{
  "mode": "dense",
  "args": {
    "mode": "dense",
    "supervision": "soft_final_query",
    "seed": 53,
    "modulus": 31,
    "observe_mod": 5,
    "observe_prob": 0.3,
    "op_family": "full",
    "query_family": "all",
    "train_min_len": 1,
    "train_max_len": 8,
    "eval_lengths": "4,8,12,16,24",
    "eval_k": "0,1,2,4,8,12,16,24",
    "eval_query_types": "all",
    "state_dim": 256,
    "instr_dim": 128,
    "heads": 4,
    "compiler_layers": 2,
    "dropout": 0.0,
    "train_steps": 2000,
    "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": 800,
    "probe_batch_size": 256,
    "probe_lr": 0.001,
    "probe_weight_decay": 0.0001,
    "probe_dropout": 0.0,
    "max_grad_norm": 1.0,
    "log_every": 250,
    "probe_log_every": 200,
    "output_dir": "experiments/dense_supervision_ladder/runs/main_soft_final_query_mod31",
    "checkpoint_dir": "large_artifacts/dense_supervision_ladder/checkpoints/main_soft_final_query_mod31"
  },
  "env": {
    "python": "3.12.3",
    "torch": "2.8.0+cu128",
    "device": "cuda",
    "cuda_name": "NVIDIA RTX 6000 Ada Generation"
  },
  "train": [
    {
      "step": 1,
      "loss": 3.45640230178833,
      "grad_norm": 0.4625813364982605,
      "query_loss": 3.45640230178833,
      "belief_loss": 0.0,
      "query_target_mass": 0.3622629642486572,
      "query_target_nll": 3.233111619949341,
      "query_top1_on_support": 0.6875,
      "mean_query_support_size": 10.6796875
    },
    {
      "step": 250,
      "loss": 3.168323040008545,
      "grad_norm": 0.4190146327018738,
      "query_loss": 3.168323040008545,
      "belief_loss": 0.0,
      "query_target_mass": 0.45320039987564087,
      "query_target_nll": 3.1465535163879395,
      "query_top1_on_support": 0.51171875,
      "mean_query_support_size": 11.55859375
    },
    {
      "step": 500,
      "loss": 3.1017894744873047,
      "grad_norm": 0.4310707747936249,
      "query_loss": 3.1017894744873047,
      "belief_loss": 0.0,
      "query_target_mass": 0.46625402569770813,
      "query_target_nll": 3.074706554412842,
      "query_top1_on_support": 0.5234375,
      "mean_query_support_size": 11.1875
    },
    {
      "step": 750,
      "loss": 3.093021869659424,
      "grad_norm": 0.48244747519493103,
      "query_loss": 3.093021869659424,
      "belief_loss": 0.0,
      "query_target_mass": 0.48931053280830383,
      "query_target_nll": 3.05346417427063,
      "query_top1_on_support": 0.5703125,
      "mean_query_support_size": 11.765625
    },
    {
      "step": 1000,
      "loss": 2.951496124267578,
      "grad_norm": 0.5099939107894897,
      "query_loss": 2.951496124267578,
      "belief_loss": 0.0,
      "query_target_mass": 0.5093703269958496,
      "query_target_nll": 2.914370059967041,
      "query_top1_on_support": 0.55859375,
      "mean_query_support_size": 11.140625
    },
    {
      "step": 1250,
      "loss": 2.866025447845459,
      "grad_norm": 0.605388879776001,
      "query_loss": 2.866025447845459,
      "belief_loss": 0.0,
      "query_target_mass": 0.507054328918457,
      "query_target_nll": 2.8142433166503906,
      "query_top1_on_support": 0.6015625,
      "mean_query_support_size": 10.265625
    },
    {
      "step": 1500,
      "loss": 2.8486828804016113,
      "grad_norm": 0.6108538508415222,
      "query_loss": 2.8486828804016113,
      "belief_loss": 0.0,
      "query_target_mass": 0.5428664088249207,
      "query_target_nll": 2.804858446121216,
      "query_top1_on_support": 0.6171875,
      "mean_query_support_size": 11.32421875
    },
    {
      "step": 1750,
      "loss": 2.6788339614868164,
      "grad_norm": 0.7558682560920715,
      "query_loss": 2.6788339614868164,
      "belief_loss": 0.0,
      "query_target_mass": 0.6073743104934692,
      "query_target_nll": 2.615793228149414,
      "query_top1_on_support": 0.640625,
      "mean_query_support_size": 11.48828125
    },
    {
      "step": 2000,
      "loss": 2.642930507659912,
      "grad_norm": 0.7429605722427368,
      "query_loss": 2.642930507659912,
      "belief_loss": 0.0,
      "query_target_mass": 0.6068724989891052,
      "query_target_nll": 2.5876150131225586,
      "query_top1_on_support": 0.69140625,
      "mean_query_support_size": 11.22265625
    }
  ],
  "probe_train": [
    {
      "step": 1,
      "loss": 6.881389617919922,
      "grad_norm": 0.21637125313282013,
      "probe_belief_target_mass": 0.019881686195731163,
      "probe_belief_target_nll": 6.881389617919922,
      "probe_belief_top1_on_support": 0.0173611119389534,
      "mean_belief_support_size": 19.06727409362793
    },
    {
      "step": 200,
      "loss": 4.239132404327393,
      "grad_norm": 0.4779316186904907,
      "probe_belief_target_mass": 0.36774152517318726,
      "probe_belief_target_nll": 4.239132404327393,
      "probe_belief_top1_on_support": 0.4869791567325592,
      "mean_belief_support_size": 17.91232681274414
    },
    {
      "step": 400,
      "loss": 4.105198860168457,
      "grad_norm": 0.44132915139198303,
      "probe_belief_target_mass": 0.4111997187137604,
      "probe_belief_target_nll": 4.105198860168457,
      "probe_belief_top1_on_support": 0.4826388955116272,
      "mean_belief_support_size": 17.484375
    },
    {
      "step": 600,
      "loss": 4.070186138153076,
      "grad_norm": 0.4261070787906647,
      "probe_belief_target_mass": 0.44417962431907654,
      "probe_belief_target_nll": 4.070186138153076,
      "probe_belief_top1_on_support": 0.534288227558136,
      "mean_belief_support_size": 17.92447853088379
    },
    {
      "step": 800,
      "loss": 3.9932243824005127,
      "grad_norm": 0.4263628125190735,
      "probe_belief_target_mass": 0.4591456651687622,
      "probe_belief_target_nll": 3.9932243824005127,
      "probe_belief_top1_on_support": 0.5412326455116272,
      "mean_belief_support_size": 18.07725715637207
    }
  ],
  "eval": [
    {
      "step": 2000,
      "rows": [
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 4,
          "query_type": "A",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.38445043563842773,
          "query_target_nll": 3.829455852508545,
          "query_top1_on_support": 0.38671875,
          "mean_query_support_size": 12.01171875,
          "probe_belief_target_mass": 0.02380499430000782,
          "probe_belief_target_nll": 19.354040145874023,
          "probe_belief_top1_on_support": 0.01953125,
          "mean_belief_support_size": 13.84765625,
          "decoder_belief_target_mass": 0.014473231509327888,
          "decoder_belief_target_nll": 6.882615089416504,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 13.84765625
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 4,
          "query_type": "A",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.4327858090400696,
          "query_target_nll": 4.842629909515381,
          "query_top1_on_support": 0.45703125,
          "mean_query_support_size": 12.01171875,
          "probe_belief_target_mass": 0.04502515122294426,
          "probe_belief_target_nll": 16.73960304260254,
          "probe_belief_top1_on_support": 0.05078125,
          "mean_belief_support_size": 13.84765625,
          "decoder_belief_target_mass": 0.01440100185573101,
          "decoder_belief_target_nll": 6.89183235168457,
          "decoder_belief_top1_on_support": 0.01953125,
          "mean_decoder_belief_support_size": 13.84765625
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 4,
          "query_type": "A",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.5138328075408936,
          "query_target_nll": 4.0847954750061035,
          "query_top1_on_support": 0.5390625,
          "mean_query_support_size": 12.01171875,
          "probe_belief_target_mass": 0.07822360843420029,
          "probe_belief_target_nll": 11.846809387207031,
          "probe_belief_top1_on_support": 0.1171875,
          "mean_belief_support_size": 13.84765625,
          "decoder_belief_target_mass": 0.0143691785633564,
          "decoder_belief_target_nll": 6.895624160766602,
          "decoder_belief_top1_on_support": 0.015625,
          "mean_decoder_belief_support_size": 13.84765625
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 4,
          "query_type": "A",
          "k": 4,
          "n": 256,
          "query_target_mass": 0.6874411106109619,
          "query_target_nll": 2.5540354251861572,
          "query_top1_on_support": 0.78125,
          "mean_query_support_size": 12.01171875,
          "probe_belief_target_mass": 0.1848699003458023,
          "probe_belief_target_nll": 4.6163482666015625,
          "probe_belief_top1_on_support": 0.296875,
          "mean_belief_support_size": 13.84765625,
          "decoder_belief_target_mass": 0.014413072727620602,
          "decoder_belief_target_nll": 6.886771202087402,
          "decoder_belief_top1_on_support": 0.015625,
          "mean_decoder_belief_support_size": 13.84765625
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 4,
          "query_type": "A",
          "k": 8,
          "n": 256,
          "query_target_mass": 0.6874411106109619,
          "query_target_nll": 2.5540354251861572,
          "query_top1_on_support": 0.78125,
          "mean_query_support_size": 12.01171875,
          "probe_belief_target_mass": 0.1848699003458023,
          "probe_belief_target_nll": 4.6163482666015625,
          "probe_belief_top1_on_support": 0.296875,
          "mean_belief_support_size": 13.84765625,
          "decoder_belief_target_mass": 0.014413072727620602,
          "decoder_belief_target_nll": 6.886771202087402,
          "decoder_belief_top1_on_support": 0.015625,
          "mean_decoder_belief_support_size": 13.84765625
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 4,
          "query_type": "A",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.6874411106109619,
          "query_target_nll": 2.5540354251861572,
          "query_top1_on_support": 0.78125,
          "mean_query_support_size": 12.01171875,
          "probe_belief_target_mass": 0.1848699003458023,
          "probe_belief_target_nll": 4.6163482666015625,
          "probe_belief_top1_on_support": 0.296875,
          "mean_belief_support_size": 13.84765625,
          "decoder_belief_target_mass": 0.014413072727620602,
          "decoder_belief_target_nll": 6.886771202087402,
          "decoder_belief_top1_on_support": 0.015625,
          "mean_decoder_belief_support_size": 13.84765625
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 4,
          "query_type": "A",
          "k": 16,
          "n": 256,
          "query_target_mass": 0.6874411106109619,
          "query_target_nll": 2.5540354251861572,
          "query_top1_on_support": 0.78125,
          "mean_query_support_size": 12.01171875,
          "probe_belief_target_mass": 0.1848699003458023,
          "probe_belief_target_nll": 4.6163482666015625,
          "probe_belief_top1_on_support": 0.296875,
          "mean_belief_support_size": 13.84765625,
          "decoder_belief_target_mass": 0.014413072727620602,
          "decoder_belief_target_nll": 6.886771202087402,
          "decoder_belief_top1_on_support": 0.015625,
          "mean_decoder_belief_support_size": 13.84765625
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 4,
          "query_type": "A",
          "k": 24,
          "n": 256,
          "query_target_mass": 0.6874411106109619,
          "query_target_nll": 2.5540354251861572,
          "query_top1_on_support": 0.78125,
          "mean_query_support_size": 12.01171875,
          "probe_belief_target_mass": 0.1848699003458023,
          "probe_belief_target_nll": 4.6163482666015625,
          "probe_belief_top1_on_support": 0.296875,
          "mean_belief_support_size": 13.84765625,
          "decoder_belief_target_mass": 0.014413072727620602,
          "decoder_belief_target_nll": 6.886771202087402,
          "decoder_belief_top1_on_support": 0.015625,
          "mean_decoder_belief_support_size": 13.84765625
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 4,
          "query_type": "B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.36701756715774536,
          "query_target_nll": 3.852468252182007,
          "query_top1_on_support": 0.40625,
          "mean_query_support_size": 11.109375,
          "probe_belief_target_mass": 0.024714376777410507,
          "probe_belief_target_nll": 18.894641876220703,
          "probe_belief_top1_on_support": 0.04296875,
          "mean_belief_support_size": 13.328125,
          "decoder_belief_target_mass": 0.013880385085940361,
          "decoder_belief_target_nll": 6.888951301574707,
          "decoder_belief_top1_on_support": 0.0078125,
          "mean_decoder_belief_support_size": 13.328125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 4,
          "query_type": "B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.4158822000026703,
          "query_target_nll": 4.89917516708374,
          "query_top1_on_support": 0.41015625,
          "mean_query_support_size": 11.109375,
          "probe_belief_target_mass": 0.07461216300725937,
          "probe_belief_target_nll": 15.791223526000977,
          "probe_belief_top1_on_support": 0.0703125,
          "mean_belief_support_size": 13.328125,
          "decoder_belief_target_mass": 0.013834385201334953,
          "decoder_belief_target_nll": 6.890925407409668,
          "decoder_belief_top1_on_support": 0.015625,
          "mean_decoder_belief_support_size": 13.328125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 4,
          "query_type": "B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.49023762345314026,
          "query_target_nll": 4.3119096755981445,
          "query_top1_on_support": 0.5234375,
          "mean_query_support_size": 11.109375,
          "probe_belief_target_mass": 0.09428292512893677,
          "probe_belief_target_nll": 12.336103439331055,
          "probe_belief_top1_on_support": 0.10546875,
          "mean_belief_support_size": 13.328125,
          "decoder_belief_target_mass": 0.013833190314471722,
          "decoder_belief_target_nll": 6.888124465942383,
          "decoder_belief_top1_on_support": 0.015625,
          "mean_decoder_belief_support_size": 13.328125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 4,
          "query_type": "B",
          "k": 4,
          "n": 256,
          "query_target_mass": 0.6988683938980103,
          "query_target_nll": 2.4297914505004883,
          "query_top1_on_support": 0.7734375,
          "mean_query_support_size": 11.109375,
          "probe_belief_target_mass": 0.22802037000656128,
          "probe_belief_target_nll": 4.319206237792969,
          "probe_belief_top1_on_support": 0.33984375,
          "mean_belief_support_size": 13.328125,
          "decoder_belief_target_mass": 0.013785714283585548,
          "decoder_belief_target_nll": 6.8930511474609375,
          "decoder_belief_top1_on_support": 0.01171875,
          "mean_decoder_belief_support_size": 13.328125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 4,
          "query_type": "B",
          "k": 8,
          "n": 256,
          "query_target_mass": 0.6988683938980103,
          "query_target_nll": 2.4297914505004883,
          "query_top1_on_support": 0.7734375,
          "mean_query_support_size": 11.109375,
