===== MB round-1 task=hs2gemma anchor=/NHNHOME/AIPR/sjkim/policy_bases/models--google--gemma-2-9b-it/snapshots/11c9b309abf73637e4b6f9a3fa1e92e615547819/ weights={"helpfulness": 0.25, "correctness": 0.25, "coherence": 0.25, "conciseness": 0.25} =====
allresp built
scores combined {'pol_42': 2500, 'pol_43': 2500, 'base_44': 2500, 'base_45': 2500}
{"surplus(2P-1)": {"helpfulness": -0.028, "correctness": -0.027, "coherence": -0.019, "conciseness": 0.002}, "P(pi>mu)": {"helpfulness": 0.486, "correctness": 0.486, "coherence": 0.491, "conciseness": 0.501}, "w_nbs": {"helpfulness": 0.333, "correctness": 0.333, "coherence": 0.333, "conciseness": 0.0}, "w_ks": {"helpfulness": 0.252, "correctness": 0.252, "coherence": 0.252, "conciseness": 0.245}, "worst": "helpfulness", "n": 5000}
RMPOOL_hs2gemma_DONE
[build_bon] 2500 pairs (train 2250/test 250) weights {'helpfulness': 0.25, 'correctness': 0.25, 'coherence': 0.25, 'conciseness': 0.25}
[prep hs2gemma EXP=exp/rmpool_mbgnb1] objs: helpfulness correctness coherence conciseness parent: /NHNHOME/AIPR/sjkim/policy_bases/models--google--gemma-2-9b-it/snapshots/11c9b309abf73637e4b6f9a3fa1e92e615547819/

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, 83560.01 examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 2250/2250 [00:00<00:00, 83030.68 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, 24180.80 examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 250/250 [00:00<00:00, 23958.14 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, 83498.65 examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 2250/2250 [00:00<00:00, 83200.51 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, 13820.33 examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 250/250 [00:00<00:00, 13756.33 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-1 task=hs2gemma anchor=/NHNHOME/AIPR/sjkim/policy_bases/models--google--gemma-2-9b-it/snapshots/11c9b309abf73637e4b6f9a3fa1e92e615547819/ weights={"helpfulness": 0.25, "correctness": 0.25, "coherence": 0.25, "conciseness": 0.25} =====
allresp built
scores combined {'pol_42': 2500, 'pol_43': 2500, 'base_44': 2500, 'base_45': 2500}
{"surplus(2P-1)": {"helpfulness": -0.028, "correctness": -0.027, "coherence": -0.019, "conciseness": 0.002}, "P(pi>mu)": {"helpfulness": 0.486, "correctness": 0.486, "coherence": 0.491, "conciseness": 0.501}, "w_nbs": {"helpfulness": 0.333, "correctness": 0.333, "coherence": 0.333, "conciseness": 0.0}, "w_ks": {"helpfulness": 0.252, "correctness": 0.252, "coherence": 0.252, "conciseness": 0.245}, "worst": "helpfulness", "n": 5000}
RMPOOL_hs2gemma_DONE
[build_bon] 2500 pairs (train 2250/test 250) weights {'helpfulness': 0.25, 'correctness': 0.25, 'coherence': 0.25, 'conciseness': 0.25}
[prep hs2gemma EXP=exp/rmpool_mbgnb1] objs: helpfulness correctness coherence conciseness parent: /NHNHOME/AIPR/sjkim/policy_bases/models--google--gemma-2-9b-it/snapshots/11c9b309abf73637e4b6f9a3fa1e92e615547819/

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, 92844.55 examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 2250/2250 [00:00<00:00, 92324.09 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, 23247.44 examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 250/250 [00:00<00:00, 23047.65 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, 96304.67 examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 2250/2250 [00:00<00:00, 95938.52 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, 24777.90 examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 250/250 [00:00<00:00, 24574.08 examples/s]
[prep] pre_pref + pre_htobj built
dpo htmnpo_coherence htmnpo_conciseness htmnpo_correctness htmnpo_helpfulness inpo ipo ks maxmin nbs simpo unif 
/NHNHOME/AIPR/sjkim/gtpref/train_gpu.sh: line 15: 963757 Killed                  CUDA_VISIBLE_DEVICES=${GA[$((g%NGA))]} $V -m accelerate.commands.launch --config_file $P/accelerate_configs/single_gpu.yaml --num_processes=1 --main_process_port=$((PB+g)) -m mnpo_scripts.run_mnpo $OUT/config.yaml > $OUT/train.log 2>&1
TRAIN_DONE aiprlab-polymer
===== MB round-1 task=hs2gemma anchor=/NHNHOME/AIPR/sjkim/policy_bases/models--google--gemma-2-9b-it/snapshots/11c9b309abf73637e4b6f9a3fa1e92e615547819/ weights={"helpfulness": 0.25, "correctness": 0.25, "coherence": 0.25, "conciseness": 0.25} =====
allresp built
scores combined {'pol_42': 2500, 'pol_43': 2500, 'base_44': 2500, 'base_45': 2500}
{"surplus(2P-1)": {"helpfulness": -0.028, "correctness": -0.027, "coherence": -0.019, "conciseness": 0.002}, "P(pi>mu)": {"helpfulness": 0.486, "correctness": 0.486, "coherence": 0.491, "conciseness": 0.501}, "w_nbs": {"helpfulness": 0.333, "correctness": 0.333, "coherence": 0.333, "conciseness": 0.0}, "w_ks": {"helpfulness": 0.252, "correctness": 0.252, "coherence": 0.252, "conciseness": 0.245}, "worst": "helpfulness", "n": 5000}
RMPOOL_hs2gemma_DONE
[build_bon] 2500 pairs (train 2250/test 250) weights {'helpfulness': 0.25, 'correctness': 0.25, 'coherence': 0.25, 'conciseness': 0.25}
[prep hs2gemma EXP=exp/rmpool_mbgnb1] objs: helpfulness correctness coherence conciseness parent: /NHNHOME/AIPR/sjkim/policy_bases/models--google--gemma-2-9b-it/snapshots/11c9b309abf73637e4b6f9a3fa1e92e615547819/

