{
  "mode": "dense",
  "args": {
    "mode": "dense",
    "supervision": "full_belief",
    "seed": 53,
    "modulus": 11,
    "observe_mod": 4,
    "observe_prob": 0.3,
    "op_family": "full",
    "query_family": "all",
    "train_min_len": 1,
    "train_max_len": 6,
    "eval_lengths": "3,6,9,12",
    "eval_k": "0,1,2,3,6,9,12",
    "eval_query_types": "all",
    "state_dim": 256,
    "instr_dim": 128,
    "heads": 4,
    "compiler_layers": 2,
    "dropout": 0.0,
    "train_steps": 1200,
    "batch_size": 256,
    "eval_batch_size": 256,
    "eval_examples": 256,
    "lr": 0.001,
    "weight_decay": 0.0001,
    "query_loss_weight": 1.0,
    "belief_loss_weight": 1.0,
    "belief_distill_prob": 0.1,
    "probe_steps": 600,
    "probe_batch_size": 256,
    "probe_lr": 0.001,
    "probe_weight_decay": 0.0001,
    "probe_dropout": 0.0,
    "max_grad_norm": 1.0,
    "log_every": 200,
    "probe_log_every": 100,
    "output_dir": "experiments/dense_supervision_ladder/runs/pilot_full_belief_mod11",
    "checkpoint_dir": "large_artifacts/dense_supervision_ladder/checkpoints/pilot_full_belief_mod11"
  },
  "env": {
    "python": "3.12.3",
    "torch": "2.8.0+cu128",
    "device": "cuda",
    "cuda_name": "NVIDIA RTX 6000 Ada Generation"
  },
  "train": [
    {
      "step": 1,
      "loss": 7.226489543914795,
      "grad_norm": 0.5122162699699402,
      "query_loss": 2.413424491882324,
      "belief_loss": 4.813065052032471,
      "query_target_mass": 0.44033393263816833,
      "query_target_nll": 2.303586959838867,
      "query_top1_on_support": 0.59375,
      "mean_query_support_size": 4.65625
    },
    {
      "step": 200,
      "loss": 4.791253089904785,
      "grad_norm": 0.8173497915267944,
      "query_loss": 1.9170970916748047,
      "belief_loss": 2.8741557598114014,
      "query_target_mass": 0.6544480323791504,
      "query_target_nll": 1.944364070892334,
      "query_top1_on_support": 0.765625,
      "mean_query_support_size": 5.20703125
    },
    {
      "step": 400,
      "loss": 3.843062400817871,
      "grad_norm": 0.644486665725708,
      "query_loss": 1.613739252090454,
      "belief_loss": 2.229323148727417,
      "query_target_mass": 0.7550052404403687,
      "query_target_nll": 1.6127427816390991,
      "query_top1_on_support": 0.84375,
      "mean_query_support_size": 4.734375
    },
    {
      "step": 600,
      "loss": 3.7295258045196533,
      "grad_norm": 0.533862292766571,
      "query_loss": 1.5852210521697998,
      "belief_loss": 2.1443047523498535,
      "query_target_mass": 0.8587566614151001,
      "query_target_nll": 1.4189207553863525,
      "query_top1_on_support": 0.93359375,
      "mean_query_support_size": 4.85546875
    },
    {
      "step": 800,
      "loss": 3.590330123901367,
      "grad_norm": 0.47575923800468445,
      "query_loss": 1.5686625242233276,
      "belief_loss": 2.021667718887329,
      "query_target_mass": 0.8622379302978516,
      "query_target_nll": 1.4819825887680054,
      "query_top1_on_support": 0.91015625,
      "mean_query_support_size": 5.1171875
    },
    {
      "step": 1000,
      "loss": 3.416537046432495,
      "grad_norm": 0.4992263615131378,
      "query_loss": 1.4648184776306152,
      "belief_loss": 1.9517185688018799,
      "query_target_mass": 0.8675879240036011,
      "query_target_nll": 1.3857005834579468,
      "query_top1_on_support": 0.921875,
      "mean_query_support_size": 4.8125
    },
    {
      "step": 1200,
      "loss": 3.4039080142974854,
      "grad_norm": 0.43736618757247925,
      "query_loss": 1.496151089668274,
      "belief_loss": 1.9077569246292114,
      "query_target_mass": 0.8766767978668213,
      "query_target_nll": 1.3544726371765137,
      "query_top1_on_support": 0.9296875,
      "mean_query_support_size": 4.62890625
    }
  ],
  "probe_train": [
    {
      "step": 1,
      "loss": 4.805259704589844,
      "grad_norm": 0.5472303032875061,
      "probe_belief_target_mass": 0.06045645847916603,
      "probe_belief_target_nll": 4.805259704589844,
      "probe_belief_top1_on_support": 0.063058041036129,
      "mean_belief_support_size": 7.3058037757873535
    },
    {
      "step": 100,
      "loss": 2.2044849395751953,
      "grad_norm": 0.3327436149120331,
      "probe_belief_target_mass": 0.8100622296333313,
      "probe_belief_target_nll": 2.2044849395751953,
      "probe_belief_top1_on_support": 0.8816964626312256,
      "mean_belief_support_size": 7.245535850524902
    },
    {
      "step": 200,
      "loss": 2.095029830932617,
      "grad_norm": 0.28763747215270996,
      "probe_belief_target_mass": 0.8677826523780823,
      "probe_belief_target_nll": 2.095029830932617,
      "probe_belief_top1_on_support": 0.9241071939468384,
      "mean_belief_support_size": 7.470424652099609
    },
    {
      "step": 300,
      "loss": 1.982529640197754,
      "grad_norm": 0.28961658477783203,
      "probe_belief_target_mass": 0.9019351005554199,
      "probe_belief_target_nll": 1.982529640197754,
      "probe_belief_top1_on_support": 0.9391741752624512,
      "mean_belief_support_size": 7.313058376312256
    },
    {
      "step": 400,
      "loss": 2.00742244720459,
      "grad_norm": 0.3328789472579956,
      "probe_belief_target_mass": 0.8787699341773987,
      "probe_belief_target_nll": 2.00742244720459,
      "probe_belief_top1_on_support": 0.926339328289032,
      "mean_belief_support_size": 7.136160850524902
    },
    {
      "step": 500,
      "loss": 1.9201231002807617,
      "grad_norm": 0.2759735584259033,
      "probe_belief_target_mass": 0.9098358750343323,
      "probe_belief_target_nll": 1.9201231002807617,
      "probe_belief_top1_on_support": 0.9553571939468384,
      "mean_belief_support_size": 7.076451301574707
    },
    {
      "step": 600,
      "loss": 1.9916402101516724,
      "grad_norm": 0.27841541171073914,
      "probe_belief_target_mass": 0.9051586985588074,
      "probe_belief_target_nll": 1.9916402101516724,
      "probe_belief_top1_on_support": 0.9436384439468384,
      "mean_belief_support_size": 7.327567100524902
    }
  ],
  "eval": [
    {
      "step": 1200,
      "rows": [
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 3,
          "query_type": "A",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.470497727394104,
          "query_target_nll": 2.4069674015045166,
          "query_top1_on_support": 0.48828125,
          "mean_query_support_size": 5.18359375,
          "probe_belief_target_mass": 0.06296727061271667,
          "probe_belief_target_nll": 14.820042610168457,
          "probe_belief_top1_on_support": 0.0625,
          "mean_belief_support_size": 5.84375,
          "decoder_belief_target_mass": 0.06389156728982925,
          "decoder_belief_target_nll": 18.714717864990234,
          "decoder_belief_top1_on_support": 0.0703125,
          "mean_decoder_belief_support_size": 5.84375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 3,
          "query_type": "A",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.5552994012832642,
          "query_target_nll": 3.796617031097412,
          "query_top1_on_support": 0.54296875,
          "mean_query_support_size": 5.18359375,
          "probe_belief_target_mass": 0.1206202358007431,
          "probe_belief_target_nll": 12.45261001586914,
          "probe_belief_top1_on_support": 0.109375,
          "mean_belief_support_size": 5.84375,
          "decoder_belief_target_mass": 0.12164768576622009,
          "decoder_belief_target_nll": 13.839357376098633,
          "decoder_belief_top1_on_support": 0.125,
          "mean_decoder_belief_support_size": 5.84375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 3,
          "query_type": "A",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.690788745880127,
          "query_target_nll": 3.174790859222412,
          "query_top1_on_support": 0.71875,
          "mean_query_support_size": 5.18359375,
          "probe_belief_target_mass": 0.2579003572463989,
          "probe_belief_target_nll": 8.520488739013672,
          "probe_belief_top1_on_support": 0.26171875,
          "mean_belief_support_size": 5.84375,
          "decoder_belief_target_mass": 0.25541239976882935,
          "decoder_belief_target_nll": 9.309812545776367,
          "decoder_belief_top1_on_support": 0.25,
          "mean_decoder_belief_support_size": 5.84375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 3,
          "query_type": "A",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.9369587302207947,
          "query_target_nll": 1.41509211063385,
          "query_top1_on_support": 0.98046875,
          "mean_query_support_size": 5.18359375,
          "probe_belief_target_mass": 0.8204923868179321,
          "probe_belief_target_nll": 1.9103777408599854,
          "probe_belief_top1_on_support": 0.8828125,
          "mean_belief_support_size": 5.84375,
          "decoder_belief_target_mass": 0.8313245177268982,
          "decoder_belief_target_nll": 1.8794565200805664,
          "decoder_belief_top1_on_support": 0.90625,
          "mean_decoder_belief_support_size": 5.84375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 3,
          "query_type": "A",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.9369587302207947,
          "query_target_nll": 1.41509211063385,
          "query_top1_on_support": 0.98046875,
          "mean_query_support_size": 5.18359375,
          "probe_belief_target_mass": 0.8204923868179321,
          "probe_belief_target_nll": 1.9103777408599854,
          "probe_belief_top1_on_support": 0.8828125,
          "mean_belief_support_size": 5.84375,
          "decoder_belief_target_mass": 0.8313245177268982,
          "decoder_belief_target_nll": 1.8794565200805664,
          "decoder_belief_top1_on_support": 0.90625,
          "mean_decoder_belief_support_size": 5.84375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 3,
          "query_type": "A",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.9369587302207947,
          "query_target_nll": 1.41509211063385,
          "query_top1_on_support": 0.98046875,
          "mean_query_support_size": 5.18359375,
          "probe_belief_target_mass": 0.8204923868179321,
          "probe_belief_target_nll": 1.9103777408599854,
          "probe_belief_top1_on_support": 0.8828125,
          "mean_belief_support_size": 5.84375,
          "decoder_belief_target_mass": 0.8313245177268982,
          "decoder_belief_target_nll": 1.8794565200805664,
          "decoder_belief_top1_on_support": 0.90625,
          "mean_decoder_belief_support_size": 5.84375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 3,
          "query_type": "A",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.9369587302207947,
          "query_target_nll": 1.41509211063385,
          "query_top1_on_support": 0.98046875,
          "mean_query_support_size": 5.18359375,
          "probe_belief_target_mass": 0.8204923868179321,
          "probe_belief_target_nll": 1.9103777408599854,
          "probe_belief_top1_on_support": 0.8828125,
          "mean_belief_support_size": 5.84375,
          "decoder_belief_target_mass": 0.8313245177268982,