          "probe_belief_target_mass": 0.22802037000656128,
          "probe_belief_target_nll": 4.319206237792969,
          "probe_belief_top1_on_support": 0.33984375,
          "mean_belief_support_size": 13.328125,
          "decoder_belief_target_mass": 0.013785714283585548,
          "decoder_belief_target_nll": 6.8930511474609375,
          "decoder_belief_top1_on_support": 0.01171875,
          "mean_decoder_belief_support_size": 13.328125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 4,
          "query_type": "B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.6988683938980103,
          "query_target_nll": 2.4297914505004883,
          "query_top1_on_support": 0.7734375,
          "mean_query_support_size": 11.109375,
          "probe_belief_target_mass": 0.22802037000656128,
          "probe_belief_target_nll": 4.319206237792969,
          "probe_belief_top1_on_support": 0.33984375,
          "mean_belief_support_size": 13.328125,
          "decoder_belief_target_mass": 0.013785714283585548,
          "decoder_belief_target_nll": 6.8930511474609375,
          "decoder_belief_top1_on_support": 0.01171875,
          "mean_decoder_belief_support_size": 13.328125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 4,
          "query_type": "B",
          "k": 16,
          "n": 256,
          "query_target_mass": 0.6988683938980103,
          "query_target_nll": 2.4297914505004883,
          "query_top1_on_support": 0.7734375,
          "mean_query_support_size": 11.109375,
          "probe_belief_target_mass": 0.22802037000656128,
          "probe_belief_target_nll": 4.319206237792969,
          "probe_belief_top1_on_support": 0.33984375,
          "mean_belief_support_size": 13.328125,
          "decoder_belief_target_mass": 0.013785714283585548,
          "decoder_belief_target_nll": 6.8930511474609375,
          "decoder_belief_top1_on_support": 0.01171875,
          "mean_decoder_belief_support_size": 13.328125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 4,
          "query_type": "B",
          "k": 24,
          "n": 256,
          "query_target_mass": 0.6988683938980103,
          "query_target_nll": 2.4297914505004883,
          "query_top1_on_support": 0.7734375,
          "mean_query_support_size": 11.109375,
          "probe_belief_target_mass": 0.22802037000656128,
          "probe_belief_target_nll": 4.319206237792969,
          "probe_belief_top1_on_support": 0.33984375,
          "mean_belief_support_size": 13.328125,
          "decoder_belief_target_mass": 0.013785714283585548,
          "decoder_belief_target_nll": 6.8930511474609375,
          "decoder_belief_top1_on_support": 0.01171875,
          "mean_decoder_belief_support_size": 13.328125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 4,
          "query_type": "A_PLUS_B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.37111997604370117,
          "query_target_nll": 3.64902663230896,
          "query_top1_on_support": 0.37109375,
          "mean_query_support_size": 11.453125,
          "probe_belief_target_mass": 0.031158631667494774,
          "probe_belief_target_nll": 18.059534072875977,
          "probe_belief_top1_on_support": 0.03125,
          "mean_belief_support_size": 12.0859375,
          "decoder_belief_target_mass": 0.012656964361667633,
          "decoder_belief_target_nll": 6.881086349487305,
          "decoder_belief_top1_on_support": 0.015625,
          "mean_decoder_belief_support_size": 12.0859375
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 4,
          "query_type": "A_PLUS_B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.38030755519866943,
          "query_target_nll": 3.9795570373535156,
          "query_top1_on_support": 0.39453125,
          "mean_query_support_size": 11.453125,
          "probe_belief_target_mass": 0.0791286751627922,
          "probe_belief_target_nll": 15.379077911376953,
          "probe_belief_top1_on_support": 0.0859375,
          "mean_belief_support_size": 12.0859375,
          "decoder_belief_target_mass": 0.012553470209240913,
          "decoder_belief_target_nll": 6.888347625732422,
          "decoder_belief_top1_on_support": 0.01171875,
          "mean_decoder_belief_support_size": 12.0859375
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 4,
          "query_type": "A_PLUS_B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.39280179142951965,
          "query_target_nll": 3.995708465576172,
          "query_top1_on_support": 0.39453125,
          "mean_query_support_size": 11.453125,
          "probe_belief_target_mass": 0.10700231790542603,
          "probe_belief_target_nll": 11.852011680603027,
          "probe_belief_top1_on_support": 0.15234375,
          "mean_belief_support_size": 12.0859375,
          "decoder_belief_target_mass": 0.01256699301302433,
          "decoder_belief_target_nll": 6.8907470703125,
          "decoder_belief_top1_on_support": 0.01171875,
          "mean_decoder_belief_support_size": 12.0859375
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 4,
          "query_type": "A_PLUS_B",
          "k": 4,
          "n": 256,
          "query_target_mass": 0.5369834303855896,
          "query_target_nll": 2.923234701156616,
          "query_top1_on_support": 0.60546875,
          "mean_query_support_size": 11.453125,
          "probe_belief_target_mass": 0.22563230991363525,
          "probe_belief_target_nll": 4.300252914428711,
          "probe_belief_top1_on_support": 0.34375,
          "mean_belief_support_size": 12.0859375,
          "decoder_belief_target_mass": 0.012538041919469833,
          "decoder_belief_target_nll": 6.889655113220215,
          "decoder_belief_top1_on_support": 0.0078125,
          "mean_decoder_belief_support_size": 12.0859375
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 4,
          "query_type": "A_PLUS_B",
          "k": 8,
          "n": 256,
          "query_target_mass": 0.5369834303855896,
          "query_target_nll": 2.923234701156616,
          "query_top1_on_support": 0.60546875,
          "mean_query_support_size": 11.453125,
          "probe_belief_target_mass": 0.22563230991363525,
          "probe_belief_target_nll": 4.300252914428711,
          "probe_belief_top1_on_support": 0.34375,
          "mean_belief_support_size": 12.0859375,
          "decoder_belief_target_mass": 0.012538041919469833,
          "decoder_belief_target_nll": 6.889655113220215,
          "decoder_belief_top1_on_support": 0.0078125,
          "mean_decoder_belief_support_size": 12.0859375
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 4,
          "query_type": "A_PLUS_B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.5369834303855896,
          "query_target_nll": 2.923234701156616,
          "query_top1_on_support": 0.60546875,
          "mean_query_support_size": 11.453125,
          "probe_belief_target_mass": 0.22563230991363525,
          "probe_belief_target_nll": 4.300252914428711,
          "probe_belief_top1_on_support": 0.34375,
          "mean_belief_support_size": 12.0859375,
          "decoder_belief_target_mass": 0.012538041919469833,
          "decoder_belief_target_nll": 6.889655113220215,
          "decoder_belief_top1_on_support": 0.0078125,
          "mean_decoder_belief_support_size": 12.0859375
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 4,
          "query_type": "A_PLUS_B",
          "k": 16,
          "n": 256,
          "query_target_mass": 0.5369834303855896,
          "query_target_nll": 2.923234701156616,
          "query_top1_on_support": 0.60546875,
          "mean_query_support_size": 11.453125,
          "probe_belief_target_mass": 0.22563230991363525,
          "probe_belief_target_nll": 4.300252914428711,
          "probe_belief_top1_on_support": 0.34375,
          "mean_belief_support_size": 12.0859375,
          "decoder_belief_target_mass": 0.012538041919469833,
          "decoder_belief_target_nll": 6.889655113220215,
          "decoder_belief_top1_on_support": 0.0078125,
          "mean_decoder_belief_support_size": 12.0859375
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 4,
          "query_type": "A_PLUS_B",
          "k": 24,
          "n": 256,
          "query_target_mass": 0.5369834303855896,
          "query_target_nll": 2.923234701156616,
          "query_top1_on_support": 0.60546875,
          "mean_query_support_size": 11.453125,
          "probe_belief_target_mass": 0.22563230991363525,
          "probe_belief_target_nll": 4.300252914428711,
          "probe_belief_top1_on_support": 0.34375,
          "mean_belief_support_size": 12.0859375,
          "decoder_belief_target_mass": 0.012538041919469833,
          "decoder_belief_target_nll": 6.889655113220215,
          "decoder_belief_top1_on_support": 0.0078125,
          "mean_decoder_belief_support_size": 12.0859375
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 4,
          "query_type": "A_MINUS_B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.42961734533309937,
          "query_target_nll": 12.487470626831055,
          "query_top1_on_support": 0.4296875,
          "mean_query_support_size": 11.60546875,
          "probe_belief_target_mass": 0.034862447530031204,
          "probe_belief_target_nll": 18.911415100097656,
          "probe_belief_top1_on_support": 0.02734375,
          "mean_belief_support_size": 14.40625,
          "decoder_belief_target_mass": 0.015012194402515888,
          "decoder_belief_target_nll": 6.885607719421387,
          "decoder_belief_top1_on_support": 0.01171875,
          "mean_decoder_belief_support_size": 14.40625
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 4,
          "query_type": "A_MINUS_B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.39935430884361267,
          "query_target_nll": 7.5761518478393555,
          "query_top1_on_support": 0.41796875,
          "mean_query_support_size": 11.60546875,
          "probe_belief_target_mass": 0.06427773833274841,
          "probe_belief_target_nll": 16.640716552734375,
          "probe_belief_top1_on_support": 0.0546875,
          "mean_belief_support_size": 14.40625,
          "decoder_belief_target_mass": 0.014906640164554119,
          "decoder_belief_target_nll": 6.893762111663818,
          "decoder_belief_top1_on_support": 0.0078125,
          "mean_decoder_belief_support_size": 14.40625
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 4,
          "query_type": "A_MINUS_B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.406169593334198,
          "query_target_nll": 5.734826564788818,
          "query_top1_on_support": 0.40625,
          "mean_query_support_size": 11.60546875,
          "probe_belief_target_mass": 0.10137493908405304,
          "probe_belief_target_nll": 11.879800796508789,
          "probe_belief_top1_on_support": 0.12109375,
          "mean_belief_support_size": 14.40625,
          "decoder_belief_target_mass": 0.014971205964684486,
          "decoder_belief_target_nll": 6.889444351196289,
          "decoder_belief_top1_on_support": 0.015625,
          "mean_decoder_belief_support_size": 14.40625
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 4,
          "query_type": "A_MINUS_B",
          "k": 4,
          "n": 256,
          "query_target_mass": 0.5567877292633057,
          "query_target_nll": 2.6992759704589844,
          "query_top1_on_support": 0.62890625,
          "mean_query_support_size": 11.60546875,
          "probe_belief_target_mass": 0.21204710006713867,
          "probe_belief_target_nll": 4.514213562011719,
          "probe_belief_top1_on_support": 0.3203125,
          "mean_belief_support_size": 14.40625,
          "decoder_belief_target_mass": 0.014978311024606228,
          "decoder_belief_target_nll": 6.886789321899414,
          "decoder_belief_top1_on_support": 0.0078125,
          "mean_decoder_belief_support_size": 14.40625
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 4,
          "query_type": "A_MINUS_B",
          "k": 8,
          "n": 256,
          "query_target_mass": 0.5567877292633057,
          "query_target_nll": 2.6992759704589844,
          "query_top1_on_support": 0.62890625,
          "mean_query_support_size": 11.60546875,
          "probe_belief_target_mass": 0.21204710006713867,
          "probe_belief_target_nll": 4.514213562011719,
          "probe_belief_top1_on_support": 0.3203125,
          "mean_belief_support_size": 14.40625,
          "decoder_belief_target_mass": 0.014978311024606228,
          "decoder_belief_target_nll": 6.886789321899414,
          "decoder_belief_top1_on_support": 0.0078125,
          "mean_decoder_belief_support_size": 14.40625
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 4,
          "query_type": "A_MINUS_B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.5567877292633057,
          "query_target_nll": 2.6992759704589844,
          "query_top1_on_support": 0.62890625,
          "mean_query_support_size": 11.60546875,
          "probe_belief_target_mass": 0.21204710006713867,
          "probe_belief_target_nll": 4.514213562011719,
          "probe_belief_top1_on_support": 0.3203125,
          "mean_belief_support_size": 14.40625,
          "decoder_belief_target_mass": 0.014978311024606228,
          "decoder_belief_target_nll": 6.886789321899414,
          "decoder_belief_top1_on_support": 0.0078125,
          "mean_decoder_belief_support_size": 14.40625
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 4,
          "query_type": "A_MINUS_B",
          "k": 16,
          "n": 256,
          "query_target_mass": 0.5567877292633057,
          "query_target_nll": 2.6992759704589844,
          "query_top1_on_support": 0.62890625,
          "mean_query_support_size": 11.60546875,
          "probe_belief_target_mass": 0.21204710006713867,
          "probe_belief_target_nll": 4.514213562011719,
          "probe_belief_top1_on_support": 0.3203125,
          "mean_belief_support_size": 14.40625,
          "decoder_belief_target_mass": 0.014978311024606228,
          "decoder_belief_target_nll": 6.886789321899414,
          "decoder_belief_top1_on_support": 0.0078125,
          "mean_decoder_belief_support_size": 14.40625
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 4,
          "query_type": "A_MINUS_B",
          "k": 24,
          "n": 256,
          "query_target_mass": 0.5567877292633057,
          "query_target_nll": 2.6992759704589844,
          "query_top1_on_support": 0.62890625,
          "mean_query_support_size": 11.60546875,
          "probe_belief_target_mass": 0.21204710006713867,
          "probe_belief_target_nll": 4.514213562011719,
          "probe_belief_top1_on_support": 0.3203125,
          "mean_belief_support_size": 14.40625,
          "decoder_belief_target_mass": 0.014978311024606228,
          "decoder_belief_target_nll": 6.886789321899414,
          "decoder_belief_top1_on_support": 0.0078125,
          "mean_decoder_belief_support_size": 14.40625
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 8,
          "query_type": "A",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.17376890778541565,
          "query_target_nll": 3.833831548690796,
          "query_top1_on_support": 0.1875,
          "mean_query_support_size": 5.2890625,
