 'coherence': 0.25, 'conciseness': 0.25}
[prep helpsteer2 EXP=exp/rmpool_mbhtcohe447] objs: helpfulness correctness coherence conciseness parent: /NHNHOME/AIPR/sjkim/gtpref/helpsteer2/exp/rmpool_mbhtcohe446/train/htmnpo_coherence

Map:   0%|          | 0/2250 [00:00<?, ? examples/s]
Map:  89%|████████▉ | 2000/2250 [00:00<00:00, 17894.67 examples/s]
Map: 100%|██████████| 2250/2250 [00:00<00:00, 17346.62 examples/s]

Map:   0%|          | 0/250 [00:00<?, ? examples/s]
Map: 100%|██████████| 250/250 [00:00<00:00, 13807.95 examples/s]

Saving the dataset (0/1 shards):   0%|          | 0/2250 [00:00<?, ? examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 2250/2250 [00:00<00:00, 205953.12 examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 2250/2250 [00:00<00:00, 204436.20 examples/s]

Saving the dataset (0/1 shards):   0%|          | 0/250 [00:00<?, ? examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 250/250 [00:00<00:00, 51312.75 examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 250/250 [00:00<00:00, 50548.40 examples/s]

Saving the dataset (0/1 shards):   0%|          | 0/2250 [00:00<?, ? examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 2250/2250 [00:00<00:00, 290956.81 examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 2250/2250 [00:00<00:00, 288043.95 examples/s]

Saving the dataset (0/1 shards):   0%|          | 0/250 [00:00<?, ? examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 250/250 [00:00<00:00, 58599.31 examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 250/250 [00:00<00:00, 57633.07 examples/s]
[prep] pre_pref + pre_htobj built
dpo htmnpo_coherence htmnpo_conciseness htmnpo_correctness htmnpo_helpfulness inpo ipo ks maxmin nbs simpo unif 
TRAIN_DONE aiprlab-polymer
{
  "label": "qwen32_helpsteer2_htmnpo_coherence",
  "n_objectives": 4,
  "surplus": {
    "helpfulness": 0.115,
    "correctness": 0.0821,
    "coherence": 0.0821,
    "conciseness": 0.0771
  },
  "hoeffding_margin": {
    "helpfulness": 0.0356,
    "correctness": 0.0359,
    "coherence": 0.0356,
    "conciseness": 0.0356
  },
  "worst_surplus": 0.0771,
  "individual_rationality": true,
  "nash_welfare_logsum": -9.724
}
EVAL_helpsteer2_DONE aiprlab-polymer
MB_STAGES_DONE