          "decoder_belief_target_nll": 1.8794565200805664,
          "decoder_belief_top1_on_support": 0.90625,
          "mean_decoder_belief_support_size": 5.84375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 3,
          "query_type": "B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.46825993061065674,
          "query_target_nll": 2.407316207885742,
          "query_top1_on_support": 0.44140625,
          "mean_query_support_size": 5.15234375,
          "probe_belief_target_mass": 0.053678251802921295,
          "probe_belief_target_nll": 14.903712272644043,
          "probe_belief_top1_on_support": 0.03515625,
          "mean_belief_support_size": 5.77734375,
          "decoder_belief_target_mass": 0.05387695133686066,
          "decoder_belief_target_nll": 18.927295684814453,
          "decoder_belief_top1_on_support": 0.046875,
          "mean_decoder_belief_support_size": 5.77734375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 3,
          "query_type": "B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.536530613899231,
          "query_target_nll": 3.9269556999206543,
          "query_top1_on_support": 0.515625,
          "mean_query_support_size": 5.15234375,
          "probe_belief_target_mass": 0.1334231197834015,
          "probe_belief_target_nll": 11.925636291503906,
          "probe_belief_top1_on_support": 0.13671875,
          "mean_belief_support_size": 5.77734375,
          "decoder_belief_target_mass": 0.13452015817165375,
          "decoder_belief_target_nll": 13.162147521972656,
          "decoder_belief_top1_on_support": 0.125,
          "mean_decoder_belief_support_size": 5.77734375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 3,
          "query_type": "B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.6807620525360107,
          "query_target_nll": 3.3976690769195557,
          "query_top1_on_support": 0.6875,
          "mean_query_support_size": 5.15234375,
          "probe_belief_target_mass": 0.2703877091407776,
          "probe_belief_target_nll": 8.57983684539795,
          "probe_belief_top1_on_support": 0.2734375,
          "mean_belief_support_size": 5.77734375,
          "decoder_belief_target_mass": 0.2695353031158447,
          "decoder_belief_target_nll": 9.448553085327148,
          "decoder_belief_top1_on_support": 0.26171875,
          "mean_decoder_belief_support_size": 5.77734375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 3,
          "query_type": "B",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.93404221534729,
          "query_target_nll": 1.3927652835845947,
          "query_top1_on_support": 0.96875,
          "mean_query_support_size": 5.15234375,
          "probe_belief_target_mass": 0.8327805995941162,
          "probe_belief_target_nll": 1.8839435577392578,
          "probe_belief_top1_on_support": 0.890625,
          "mean_belief_support_size": 5.77734375,
          "decoder_belief_target_mass": 0.8411904573440552,
          "decoder_belief_target_nll": 1.8533508777618408,
          "decoder_belief_top1_on_support": 0.89453125,
          "mean_decoder_belief_support_size": 5.77734375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 3,
          "query_type": "B",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.93404221534729,
          "query_target_nll": 1.3927652835845947,
          "query_top1_on_support": 0.96875,
          "mean_query_support_size": 5.15234375,
          "probe_belief_target_mass": 0.8327805995941162,
          "probe_belief_target_nll": 1.8839435577392578,
          "probe_belief_top1_on_support": 0.890625,
          "mean_belief_support_size": 5.77734375,
          "decoder_belief_target_mass": 0.8411904573440552,
          "decoder_belief_target_nll": 1.8533508777618408,
          "decoder_belief_top1_on_support": 0.89453125,
          "mean_decoder_belief_support_size": 5.77734375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 3,
          "query_type": "B",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.93404221534729,
          "query_target_nll": 1.3927652835845947,
          "query_top1_on_support": 0.96875,
          "mean_query_support_size": 5.15234375,
          "probe_belief_target_mass": 0.8327805995941162,
          "probe_belief_target_nll": 1.8839435577392578,
          "probe_belief_top1_on_support": 0.890625,
          "mean_belief_support_size": 5.77734375,
          "decoder_belief_target_mass": 0.8411904573440552,
          "decoder_belief_target_nll": 1.8533508777618408,
          "decoder_belief_top1_on_support": 0.89453125,
          "mean_decoder_belief_support_size": 5.77734375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 3,
          "query_type": "B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.93404221534729,
          "query_target_nll": 1.3927652835845947,
          "query_top1_on_support": 0.96875,
          "mean_query_support_size": 5.15234375,
          "probe_belief_target_mass": 0.8327805995941162,
          "probe_belief_target_nll": 1.8839435577392578,
          "probe_belief_top1_on_support": 0.890625,
          "mean_belief_support_size": 5.77734375,
          "decoder_belief_target_mass": 0.8411904573440552,
          "decoder_belief_target_nll": 1.8533508777618408,
          "decoder_belief_top1_on_support": 0.89453125,
          "mean_decoder_belief_support_size": 5.77734375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 3,
          "query_type": "A_PLUS_B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.48862123489379883,
          "query_target_nll": 2.402876377105713,
          "query_top1_on_support": 0.49609375,
          "mean_query_support_size": 5.3671875,
          "probe_belief_target_mass": 0.0657416507601738,
          "probe_belief_target_nll": 14.61894416809082,
          "probe_belief_top1_on_support": 0.05859375,
          "mean_belief_support_size": 5.484375,
          "decoder_belief_target_mass": 0.06577613949775696,
          "decoder_belief_target_nll": 18.591270446777344,
          "decoder_belief_top1_on_support": 0.0703125,
          "mean_decoder_belief_support_size": 5.484375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 3,
          "query_type": "A_PLUS_B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.4932786822319031,
          "query_target_nll": 3.5241379737854004,
          "query_top1_on_support": 0.515625,
          "mean_query_support_size": 5.3671875,
          "probe_belief_target_mass": 0.1547054797410965,
          "probe_belief_target_nll": 12.336336135864258,
          "probe_belief_top1_on_support": 0.140625,
          "mean_belief_support_size": 5.484375,
          "decoder_belief_target_mass": 0.15349310636520386,
          "decoder_belief_target_nll": 13.851818084716797,
          "decoder_belief_top1_on_support": 0.13671875,
          "mean_decoder_belief_support_size": 5.484375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 3,
          "query_type": "A_PLUS_B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.5344958901405334,
          "query_target_nll": 3.503467321395874,
          "query_top1_on_support": 0.546875,
          "mean_query_support_size": 5.3671875,
          "probe_belief_target_mass": 0.24473384022712708,
          "probe_belief_target_nll": 9.66826057434082,
          "probe_belief_top1_on_support": 0.2265625,
          "mean_belief_support_size": 5.484375,
          "decoder_belief_target_mass": 0.24629336595535278,
          "decoder_belief_target_nll": 10.501893043518066,
          "decoder_belief_top1_on_support": 0.23828125,
          "mean_decoder_belief_support_size": 5.484375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 3,
          "query_type": "A_PLUS_B",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.8528953790664673,
          "query_target_nll": 1.6584852933883667,
          "query_top1_on_support": 0.92578125,
          "mean_query_support_size": 5.3671875,
          "probe_belief_target_mass": 0.880492091178894,
          "probe_belief_target_nll": 1.674355387687683,
          "probe_belief_top1_on_support": 0.953125,
          "mean_belief_support_size": 5.484375,
          "decoder_belief_target_mass": 0.8905535936355591,
          "decoder_belief_target_nll": 1.665908932685852,
          "decoder_belief_top1_on_support": 0.953125,
          "mean_decoder_belief_support_size": 5.484375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 3,
          "query_type": "A_PLUS_B",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.8528953790664673,
          "query_target_nll": 1.6584852933883667,
          "query_top1_on_support": 0.92578125,
          "mean_query_support_size": 5.3671875,
          "probe_belief_target_mass": 0.880492091178894,
          "probe_belief_target_nll": 1.674355387687683,
          "probe_belief_top1_on_support": 0.953125,
          "mean_belief_support_size": 5.484375,
          "decoder_belief_target_mass": 0.8905535936355591,
          "decoder_belief_target_nll": 1.665908932685852,
          "decoder_belief_top1_on_support": 0.953125,
          "mean_decoder_belief_support_size": 5.484375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 3,
          "query_type": "A_PLUS_B",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.8528953790664673,
          "query_target_nll": 1.6584852933883667,
          "query_top1_on_support": 0.92578125,
          "mean_query_support_size": 5.3671875,
          "probe_belief_target_mass": 0.880492091178894,
          "probe_belief_target_nll": 1.674355387687683,
          "probe_belief_top1_on_support": 0.953125,
          "mean_belief_support_size": 5.484375,
          "decoder_belief_target_mass": 0.8905535936355591,
          "decoder_belief_target_nll": 1.665908932685852,
          "decoder_belief_top1_on_support": 0.953125,
          "mean_decoder_belief_support_size": 5.484375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 3,
          "query_type": "A_PLUS_B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.8528953790664673,
          "query_target_nll": 1.6584852933883667,
          "query_top1_on_support": 0.92578125,
          "mean_query_support_size": 5.3671875,
          "probe_belief_target_mass": 0.880492091178894,
          "probe_belief_target_nll": 1.674355387687683,
          "probe_belief_top1_on_support": 0.953125,
          "mean_belief_support_size": 5.484375,
          "decoder_belief_target_mass": 0.8905535936355591,
          "decoder_belief_target_nll": 1.665908932685852,
          "decoder_belief_top1_on_support": 0.953125,
          "mean_decoder_belief_support_size": 5.484375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 3,
          "query_type": "A_MINUS_B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.472655713558197,
          "query_target_nll": 14.848567008972168,
          "query_top1_on_support": 0.47265625,
          "mean_query_support_size": 4.671875,
          "probe_belief_target_mass": 0.05895450338721275,
          "probe_belief_target_nll": 14.408632278442383,
          "probe_belief_top1_on_support": 0.07421875,
          "mean_belief_support_size": 5.6015625,