          "probe_belief_target_mass": 0.007923955097794533,
          "probe_belief_target_nll": 19.82428550720215,
          "probe_belief_top1_on_support": 0.01171875,
          "mean_belief_support_size": 7.140625,
          "decoder_belief_target_mass": 0.007427264470607042,
          "decoder_belief_target_nll": 6.892525672912598,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 7.140625
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 8,
          "query_type": "A",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.19700828194618225,
          "query_target_nll": 5.944962024688721,
          "query_top1_on_support": 0.203125,
          "mean_query_support_size": 5.2890625,
          "probe_belief_target_mass": 0.012759552337229252,
          "probe_belief_target_nll": 19.305845260620117,
          "probe_belief_top1_on_support": 0.015625,
          "mean_belief_support_size": 7.140625,
          "decoder_belief_target_mass": 0.0074480148032307625,
          "decoder_belief_target_nll": 6.8873701095581055,
          "decoder_belief_top1_on_support": 0.0078125,
          "mean_decoder_belief_support_size": 7.140625
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 8,
          "query_type": "A",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.20496457815170288,
          "query_target_nll": 5.839746475219727,
          "query_top1_on_support": 0.21875,
          "mean_query_support_size": 5.2890625,
          "probe_belief_target_mass": 0.012266919016838074,
          "probe_belief_target_nll": 15.938765525817871,
          "probe_belief_top1_on_support": 0.015625,
          "mean_belief_support_size": 7.140625,
          "decoder_belief_target_mass": 0.007451231591403484,
          "decoder_belief_target_nll": 6.888549327850342,
          "decoder_belief_top1_on_support": 0.01171875,
          "mean_decoder_belief_support_size": 7.140625
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 8,
          "query_type": "A",
          "k": 4,
          "n": 256,
          "query_target_mass": 0.20835761725902557,
          "query_target_nll": 6.13176965713501,
          "query_top1_on_support": 0.23046875,
          "mean_query_support_size": 5.2890625,
          "probe_belief_target_mass": 0.021832898259162903,
          "probe_belief_target_nll": 13.043587684631348,
          "probe_belief_top1_on_support": 0.03515625,
          "mean_belief_support_size": 7.140625,
          "decoder_belief_target_mass": 0.007455992978066206,
          "decoder_belief_target_nll": 6.884099960327148,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 7.140625
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 8,
          "query_type": "A",
          "k": 8,
          "n": 256,
          "query_target_mass": 0.42524829506874084,
          "query_target_nll": 2.338374376296997,
          "query_top1_on_support": 0.5078125,
          "mean_query_support_size": 5.2890625,
          "probe_belief_target_mass": 0.09329772740602493,
          "probe_belief_target_nll": 4.48843240737915,
          "probe_belief_top1_on_support": 0.140625,
          "mean_belief_support_size": 7.140625,
          "decoder_belief_target_mass": 0.007459722459316254,
          "decoder_belief_target_nll": 6.879897117614746,
          "decoder_belief_top1_on_support": 0.0234375,
          "mean_decoder_belief_support_size": 7.140625
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 8,
          "query_type": "A",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.42524829506874084,
          "query_target_nll": 2.338374376296997,
          "query_top1_on_support": 0.5078125,
          "mean_query_support_size": 5.2890625,
          "probe_belief_target_mass": 0.09329772740602493,
          "probe_belief_target_nll": 4.48843240737915,
          "probe_belief_top1_on_support": 0.140625,
          "mean_belief_support_size": 7.140625,
          "decoder_belief_target_mass": 0.007459722459316254,
          "decoder_belief_target_nll": 6.879897117614746,
          "decoder_belief_top1_on_support": 0.0234375,
          "mean_decoder_belief_support_size": 7.140625
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 8,
          "query_type": "A",
          "k": 16,
          "n": 256,
          "query_target_mass": 0.42524829506874084,
          "query_target_nll": 2.338374376296997,
          "query_top1_on_support": 0.5078125,
          "mean_query_support_size": 5.2890625,
          "probe_belief_target_mass": 0.09329772740602493,
          "probe_belief_target_nll": 4.48843240737915,
          "probe_belief_top1_on_support": 0.140625,
          "mean_belief_support_size": 7.140625,
          "decoder_belief_target_mass": 0.007459722459316254,
          "decoder_belief_target_nll": 6.879897117614746,
          "decoder_belief_top1_on_support": 0.0234375,
          "mean_decoder_belief_support_size": 7.140625
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 8,
          "query_type": "A",
          "k": 24,
          "n": 256,
          "query_target_mass": 0.42524829506874084,
          "query_target_nll": 2.338374376296997,
          "query_top1_on_support": 0.5078125,
          "mean_query_support_size": 5.2890625,
          "probe_belief_target_mass": 0.09329772740602493,
          "probe_belief_target_nll": 4.48843240737915,
          "probe_belief_top1_on_support": 0.140625,
          "mean_belief_support_size": 7.140625,
          "decoder_belief_target_mass": 0.007459722459316254,
          "decoder_belief_target_nll": 6.879897117614746,
          "decoder_belief_top1_on_support": 0.0234375,
          "mean_decoder_belief_support_size": 7.140625
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 8,
          "query_type": "B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.20010113716125488,
          "query_target_nll": 3.819891929626465,
          "query_top1_on_support": 0.203125,
          "mean_query_support_size": 5.84765625,
          "probe_belief_target_mass": 0.007543712388724089,
          "probe_belief_target_nll": 19.785362243652344,
          "probe_belief_top1_on_support": 0.0078125,
          "mean_belief_support_size": 7.09765625,
          "decoder_belief_target_mass": 0.0073592206463217735,
          "decoder_belief_target_nll": 6.897022724151611,
          "decoder_belief_top1_on_support": 0.0078125,
          "mean_decoder_belief_support_size": 7.09765625
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 8,
          "query_type": "B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.2009735256433487,
          "query_target_nll": 5.87109899520874,
          "query_top1_on_support": 0.1953125,
          "mean_query_support_size": 5.84765625,
          "probe_belief_target_mass": 0.011164918541908264,
          "probe_belief_target_nll": 18.612525939941406,
          "probe_belief_top1_on_support": 0.0078125,
          "mean_belief_support_size": 7.09765625,
          "decoder_belief_target_mass": 0.0073613254353404045,
          "decoder_belief_target_nll": 6.896054744720459,
          "decoder_belief_top1_on_support": 0.015625,
          "mean_decoder_belief_support_size": 7.09765625
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 8,
          "query_type": "B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.2184755802154541,
          "query_target_nll": 5.909079551696777,
          "query_top1_on_support": 0.25,
          "mean_query_support_size": 5.84765625,
          "probe_belief_target_mass": 0.0163312628865242,
          "probe_belief_target_nll": 15.220398902893066,
          "probe_belief_top1_on_support": 0.01953125,
          "mean_belief_support_size": 7.09765625,
          "decoder_belief_target_mass": 0.007377458270639181,
          "decoder_belief_target_nll": 6.88676643371582,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 7.09765625
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 8,
          "query_type": "B",
          "k": 4,
          "n": 256,
          "query_target_mass": 0.2501997947692871,
          "query_target_nll": 5.571483612060547,
          "query_top1_on_support": 0.2734375,
          "mean_query_support_size": 5.84765625,
          "probe_belief_target_mass": 0.022072773426771164,
          "probe_belief_target_nll": 12.985227584838867,
          "probe_belief_top1_on_support": 0.02734375,
          "mean_belief_support_size": 7.09765625,
          "decoder_belief_target_mass": 0.00738498754799366,
          "decoder_belief_target_nll": 6.887648582458496,
          "decoder_belief_top1_on_support": 0.0078125,
          "mean_decoder_belief_support_size": 7.09765625
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 8,
          "query_type": "B",
          "k": 8,
          "n": 256,
          "query_target_mass": 0.46577727794647217,
          "query_target_nll": 2.380614995956421,
          "query_top1_on_support": 0.5625,
          "mean_query_support_size": 5.84765625,
          "probe_belief_target_mass": 0.09683775901794434,
          "probe_belief_target_nll": 4.518664360046387,
          "probe_belief_top1_on_support": 0.15234375,
          "mean_belief_support_size": 7.09765625,
          "decoder_belief_target_mass": 0.007429054006934166,
          "decoder_belief_target_nll": 6.872107982635498,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 7.09765625
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 8,
          "query_type": "B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.46577727794647217,
          "query_target_nll": 2.380614995956421,
          "query_top1_on_support": 0.5625,
          "mean_query_support_size": 5.84765625,
          "probe_belief_target_mass": 0.09683775901794434,
          "probe_belief_target_nll": 4.518664360046387,
          "probe_belief_top1_on_support": 0.15234375,
          "mean_belief_support_size": 7.09765625,
          "decoder_belief_target_mass": 0.007429054006934166,
          "decoder_belief_target_nll": 6.872107982635498,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 7.09765625
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 8,
          "query_type": "B",
          "k": 16,
          "n": 256,
          "query_target_mass": 0.46577727794647217,
          "query_target_nll": 2.380614995956421,
          "query_top1_on_support": 0.5625,
          "mean_query_support_size": 5.84765625,
          "probe_belief_target_mass": 0.09683775901794434,
          "probe_belief_target_nll": 4.518664360046387,
          "probe_belief_top1_on_support": 0.15234375,
          "mean_belief_support_size": 7.09765625,
          "decoder_belief_target_mass": 0.007429054006934166,
          "decoder_belief_target_nll": 6.872107982635498,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 7.09765625
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 8,
          "query_type": "B",
          "k": 24,
          "n": 256,
          "query_target_mass": 0.46577727794647217,
          "query_target_nll": 2.380614995956421,
          "query_top1_on_support": 0.5625,
          "mean_query_support_size": 5.84765625,
          "probe_belief_target_mass": 0.09683775901794434,
          "probe_belief_target_nll": 4.518664360046387,
          "probe_belief_top1_on_support": 0.15234375,
          "mean_belief_support_size": 7.09765625,
          "decoder_belief_target_mass": 0.007429054006934166,
          "decoder_belief_target_nll": 6.872107982635498,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 7.09765625
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 8,
          "query_type": "A_PLUS_B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.17086197435855865,
          "query_target_nll": 3.681385040283203,
          "query_top1_on_support": 0.1875,
          "mean_query_support_size": 5.3515625,
          "probe_belief_target_mass": 0.00980805978178978,
          "probe_belief_target_nll": 20.073537826538086,
          "probe_belief_top1_on_support": 0.0078125,
          "mean_belief_support_size": 5.765625,
          "decoder_belief_target_mass": 0.006011884659528732,
          "decoder_belief_target_nll": 6.883800029754639,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 5.765625
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 8,
          "query_type": "A_PLUS_B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.17860326170921326,
          "query_target_nll": 4.064939975738525,
          "query_top1_on_support": 0.17578125,
          "mean_query_support_size": 5.3515625,
          "probe_belief_target_mass": 0.0075708357617259026,
          "probe_belief_target_nll": 19.229480743408203,
          "probe_belief_top1_on_support": 0.0078125,
          "mean_belief_support_size": 5.765625,
          "decoder_belief_target_mass": 0.006000726018100977,
          "decoder_belief_target_nll": 6.891666412353516,
          "decoder_belief_top1_on_support": 0.01171875,
          "mean_decoder_belief_support_size": 5.765625
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 8,
          "query_type": "A_PLUS_B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.17340299487113953,
          "query_target_nll": 4.247125148773193,
          "query_top1_on_support": 0.19140625,
          "mean_query_support_size": 5.3515625,
          "probe_belief_target_mass": 0.013416605070233345,
          "probe_belief_target_nll": 15.936019897460938,
          "probe_belief_top1_on_support": 0.01953125,
          "mean_belief_support_size": 5.765625,
          "decoder_belief_target_mass": 0.006008876487612724,
          "decoder_belief_target_nll": 6.889143466949463,
          "decoder_belief_top1_on_support": 0.01953125,
          "mean_decoder_belief_support_size": 5.765625
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 8,
          "query_type": "A_PLUS_B",
          "k": 4,
          "n": 256,
          "query_target_mass": 0.1754649579524994,
          "query_target_nll": 4.462127685546875,
          "query_top1_on_support": 0.1875,
          "mean_query_support_size": 5.3515625,
          "probe_belief_target_mass": 0.024081770330667496,
          "probe_belief_target_nll": 13.077432632446289,
          "probe_belief_top1_on_support": 0.04296875,
          "mean_belief_support_size": 5.765625,
          "decoder_belief_target_mass": 0.005942005664110184,
          "decoder_belief_target_nll": 6.900018215179443,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 5.765625
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 8,
          "query_type": "A_PLUS_B",
          "k": 8,
          "n": 256,
          "query_target_mass": 0.31181585788726807,
          "query_target_nll": 2.7950942516326904,
          "query_top1_on_support": 0.40234375,
          "mean_query_support_size": 5.3515625,
          "probe_belief_target_mass": 0.08166833221912384,
          "probe_belief_target_nll": 4.5845947265625,
          "probe_belief_top1_on_support": 0.1328125,
          "mean_belief_support_size": 5.765625,
          "decoder_belief_target_mass": 0.005957599729299545,
          "decoder_belief_target_nll": 6.903252601623535,
          "decoder_belief_top1_on_support": 0.01171875,
          "mean_decoder_belief_support_size": 5.765625
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 8,
          "query_type": "A_PLUS_B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.31181585788726807,