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, 158005.32 examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 2250/2250 [00:00<00:00, 156519.45 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, 50334.87 examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 250/250 [00:00<00:00, 49456.47 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, 204951.22 examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 2250/2250 [00:00<00:00, 203396.35 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, 55000.05 examples/s]
Saving the dataset (1/1 shards): 100%|██████████| 250/250 [00:00<00:00, 54030.81 examples/s]
[prep] pre_pref + pre_htobj built
dpo htmnpo_coherence htmnpo_conciseness htmnpo_correctness htmnpo_helpfulness inpo ipo ks maxmin nbs simpo unif 
===== MB round-1 task=hs2gemma anchor=/NHNHOME/AIPR/sjkim/policy_bases/models--google--gemma-2-9b-it/snapshots/11c9b309abf73637e4b6f9a3fa1e92e615547819/ weights={"helpfulness": 0.25, "correctness": 0.25, "coherence": 0.25, "conciseness": 0.25} =====
allresp built
/NHNHOME/AIPR/sjkim/gtpref/train_gpu.sh: line 17: syntax error near unexpected token `)'
/NHNHOME/AIPR/sjkim/gtpref/train_gpu.sh: line 17: `q 1 $MAXSEM); do'
cp: cannot stat '/NHNHOME/AIPR/sjkim/gtpref/hs2gemma/exp/rmpool_mbgnb1/eval/nbs/xj_qwen32/vd/shard0.jsonl': No such file or directory
Traceback (most recent call last):
  File "/NHNHOME/AIPR/sjkim/MNPO_rev_20260710/scripts/bpo/eval_bpo_surplus.py", line 55, in <module>
    main()
  File "/NHNHOME/AIPR/sjkim/MNPO_rev_20260710/scripts/bpo/eval_bpo_surplus.py", line 23, in main
    rows = [json.loads(l) for l in args.verdicts.read_text().splitlines() if l.strip()]
                                   ^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/lib/python3.12/pathlib.py", line 1029, in read_text
    with self.open(mode='r', encoding=encoding, errors=errors) as f:
         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/lib/python3.12/pathlib.py", line 1015, in open
    return io.open(self, mode, buffering, encoding, errors, newline)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
FileNotFoundError: [Errno 2] No such file or directory: '/NHNHOME/AIPR/sjkim/gtpref/hs2gemma/exp/rmpool_mbgnb1/eval/nbs/xj_qwen32/verdicts.jsonl'
EVAL_hs2gemma_DONE aiprlab-polymer
Traceback (most recent call last):
  File "/NHNHOME/AIPR/sjkim/gtpref/ma_weights.py", line 4, in <module>
    s = json.load(open(sys.argv[1]))["surplus"]
                  ^^^^^^^^^^^^^^^^^
FileNotFoundError: [Errno 2] No such file or directory: '/NHNHOME/AIPR/sjkim/gtpref/hs2gemma/exp/rmpool_mbgnb1/eval/nbs/xj_qwen32/surplus.json'
===== MB round-2 task=hs2gemma anchor=/NHNHOME/AIPR/sjkim/gtpref/hs2gemma/exp/rmpool_mbgnb1/train/nbs weights= =====
manifest 2500
allresp built
/NHNHOME/AIPR/sjkim/gtpref/rmpool_seq.sh: line 56: 1599415 Killed                  CUDA_VISIBLE_DEVICES=${G[$((i%NG))]} $V $P/on_policy_data_gen/rm_armo.py --cache_dir /NHNHOME/AIPR/sjkim/xj_judges --input_file $POOL/allresp.jsonl --output_file $POOL/sc_$obj.jsonl --reward_attribute_name $attr --batch_size 16 --local_files_only > $POOL/rm_$obj.log 2>&1
scores combined {'pol_42': 2500, 'pol_43': 2500, 'base_44': 2500, 'base_45': 2500}
Traceback (most recent call last):
  File "/NHNHOME/AIPR/sjkim/gtpref/build_rmpool.py", line 75, in <module>
    main()
  File "/NHNHOME/AIPR/sjkim/gtpref/build_rmpool.py", line 38, in main
    dv = pol[pid]["scores"][o] - base[pid]["scores"][o]
                                 ~~~~~~~~~~~~~~~~~~~^^^
KeyError: 'conciseness'
RMPOOL_hs2gemma_DONE