          "decoder_belief_target_mass": 0.05932197719812393,
          "decoder_belief_target_nll": 18.234405517578125,
          "decoder_belief_top1_on_support": 0.06640625,
          "mean_decoder_belief_support_size": 5.6015625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 3,
          "query_type": "A_MINUS_B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.5288023948669434,
          "query_target_nll": 7.672093391418457,
          "query_top1_on_support": 0.53125,
          "mean_query_support_size": 4.671875,
          "probe_belief_target_mass": 0.15690022706985474,
          "probe_belief_target_nll": 11.519844055175781,
          "probe_belief_top1_on_support": 0.140625,
          "mean_belief_support_size": 5.6015625,
          "decoder_belief_target_mass": 0.15784567594528198,
          "decoder_belief_target_nll": 12.939579010009766,
          "decoder_belief_top1_on_support": 0.1640625,
          "mean_decoder_belief_support_size": 5.6015625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 3,
          "query_type": "A_MINUS_B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.5883186459541321,
          "query_target_nll": 5.417423248291016,
          "query_top1_on_support": 0.59765625,
          "mean_query_support_size": 4.671875,
          "probe_belief_target_mass": 0.28857624530792236,
          "probe_belief_target_nll": 8.264647483825684,
          "probe_belief_top1_on_support": 0.26953125,
          "mean_belief_support_size": 5.6015625,
          "decoder_belief_target_mass": 0.2905160188674927,
          "decoder_belief_target_nll": 8.900525093078613,
          "decoder_belief_top1_on_support": 0.265625,
          "mean_decoder_belief_support_size": 5.6015625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 3,
          "query_type": "A_MINUS_B",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.9120296835899353,
          "query_target_nll": 1.2457325458526611,
          "query_top1_on_support": 0.95703125,
          "mean_query_support_size": 4.671875,
          "probe_belief_target_mass": 0.8496957421302795,
          "probe_belief_target_nll": 1.7985081672668457,
          "probe_belief_top1_on_support": 0.91796875,
          "mean_belief_support_size": 5.6015625,
          "decoder_belief_target_mass": 0.8591122627258301,
          "decoder_belief_target_nll": 1.767818808555603,
          "decoder_belief_top1_on_support": 0.921875,
          "mean_decoder_belief_support_size": 5.6015625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 3,
          "query_type": "A_MINUS_B",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.9120296835899353,
          "query_target_nll": 1.2457325458526611,
          "query_top1_on_support": 0.95703125,
          "mean_query_support_size": 4.671875,
          "probe_belief_target_mass": 0.8496957421302795,
          "probe_belief_target_nll": 1.7985081672668457,
          "probe_belief_top1_on_support": 0.91796875,
          "mean_belief_support_size": 5.6015625,
          "decoder_belief_target_mass": 0.8591122627258301,
          "decoder_belief_target_nll": 1.767818808555603,
          "decoder_belief_top1_on_support": 0.921875,
          "mean_decoder_belief_support_size": 5.6015625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 3,
          "query_type": "A_MINUS_B",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.9120296835899353,
          "query_target_nll": 1.2457325458526611,
          "query_top1_on_support": 0.95703125,
          "mean_query_support_size": 4.671875,
          "probe_belief_target_mass": 0.8496957421302795,
          "probe_belief_target_nll": 1.7985081672668457,
          "probe_belief_top1_on_support": 0.91796875,
          "mean_belief_support_size": 5.6015625,
          "decoder_belief_target_mass": 0.8591122627258301,
          "decoder_belief_target_nll": 1.767818808555603,
          "decoder_belief_top1_on_support": 0.921875,
          "mean_decoder_belief_support_size": 5.6015625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 3,
          "query_type": "A_MINUS_B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.9120296835899353,
          "query_target_nll": 1.2457325458526611,
          "query_top1_on_support": 0.95703125,
          "mean_query_support_size": 4.671875,
          "probe_belief_target_mass": 0.8496957421302795,
          "probe_belief_target_nll": 1.7985081672668457,
          "probe_belief_top1_on_support": 0.91796875,
          "mean_belief_support_size": 5.6015625,
          "decoder_belief_target_mass": 0.8591122627258301,
          "decoder_belief_target_nll": 1.767818808555603,
          "decoder_belief_top1_on_support": 0.921875,
          "mean_decoder_belief_support_size": 5.6015625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 6,
          "query_type": "A",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.2530818581581116,
          "query_target_nll": 2.404404401779175,
          "query_top1_on_support": 0.265625,
          "mean_query_support_size": 2.78125,
          "probe_belief_target_mass": 0.030176561325788498,
          "probe_belief_target_nll": 14.971030235290527,
          "probe_belief_top1_on_support": 0.03125,
          "mean_belief_support_size": 3.45703125,
          "decoder_belief_target_mass": 0.030349809676408768,
          "decoder_belief_target_nll": 19.045269012451172,
          "decoder_belief_top1_on_support": 0.03515625,
          "mean_decoder_belief_support_size": 3.45703125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 6,
          "query_type": "A",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.28165149688720703,
          "query_target_nll": 4.507490158081055,
          "query_top1_on_support": 0.2890625,
          "mean_query_support_size": 2.78125,
          "probe_belief_target_mass": 0.0705072432756424,
          "probe_belief_target_nll": 13.301746368408203,
          "probe_belief_top1_on_support": 0.05859375,
          "mean_belief_support_size": 3.45703125,
          "decoder_belief_target_mass": 0.07161024212837219,
          "decoder_belief_target_nll": 14.797473907470703,
          "decoder_belief_top1_on_support": 0.09375,
          "mean_decoder_belief_support_size": 3.45703125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 6,
          "query_type": "A",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.35068491101264954,
          "query_target_nll": 4.551787853240967,
          "query_top1_on_support": 0.38671875,
          "mean_query_support_size": 2.78125,
          "probe_belief_target_mass": 0.08699190616607666,
          "probe_belief_target_nll": 12.522146224975586,
          "probe_belief_top1_on_support": 0.078125,
          "mean_belief_support_size": 3.45703125,
          "decoder_belief_target_mass": 0.08806543052196503,
          "decoder_belief_target_nll": 13.912010192871094,
          "decoder_belief_top1_on_support": 0.0859375,
          "mean_decoder_belief_support_size": 3.45703125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 6,
          "query_type": "A",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.40584924817085266,
          "query_target_nll": 4.426654815673828,
          "query_top1_on_support": 0.4140625,
          "mean_query_support_size": 2.78125,
          "probe_belief_target_mass": 0.11077713966369629,
          "probe_belief_target_nll": 11.535524368286133,
          "probe_belief_top1_on_support": 0.125,
          "mean_belief_support_size": 3.45703125,
          "decoder_belief_target_mass": 0.11198648810386658,
          "decoder_belief_target_nll": 12.82072925567627,
          "decoder_belief_top1_on_support": 0.11328125,
          "mean_decoder_belief_support_size": 3.45703125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 6,
          "query_type": "A",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.7466932535171509,
          "query_target_nll": 1.2038761377334595,
          "query_top1_on_support": 0.82421875,
          "mean_query_support_size": 2.78125,
          "probe_belief_target_mass": 0.6374768018722534,
          "probe_belief_target_nll": 1.8530912399291992,
          "probe_belief_top1_on_support": 0.76171875,
          "mean_belief_support_size": 3.45703125,
          "decoder_belief_target_mass": 0.6278401017189026,
          "decoder_belief_target_nll": 1.8593370914459229,
          "decoder_belief_top1_on_support": 0.75390625,
          "mean_decoder_belief_support_size": 3.45703125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 6,
          "query_type": "A",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.7466932535171509,
          "query_target_nll": 1.2038761377334595,
          "query_top1_on_support": 0.82421875,
          "mean_query_support_size": 2.78125,
          "probe_belief_target_mass": 0.6374768018722534,
          "probe_belief_target_nll": 1.8530912399291992,
          "probe_belief_top1_on_support": 0.76171875,
          "mean_belief_support_size": 3.45703125,
          "decoder_belief_target_mass": 0.6278401017189026,
          "decoder_belief_target_nll": 1.8593370914459229,
          "decoder_belief_top1_on_support": 0.75390625,
          "mean_decoder_belief_support_size": 3.45703125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 6,
          "query_type": "A",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.7466932535171509,
          "query_target_nll": 1.2038761377334595,
          "query_top1_on_support": 0.82421875,
          "mean_query_support_size": 2.78125,
          "probe_belief_target_mass": 0.6374768018722534,
          "probe_belief_target_nll": 1.8530912399291992,
          "probe_belief_top1_on_support": 0.76171875,
          "mean_belief_support_size": 3.45703125,
          "decoder_belief_target_mass": 0.6278401017189026,
          "decoder_belief_target_nll": 1.8593370914459229,
          "decoder_belief_top1_on_support": 0.75390625,
          "mean_decoder_belief_support_size": 3.45703125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 6,
          "query_type": "B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.3105168342590332,
          "query_target_nll": 2.4023823738098145,
          "query_top1_on_support": 0.28125,
          "mean_query_support_size": 3.4140625,
          "probe_belief_target_mass": 0.034212976694107056,
          "probe_belief_target_nll": 15.355527877807617,
          "probe_belief_top1_on_support": 0.03515625,
          "mean_belief_support_size": 4.03125,
          "decoder_belief_target_mass": 0.03430630639195442,
          "decoder_belief_target_nll": 19.634836196899414,
          "decoder_belief_top1_on_support": 0.0234375,
          "mean_decoder_belief_support_size": 4.03125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 6,
          "query_type": "B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.34078216552734375,
          "query_target_nll": 5.117256164550781,
          "query_top1_on_support": 0.33984375,
          "mean_query_support_size": 3.4140625,
          "probe_belief_target_mass": 0.04496061056852341,
          "probe_belief_target_nll": 14.54107666015625,