          "query_target_nll": 2.7950942516326904,
          "query_top1_on_support": 0.40234375,
          "mean_query_support_size": 5.3515625,
          "probe_belief_target_mass": 0.08166833221912384,
          "probe_belief_target_nll": 4.5845947265625,
          "probe_belief_top1_on_support": 0.1328125,
          "mean_belief_support_size": 5.765625,
          "decoder_belief_target_mass": 0.005957599729299545,
          "decoder_belief_target_nll": 6.903252601623535,
          "decoder_belief_top1_on_support": 0.01171875,
          "mean_decoder_belief_support_size": 5.765625
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 8,
          "query_type": "A_PLUS_B",
          "k": 16,
          "n": 256,
          "query_target_mass": 0.31181585788726807,
          "query_target_nll": 2.7950942516326904,
          "query_top1_on_support": 0.40234375,
          "mean_query_support_size": 5.3515625,
          "probe_belief_target_mass": 0.08166833221912384,
          "probe_belief_target_nll": 4.5845947265625,
          "probe_belief_top1_on_support": 0.1328125,
          "mean_belief_support_size": 5.765625,
          "decoder_belief_target_mass": 0.005957599729299545,
          "decoder_belief_target_nll": 6.903252601623535,
          "decoder_belief_top1_on_support": 0.01171875,
          "mean_decoder_belief_support_size": 5.765625
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 8,
          "query_type": "A_PLUS_B",
          "k": 24,
          "n": 256,
          "query_target_mass": 0.31181585788726807,
          "query_target_nll": 2.7950942516326904,
          "query_top1_on_support": 0.40234375,
          "mean_query_support_size": 5.3515625,
          "probe_belief_target_mass": 0.08166833221912384,
          "probe_belief_target_nll": 4.5845947265625,
          "probe_belief_top1_on_support": 0.1328125,
          "mean_belief_support_size": 5.765625,
          "decoder_belief_target_mass": 0.005957599729299545,
          "decoder_belief_target_nll": 6.903252601623535,
          "decoder_belief_top1_on_support": 0.01171875,
          "mean_decoder_belief_support_size": 5.765625
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 8,
          "query_type": "A_MINUS_B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.17967453598976135,
          "query_target_nll": 13.0963134765625,
          "query_top1_on_support": 0.1796875,
          "mean_query_support_size": 6.21875,
          "probe_belief_target_mass": 0.006175381131470203,
          "probe_belief_target_nll": 20.31393051147461,
          "probe_belief_top1_on_support": 0.01171875,
          "mean_belief_support_size": 6.9609375,
          "decoder_belief_target_mass": 0.007292508613318205,
          "decoder_belief_target_nll": 6.871485233306885,
          "decoder_belief_top1_on_support": 0.015625,
          "mean_decoder_belief_support_size": 6.9609375
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 8,
          "query_type": "A_MINUS_B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.1924166977405548,
          "query_target_nll": 8.250945091247559,
          "query_top1_on_support": 0.16796875,
          "mean_query_support_size": 6.21875,
          "probe_belief_target_mass": 0.015342069789767265,
          "probe_belief_target_nll": 18.82648277282715,
          "probe_belief_top1_on_support": 0.0234375,
          "mean_belief_support_size": 6.9609375,
          "decoder_belief_target_mass": 0.007240187842398882,
          "decoder_belief_target_nll": 6.881553649902344,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 6.9609375
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 8,
          "query_type": "A_MINUS_B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.1992175281047821,
          "query_target_nll": 6.629185199737549,
          "query_top1_on_support": 0.17578125,
          "mean_query_support_size": 6.21875,
          "probe_belief_target_mass": 0.017260901629924774,
          "probe_belief_target_nll": 16.193923950195312,
          "probe_belief_top1_on_support": 0.03515625,
          "mean_belief_support_size": 6.9609375,
          "decoder_belief_target_mass": 0.007256620563566685,
          "decoder_belief_target_nll": 6.875842571258545,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 6.9609375
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 8,
          "query_type": "A_MINUS_B",
          "k": 4,
          "n": 256,
          "query_target_mass": 0.21579742431640625,
          "query_target_nll": 5.24411153793335,
          "query_top1_on_support": 0.18359375,
          "mean_query_support_size": 6.21875,
          "probe_belief_target_mass": 0.01820828765630722,
          "probe_belief_target_nll": 13.052781105041504,
          "probe_belief_top1_on_support": 0.015625,
          "mean_belief_support_size": 6.9609375,
          "decoder_belief_target_mass": 0.007183297537267208,
          "decoder_belief_target_nll": 6.890161037445068,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 6.9609375
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 8,
          "query_type": "A_MINUS_B",
          "k": 8,
          "n": 256,
          "query_target_mass": 0.374320924282074,
          "query_target_nll": 2.7187750339508057,
          "query_top1_on_support": 0.40625,
          "mean_query_support_size": 6.21875,
          "probe_belief_target_mass": 0.09329825639724731,
          "probe_belief_target_nll": 4.524688720703125,
          "probe_belief_top1_on_support": 0.16796875,
          "mean_belief_support_size": 6.9609375,
          "decoder_belief_target_mass": 0.007233332842588425,
          "decoder_belief_target_nll": 6.887870788574219,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 6.9609375
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 8,
          "query_type": "A_MINUS_B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.374320924282074,
          "query_target_nll": 2.7187750339508057,
          "query_top1_on_support": 0.40625,
          "mean_query_support_size": 6.21875,
          "probe_belief_target_mass": 0.09329825639724731,
          "probe_belief_target_nll": 4.524688720703125,
          "probe_belief_top1_on_support": 0.16796875,
          "mean_belief_support_size": 6.9609375,
          "decoder_belief_target_mass": 0.007233332842588425,
          "decoder_belief_target_nll": 6.887870788574219,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 6.9609375
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 8,
          "query_type": "A_MINUS_B",
          "k": 16,
          "n": 256,
          "query_target_mass": 0.374320924282074,
          "query_target_nll": 2.7187750339508057,
          "query_top1_on_support": 0.40625,
          "mean_query_support_size": 6.21875,
          "probe_belief_target_mass": 0.09329825639724731,
          "probe_belief_target_nll": 4.524688720703125,
          "probe_belief_top1_on_support": 0.16796875,
          "mean_belief_support_size": 6.9609375,
          "decoder_belief_target_mass": 0.007233332842588425,
          "decoder_belief_target_nll": 6.887870788574219,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 6.9609375
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 8,
          "query_type": "A_MINUS_B",
          "k": 24,
          "n": 256,
          "query_target_mass": 0.374320924282074,
          "query_target_nll": 2.7187750339508057,
          "query_top1_on_support": 0.40625,
          "mean_query_support_size": 6.21875,
          "probe_belief_target_mass": 0.09329825639724731,
          "probe_belief_target_nll": 4.524688720703125,
          "probe_belief_top1_on_support": 0.16796875,
          "mean_belief_support_size": 6.9609375,
          "decoder_belief_target_mass": 0.007233332842588425,
          "decoder_belief_target_nll": 6.887870788574219,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 6.9609375
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 12,
          "query_type": "A",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.11209457367658615,
          "query_target_nll": 3.7696402072906494,
          "query_top1_on_support": 0.09375,
          "mean_query_support_size": 3.5078125,
          "probe_belief_target_mass": 0.004933658055961132,
          "probe_belief_target_nll": 19.759456634521484,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 3.8125,
          "decoder_belief_target_mass": 0.003918082918971777,
          "decoder_belief_target_nll": 6.897935390472412,
          "decoder_belief_top1_on_support": 0.0078125,
          "mean_decoder_belief_support_size": 3.8125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 12,
          "query_type": "A",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.11888817697763443,
          "query_target_nll": 5.991726875305176,
          "query_top1_on_support": 0.12109375,
          "mean_query_support_size": 3.5078125,
          "probe_belief_target_mass": 0.003652026876807213,
          "probe_belief_target_nll": 18.520753860473633,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 3.8125,
          "decoder_belief_target_mass": 0.003964750096201897,
          "decoder_belief_target_nll": 6.876324653625488,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 3.8125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 12,
          "query_type": "A",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.1331070363521576,
          "query_target_nll": 5.67323637008667,
          "query_top1_on_support": 0.13671875,
          "mean_query_support_size": 3.5078125,
          "probe_belief_target_mass": 0.006597808096557856,
          "probe_belief_target_nll": 15.33758544921875,
          "probe_belief_top1_on_support": 0.00390625,
          "mean_belief_support_size": 3.8125,
          "decoder_belief_target_mass": 0.003968417178839445,
          "decoder_belief_target_nll": 6.877453804016113,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 3.8125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 12,
          "query_type": "A",
          "k": 4,
          "n": 256,
          "query_target_mass": 0.12681865692138672,
          "query_target_nll": 6.115803241729736,
          "query_top1_on_support": 0.1015625,
          "mean_query_support_size": 3.5078125,
          "probe_belief_target_mass": 0.004365932662039995,
          "probe_belief_target_nll": 13.928616523742676,
          "probe_belief_top1_on_support": 0.0078125,
          "mean_belief_support_size": 3.8125,
          "decoder_belief_target_mass": 0.0039481110870838165,
          "decoder_belief_target_nll": 6.887622833251953,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 3.8125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 12,
          "query_type": "A",
          "k": 8,
          "n": 256,
          "query_target_mass": 0.14937134087085724,
          "query_target_nll": 5.563243865966797,
          "query_top1_on_support": 0.16796875,
          "mean_query_support_size": 3.5078125,
          "probe_belief_target_mass": 0.007996102795004845,
          "probe_belief_target_nll": 11.91664981842041,
          "probe_belief_top1_on_support": 0.01171875,
          "mean_belief_support_size": 3.8125,
          "decoder_belief_target_mass": 0.003904665121808648,
          "decoder_belief_target_nll": 6.891603946685791,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 3.8125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 12,
          "query_type": "A",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.3169350326061249,
          "query_target_nll": 2.474381685256958,
          "query_top1_on_support": 0.421875,
          "mean_query_support_size": 3.5078125,
          "probe_belief_target_mass": 0.056898295879364014,
          "probe_belief_target_nll": 4.618937015533447,
          "probe_belief_top1_on_support": 0.08203125,
          "mean_belief_support_size": 3.8125,
          "decoder_belief_target_mass": 0.003948054276406765,
          "decoder_belief_target_nll": 6.887016773223877,
          "decoder_belief_top1_on_support": 0.01171875,
          "mean_decoder_belief_support_size": 3.8125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 12,
          "query_type": "A",
          "k": 16,
          "n": 256,
          "query_target_mass": 0.3169350326061249,
          "query_target_nll": 2.474381685256958,
          "query_top1_on_support": 0.421875,
          "mean_query_support_size": 3.5078125,
          "probe_belief_target_mass": 0.056898295879364014,
          "probe_belief_target_nll": 4.618937015533447,
          "probe_belief_top1_on_support": 0.08203125,
          "mean_belief_support_size": 3.8125,
          "decoder_belief_target_mass": 0.003948054276406765,
          "decoder_belief_target_nll": 6.887016773223877,
          "decoder_belief_top1_on_support": 0.01171875,
          "mean_decoder_belief_support_size": 3.8125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 12,
          "query_type": "A",
          "k": 24,
          "n": 256,
          "query_target_mass": 0.3169350326061249,
          "query_target_nll": 2.474381685256958,
          "query_top1_on_support": 0.421875,
          "mean_query_support_size": 3.5078125,
          "probe_belief_target_mass": 0.056898295879364014,
          "probe_belief_target_nll": 4.618937015533447,
          "probe_belief_top1_on_support": 0.08203125,
          "mean_belief_support_size": 3.8125,
          "decoder_belief_target_mass": 0.003948054276406765,
          "decoder_belief_target_nll": 6.887016773223877,
          "decoder_belief_top1_on_support": 0.01171875,
          "mean_decoder_belief_support_size": 3.8125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 12,
          "query_type": "B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.12125901877880096,
          "query_target_nll": 3.8546223640441895,
          "query_top1_on_support": 0.14453125,
          "mean_query_support_size": 3.7265625,
          "probe_belief_target_mass": 0.005588080734014511,
          "probe_belief_target_nll": 20.070796966552734,
          "probe_belief_top1_on_support": 0.00390625,
          "mean_belief_support_size": 4.64453125,
          "decoder_belief_target_mass": 0.004827468656003475,
          "decoder_belief_target_nll": 6.89847469329834,
          "decoder_belief_top1_on_support": 0.0078125,
          "mean_decoder_belief_support_size": 4.64453125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 12,
          "query_type": "B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.12546595931053162,
          "query_target_nll": 5.84309196472168,
          "query_top1_on_support": 0.1171875,
          "mean_query_support_size": 3.7265625,
          "probe_belief_target_mass": 0.005608836654573679,
          "probe_belief_target_nll": 18.37398910522461,
          "probe_belief_top1_on_support": 0.01171875,
          "mean_belief_support_size": 4.64453125,
          "decoder_belief_target_mass": 0.004821313079446554,
          "decoder_belief_target_nll": 6.895608901977539,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 4.64453125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 12,
          "query_type": "B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.10767471790313721,