          "probe_belief_top1_on_support": 0.03515625,
          "mean_belief_support_size": 4.03125,
          "decoder_belief_target_mass": 0.04506306350231171,
          "decoder_belief_target_nll": 16.228254318237305,
          "decoder_belief_top1_on_support": 0.0625,
          "mean_decoder_belief_support_size": 4.03125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 6,
          "query_type": "B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.38834109902381897,
          "query_target_nll": 4.98013973236084,
          "query_top1_on_support": 0.4140625,
          "mean_query_support_size": 3.4140625,
          "probe_belief_target_mass": 0.07256152480840683,
          "probe_belief_target_nll": 13.031356811523438,
          "probe_belief_top1_on_support": 0.0703125,
          "mean_belief_support_size": 4.03125,
          "decoder_belief_target_mass": 0.07221895456314087,
          "decoder_belief_target_nll": 14.627594947814941,
          "decoder_belief_top1_on_support": 0.08984375,
          "mean_decoder_belief_support_size": 4.03125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 6,
          "query_type": "B",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.4273007810115814,
          "query_target_nll": 5.431184768676758,
          "query_top1_on_support": 0.4453125,
          "mean_query_support_size": 3.4140625,
          "probe_belief_target_mass": 0.10644866526126862,
          "probe_belief_target_nll": 11.98141860961914,
          "probe_belief_top1_on_support": 0.10546875,
          "mean_belief_support_size": 4.03125,
          "decoder_belief_target_mass": 0.10465620458126068,
          "decoder_belief_target_nll": 13.444873809814453,
          "decoder_belief_top1_on_support": 0.09765625,
          "mean_decoder_belief_support_size": 4.03125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 6,
          "query_type": "B",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.7715412378311157,
          "query_target_nll": 1.3125643730163574,
          "query_top1_on_support": 0.84375,
          "mean_query_support_size": 3.4140625,
          "probe_belief_target_mass": 0.5964440107345581,
          "probe_belief_target_nll": 2.068345308303833,
          "probe_belief_top1_on_support": 0.72265625,
          "mean_belief_support_size": 4.03125,
          "decoder_belief_target_mass": 0.5864962935447693,
          "decoder_belief_target_nll": 2.054654836654663,
          "decoder_belief_top1_on_support": 0.703125,
          "mean_decoder_belief_support_size": 4.03125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 6,
          "query_type": "B",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.7715412378311157,
          "query_target_nll": 1.3125643730163574,
          "query_top1_on_support": 0.84375,
          "mean_query_support_size": 3.4140625,
          "probe_belief_target_mass": 0.5964440107345581,
          "probe_belief_target_nll": 2.068345308303833,
          "probe_belief_top1_on_support": 0.72265625,
          "mean_belief_support_size": 4.03125,
          "decoder_belief_target_mass": 0.5864962935447693,
          "decoder_belief_target_nll": 2.054654836654663,
          "decoder_belief_top1_on_support": 0.703125,
          "mean_decoder_belief_support_size": 4.03125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 6,
          "query_type": "B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.7715412378311157,
          "query_target_nll": 1.3125643730163574,
          "query_top1_on_support": 0.84375,
          "mean_query_support_size": 3.4140625,
          "probe_belief_target_mass": 0.5964440107345581,
          "probe_belief_target_nll": 2.068345308303833,
          "probe_belief_top1_on_support": 0.72265625,
          "mean_belief_support_size": 4.03125,
          "decoder_belief_target_mass": 0.5864962935447693,
          "decoder_belief_target_nll": 2.054654836654663,
          "decoder_belief_top1_on_support": 0.703125,
          "mean_decoder_belief_support_size": 4.03125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 6,
          "query_type": "A_PLUS_B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.33000099658966064,
          "query_target_nll": 2.4036169052124023,
          "query_top1_on_support": 0.34765625,
          "mean_query_support_size": 3.6171875,
          "probe_belief_target_mass": 0.033164046704769135,
          "probe_belief_target_nll": 15.194226264953613,
          "probe_belief_top1_on_support": 0.0234375,
          "mean_belief_support_size": 3.97265625,
          "decoder_belief_target_mass": 0.03338306397199631,
          "decoder_belief_target_nll": 19.33576011657715,
          "decoder_belief_top1_on_support": 0.03125,
          "mean_decoder_belief_support_size": 3.97265625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 6,
          "query_type": "A_PLUS_B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.337532103061676,
          "query_target_nll": 3.2034106254577637,
          "query_top1_on_support": 0.375,
          "mean_query_support_size": 3.6171875,
          "probe_belief_target_mass": 0.05043330788612366,
          "probe_belief_target_nll": 13.813321113586426,
          "probe_belief_top1_on_support": 0.05078125,
          "mean_belief_support_size": 3.97265625,
          "decoder_belief_target_mass": 0.050717540085315704,
          "decoder_belief_target_nll": 15.517746925354004,
          "decoder_belief_top1_on_support": 0.05078125,
          "mean_decoder_belief_support_size": 3.97265625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 6,
          "query_type": "A_PLUS_B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.3313302993774414,
          "query_target_nll": 3.637758493423462,
          "query_top1_on_support": 0.31640625,
          "mean_query_support_size": 3.6171875,
          "probe_belief_target_mass": 0.0731259435415268,
          "probe_belief_target_nll": 12.677604675292969,
          "probe_belief_top1_on_support": 0.05859375,
          "mean_belief_support_size": 3.97265625,
          "decoder_belief_target_mass": 0.07514117658138275,
          "decoder_belief_target_nll": 13.989625930786133,
          "decoder_belief_top1_on_support": 0.0859375,
          "mean_decoder_belief_support_size": 3.97265625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 6,
          "query_type": "A_PLUS_B",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.3311459720134735,
          "query_target_nll": 3.8029212951660156,
          "query_top1_on_support": 0.34375,
          "mean_query_support_size": 3.6171875,
          "probe_belief_target_mass": 0.10255686938762665,
          "probe_belief_target_nll": 11.298919677734375,
          "probe_belief_top1_on_support": 0.08984375,
          "mean_belief_support_size": 3.97265625,
          "decoder_belief_target_mass": 0.10187555104494095,
          "decoder_belief_target_nll": 12.523211479187012,
          "decoder_belief_top1_on_support": 0.0859375,
          "mean_decoder_belief_support_size": 3.97265625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 6,
          "query_type": "A_PLUS_B",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.6417573690414429,
          "query_target_nll": 1.6554900407791138,
          "query_top1_on_support": 0.75,
          "mean_query_support_size": 3.6171875,
          "probe_belief_target_mass": 0.5943344831466675,
          "probe_belief_target_nll": 2.0607547760009766,
          "probe_belief_top1_on_support": 0.69921875,
          "mean_belief_support_size": 3.97265625,
          "decoder_belief_target_mass": 0.5796753168106079,
          "decoder_belief_target_nll": 2.0340142250061035,
          "decoder_belief_top1_on_support": 0.7109375,
          "mean_decoder_belief_support_size": 3.97265625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 6,
          "query_type": "A_PLUS_B",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.6417573690414429,
          "query_target_nll": 1.6554900407791138,
          "query_top1_on_support": 0.75,
          "mean_query_support_size": 3.6171875,
          "probe_belief_target_mass": 0.5943344831466675,
          "probe_belief_target_nll": 2.0607547760009766,
          "probe_belief_top1_on_support": 0.69921875,
          "mean_belief_support_size": 3.97265625,
          "decoder_belief_target_mass": 0.5796753168106079,
          "decoder_belief_target_nll": 2.0340142250061035,
          "decoder_belief_top1_on_support": 0.7109375,
          "mean_decoder_belief_support_size": 3.97265625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 6,
          "query_type": "A_PLUS_B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.6417573690414429,
          "query_target_nll": 1.6554900407791138,
          "query_top1_on_support": 0.75,
          "mean_query_support_size": 3.6171875,
          "probe_belief_target_mass": 0.5943344831466675,
          "probe_belief_target_nll": 2.0607547760009766,
          "probe_belief_top1_on_support": 0.69921875,
          "mean_belief_support_size": 3.97265625,
          "decoder_belief_target_mass": 0.5796753168106079,
          "decoder_belief_target_nll": 2.0340142250061035,
          "decoder_belief_top1_on_support": 0.7109375,
          "mean_decoder_belief_support_size": 3.97265625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 6,
          "query_type": "A_MINUS_B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.3124997615814209,
          "query_target_nll": 15.581740379333496,
          "query_top1_on_support": 0.3125,
          "mean_query_support_size": 3.2265625,
          "probe_belief_target_mass": 0.03164301812648773,
          "probe_belief_target_nll": 15.008096694946289,
          "probe_belief_top1_on_support": 0.046875,
          "mean_belief_support_size": 3.76171875,
          "decoder_belief_target_mass": 0.03126494586467743,
          "decoder_belief_target_nll": 18.910160064697266,
          "decoder_belief_top1_on_support": 0.01953125,
          "mean_decoder_belief_support_size": 3.76171875
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 6,
          "query_type": "A_MINUS_B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.3015841543674469,
          "query_target_nll": 8.701509475708008,
          "query_top1_on_support": 0.28125,
          "mean_query_support_size": 3.2265625,
          "probe_belief_target_mass": 0.0440542995929718,
          "probe_belief_target_nll": 13.959927558898926,
          "probe_belief_top1_on_support": 0.05859375,
          "mean_belief_support_size": 3.76171875,
          "decoder_belief_target_mass": 0.04485006630420685,
          "decoder_belief_target_nll": 15.66829776763916,
          "decoder_belief_top1_on_support": 0.05859375,
          "mean_decoder_belief_support_size": 3.76171875
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 6,
          "query_type": "A_MINUS_B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.333312451839447,
          "query_target_nll": 7.166574478149414,
          "query_top1_on_support": 0.328125,
          "mean_query_support_size": 3.2265625,
          "probe_belief_target_mass": 0.0706152692437172,
          "probe_belief_target_nll": 12.826810836791992,
          "probe_belief_top1_on_support": 0.0859375,
          "mean_belief_support_size": 3.76171875,
          "decoder_belief_target_mass": 0.06791658699512482,
          "decoder_belief_target_nll": 14.253889083862305,
          "decoder_belief_top1_on_support": 0.06640625,
          "mean_decoder_belief_support_size": 3.76171875
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 6,
          "query_type": "A_MINUS_B",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.31731539964675903,
          "query_target_nll": 6.327928066253662,
          "query_top1_on_support": 0.328125,
          "mean_query_support_size": 3.2265625,
          "probe_belief_target_mass": 0.07863132655620575,
          "probe_belief_target_nll": 11.730693817138672,
          "probe_belief_top1_on_support": 0.08984375,
          "mean_belief_support_size": 3.76171875,
          "decoder_belief_target_mass": 0.076526939868927,
          "decoder_belief_target_nll": 13.097808837890625,
          "decoder_belief_top1_on_support": 0.06640625,
          "mean_decoder_belief_support_size": 3.76171875
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 6,
          "query_type": "A_MINUS_B",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.6900098323822021,
          "query_target_nll": 1.413759469985962,
          "query_top1_on_support": 0.78125,
          "mean_query_support_size": 3.2265625,
          "probe_belief_target_mass": 0.6027534604072571,
          "probe_belief_target_nll": 2.0204200744628906,
          "probe_belief_top1_on_support": 0.70703125,
          "mean_belief_support_size": 3.76171875,
          "decoder_belief_target_mass": 0.5942018032073975,
          "decoder_belief_target_nll": 1.990920066833496,
          "decoder_belief_top1_on_support": 0.703125,
          "mean_decoder_belief_support_size": 3.76171875
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 6,
          "query_type": "A_MINUS_B",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.6900098323822021,
          "query_target_nll": 1.413759469985962,
          "query_top1_on_support": 0.78125,
          "mean_query_support_size": 3.2265625,
          "probe_belief_target_mass": 0.6027534604072571,
          "probe_belief_target_nll": 2.0204200744628906,
          "probe_belief_top1_on_support": 0.70703125,
          "mean_belief_support_size": 3.76171875,
          "decoder_belief_target_mass": 0.5942018032073975,
          "decoder_belief_target_nll": 1.990920066833496,
          "decoder_belief_top1_on_support": 0.703125,
          "mean_decoder_belief_support_size": 3.76171875
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 6,
          "query_type": "A_MINUS_B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.6900098323822021,
          "query_target_nll": 1.413759469985962,
          "query_top1_on_support": 0.78125,
          "mean_query_support_size": 3.2265625,
          "probe_belief_target_mass": 0.6027534604072571,
          "probe_belief_target_nll": 2.0204200744628906,
          "probe_belief_top1_on_support": 0.70703125,
          "mean_belief_support_size": 3.76171875,
          "decoder_belief_target_mass": 0.5942018032073975,
          "decoder_belief_target_nll": 1.990920066833496,
          "decoder_belief_top1_on_support": 0.703125,
          "mean_decoder_belief_support_size": 3.76171875
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 9,
          "query_type": "A",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.21917998790740967,
          "query_target_nll": 2.4024229049682617,
          "query_top1_on_support": 0.2421875,
          "mean_query_support_size": 2.40625,
          "probe_belief_target_mass": 0.021715233102440834,
          "probe_belief_target_nll": 15.21967601776123,
          "probe_belief_top1_on_support": 0.01171875,
          "mean_belief_support_size": 2.73828125,
          "decoder_belief_target_mass": 0.021676113829016685,
          "decoder_belief_target_nll": 19.50696563720703,
          "decoder_belief_top1_on_support": 0.03125,
          "mean_decoder_belief_support_size": 2.73828125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 9,
          "query_type": "A",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.24174827337265015,
          "query_target_nll": 4.608434677124023,
          "query_top1_on_support": 0.265625,
          "mean_query_support_size": 2.40625,
          "probe_belief_target_mass": 0.03576366975903511,
          "probe_belief_target_nll": 13.903057098388672,
          "probe_belief_top1_on_support": 0.04296875,
          "mean_belief_support_size": 2.73828125,
          "decoder_belief_target_mass": 0.035209815949201584,
          "decoder_belief_target_nll": 15.44052505493164,
          "decoder_belief_top1_on_support": 0.0234375,
          "mean_decoder_belief_support_size": 2.73828125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 9,
          "query_type": "A",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.2620830237865448,
          "query_target_nll": 5.408433437347412,
          "query_top1_on_support": 0.25390625,
          "mean_query_support_size": 2.40625,
          "probe_belief_target_mass": 0.03351844847202301,
          "probe_belief_target_nll": 13.829469680786133,
          "probe_belief_top1_on_support": 0.02734375,
          "mean_belief_support_size": 2.73828125,
          "decoder_belief_target_mass": 0.03451690822839737,
          "decoder_belief_target_nll": 15.398207664489746,
          "decoder_belief_top1_on_support": 0.03515625,
          "mean_decoder_belief_support_size": 2.73828125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 9,
          "query_type": "A",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.269538551568985,
          "query_target_nll": 5.950573444366455,
          "query_top1_on_support": 0.25,
          "mean_query_support_size": 2.40625,
          "probe_belief_target_mass": 0.04069415107369423,
          "probe_belief_target_nll": 13.221373558044434,
          "probe_belief_top1_on_support": 0.03125,
          "mean_belief_support_size": 2.73828125,
          "decoder_belief_target_mass": 0.03948362171649933,
          "decoder_belief_target_nll": 14.690391540527344,
          "decoder_belief_top1_on_support": 0.03515625,
          "mean_decoder_belief_support_size": 2.73828125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 9,
          "query_type": "A",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.33043646812438965,
          "query_target_nll": 5.286316394805908,
          "query_top1_on_support": 0.32421875,
          "mean_query_support_size": 2.40625,
          "probe_belief_target_mass": 0.07649917155504227,
          "probe_belief_target_nll": 10.541196823120117,
          "probe_belief_top1_on_support": 0.0859375,
          "mean_belief_support_size": 2.73828125,
          "decoder_belief_target_mass": 0.07648133486509323,
          "decoder_belief_target_nll": 11.51611614227295,
          "decoder_belief_top1_on_support": 0.07421875,
          "mean_decoder_belief_support_size": 2.73828125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 9,
          "query_type": "A",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.6417803764343262,
          "query_target_nll": 1.2733101844787598,
          "query_top1_on_support": 0.703125,
          "mean_query_support_size": 2.40625,
          "probe_belief_target_mass": 0.43091967701911926,
          "probe_belief_target_nll": 2.2110114097595215,
          "probe_belief_top1_on_support": 0.54296875,
          "mean_belief_support_size": 2.73828125,
          "decoder_belief_target_mass": 0.4220062494277954,
          "decoder_belief_target_nll": 2.1934890747070312,
          "decoder_belief_top1_on_support": 0.51171875,
          "mean_decoder_belief_support_size": 2.73828125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 9,
          "query_type": "A",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.6417803764343262,
          "query_target_nll": 1.2733101844787598,
          "query_top1_on_support": 0.703125,
          "mean_query_support_size": 2.40625,
          "probe_belief_target_mass": 0.43091967701911926,
          "probe_belief_target_nll": 2.2110114097595215,
          "probe_belief_top1_on_support": 0.54296875,
          "mean_belief_support_size": 2.73828125,
          "decoder_belief_target_mass": 0.4220062494277954,
          "decoder_belief_target_nll": 2.1934890747070312,
          "decoder_belief_top1_on_support": 0.51171875,
          "mean_decoder_belief_support_size": 2.73828125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 9,
          "query_type": "B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.19168782234191895,
          "query_target_nll": 2.409575939178467,
          "query_top1_on_support": 0.22265625,
          "mean_query_support_size": 2.1171875,
          "probe_belief_target_mass": 0.024687521159648895,
          "probe_belief_target_nll": 15.057048797607422,
          "probe_belief_top1_on_support": 0.01953125,
          "mean_belief_support_size": 2.37109375,
          "decoder_belief_target_mass": 0.02497500739991665,
          "decoder_belief_target_nll": 19.34463119506836,
          "decoder_belief_top1_on_support": 0.03125,
          "mean_decoder_belief_support_size": 2.37109375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 9,
          "query_type": "B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.1965523213148117,
          "query_target_nll": 5.039180278778076,
          "query_top1_on_support": 0.19140625,
          "mean_query_support_size": 2.1171875,
          "probe_belief_target_mass": 0.025244569405913353,
          "probe_belief_target_nll": 14.23426342010498,
          "probe_belief_top1_on_support": 0.03125,
          "mean_belief_support_size": 2.37109375,
          "decoder_belief_target_mass": 0.0248072762042284,
          "decoder_belief_target_nll": 16.01129913330078,
          "decoder_belief_top1_on_support": 0.02734375,
          "mean_decoder_belief_support_size": 2.37109375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 9,
          "query_type": "B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.21195119619369507,
          "query_target_nll": 5.326318740844727,
          "query_top1_on_support": 0.19140625,
          "mean_query_support_size": 2.1171875,
          "probe_belief_target_mass": 0.0302311722189188,
          "probe_belief_target_nll": 13.406440734863281,
          "probe_belief_top1_on_support": 0.0234375,
          "mean_belief_support_size": 2.37109375,
          "decoder_belief_target_mass": 0.030804984271526337,
          "decoder_belief_target_nll": 15.102956771850586,
          "decoder_belief_top1_on_support": 0.03125,
          "mean_decoder_belief_support_size": 2.37109375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 9,
          "query_type": "B",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.22881698608398438,
          "query_target_nll": 5.830305576324463,