          "query_target_nll": 6.396644592285156,
          "query_top1_on_support": 0.109375,
          "mean_query_support_size": 3.7265625,
          "probe_belief_target_mass": 0.009036178700625896,
          "probe_belief_target_nll": 16.290332794189453,
          "probe_belief_top1_on_support": 0.015625,
          "mean_belief_support_size": 4.64453125,
          "decoder_belief_target_mass": 0.00482553243637085,
          "decoder_belief_target_nll": 6.894772529602051,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 4.64453125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 12,
          "query_type": "B",
          "k": 4,
          "n": 256,
          "query_target_mass": 0.13246236741542816,
          "query_target_nll": 6.1133832931518555,
          "query_top1_on_support": 0.140625,
          "mean_query_support_size": 3.7265625,
          "probe_belief_target_mass": 0.008951460942626,
          "probe_belief_target_nll": 13.631153106689453,
          "probe_belief_top1_on_support": 0.01171875,
          "mean_belief_support_size": 4.64453125,
          "decoder_belief_target_mass": 0.004863709211349487,
          "decoder_belief_target_nll": 6.884429454803467,
          "decoder_belief_top1_on_support": 0.0078125,
          "mean_decoder_belief_support_size": 4.64453125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 12,
          "query_type": "B",
          "k": 8,
          "n": 256,
          "query_target_mass": 0.16722536087036133,
          "query_target_nll": 5.5045366287231445,
          "query_top1_on_support": 0.16015625,
          "mean_query_support_size": 3.7265625,
          "probe_belief_target_mass": 0.013295907527208328,
          "probe_belief_target_nll": 11.910564422607422,
          "probe_belief_top1_on_support": 0.0390625,
          "mean_belief_support_size": 4.64453125,
          "decoder_belief_target_mass": 0.004865796305239201,
          "decoder_belief_target_nll": 6.880838871002197,
          "decoder_belief_top1_on_support": 0.0078125,
          "mean_decoder_belief_support_size": 4.64453125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 12,
          "query_type": "B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.34104469418525696,
          "query_target_nll": 2.3705921173095703,
          "query_top1_on_support": 0.38671875,
          "mean_query_support_size": 3.7265625,
          "probe_belief_target_mass": 0.06094701960682869,
          "probe_belief_target_nll": 4.6635661125183105,
          "probe_belief_top1_on_support": 0.08203125,
          "mean_belief_support_size": 4.64453125,
          "decoder_belief_target_mass": 0.004840826150029898,
          "decoder_belief_target_nll": 6.882895469665527,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 4.64453125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 12,
          "query_type": "B",
          "k": 16,
          "n": 256,
          "query_target_mass": 0.34104469418525696,
          "query_target_nll": 2.3705921173095703,
          "query_top1_on_support": 0.38671875,
          "mean_query_support_size": 3.7265625,
          "probe_belief_target_mass": 0.06094701960682869,
          "probe_belief_target_nll": 4.6635661125183105,
          "probe_belief_top1_on_support": 0.08203125,
          "mean_belief_support_size": 4.64453125,
          "decoder_belief_target_mass": 0.004840826150029898,
          "decoder_belief_target_nll": 6.882895469665527,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 4.64453125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 12,
          "query_type": "B",
          "k": 24,
          "n": 256,
          "query_target_mass": 0.34104469418525696,
          "query_target_nll": 2.3705921173095703,
          "query_top1_on_support": 0.38671875,
          "mean_query_support_size": 3.7265625,
          "probe_belief_target_mass": 0.06094701960682869,
          "probe_belief_target_nll": 4.6635661125183105,
          "probe_belief_top1_on_support": 0.08203125,
          "mean_belief_support_size": 4.64453125,
          "decoder_belief_target_mass": 0.004840826150029898,
          "decoder_belief_target_nll": 6.882895469665527,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 4.64453125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 12,
          "query_type": "A_PLUS_B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.13095100224018097,
          "query_target_nll": 3.619675397872925,
          "query_top1_on_support": 0.125,
          "mean_query_support_size": 4.0,
          "probe_belief_target_mass": 0.0035162288695573807,
          "probe_belief_target_nll": 20.261093139648438,
          "probe_belief_top1_on_support": 0.00390625,
          "mean_belief_support_size": 4.4140625,
          "decoder_belief_target_mass": 0.0045907944440841675,
          "decoder_belief_target_nll": 6.894833564758301,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 4.4140625
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 12,
          "query_type": "A_PLUS_B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.12457214295864105,
          "query_target_nll": 4.131984710693359,
          "query_top1_on_support": 0.13671875,
          "mean_query_support_size": 4.0,
          "probe_belief_target_mass": 0.006129232235252857,
          "probe_belief_target_nll": 19.617931365966797,
          "probe_belief_top1_on_support": 0.0078125,
          "mean_belief_support_size": 4.4140625,
          "decoder_belief_target_mass": 0.004543661139905453,
          "decoder_belief_target_nll": 6.891878128051758,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 4.4140625
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 12,
          "query_type": "A_PLUS_B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.1293846070766449,
          "query_target_nll": 4.283117294311523,
          "query_top1_on_support": 0.12109375,
          "mean_query_support_size": 4.0,
          "probe_belief_target_mass": 0.007182661909610033,
          "probe_belief_target_nll": 16.676408767700195,
          "probe_belief_top1_on_support": 0.00390625,
          "mean_belief_support_size": 4.4140625,
          "decoder_belief_target_mass": 0.004587114788591862,
          "decoder_belief_target_nll": 6.888367652893066,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 4.4140625
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 12,
          "query_type": "A_PLUS_B",
          "k": 4,
          "n": 256,
          "query_target_mass": 0.1326920986175537,
          "query_target_nll": 4.370588302612305,
          "query_top1_on_support": 0.1015625,
          "mean_query_support_size": 4.0,
          "probe_belief_target_mass": 0.004700963385403156,
          "probe_belief_target_nll": 14.738424301147461,
          "probe_belief_top1_on_support": 0.00390625,
          "mean_belief_support_size": 4.4140625,
          "decoder_belief_target_mass": 0.004616330377757549,
          "decoder_belief_target_nll": 6.873518943786621,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 4.4140625
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 12,
          "query_type": "A_PLUS_B",
          "k": 8,
          "n": 256,
          "query_target_mass": 0.13342779874801636,
          "query_target_nll": 4.594979286193848,
          "query_top1_on_support": 0.109375,
          "mean_query_support_size": 4.0,
          "probe_belief_target_mass": 0.011220106855034828,
          "probe_belief_target_nll": 12.517699241638184,
          "probe_belief_top1_on_support": 0.01953125,
          "mean_belief_support_size": 4.4140625,
          "decoder_belief_target_mass": 0.004566662013530731,
          "decoder_belief_target_nll": 6.888083457946777,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 4.4140625
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 12,
          "query_type": "A_PLUS_B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.2594352960586548,
          "query_target_nll": 2.754641532897949,
          "query_top1_on_support": 0.3515625,
          "mean_query_support_size": 4.0,
          "probe_belief_target_mass": 0.06651422381401062,
          "probe_belief_target_nll": 4.5359649658203125,
          "probe_belief_top1_on_support": 0.10546875,
          "mean_belief_support_size": 4.4140625,
          "decoder_belief_target_mass": 0.004595110192894936,
          "decoder_belief_target_nll": 6.8848114013671875,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 4.4140625
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 12,
          "query_type": "A_PLUS_B",
          "k": 16,
          "n": 256,
          "query_target_mass": 0.2594352960586548,
          "query_target_nll": 2.754641532897949,
          "query_top1_on_support": 0.3515625,
          "mean_query_support_size": 4.0,
          "probe_belief_target_mass": 0.06651422381401062,
          "probe_belief_target_nll": 4.5359649658203125,
          "probe_belief_top1_on_support": 0.10546875,
          "mean_belief_support_size": 4.4140625,
          "decoder_belief_target_mass": 0.004595110192894936,
          "decoder_belief_target_nll": 6.8848114013671875,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 4.4140625
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 12,
          "query_type": "A_PLUS_B",
          "k": 24,
          "n": 256,
          "query_target_mass": 0.2594352960586548,
          "query_target_nll": 2.754641532897949,
          "query_top1_on_support": 0.3515625,
          "mean_query_support_size": 4.0,
          "probe_belief_target_mass": 0.06651422381401062,
          "probe_belief_target_nll": 4.5359649658203125,
          "probe_belief_top1_on_support": 0.10546875,
          "mean_belief_support_size": 4.4140625,
          "decoder_belief_target_mass": 0.004595110192894936,
          "decoder_belief_target_nll": 6.8848114013671875,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 4.4140625
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 12,
          "query_type": "A_MINUS_B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.15625932812690735,
          "query_target_nll": 12.903413772583008,
          "query_top1_on_support": 0.15625,
          "mean_query_support_size": 4.078125,
          "probe_belief_target_mass": 0.004965455271303654,
          "probe_belief_target_nll": 19.934223175048828,
          "probe_belief_top1_on_support": 0.01171875,
          "mean_belief_support_size": 4.32421875,
          "decoder_belief_target_mass": 0.004465956706553698,
          "decoder_belief_target_nll": 6.890998363494873,
          "decoder_belief_top1_on_support": 0.0078125,
          "mean_decoder_belief_support_size": 4.32421875
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 12,
          "query_type": "A_MINUS_B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.14437691867351532,
          "query_target_nll": 8.028491020202637,
          "query_top1_on_support": 0.12890625,
          "mean_query_support_size": 4.078125,
          "probe_belief_target_mass": 0.006833328865468502,
          "probe_belief_target_nll": 17.97233772277832,
          "probe_belief_top1_on_support": 0.00390625,
          "mean_belief_support_size": 4.32421875,
          "decoder_belief_target_mass": 0.004490828141570091,
          "decoder_belief_target_nll": 6.890607833862305,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 4.32421875
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 12,
          "query_type": "A_MINUS_B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.1301761120557785,
          "query_target_nll": 6.407233715057373,
          "query_top1_on_support": 0.12109375,
          "mean_query_support_size": 4.078125,
          "probe_belief_target_mass": 0.00473964586853981,
          "probe_belief_target_nll": 15.744800567626953,
          "probe_belief_top1_on_support": 0.00390625,
          "mean_belief_support_size": 4.32421875,
          "decoder_belief_target_mass": 0.0045250882394611835,
          "decoder_belief_target_nll": 6.881946086883545,
          "decoder_belief_top1_on_support": 0.0078125,
          "mean_decoder_belief_support_size": 4.32421875
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 12,
          "query_type": "A_MINUS_B",
          "k": 4,
          "n": 256,
          "query_target_mass": 0.12954024970531464,
          "query_target_nll": 5.657330513000488,
          "query_top1_on_support": 0.1484375,
          "mean_query_support_size": 4.078125,
          "probe_belief_target_mass": 0.005286814644932747,
          "probe_belief_target_nll": 14.14126205444336,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 4.32421875,
          "decoder_belief_target_mass": 0.00448469165712595,
          "decoder_belief_target_nll": 6.899601459503174,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 4.32421875
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 12,
          "query_type": "A_MINUS_B",
          "k": 8,
          "n": 256,
          "query_target_mass": 0.14293113350868225,
          "query_target_nll": 5.18736457824707,
          "query_top1_on_support": 0.13671875,
          "mean_query_support_size": 4.078125,
          "probe_belief_target_mass": 0.011952517554163933,
          "probe_belief_target_nll": 11.878715515136719,
          "probe_belief_top1_on_support": 0.0078125,
          "mean_belief_support_size": 4.32421875,
          "decoder_belief_target_mass": 0.004472319036722183,
          "decoder_belief_target_nll": 6.8916015625,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 4.32421875
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 12,
          "query_type": "A_MINUS_B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.290775865316391,
          "query_target_nll": 2.700913906097412,
          "query_top1_on_support": 0.3671875,
          "mean_query_support_size": 4.078125,
          "probe_belief_target_mass": 0.06559297442436218,
          "probe_belief_target_nll": 4.503450393676758,
          "probe_belief_top1_on_support": 0.10546875,
          "mean_belief_support_size": 4.32421875,
          "decoder_belief_target_mass": 0.004490852355957031,
          "decoder_belief_target_nll": 6.891037940979004,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 4.32421875
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 12,
          "query_type": "A_MINUS_B",
          "k": 16,
          "n": 256,
          "query_target_mass": 0.290775865316391,
          "query_target_nll": 2.700913906097412,
          "query_top1_on_support": 0.3671875,
          "mean_query_support_size": 4.078125,
          "probe_belief_target_mass": 0.06559297442436218,
          "probe_belief_target_nll": 4.503450393676758,
          "probe_belief_top1_on_support": 0.10546875,
          "mean_belief_support_size": 4.32421875,
          "decoder_belief_target_mass": 0.004490852355957031,
          "decoder_belief_target_nll": 6.891037940979004,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 4.32421875
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 12,
          "query_type": "A_MINUS_B",
          "k": 24,
          "n": 256,
          "query_target_mass": 0.290775865316391,
          "query_target_nll": 2.700913906097412,
          "query_top1_on_support": 0.3671875,