          "query_top1_on_support": 0.23828125,
          "mean_query_support_size": 2.1171875,
          "probe_belief_target_mass": 0.030832631513476372,
          "probe_belief_target_nll": 13.046674728393555,
          "probe_belief_top1_on_support": 0.0234375,
          "mean_belief_support_size": 2.37109375,
          "decoder_belief_target_mass": 0.03319378197193146,
          "decoder_belief_target_nll": 14.51974105834961,
          "decoder_belief_top1_on_support": 0.04296875,
          "mean_decoder_belief_support_size": 2.37109375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 9,
          "query_type": "B",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.2891945242881775,
          "query_target_nll": 5.808460712432861,
          "query_top1_on_support": 0.30078125,
          "mean_query_support_size": 2.1171875,
          "probe_belief_target_mass": 0.05797610804438591,
          "probe_belief_target_nll": 11.364771842956543,
          "probe_belief_top1_on_support": 0.06640625,
          "mean_belief_support_size": 2.37109375,
          "decoder_belief_target_mass": 0.05461960285902023,
          "decoder_belief_target_nll": 12.551033973693848,
          "decoder_belief_top1_on_support": 0.0546875,
          "mean_decoder_belief_support_size": 2.37109375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 9,
          "query_type": "B",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.6164610385894775,
          "query_target_nll": 1.2572312355041504,
          "query_top1_on_support": 0.6796875,
          "mean_query_support_size": 2.1171875,
          "probe_belief_target_mass": 0.42830801010131836,
          "probe_belief_target_nll": 2.121614694595337,
          "probe_belief_top1_on_support": 0.5234375,
          "mean_belief_support_size": 2.37109375,
          "decoder_belief_target_mass": 0.41745680570602417,
          "decoder_belief_target_nll": 2.10146427154541,
          "decoder_belief_top1_on_support": 0.55078125,
          "mean_decoder_belief_support_size": 2.37109375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 9,
          "query_type": "B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.6164610385894775,
          "query_target_nll": 1.2572312355041504,
          "query_top1_on_support": 0.6796875,
          "mean_query_support_size": 2.1171875,
          "probe_belief_target_mass": 0.42830801010131836,
          "probe_belief_target_nll": 2.121614694595337,
          "probe_belief_top1_on_support": 0.5234375,
          "mean_belief_support_size": 2.37109375,
          "decoder_belief_target_mass": 0.41745680570602417,
          "decoder_belief_target_nll": 2.10146427154541,
          "decoder_belief_top1_on_support": 0.55078125,
          "mean_decoder_belief_support_size": 2.37109375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 9,
          "query_type": "A_PLUS_B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.22652092576026917,
          "query_target_nll": 2.4159414768218994,
          "query_top1_on_support": 0.1875,
          "mean_query_support_size": 2.50390625,
          "probe_belief_target_mass": 0.02208864688873291,
          "probe_belief_target_nll": 15.388028144836426,
          "probe_belief_top1_on_support": 0.0078125,
          "mean_belief_support_size": 2.71875,
          "decoder_belief_target_mass": 0.02228963002562523,
          "decoder_belief_target_nll": 19.575241088867188,
          "decoder_belief_top1_on_support": 0.01953125,
          "mean_decoder_belief_support_size": 2.71875
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 9,
          "query_type": "A_PLUS_B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.23303212225437164,
          "query_target_nll": 3.2214877605438232,
          "query_top1_on_support": 0.234375,
          "mean_query_support_size": 2.50390625,
          "probe_belief_target_mass": 0.030808016657829285,
          "probe_belief_target_nll": 14.106508255004883,
          "probe_belief_top1_on_support": 0.0546875,
          "mean_belief_support_size": 2.71875,
          "decoder_belief_target_mass": 0.030065327882766724,
          "decoder_belief_target_nll": 15.824604034423828,
          "decoder_belief_top1_on_support": 0.02734375,
          "mean_decoder_belief_support_size": 2.71875
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 9,
          "query_type": "A_PLUS_B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.21311607956886292,
          "query_target_nll": 3.7770209312438965,
          "query_top1_on_support": 0.19140625,
          "mean_query_support_size": 2.50390625,
          "probe_belief_target_mass": 0.0271175317466259,
          "probe_belief_target_nll": 13.551761627197266,
          "probe_belief_top1_on_support": 0.0390625,
          "mean_belief_support_size": 2.71875,
          "decoder_belief_target_mass": 0.02625294402241707,
          "decoder_belief_target_nll": 14.924287796020508,
          "decoder_belief_top1_on_support": 0.01953125,
          "mean_decoder_belief_support_size": 2.71875
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 9,
          "query_type": "A_PLUS_B",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.2252829670906067,
          "query_target_nll": 3.8018128871917725,
          "query_top1_on_support": 0.22265625,
          "mean_query_support_size": 2.50390625,
          "probe_belief_target_mass": 0.040417157113552094,
          "probe_belief_target_nll": 12.469825744628906,
          "probe_belief_top1_on_support": 0.05078125,
          "mean_belief_support_size": 2.71875,
          "decoder_belief_target_mass": 0.04031673073768616,
          "decoder_belief_target_nll": 13.80705451965332,
          "decoder_belief_top1_on_support": 0.03515625,
          "mean_decoder_belief_support_size": 2.71875
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 9,
          "query_type": "A_PLUS_B",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.21583731472492218,
          "query_target_nll": 3.853325128555298,
          "query_top1_on_support": 0.21875,
          "mean_query_support_size": 2.50390625,
          "probe_belief_target_mass": 0.05686970055103302,
          "probe_belief_target_nll": 10.521811485290527,
          "probe_belief_top1_on_support": 0.06640625,
          "mean_belief_support_size": 2.71875,
          "decoder_belief_target_mass": 0.05790994316339493,
          "decoder_belief_target_nll": 11.483793258666992,
          "decoder_belief_top1_on_support": 0.06640625,
          "mean_decoder_belief_support_size": 2.71875
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 9,
          "query_type": "A_PLUS_B",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.5060155987739563,
          "query_target_nll": 1.653557300567627,
          "query_top1_on_support": 0.6328125,
          "mean_query_support_size": 2.50390625,
          "probe_belief_target_mass": 0.4344879984855652,
          "probe_belief_target_nll": 2.238893508911133,
          "probe_belief_top1_on_support": 0.54296875,
          "mean_belief_support_size": 2.71875,
          "decoder_belief_target_mass": 0.41087186336517334,
          "decoder_belief_target_nll": 2.263479709625244,
          "decoder_belief_top1_on_support": 0.5390625,
          "mean_decoder_belief_support_size": 2.71875
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 9,
          "query_type": "A_PLUS_B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.5060155987739563,
          "query_target_nll": 1.653557300567627,
          "query_top1_on_support": 0.6328125,
          "mean_query_support_size": 2.50390625,
          "probe_belief_target_mass": 0.4344879984855652,
          "probe_belief_target_nll": 2.238893508911133,
          "probe_belief_top1_on_support": 0.54296875,
          "mean_belief_support_size": 2.71875,
          "decoder_belief_target_mass": 0.41087186336517334,
          "decoder_belief_target_nll": 2.263479709625244,
          "decoder_belief_top1_on_support": 0.5390625,
          "mean_decoder_belief_support_size": 2.71875
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 9,
          "query_type": "A_MINUS_B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.18359482288360596,
          "query_target_nll": 16.045299530029297,
          "query_top1_on_support": 0.18359375,
          "mean_query_support_size": 2.35546875,
          "probe_belief_target_mass": 0.016992513090372086,
          "probe_belief_target_nll": 15.571773529052734,
          "probe_belief_top1_on_support": 0.01953125,
          "mean_belief_support_size": 2.48046875,
          "decoder_belief_target_mass": 0.016747303307056427,
          "decoder_belief_target_nll": 19.844669342041016,
          "decoder_belief_top1_on_support": 0.0078125,
          "mean_decoder_belief_support_size": 2.48046875
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 9,
          "query_type": "A_MINUS_B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.19973063468933105,
          "query_target_nll": 7.990225791931152,
          "query_top1_on_support": 0.19140625,
          "mean_query_support_size": 2.35546875,
          "probe_belief_target_mass": 0.021488916128873825,
          "probe_belief_target_nll": 14.143263816833496,
          "probe_belief_top1_on_support": 0.03515625,
          "mean_belief_support_size": 2.48046875,
          "decoder_belief_target_mass": 0.02182084694504738,
          "decoder_belief_target_nll": 15.66469955444336,
          "decoder_belief_top1_on_support": 0.015625,
          "mean_decoder_belief_support_size": 2.48046875
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 9,
          "query_type": "A_MINUS_B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.2352021485567093,
          "query_target_nll": 7.007611274719238,
          "query_top1_on_support": 0.23828125,
          "mean_query_support_size": 2.35546875,
          "probe_belief_target_mass": 0.02343781851232052,
          "probe_belief_target_nll": 13.538871765136719,
          "probe_belief_top1_on_support": 0.02734375,
          "mean_belief_support_size": 2.48046875,
          "decoder_belief_target_mass": 0.024633724242448807,
          "decoder_belief_target_nll": 14.981553077697754,
          "decoder_belief_top1_on_support": 0.03125,
          "mean_decoder_belief_support_size": 2.48046875
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 9,
          "query_type": "A_MINUS_B",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.23492302000522614,
          "query_target_nll": 6.563282489776611,
          "query_top1_on_support": 0.234375,
          "mean_query_support_size": 2.35546875,
          "probe_belief_target_mass": 0.027929481118917465,
          "probe_belief_target_nll": 13.128324508666992,
          "probe_belief_top1_on_support": 0.01953125,
          "mean_belief_support_size": 2.48046875,
          "decoder_belief_target_mass": 0.03020620346069336,
          "decoder_belief_target_nll": 14.619006156921387,
          "decoder_belief_top1_on_support": 0.03125,
          "mean_decoder_belief_support_size": 2.48046875
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 9,
          "query_type": "A_MINUS_B",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.2429119050502777,
          "query_target_nll": 5.612887382507324,
          "query_top1_on_support": 0.26953125,
          "mean_query_support_size": 2.35546875,
          "probe_belief_target_mass": 0.0671347975730896,
          "probe_belief_target_nll": 10.85142993927002,
          "probe_belief_top1_on_support": 0.0703125,
          "mean_belief_support_size": 2.48046875,
          "decoder_belief_target_mass": 0.06462427228689194,
          "decoder_belief_target_nll": 11.950287818908691,
          "decoder_belief_top1_on_support": 0.06640625,
          "mean_decoder_belief_support_size": 2.48046875
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 9,
          "query_type": "A_MINUS_B",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.5428646802902222,
          "query_target_nll": 1.5724856853485107,
          "query_top1_on_support": 0.6328125,
          "mean_query_support_size": 2.35546875,
          "probe_belief_target_mass": 0.43422770500183105,
          "probe_belief_target_nll": 2.1885383129119873,
          "probe_belief_top1_on_support": 0.51953125,
          "mean_belief_support_size": 2.48046875,
          "decoder_belief_target_mass": 0.4092726707458496,
          "decoder_belief_target_nll": 2.201195001602173,
          "decoder_belief_top1_on_support": 0.51171875,
          "mean_decoder_belief_support_size": 2.48046875
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 9,
          "query_type": "A_MINUS_B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.5428646802902222,
          "query_target_nll": 1.5724856853485107,
          "query_top1_on_support": 0.6328125,
          "mean_query_support_size": 2.35546875,
          "probe_belief_target_mass": 0.43422770500183105,
          "probe_belief_target_nll": 2.1885383129119873,
          "probe_belief_top1_on_support": 0.51953125,
          "mean_belief_support_size": 2.48046875,
          "decoder_belief_target_mass": 0.4092726707458496,
          "decoder_belief_target_nll": 2.201195001602173,
          "decoder_belief_top1_on_support": 0.51171875,
          "mean_decoder_belief_support_size": 2.48046875
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "A",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.1728089451789856,
          "query_target_nll": 2.40132999420166,
          "query_top1_on_support": 0.19921875,
          "mean_query_support_size": 1.90625,
          "probe_belief_target_mass": 0.01759195700287819,
          "probe_belief_target_nll": 15.309364318847656,
          "probe_belief_top1_on_support": 0.015625,
          "mean_belief_support_size": 2.21484375,
          "decoder_belief_target_mass": 0.0179782435297966,
          "decoder_belief_target_nll": 19.38961410522461,
          "decoder_belief_top1_on_support": 0.02734375,
          "mean_decoder_belief_support_size": 2.21484375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "A",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.1895776093006134,
          "query_target_nll": 4.596230506896973,
          "query_top1_on_support": 0.23828125,
          "mean_query_support_size": 1.90625,
          "probe_belief_target_mass": 0.029960421845316887,
          "probe_belief_target_nll": 14.00480842590332,
          "probe_belief_top1_on_support": 0.0390625,
          "mean_belief_support_size": 2.21484375,
          "decoder_belief_target_mass": 0.02935856580734253,
          "decoder_belief_target_nll": 15.632859230041504,
          "decoder_belief_top1_on_support": 0.03125,
          "mean_decoder_belief_support_size": 2.21484375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "A",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.18346244096755981,
          "query_target_nll": 5.49724006652832,
          "query_top1_on_support": 0.1875,
          "mean_query_support_size": 1.90625,
          "probe_belief_target_mass": 0.021345149725675583,
          "probe_belief_target_nll": 13.994073867797852,
          "probe_belief_top1_on_support": 0.03125,
          "mean_belief_support_size": 2.21484375,
          "decoder_belief_target_mass": 0.020256729796528816,
          "decoder_belief_target_nll": 15.665922164916992,
          "decoder_belief_top1_on_support": 0.02734375,
          "mean_decoder_belief_support_size": 2.21484375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "A",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.18111692368984222,
          "query_target_nll": 6.0521345138549805,
          "query_top1_on_support": 0.16796875,
          "mean_query_support_size": 1.90625,
          "probe_belief_target_mass": 0.024726849049329758,
          "probe_belief_target_nll": 13.187455177307129,
          "probe_belief_top1_on_support": 0.02734375,
          "mean_belief_support_size": 2.21484375,
          "decoder_belief_target_mass": 0.02456977590918541,
          "decoder_belief_target_nll": 14.699136734008789,
          "decoder_belief_top1_on_support": 0.02734375,
          "mean_decoder_belief_support_size": 2.21484375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "A",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.19648991525173187,
          "query_target_nll": 5.952618598937988,
          "query_top1_on_support": 0.21484375,
          "mean_query_support_size": 1.90625,
          "probe_belief_target_mass": 0.02175248973071575,
          "probe_belief_target_nll": 12.157074928283691,
          "probe_belief_top1_on_support": 0.02734375,
          "mean_belief_support_size": 2.21484375,
          "decoder_belief_target_mass": 0.020032137632369995,
          "decoder_belief_target_nll": 13.398677825927734,
          "decoder_belief_top1_on_support": 0.015625,
          "mean_decoder_belief_support_size": 2.21484375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "A",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.2863420844078064,
          "query_target_nll": 4.814497947692871,
          "query_top1_on_support": 0.32421875,
          "mean_query_support_size": 1.90625,
          "probe_belief_target_mass": 0.05659432336688042,
          "probe_belief_target_nll": 10.02192211151123,
          "probe_belief_top1_on_support": 0.05859375,
          "mean_belief_support_size": 2.21484375,
          "decoder_belief_target_mass": 0.054248347878456116,
          "decoder_belief_target_nll": 11.05203914642334,
          "decoder_belief_top1_on_support": 0.0546875,
          "mean_decoder_belief_support_size": 2.21484375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "A",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.5252021551132202,
          "query_target_nll": 1.350102186203003,
          "query_top1_on_support": 0.5859375,
          "mean_query_support_size": 1.90625,
          "probe_belief_target_mass": 0.3174988627433777,
          "probe_belief_target_nll": 2.4939723014831543,
          "probe_belief_top1_on_support": 0.40625,
          "mean_belief_support_size": 2.21484375,
          "decoder_belief_target_mass": 0.3075821101665497,
          "decoder_belief_target_nll": 2.5038814544677734,
          "decoder_belief_top1_on_support": 0.36328125,
          "mean_decoder_belief_support_size": 2.21484375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.1535613089799881,
          "query_target_nll": 2.395480155944824,
          "query_top1_on_support": 0.1796875,
          "mean_query_support_size": 1.6796875,
          "probe_belief_target_mass": 0.01352666411548853,
          "probe_belief_target_nll": 15.342855453491211,
          "probe_belief_top1_on_support": 0.0078125,
          "mean_belief_support_size": 1.84765625,
          "decoder_belief_target_mass": 0.013161150738596916,
          "decoder_belief_target_nll": 19.677635192871094,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 1.84765625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.1536579132080078,
          "query_target_nll": 4.527849197387695,
          "query_top1_on_support": 0.171875,
          "mean_query_support_size": 1.6796875,
          "probe_belief_target_mass": 0.014450552873313427,
          "probe_belief_target_nll": 14.38009262084961,
          "probe_belief_top1_on_support": 0.015625,
          "mean_belief_support_size": 1.84765625,
          "decoder_belief_target_mass": 0.015076767653226852,
          "decoder_belief_target_nll": 16.070384979248047,
          "decoder_belief_top1_on_support": 0.015625,
          "mean_decoder_belief_support_size": 1.84765625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.17384758591651917,
          "query_target_nll": 5.46311616897583,
          "query_top1_on_support": 0.15625,
          "mean_query_support_size": 1.6796875,
          "probe_belief_target_mass": 0.014373620972037315,
          "probe_belief_target_nll": 13.539337158203125,
          "probe_belief_top1_on_support": 0.015625,
          "mean_belief_support_size": 1.84765625,
          "decoder_belief_target_mass": 0.014749206602573395,
          "decoder_belief_target_nll": 14.885804176330566,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 1.84765625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "B",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.17379039525985718,
          "query_target_nll": 5.8033342361450195,
          "query_top1_on_support": 0.19140625,
          "mean_query_support_size": 1.6796875,
          "probe_belief_target_mass": 0.012541367672383785,
          "probe_belief_target_nll": 13.163748741149902,
          "probe_belief_top1_on_support": 0.01171875,
          "mean_belief_support_size": 1.84765625,
          "decoder_belief_target_mass": 0.012039357796311378,
          "decoder_belief_target_nll": 14.5648775100708,
          "decoder_belief_top1_on_support": 0.01171875,
          "mean_decoder_belief_support_size": 1.84765625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "B",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.16940271854400635,
          "query_target_nll": 6.808175563812256,
          "query_top1_on_support": 0.1953125,
          "mean_query_support_size": 1.6796875,
          "probe_belief_target_mass": 0.01922665908932686,
          "probe_belief_target_nll": 12.4888916015625,
          "probe_belief_top1_on_support": 0.01953125,
          "mean_belief_support_size": 1.84765625,
          "decoder_belief_target_mass": 0.01830964721739292,
          "decoder_belief_target_nll": 13.906400680541992,
          "decoder_belief_top1_on_support": 0.015625,
          "mean_decoder_belief_support_size": 1.84765625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "B",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.24219420552253723,
          "query_target_nll": 5.160508632659912,
          "query_top1_on_support": 0.24609375,
          "mean_query_support_size": 1.6796875,
          "probe_belief_target_mass": 0.05537017807364464,
          "probe_belief_target_nll": 10.113310813903809,
          "probe_belief_top1_on_support": 0.078125,
          "mean_belief_support_size": 1.84765625,
          "decoder_belief_target_mass": 0.05131060630083084,