          "mean_query_support_size": 4.078125,
          "probe_belief_target_mass": 0.06559297442436218,
          "probe_belief_target_nll": 4.503450393676758,
          "probe_belief_top1_on_support": 0.10546875,
          "mean_belief_support_size": 4.32421875,
          "decoder_belief_target_mass": 0.004490852355957031,
          "decoder_belief_target_nll": 6.891037940979004,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 4.32421875
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 16,
          "query_type": "A",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.07154066860675812,
          "query_target_nll": 3.858095169067383,
          "query_top1_on_support": 0.08203125,
          "mean_query_support_size": 2.296875,
          "probe_belief_target_mass": 0.0030901494901627302,
          "probe_belief_target_nll": 20.51943588256836,
          "probe_belief_top1_on_support": 0.0078125,
          "mean_belief_support_size": 2.58984375,
          "decoder_belief_target_mass": 0.0026618149131536484,
          "decoder_belief_target_nll": 6.9031171798706055,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 2.58984375
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 16,
          "query_type": "A",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.07434789836406708,
          "query_target_nll": 5.987098693847656,
          "query_top1_on_support": 0.05078125,
          "mean_query_support_size": 2.296875,
          "probe_belief_target_mass": 0.0032926900312304497,
          "probe_belief_target_nll": 19.472000122070312,
          "probe_belief_top1_on_support": 0.00390625,
          "mean_belief_support_size": 2.58984375,
          "decoder_belief_target_mass": 0.002686896827071905,
          "decoder_belief_target_nll": 6.900522708892822,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 2.58984375
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 16,
          "query_type": "A",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.07354198396205902,
          "query_target_nll": 6.1504974365234375,
          "query_top1_on_support": 0.0546875,
          "mean_query_support_size": 2.296875,
          "probe_belief_target_mass": 0.0025716512463986874,
          "probe_belief_target_nll": 15.733414649963379,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 2.58984375,
          "decoder_belief_target_mass": 0.002705290913581848,
          "decoder_belief_target_nll": 6.889202117919922,
          "decoder_belief_top1_on_support": 0.0078125,
          "mean_decoder_belief_support_size": 2.58984375
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 16,
          "query_type": "A",
          "k": 4,
          "n": 256,
          "query_target_mass": 0.07236681133508682,
          "query_target_nll": 6.115816593170166,
          "query_top1_on_support": 0.078125,
          "mean_query_support_size": 2.296875,
          "probe_belief_target_mass": 0.0022860642056912184,
          "probe_belief_target_nll": 13.851554870605469,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 2.58984375,
          "decoder_belief_target_mass": 0.002707995241507888,
          "decoder_belief_target_nll": 6.875871181488037,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 2.58984375
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 16,
          "query_type": "A",
          "k": 8,
          "n": 256,
          "query_target_mass": 0.08751629292964935,
          "query_target_nll": 5.9234395027160645,
          "query_top1_on_support": 0.1015625,
          "mean_query_support_size": 2.296875,
          "probe_belief_target_mass": 0.003137071616947651,
          "probe_belief_target_nll": 13.288895606994629,
          "probe_belief_top1_on_support": 0.0078125,
          "mean_belief_support_size": 2.58984375,
          "decoder_belief_target_mass": 0.002676496049389243,
          "decoder_belief_target_nll": 6.900239944458008,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 2.58984375
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 16,
          "query_type": "A",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.11376326531171799,
          "query_target_nll": 5.777548789978027,
          "query_top1_on_support": 0.14453125,
          "mean_query_support_size": 2.296875,
          "probe_belief_target_mass": 0.006503182463347912,
          "probe_belief_target_nll": 12.365018844604492,
          "probe_belief_top1_on_support": 0.01953125,
          "mean_belief_support_size": 2.58984375,
          "decoder_belief_target_mass": 0.0027178460732102394,
          "decoder_belief_target_nll": 6.886316776275635,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 2.58984375
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 16,
          "query_type": "A",
          "k": 16,
          "n": 256,
          "query_target_mass": 0.23743702471256256,
          "query_target_nll": 2.4255521297454834,
          "query_top1_on_support": 0.28125,
          "mean_query_support_size": 2.296875,
          "probe_belief_target_mass": 0.04557390138506889,
          "probe_belief_target_nll": 4.603466987609863,
          "probe_belief_top1_on_support": 0.0703125,
          "mean_belief_support_size": 2.58984375,
          "decoder_belief_target_mass": 0.002711088862270117,
          "decoder_belief_target_nll": 6.886417865753174,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 2.58984375
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 16,
          "query_type": "A",
          "k": 24,
          "n": 256,
          "query_target_mass": 0.23743702471256256,
          "query_target_nll": 2.4255521297454834,
          "query_top1_on_support": 0.28125,
          "mean_query_support_size": 2.296875,
          "probe_belief_target_mass": 0.04557390138506889,
          "probe_belief_target_nll": 4.603466987609863,
          "probe_belief_top1_on_support": 0.0703125,
          "mean_belief_support_size": 2.58984375,
          "decoder_belief_target_mass": 0.002711088862270117,
          "decoder_belief_target_nll": 6.886417865753174,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 2.58984375
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 16,
          "query_type": "B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.07178057730197906,
          "query_target_nll": 3.9777157306671143,
          "query_top1_on_support": 0.06640625,
          "mean_query_support_size": 2.44140625,
          "probe_belief_target_mass": 0.0035907200071960688,
          "probe_belief_target_nll": 20.68081283569336,
          "probe_belief_top1_on_support": 0.00390625,
          "mean_belief_support_size": 2.828125,
          "decoder_belief_target_mass": 0.00290731736458838,
          "decoder_belief_target_nll": 6.902312755584717,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 2.828125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 16,
          "query_type": "B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.0779876708984375,
          "query_target_nll": 5.950278282165527,
          "query_top1_on_support": 0.0546875,
          "mean_query_support_size": 2.44140625,
          "probe_belief_target_mass": 0.003203662345185876,
          "probe_belief_target_nll": 18.89975357055664,
          "probe_belief_top1_on_support": 0.00390625,
          "mean_belief_support_size": 2.828125,
          "decoder_belief_target_mass": 0.002935483120381832,
          "decoder_belief_target_nll": 6.890002727508545,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 2.828125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 16,
          "query_type": "B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.08011247962713242,
          "query_target_nll": 6.174023628234863,
          "query_top1_on_support": 0.09765625,
          "mean_query_support_size": 2.44140625,
          "probe_belief_target_mass": 0.0036413324996829033,
          "probe_belief_target_nll": 15.841962814331055,
          "probe_belief_top1_on_support": 0.00390625,
          "mean_belief_support_size": 2.828125,
          "decoder_belief_target_mass": 0.002946023130789399,
          "decoder_belief_target_nll": 6.893389701843262,
          "decoder_belief_top1_on_support": 0.0078125,
          "mean_decoder_belief_support_size": 2.828125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 16,
          "query_type": "B",
          "k": 4,
          "n": 256,
          "query_target_mass": 0.07682512700557709,
          "query_target_nll": 6.378631591796875,
          "query_top1_on_support": 0.09375,
          "mean_query_support_size": 2.44140625,
          "probe_belief_target_mass": 0.003401717636734247,
          "probe_belief_target_nll": 13.489469528198242,
          "probe_belief_top1_on_support": 0.00390625,
          "mean_belief_support_size": 2.828125,
          "decoder_belief_target_mass": 0.002954012481495738,
          "decoder_belief_target_nll": 6.885918617248535,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 2.828125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 16,
          "query_type": "B",
          "k": 8,
          "n": 256,
          "query_target_mass": 0.08533190190792084,
          "query_target_nll": 6.486554145812988,
          "query_top1_on_support": 0.078125,
          "mean_query_support_size": 2.44140625,
          "probe_belief_target_mass": 0.003140618558973074,
          "probe_belief_target_nll": 13.604873657226562,
          "probe_belief_top1_on_support": 0.00390625,
          "mean_belief_support_size": 2.828125,
          "decoder_belief_target_mass": 0.0029274453409016132,
          "decoder_belief_target_nll": 6.893141746520996,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 2.828125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 16,
          "query_type": "B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.1146620362997055,
          "query_target_nll": 5.9095964431762695,
          "query_top1_on_support": 0.12890625,
          "mean_query_support_size": 2.44140625,
          "probe_belief_target_mass": 0.005716951563954353,
          "probe_belief_target_nll": 12.622091293334961,
          "probe_belief_top1_on_support": 0.015625,
          "mean_belief_support_size": 2.828125,
          "decoder_belief_target_mass": 0.0029342686757445335,
          "decoder_belief_target_nll": 6.894593238830566,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 2.828125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 16,
          "query_type": "B",
          "k": 16,
          "n": 256,
          "query_target_mass": 0.2670396864414215,
          "query_target_nll": 2.310004472732544,
          "query_top1_on_support": 0.35546875,
          "mean_query_support_size": 2.44140625,
          "probe_belief_target_mass": 0.0446193628013134,
          "probe_belief_target_nll": 4.575182914733887,
          "probe_belief_top1_on_support": 0.07421875,
          "mean_belief_support_size": 2.828125,
          "decoder_belief_target_mass": 0.002946687862277031,
          "decoder_belief_target_nll": 6.889518737792969,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 2.828125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 16,
          "query_type": "B",
          "k": 24,
          "n": 256,
          "query_target_mass": 0.2670396864414215,
          "query_target_nll": 2.310004472732544,
          "query_top1_on_support": 0.35546875,
          "mean_query_support_size": 2.44140625,
          "probe_belief_target_mass": 0.0446193628013134,
          "probe_belief_target_nll": 4.575182914733887,
          "probe_belief_top1_on_support": 0.07421875,
          "mean_belief_support_size": 2.828125,
          "decoder_belief_target_mass": 0.002946687862277031,
          "decoder_belief_target_nll": 6.889518737792969,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 2.828125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 16,
          "query_type": "A_PLUS_B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.07476130872964859,
          "query_target_nll": 3.644373893737793,
          "query_top1_on_support": 0.05859375,
          "mean_query_support_size": 2.31640625,
          "probe_belief_target_mass": 0.002565148752182722,
          "probe_belief_target_nll": 20.668052673339844,
          "probe_belief_top1_on_support": 0.00390625,
          "mean_belief_support_size": 2.46484375,
          "decoder_belief_target_mass": 0.0025806156918406487,
          "decoder_belief_target_nll": 6.881160736083984,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 2.46484375
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 16,
          "query_type": "A_PLUS_B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.06732121109962463,
          "query_target_nll": 4.251317024230957,
          "query_top1_on_support": 0.0546875,
          "mean_query_support_size": 2.31640625,
          "probe_belief_target_mass": 0.0033156704157590866,
          "probe_belief_target_nll": 19.781368255615234,
          "probe_belief_top1_on_support": 0.00390625,
          "mean_belief_support_size": 2.46484375,
          "decoder_belief_target_mass": 0.002596211154013872,
          "decoder_belief_target_nll": 6.880867958068848,
          "decoder_belief_top1_on_support": 0.0078125,
          "mean_decoder_belief_support_size": 2.46484375
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 16,
          "query_type": "A_PLUS_B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.06846418976783752,
          "query_target_nll": 4.328859329223633,
          "query_top1_on_support": 0.08203125,
          "mean_query_support_size": 2.31640625,
          "probe_belief_target_mass": 0.002613271586596966,
          "probe_belief_target_nll": 16.845626831054688,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 2.46484375,
          "decoder_belief_target_mass": 0.0026079139206558466,
          "decoder_belief_target_nll": 6.873122215270996,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 2.46484375
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 16,
          "query_type": "A_PLUS_B",
          "k": 4,
          "n": 256,
          "query_target_mass": 0.08037738502025604,
          "query_target_nll": 4.2711591720581055,
          "query_top1_on_support": 0.09375,
          "mean_query_support_size": 2.31640625,
          "probe_belief_target_mass": 0.002680657897144556,
          "probe_belief_target_nll": 14.303131103515625,
          "probe_belief_top1_on_support": 0.00390625,
          "mean_belief_support_size": 2.46484375,
          "decoder_belief_target_mass": 0.0025862930342555046,
          "decoder_belief_target_nll": 6.8817925453186035,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 2.46484375
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 16,
          "query_type": "A_PLUS_B",
          "k": 8,
          "n": 256,
          "query_target_mass": 0.08330920338630676,
          "query_target_nll": 4.674090385437012,
          "query_top1_on_support": 0.09765625,
          "mean_query_support_size": 2.31640625,
          "probe_belief_target_mass": 0.004169901832938194,
          "probe_belief_target_nll": 13.213382720947266,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 2.46484375,
          "decoder_belief_target_mass": 0.002566450973972678,