          "decoder_belief_target_nll": 11.004091262817383,
          "decoder_belief_top1_on_support": 0.0625,
          "mean_decoder_belief_support_size": 1.84765625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.4800485372543335,
          "query_target_nll": 1.5266087055206299,
          "query_top1_on_support": 0.5546875,
          "mean_query_support_size": 1.6796875,
          "probe_belief_target_mass": 0.28447046875953674,
          "probe_belief_target_nll": 2.467773914337158,
          "probe_belief_top1_on_support": 0.33203125,
          "mean_belief_support_size": 1.84765625,
          "decoder_belief_target_mass": 0.2732270359992981,
          "decoder_belief_target_nll": 2.488243579864502,
          "decoder_belief_top1_on_support": 0.33984375,
          "mean_decoder_belief_support_size": 1.84765625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "A_PLUS_B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.17559148371219635,
          "query_target_nll": 2.406614065170288,
          "query_top1_on_support": 0.15625,
          "mean_query_support_size": 1.93359375,
          "probe_belief_target_mass": 0.02537117898464203,
          "probe_belief_target_nll": 14.827905654907227,
          "probe_belief_top1_on_support": 0.03125,
          "mean_belief_support_size": 2.0,
          "decoder_belief_target_mass": 0.024884548038244247,
          "decoder_belief_target_nll": 19.104557037353516,
          "decoder_belief_top1_on_support": 0.04296875,
          "mean_decoder_belief_support_size": 2.0
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "A_PLUS_B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.18435955047607422,
          "query_target_nll": 3.2988853454589844,
          "query_top1_on_support": 0.1875,
          "mean_query_support_size": 1.93359375,
          "probe_belief_target_mass": 0.015134558081626892,
          "probe_belief_target_nll": 14.39096736907959,
          "probe_belief_top1_on_support": 0.01171875,
          "mean_belief_support_size": 2.0,
          "decoder_belief_target_mass": 0.015187263488769531,
          "decoder_belief_target_nll": 16.175838470458984,
          "decoder_belief_top1_on_support": 0.02734375,
          "mean_decoder_belief_support_size": 2.0
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "A_PLUS_B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.1757778525352478,
          "query_target_nll": 3.688588857650757,
          "query_top1_on_support": 0.17578125,
          "mean_query_support_size": 1.93359375,
          "probe_belief_target_mass": 0.020494600757956505,
          "probe_belief_target_nll": 13.951269149780273,
          "probe_belief_top1_on_support": 0.01953125,
          "mean_belief_support_size": 2.0,
          "decoder_belief_target_mass": 0.01849287934601307,
          "decoder_belief_target_nll": 15.838634490966797,
          "decoder_belief_top1_on_support": 0.0078125,
          "mean_decoder_belief_support_size": 2.0
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "A_PLUS_B",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.17734351754188538,
          "query_target_nll": 3.8751604557037354,
          "query_top1_on_support": 0.17578125,
          "mean_query_support_size": 1.93359375,
          "probe_belief_target_mass": 0.016852181404829025,
          "probe_belief_target_nll": 13.989749908447266,
          "probe_belief_top1_on_support": 0.01953125,
          "mean_belief_support_size": 2.0,
          "decoder_belief_target_mass": 0.01712791621685028,
          "decoder_belief_target_nll": 15.595837593078613,
          "decoder_belief_top1_on_support": 0.0234375,
          "mean_decoder_belief_support_size": 2.0
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "A_PLUS_B",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.16880816221237183,
          "query_target_nll": 3.903651714324951,
          "query_top1_on_support": 0.15234375,
          "mean_query_support_size": 1.93359375,
          "probe_belief_target_mass": 0.025484975427389145,
          "probe_belief_target_nll": 12.56014633178711,
          "probe_belief_top1_on_support": 0.01953125,
          "mean_belief_support_size": 2.0,
          "decoder_belief_target_mass": 0.025885585695505142,
          "decoder_belief_target_nll": 13.867462158203125,
          "decoder_belief_top1_on_support": 0.0234375,
          "mean_decoder_belief_support_size": 2.0
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "A_PLUS_B",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.18746240437030792,
          "query_target_nll": 3.831800937652588,
          "query_top1_on_support": 0.1953125,
          "mean_query_support_size": 1.93359375,
          "probe_belief_target_mass": 0.06740802526473999,
          "probe_belief_target_nll": 10.379791259765625,
          "probe_belief_top1_on_support": 0.078125,
          "mean_belief_support_size": 2.0,
          "decoder_belief_target_mass": 0.06268210709095001,
          "decoder_belief_target_nll": 11.36407470703125,
          "decoder_belief_top1_on_support": 0.078125,
          "mean_decoder_belief_support_size": 2.0
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "A_PLUS_B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.3927243947982788,
          "query_target_nll": 1.7419097423553467,
          "query_top1_on_support": 0.42578125,
          "mean_query_support_size": 1.93359375,
          "probe_belief_target_mass": 0.3264416456222534,
          "probe_belief_target_nll": 2.245014190673828,
          "probe_belief_top1_on_support": 0.390625,
          "mean_belief_support_size": 2.0,
          "decoder_belief_target_mass": 0.3091386556625366,
          "decoder_belief_target_nll": 2.304633140563965,
          "decoder_belief_top1_on_support": 0.375,
          "mean_decoder_belief_support_size": 2.0
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "A_MINUS_B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.2148425579071045,
          "query_target_nll": 15.578904151916504,
          "query_top1_on_support": 0.21484375,
          "mean_query_support_size": 1.95703125,
          "probe_belief_target_mass": 0.02023732289671898,
          "probe_belief_target_nll": 15.100421905517578,
          "probe_belief_top1_on_support": 0.01171875,
          "mean_belief_support_size": 2.17578125,
          "decoder_belief_target_mass": 0.020533373579382896,
          "decoder_belief_target_nll": 19.16492462158203,
          "decoder_belief_top1_on_support": 0.01953125,
          "mean_decoder_belief_support_size": 2.17578125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "A_MINUS_B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.1829468309879303,
          "query_target_nll": 8.169305801391602,
          "query_top1_on_support": 0.18359375,
          "mean_query_support_size": 1.95703125,
          "probe_belief_target_mass": 0.01613261178135872,
          "probe_belief_target_nll": 14.887609481811523,
          "probe_belief_top1_on_support": 0.00390625,
          "mean_belief_support_size": 2.17578125,
          "decoder_belief_target_mass": 0.016188494861125946,
          "decoder_belief_target_nll": 16.503551483154297,
          "decoder_belief_top1_on_support": 0.01171875,
          "mean_decoder_belief_support_size": 2.17578125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "A_MINUS_B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.15656763315200806,
          "query_target_nll": 7.057192802429199,
          "query_top1_on_support": 0.15625,
          "mean_query_support_size": 1.95703125,
          "probe_belief_target_mass": 0.01606237143278122,
          "probe_belief_target_nll": 14.289234161376953,
          "probe_belief_top1_on_support": 0.015625,
          "mean_belief_support_size": 2.17578125,
          "decoder_belief_target_mass": 0.01659972220659256,
          "decoder_belief_target_nll": 15.74891471862793,
          "decoder_belief_top1_on_support": 0.01171875,
          "mean_decoder_belief_support_size": 2.17578125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "A_MINUS_B",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.1642453521490097,
          "query_target_nll": 6.426852226257324,
          "query_top1_on_support": 0.16015625,
          "mean_query_support_size": 1.95703125,
          "probe_belief_target_mass": 0.013694627210497856,
          "probe_belief_target_nll": 13.645438194274902,
          "probe_belief_top1_on_support": 0.015625,
          "mean_belief_support_size": 2.17578125,
          "decoder_belief_target_mass": 0.013592161238193512,
          "decoder_belief_target_nll": 15.117996215820312,
          "decoder_belief_top1_on_support": 0.01171875,
          "mean_decoder_belief_support_size": 2.17578125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "A_MINUS_B",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.17445896565914154,
          "query_target_nll": 5.339714050292969,
          "query_top1_on_support": 0.171875,
          "mean_query_support_size": 1.95703125,
          "probe_belief_target_mass": 0.019420607015490532,
          "probe_belief_target_nll": 12.755434036254883,
          "probe_belief_top1_on_support": 0.0078125,
          "mean_belief_support_size": 2.17578125,
          "decoder_belief_target_mass": 0.021890345960855484,
          "decoder_belief_target_nll": 13.930307388305664,
          "decoder_belief_top1_on_support": 0.03125,
          "mean_decoder_belief_support_size": 2.17578125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "A_MINUS_B",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.19035300612449646,
          "query_target_nll": 5.084627628326416,
          "query_top1_on_support": 0.1875,
          "mean_query_support_size": 1.95703125,
          "probe_belief_target_mass": 0.05562637001276016,
          "probe_belief_target_nll": 10.047319412231445,
          "probe_belief_top1_on_support": 0.0625,
          "mean_belief_support_size": 2.17578125,
          "decoder_belief_target_mass": 0.05259128659963608,
          "decoder_belief_target_nll": 10.998791694641113,
          "decoder_belief_top1_on_support": 0.05859375,
          "mean_decoder_belief_support_size": 2.17578125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "A_MINUS_B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.42848414182662964,
          "query_target_nll": 1.6640095710754395,
          "query_top1_on_support": 0.50390625,
          "mean_query_support_size": 1.95703125,
          "probe_belief_target_mass": 0.27828094363212585,
          "probe_belief_target_nll": 2.564878225326538,
          "probe_belief_top1_on_support": 0.34375,
          "mean_belief_support_size": 2.17578125,
          "decoder_belief_target_mass": 0.26676586270332336,
          "decoder_belief_target_nll": 2.5975897312164307,
          "decoder_belief_top1_on_support": 0.34765625,
          "mean_decoder_belief_support_size": 2.17578125
        }
      ]
    }
  ],
  "checkpoints": [
    "large_artifacts/dense_supervision_ladder/checkpoints/pilot_full_belief_mod11/checkpoint_final.pt"
  ]
}