          "decoder_belief_target_nll": 6.88157320022583,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 2.46484375
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 16,
          "query_type": "A_PLUS_B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.0745505690574646,
          "query_target_nll": 4.825170516967773,
          "query_top1_on_support": 0.0703125,
          "mean_query_support_size": 2.31640625,
          "probe_belief_target_mass": 0.005196545738726854,
          "probe_belief_target_nll": 12.842607498168945,
          "probe_belief_top1_on_support": 0.00390625,
          "mean_belief_support_size": 2.46484375,
          "decoder_belief_target_mass": 0.0025461178738623857,
          "decoder_belief_target_nll": 6.887135982513428,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 2.46484375
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 16,
          "query_type": "A_PLUS_B",
          "k": 16,
          "n": 256,
          "query_target_mass": 0.1701592355966568,
          "query_target_nll": 2.7730190753936768,
          "query_top1_on_support": 0.23046875,
          "mean_query_support_size": 2.31640625,
          "probe_belief_target_mass": 0.04917971044778824,
          "probe_belief_target_nll": 4.545315742492676,
          "probe_belief_top1_on_support": 0.0625,
          "mean_belief_support_size": 2.46484375,
          "decoder_belief_target_mass": 0.002565734088420868,
          "decoder_belief_target_nll": 6.880471229553223,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 2.46484375
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 16,
          "query_type": "A_PLUS_B",
          "k": 24,
          "n": 256,
          "query_target_mass": 0.1701592355966568,
          "query_target_nll": 2.7730190753936768,
          "query_top1_on_support": 0.23046875,
          "mean_query_support_size": 2.31640625,
          "probe_belief_target_mass": 0.04917971044778824,
          "probe_belief_target_nll": 4.545315742492676,
          "probe_belief_top1_on_support": 0.0625,
          "mean_belief_support_size": 2.46484375,
          "decoder_belief_target_mass": 0.002565734088420868,
          "decoder_belief_target_nll": 6.880471229553223,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 2.46484375
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 16,
          "query_type": "A_MINUS_B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.09764617681503296,
          "query_target_nll": 13.247203826904297,
          "query_top1_on_support": 0.09765625,
          "mean_query_support_size": 2.5546875,
          "probe_belief_target_mass": 0.002990087028592825,
          "probe_belief_target_nll": 20.34508514404297,
          "probe_belief_top1_on_support": 0.00390625,
          "mean_belief_support_size": 2.69921875,
          "decoder_belief_target_mass": 0.0028213635087013245,
          "decoder_belief_target_nll": 6.8812947273254395,
          "decoder_belief_top1_on_support": 0.01171875,
          "mean_decoder_belief_support_size": 2.69921875
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 16,
          "query_type": "A_MINUS_B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.07924827933311462,
          "query_target_nll": 8.810505867004395,
          "query_top1_on_support": 0.08203125,
          "mean_query_support_size": 2.5546875,
          "probe_belief_target_mass": 0.002742427634075284,
          "probe_belief_target_nll": 20.074974060058594,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 2.69921875,
          "decoder_belief_target_mass": 0.0027946210466325283,
          "decoder_belief_target_nll": 6.8983612060546875,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 2.69921875
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 16,
          "query_type": "A_MINUS_B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.08670412003993988,
          "query_target_nll": 6.599215507507324,
          "query_top1_on_support": 0.078125,
          "mean_query_support_size": 2.5546875,
          "probe_belief_target_mass": 0.003030585590749979,
          "probe_belief_target_nll": 16.940990447998047,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 2.69921875,
          "decoder_belief_target_mass": 0.0027943719178438187,
          "decoder_belief_target_nll": 6.902510643005371,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 2.69921875
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 16,
          "query_type": "A_MINUS_B",
          "k": 4,
          "n": 256,
          "query_target_mass": 0.07594315707683563,
          "query_target_nll": 5.687929630279541,
          "query_top1_on_support": 0.0625,
          "mean_query_support_size": 2.5546875,
          "probe_belief_target_mass": 0.0017145926831290126,
          "probe_belief_target_nll": 14.857439994812012,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 2.69921875,
          "decoder_belief_target_mass": 0.0028259160462766886,
          "decoder_belief_target_nll": 6.881147861480713,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 2.69921875
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 16,
          "query_type": "A_MINUS_B",
          "k": 8,
          "n": 256,
          "query_target_mass": 0.07869087904691696,
          "query_target_nll": 5.735628128051758,
          "query_top1_on_support": 0.07421875,
          "mean_query_support_size": 2.5546875,
          "probe_belief_target_mass": 0.003078110981732607,
          "probe_belief_target_nll": 13.515539169311523,
          "probe_belief_top1_on_support": 0.00390625,
          "mean_belief_support_size": 2.69921875,
          "decoder_belief_target_mass": 0.0028564271051436663,
          "decoder_belief_target_nll": 6.878608703613281,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 2.69921875
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 16,
          "query_type": "A_MINUS_B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.07633817195892334,
          "query_target_nll": 5.462204933166504,
          "query_top1_on_support": 0.0859375,
          "mean_query_support_size": 2.5546875,
          "probe_belief_target_mass": 0.006558967754244804,
          "probe_belief_target_nll": 12.405632972717285,
          "probe_belief_top1_on_support": 0.015625,
          "mean_belief_support_size": 2.69921875,
          "decoder_belief_target_mass": 0.002848302712664008,
          "decoder_belief_target_nll": 6.873403072357178,
          "decoder_belief_top1_on_support": 0.01171875,
          "mean_decoder_belief_support_size": 2.69921875
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 16,
          "query_type": "A_MINUS_B",
          "k": 16,
          "n": 256,
          "query_target_mass": 0.21172885596752167,
          "query_target_nll": 2.7177882194519043,
          "query_top1_on_support": 0.26953125,
          "mean_query_support_size": 2.5546875,
          "probe_belief_target_mass": 0.047197964042425156,
          "probe_belief_target_nll": 4.657375812530518,
          "probe_belief_top1_on_support": 0.06640625,
          "mean_belief_support_size": 2.69921875,
          "decoder_belief_target_mass": 0.0028494009748101234,
          "decoder_belief_target_nll": 6.88134765625,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 2.69921875
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 16,
          "query_type": "A_MINUS_B",
          "k": 24,
          "n": 256,
          "query_target_mass": 0.21172885596752167,
          "query_target_nll": 2.7177882194519043,
          "query_top1_on_support": 0.26953125,
          "mean_query_support_size": 2.5546875,
          "probe_belief_target_mass": 0.047197964042425156,
          "probe_belief_target_nll": 4.657375812530518,
          "probe_belief_top1_on_support": 0.06640625,
          "mean_belief_support_size": 2.69921875,
          "decoder_belief_target_mass": 0.0028494009748101234,
          "decoder_belief_target_nll": 6.88134765625,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 2.69921875
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 24,
          "query_type": "A",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.05356120318174362,
          "query_target_nll": 3.773444652557373,
          "query_top1_on_support": 0.0625,
          "mean_query_support_size": 1.56640625,
          "probe_belief_target_mass": 0.0015640450874343514,
          "probe_belief_target_nll": 19.94533920288086,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 1.64453125,
          "decoder_belief_target_mass": 0.0016989100258797407,
          "decoder_belief_target_nll": 6.891861915588379,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 1.64453125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 24,
          "query_type": "A",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.04899047315120697,
          "query_target_nll": 6.968587875366211,
          "query_top1_on_support": 0.04296875,
          "mean_query_support_size": 1.56640625,
          "probe_belief_target_mass": 0.0014283980708569288,
          "probe_belief_target_nll": 20.093738555908203,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 1.64453125,
          "decoder_belief_target_mass": 0.0016899704933166504,
          "decoder_belief_target_nll": 6.901495933532715,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 1.64453125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 24,
          "query_type": "A",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.04507497698068619,
          "query_target_nll": 6.523509502410889,
          "query_top1_on_support": 0.03515625,
          "mean_query_support_size": 1.56640625,
          "probe_belief_target_mass": 0.0013042017817497253,
          "probe_belief_target_nll": 16.62494659423828,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 1.64453125,
          "decoder_belief_target_mass": 0.0016803005710244179,
          "decoder_belief_target_nll": 6.903031349182129,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 1.64453125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 24,
          "query_type": "A",
          "k": 4,
          "n": 256,
          "query_target_mass": 0.05047840625047684,
          "query_target_nll": 6.220879554748535,
          "query_top1_on_support": 0.0390625,
          "mean_query_support_size": 1.56640625,
          "probe_belief_target_mass": 0.0023823631927371025,
          "probe_belief_target_nll": 13.952962875366211,
          "probe_belief_top1_on_support": 0.0078125,
          "mean_belief_support_size": 1.64453125,
          "decoder_belief_target_mass": 0.0016569581348448992,
          "decoder_belief_target_nll": 6.915121555328369,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 1.64453125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 24,
          "query_type": "A",
          "k": 8,
          "n": 256,
          "query_target_mass": 0.04829774051904678,
          "query_target_nll": 6.557246208190918,
          "query_top1_on_support": 0.04296875,
          "mean_query_support_size": 1.56640625,
          "probe_belief_target_mass": 0.0019921204075217247,
          "probe_belief_target_nll": 13.524858474731445,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 1.64453125,
          "decoder_belief_target_mass": 0.0017041624523699284,
          "decoder_belief_target_nll": 6.892889976501465,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 1.64453125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 24,
          "query_type": "A",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.05454334616661072,
          "query_target_nll": 6.6922807693481445,
          "query_top1_on_support": 0.07421875,
          "mean_query_support_size": 1.56640625,
          "probe_belief_target_mass": 0.0024565206840634346,
          "probe_belief_target_nll": 13.385371208190918,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 1.64453125,
          "decoder_belief_target_mass": 0.0016998782521113753,
          "decoder_belief_target_nll": 6.897465229034424,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 1.64453125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 24,
          "query_type": "A",
          "k": 16,
          "n": 256,
          "query_target_mass": 0.056303538382053375,
          "query_target_nll": 6.577376365661621,
          "query_top1_on_support": 0.04296875,
          "mean_query_support_size": 1.56640625,
          "probe_belief_target_mass": 0.002722682897001505,
          "probe_belief_target_nll": 13.433868408203125,
          "probe_belief_top1_on_support": 0.0078125,
          "mean_belief_support_size": 1.64453125,
          "decoder_belief_target_mass": 0.0016919602639973164,
          "decoder_belief_target_nll": 6.901617050170898,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 1.64453125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 24,
          "query_type": "A",
          "k": 24,
          "n": 256,
          "query_target_mass": 0.1892852485179901,
          "query_target_nll": 2.329329252243042,
          "query_top1_on_support": 0.2421875,
          "mean_query_support_size": 1.56640625,
          "probe_belief_target_mass": 0.032554734498262405,
          "probe_belief_target_nll": 4.461655616760254,
          "probe_belief_top1_on_support": 0.0625,
          "mean_belief_support_size": 1.64453125,
          "decoder_belief_target_mass": 0.001705085625872016,
          "decoder_belief_target_nll": 6.885428428649902,
          "decoder_belief_top1_on_support": 0.0078125,
          "mean_decoder_belief_support_size": 1.64453125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 24,
          "query_type": "B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.04299907386302948,
          "query_target_nll": 4.018581390380859,
          "query_top1_on_support": 0.0390625,
          "mean_query_support_size": 1.578125,
          "probe_belief_target_mass": 0.0025920956395566463,
          "probe_belief_target_nll": 19.999855041503906,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 1.78125,
          "decoder_belief_target_mass": 0.001866664970293641,
          "decoder_belief_target_nll": 6.8882598876953125,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 1.78125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 24,
          "query_type": "B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.05027741566300392,
          "query_target_nll": 5.447389602661133,
          "query_top1_on_support": 0.0546875,
          "mean_query_support_size": 1.578125,
          "probe_belief_target_mass": 0.0022401120513677597,
          "probe_belief_target_nll": 18.78745460510254,
          "probe_belief_top1_on_support": 0.00390625,
          "mean_belief_support_size": 1.78125,
          "decoder_belief_target_mass": 0.0018671154975891113,
          "decoder_belief_target_nll": 6.881349563598633,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 1.78125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 24,
          "query_type": "B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.05237879976630211,
          "query_target_nll": 5.65531063079834,
          "query_top1_on_support": 0.046875,
          "mean_query_support_size": 1.578125,
          "probe_belief_target_mass": 0.0031762197613716125,
          "probe_belief_target_nll": 15.423518180847168,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 1.78125,
          "decoder_belief_target_mass": 0.001876786001957953,
          "decoder_belief_target_nll": 6.873655319213867,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 1.78125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 24,
          "query_type": "B",
          "k": 4,
          "n": 256,
          "query_target_mass": 0.05141102895140648,
          "query_target_nll": 5.858893871307373,
          "query_top1_on_support": 0.046875,
          "mean_query_support_size": 1.578125,
          "probe_belief_target_mass": 0.0025490904226899147,
          "probe_belief_target_nll": 13.773096084594727,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 1.78125,
          "decoder_belief_target_mass": 0.0018841720884665847,
          "decoder_belief_target_nll": 6.863466262817383,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 1.78125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 24,
          "query_type": "B",
          "k": 8,
          "n": 256,
          "query_target_mass": 0.05348841845989227,
          "query_target_nll": 6.345643997192383,
          "query_top1_on_support": 0.06640625,
          "mean_query_support_size": 1.578125,
          "probe_belief_target_mass": 0.0014785690000280738,
          "probe_belief_target_nll": 13.924027442932129,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 1.78125,
          "decoder_belief_target_mass": 0.0018596790032461286,
          "decoder_belief_target_nll": 6.880334377288818,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 1.78125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 24,
          "query_type": "B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.04919765144586563,
          "query_target_nll": 6.445404529571533,
          "query_top1_on_support": 0.04296875,
          "mean_query_support_size": 1.578125,
          "probe_belief_target_mass": 0.0008923178538680077,
          "probe_belief_target_nll": 13.836673736572266,
          "probe_belief_top1_on_support": 0.00390625,
          "mean_belief_support_size": 1.78125,
          "decoder_belief_target_mass": 0.0018508981447666883,
          "decoder_belief_target_nll": 6.8779401779174805,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 1.78125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 24,
          "query_type": "B",
          "k": 16,
          "n": 256,
          "query_target_mass": 0.04994913190603256,
          "query_target_nll": 6.720128536224365,
          "query_top1_on_support": 0.04296875,
          "mean_query_support_size": 1.578125,
          "probe_belief_target_mass": 0.0021893084049224854,
          "probe_belief_target_nll": 13.841207504272461,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 1.78125,
          "decoder_belief_target_mass": 0.0018541747704148293,
          "decoder_belief_target_nll": 6.8806352615356445,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 1.78125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 24,
          "query_type": "B",
          "k": 24,
          "n": 256,
          "query_target_mass": 0.18789809942245483,
          "query_target_nll": 2.3396987915039062,
          "query_top1_on_support": 0.2578125,
          "mean_query_support_size": 1.578125,
          "probe_belief_target_mass": 0.03175080940127373,
          "probe_belief_target_nll": 4.668039798736572,
          "probe_belief_top1_on_support": 0.046875,
          "mean_belief_support_size": 1.78125,
          "decoder_belief_target_mass": 0.0018531294772401452,
          "decoder_belief_target_nll": 6.887885093688965,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 1.78125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 24,
          "query_type": "A_PLUS_B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.05332929268479347,
          "query_target_nll": 3.60904598236084,
          "query_top1_on_support": 0.06640625,
          "mean_query_support_size": 1.515625,
          "probe_belief_target_mass": 0.0016857562586665154,
          "probe_belief_target_nll": 21.124784469604492,
          "probe_belief_top1_on_support": 0.0078125,
          "mean_belief_support_size": 1.578125,
          "decoder_belief_target_mass": 0.0016586501151323318,
          "decoder_belief_target_nll": 6.877699851989746,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 1.578125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 24,
          "query_type": "A_PLUS_B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.05149093642830849,
          "query_target_nll": 3.988476276397705,
          "query_top1_on_support": 0.04296875,
          "mean_query_support_size": 1.515625,
          "probe_belief_target_mass": 0.0005645185010507703,
          "probe_belief_target_nll": 19.940654754638672,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 1.578125,
          "decoder_belief_target_mass": 0.001642199233174324,
          "decoder_belief_target_nll": 6.887428283691406,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 1.578125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 24,
          "query_type": "A_PLUS_B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.04576847702264786,
          "query_target_nll": 4.1712141036987305,
          "query_top1_on_support": 0.03125,
          "mean_query_support_size": 1.515625,
          "probe_belief_target_mass": 0.00095287652220577,
          "probe_belief_target_nll": 16.964353561401367,
          "probe_belief_top1_on_support": 0.00390625,
          "mean_belief_support_size": 1.578125,
          "decoder_belief_target_mass": 0.001663220813497901,
          "decoder_belief_target_nll": 6.871547698974609,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 1.578125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 24,
          "query_type": "A_PLUS_B",
          "k": 4,
          "n": 256,
          "query_target_mass": 0.04874414950609207,
          "query_target_nll": 4.604409694671631,
          "query_top1_on_support": 0.05859375,
          "mean_query_support_size": 1.515625,
          "probe_belief_target_mass": 0.0013869344256818295,
          "probe_belief_target_nll": 14.410597801208496,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 1.578125,
          "decoder_belief_target_mass": 0.0016488900873810053,
          "decoder_belief_target_nll": 6.878962993621826,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 1.578125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 24,
          "query_type": "A_PLUS_B",
          "k": 8,
          "n": 256,
          "query_target_mass": 0.048017941415309906,
          "query_target_nll": 4.639918327331543,
          "query_top1_on_support": 0.0390625,
          "mean_query_support_size": 1.515625,
          "probe_belief_target_mass": 0.0016066611278802156,
          "probe_belief_target_nll": 13.496642112731934,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 1.578125,
          "decoder_belief_target_mass": 0.0016571714077144861,
          "decoder_belief_target_nll": 6.8763427734375,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 1.578125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 24,
          "query_type": "A_PLUS_B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.049625951796770096,
          "query_target_nll": 4.565080642700195,
          "query_top1_on_support": 0.046875,
          "mean_query_support_size": 1.515625,
          "probe_belief_target_mass": 0.001410499680787325,
          "probe_belief_target_nll": 13.971654891967773,
          "probe_belief_top1_on_support": 0.00390625,
          "mean_belief_support_size": 1.578125,
          "decoder_belief_target_mass": 0.0016582116950303316,
          "decoder_belief_target_nll": 6.884101867675781,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 1.578125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 24,
          "query_type": "A_PLUS_B",
          "k": 16,
          "n": 256,
          "query_target_mass": 0.05580977723002434,
          "query_target_nll": 4.528255462646484,
          "query_top1_on_support": 0.0625,
          "mean_query_support_size": 1.515625,
          "probe_belief_target_mass": 0.0019893210846930742,
          "probe_belief_target_nll": 13.539154052734375,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 1.578125,
          "decoder_belief_target_mass": 0.0016509097767993808,
          "decoder_belief_target_nll": 6.883213996887207,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 1.578125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 24,
          "query_type": "A_PLUS_B",
          "k": 24,
          "n": 256,
          "query_target_mass": 0.12085740268230438,
          "query_target_nll": 2.768294334411621,
          "query_top1_on_support": 0.16015625,
          "mean_query_support_size": 1.515625,
          "probe_belief_target_mass": 0.030863087624311447,
          "probe_belief_target_nll": 4.67378568649292,
          "probe_belief_top1_on_support": 0.05859375,
          "mean_belief_support_size": 1.578125,
          "decoder_belief_target_mass": 0.0016400471795350313,
          "decoder_belief_target_nll": 6.884096145629883,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 1.578125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 24,
          "query_type": "A_MINUS_B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.05858006700873375,
          "query_target_nll": 12.944475173950195,
          "query_top1_on_support": 0.05859375,
          "mean_query_support_size": 1.51953125,
          "probe_belief_target_mass": 0.0016440385952591896,
          "probe_belief_target_nll": 19.789066314697266,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 1.55078125,
          "decoder_belief_target_mass": 0.0016313712112605572,
          "decoder_belief_target_nll": 6.873848915100098,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 1.55078125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 24,
          "query_type": "A_MINUS_B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.05834566056728363,
          "query_target_nll": 8.22079086303711,
          "query_top1_on_support": 0.0625,
          "mean_query_support_size": 1.51953125,
          "probe_belief_target_mass": 0.0025807986967265606,
          "probe_belief_target_nll": 19.021146774291992,
          "probe_belief_top1_on_support": 0.00390625,
          "mean_belief_support_size": 1.55078125,
          "decoder_belief_target_mass": 0.0016209841705858707,
          "decoder_belief_target_nll": 6.878856658935547,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 1.55078125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 24,
          "query_type": "A_MINUS_B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.04577682912349701,
          "query_target_nll": 6.842116355895996,
          "query_top1_on_support": 0.02734375,
          "mean_query_support_size": 1.51953125,
          "probe_belief_target_mass": 0.0007767381030134857,
          "probe_belief_target_nll": 16.795392990112305,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 1.55078125,
          "decoder_belief_target_mass": 0.0016189042944461107,
          "decoder_belief_target_nll": 6.880967617034912,
          "decoder_belief_top1_on_support": 0.0078125,
          "mean_decoder_belief_support_size": 1.55078125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 24,
          "query_type": "A_MINUS_B",
          "k": 4,
          "n": 256,
          "query_target_mass": 0.05197557806968689,
          "query_target_nll": 5.41538143157959,
          "query_top1_on_support": 0.04296875,
          "mean_query_support_size": 1.51953125,
          "probe_belief_target_mass": 0.002638137899339199,
          "probe_belief_target_nll": 13.836291313171387,
          "probe_belief_top1_on_support": 0.00390625,
          "mean_belief_support_size": 1.55078125,
          "decoder_belief_target_mass": 0.001608449500054121,
          "decoder_belief_target_nll": 6.884507656097412,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 1.55078125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 24,
          "query_type": "A_MINUS_B",
          "k": 8,
          "n": 256,
          "query_target_mass": 0.04884443059563637,
          "query_target_nll": 5.539342403411865,
          "query_top1_on_support": 0.05859375,
          "mean_query_support_size": 1.51953125,
          "probe_belief_target_mass": 0.0016760823782533407,
          "probe_belief_target_nll": 13.354022026062012,
          "probe_belief_top1_on_support": 0.00390625,
          "mean_belief_support_size": 1.55078125,
          "decoder_belief_target_mass": 0.0016462974017485976,
          "decoder_belief_target_nll": 6.8584370613098145,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 1.55078125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 24,
          "query_type": "A_MINUS_B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.06106045842170715,
          "query_target_nll": 5.331088066101074,
          "query_top1_on_support": 0.06640625,
          "mean_query_support_size": 1.51953125,
          "probe_belief_target_mass": 0.0014361938228830695,
          "probe_belief_target_nll": 13.535706520080566,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 1.55078125,
          "decoder_belief_target_mass": 0.001622871495783329,
          "decoder_belief_target_nll": 6.884950637817383,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 1.55078125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 24,
          "query_type": "A_MINUS_B",
          "k": 16,
          "n": 256,
          "query_target_mass": 0.05541466921567917,
          "query_target_nll": 5.1992106437683105,
          "query_top1_on_support": 0.078125,
          "mean_query_support_size": 1.51953125,
          "probe_belief_target_mass": 0.0023676943965256214,
          "probe_belief_target_nll": 13.51623249053955,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 1.55078125,
          "decoder_belief_target_mass": 0.001639246242120862,
          "decoder_belief_target_nll": 6.873054027557373,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 1.55078125
        },
        {
          "model": "dense",
          "supervision": "soft_final_query",
          "length": 24,
          "query_type": "A_MINUS_B",
          "k": 24,
          "n": 256,
          "query_target_mass": 0.14547324180603027,
          "query_target_nll": 2.7905662059783936,
          "query_top1_on_support": 0.16796875,
          "mean_query_support_size": 1.51953125,
          "probe_belief_target_mass": 0.031711921095848083,
          "probe_belief_target_nll": 4.729643821716309,
          "probe_belief_top1_on_support": 0.0625,
          "mean_belief_support_size": 1.55078125,
          "decoder_belief_target_mass": 0.0016054240986704826,
          "decoder_belief_target_nll": 6.894874095916748,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 1.55078125
        }
      ]
    }
  ],
  "checkpoints": [
    "large_artifacts/dense_supervision_ladder/checkpoints/main_soft_final_query_mod31/checkpoint_final.pt"
  ]
}