{
  "mode": "dense",
  "args": {
    "mode": "dense",
    "supervision": "full_belief",
    "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_full_belief_mod31",
    "checkpoint_dir": "large_artifacts/dense_supervision_ladder/checkpoints/main_full_belief_mod31"
  },
  "env": {
    "python": "3.12.3",
    "torch": "2.8.0+cu128",
    "device": "cuda",
    "cuda_name": "NVIDIA RTX 6000 Ada Generation"
  },
  "train": [
    {
      "step": 1,
      "loss": 10.34587287902832,
      "grad_norm": 0.33807992935180664,
      "query_loss": 3.4527366161346436,
      "belief_loss": 6.893136024475098,
      "query_target_mass": 0.353519082069397,
      "query_target_nll": 3.3260269165039062,
      "query_top1_on_support": 0.50390625,
      "mean_query_support_size": 10.6796875
    },
    {
      "step": 250,
      "loss": 7.6970930099487305,
      "grad_norm": 0.4524201452732086,
      "query_loss": 2.9392197132110596,
      "belief_loss": 4.75787353515625,
      "query_target_mass": 0.48771774768829346,
      "query_target_nll": 3.045257091522217,
      "query_top1_on_support": 0.53125,
      "mean_query_support_size": 11.55859375
    },
    {
      "step": 500,
      "loss": 7.440953731536865,
      "grad_norm": 0.49368855357170105,
      "query_loss": 2.828256607055664,
      "belief_loss": 4.612697124481201,
      "query_target_mass": 0.5264670252799988,
      "query_target_nll": 2.9144906997680664,
      "query_top1_on_support": 0.55859375,
      "mean_query_support_size": 11.1875
    },
    {
      "step": 750,
      "loss": 6.831553936004639,
      "grad_norm": 1.1294127702713013,
      "query_loss": 2.6547465324401855,
      "belief_loss": 4.176807403564453,
      "query_target_mass": 0.5697920322418213,
      "query_target_nll": 2.73593807220459,
      "query_top1_on_support": 0.6640625,
      "mean_query_support_size": 11.765625
    },
    {
      "step": 1000,
      "loss": 6.16663932800293,
      "grad_norm": 1.2107579708099365,
      "query_loss": 2.4895098209381104,
      "belief_loss": 3.6771295070648193,
      "query_target_mass": 0.6455705761909485,
      "query_target_nll": 2.419189691543579,
      "query_top1_on_support": 0.78125,
      "mean_query_support_size": 11.140625
    },
    {
      "step": 1250,
      "loss": 6.0213470458984375,
      "grad_norm": 1.0897949934005737,
      "query_loss": 2.4552273750305176,
      "belief_loss": 3.566119909286499,
      "query_target_mass": 0.6588866710662842,
      "query_target_nll": 2.308969497680664,
      "query_top1_on_support": 0.79296875,
      "mean_query_support_size": 10.265625
    },
    {
      "step": 1500,
      "loss": 5.755584716796875,
      "grad_norm": 1.1078656911849976,
      "query_loss": 2.3465585708618164,
      "belief_loss": 3.4090263843536377,
      "query_target_mass": 0.7011427879333496,
      "query_target_nll": 2.287188768386841,
      "query_top1_on_support": 0.78515625,
      "mean_query_support_size": 11.32421875
    },
    {
      "step": 1750,
      "loss": 5.798122406005859,
      "grad_norm": 1.2386200428009033,
      "query_loss": 2.409600257873535,
      "belief_loss": 3.3885223865509033,
      "query_target_mass": 0.7358611226081848,
      "query_target_nll": 2.290496587753296,
      "query_top1_on_support": 0.82421875,
      "mean_query_support_size": 11.48828125
    },
    {
      "step": 2000,
      "loss": 5.632470607757568,
      "grad_norm": 1.298746109008789,
      "query_loss": 2.367485523223877,
      "belief_loss": 3.2649850845336914,
      "query_target_mass": 0.7159945964813232,
      "query_target_nll": 2.2957873344421387,
      "query_top1_on_support": 0.83203125,
      "mean_query_support_size": 11.22265625
    }
  ],
  "probe_train": [
    {
      "step": 1,
      "loss": 6.884596347808838,
      "grad_norm": 0.22687296569347382,
      "probe_belief_target_mass": 0.019821051508188248,
      "probe_belief_target_nll": 6.884596347808838,
      "probe_belief_top1_on_support": 0.0047743055038154125,
      "mean_belief_support_size": 19.06727409362793
    },
    {
      "step": 200,
      "loss": 3.6011803150177,
      "grad_norm": 0.6067706942558289,
      "probe_belief_target_mass": 0.57196444272995,
      "probe_belief_target_nll": 3.6011803150177,
      "probe_belief_top1_on_support": 0.7261284589767456,
      "mean_belief_support_size": 17.91232681274414
    },
    {
      "step": 400,
      "loss": 3.42413330078125,
      "grad_norm": 0.5151277780532837,
      "probe_belief_target_mass": 0.6495693325996399,
      "probe_belief_target_nll": 3.42413330078125,
      "probe_belief_top1_on_support": 0.7630208134651184,
      "mean_belief_support_size": 17.484375
    },
    {
      "step": 600,
      "loss": 3.4000189304351807,
      "grad_norm": 0.510934054851532,
      "probe_belief_target_mass": 0.6812030673027039,
      "probe_belief_target_nll": 3.4000189304351807,
      "probe_belief_top1_on_support": 0.7855902910232544,
      "mean_belief_support_size": 17.92447853088379
    },
    {
      "step": 800,
      "loss": 3.4028329849243164,
      "grad_norm": 0.5036962032318115,
      "probe_belief_target_mass": 0.6883137822151184,
      "probe_belief_target_nll": 3.4028329849243164,
      "probe_belief_top1_on_support": 0.7690972089767456,
      "mean_belief_support_size": 18.07725715637207
    }
  ],
  "eval": [
    {
      "step": 2000,
      "rows": [
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 4,
          "query_type": "A",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.38670164346694946,
          "query_target_nll": 3.445953369140625,
          "query_top1_on_support": 0.37109375,
          "mean_query_support_size": 12.01171875,
          "probe_belief_target_mass": 0.024177201092243195,
          "probe_belief_target_nll": 20.435028076171875,
          "probe_belief_top1_on_support": 0.01953125,
          "mean_belief_support_size": 13.84765625,
          "decoder_belief_target_mass": 0.023622898384928703,
          "decoder_belief_target_nll": 24.720022201538086,
          "decoder_belief_top1_on_support": 0.01953125,
          "mean_decoder_belief_support_size": 13.84765625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 4,
          "query_type": "A",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.4342309832572937,
          "query_target_nll": 5.421143531799316,
          "query_top1_on_support": 0.453125,
          "mean_query_support_size": 12.01171875,
          "probe_belief_target_mass": 0.04790434241294861,
          "probe_belief_target_nll": 23.54358673095703,
          "probe_belief_top1_on_support": 0.05078125,
          "mean_belief_support_size": 13.84765625,
          "decoder_belief_target_mass": 0.0474700853228569,
          "decoder_belief_target_nll": 27.42455291748047,
          "decoder_belief_top1_on_support": 0.06640625,
          "mean_decoder_belief_support_size": 13.84765625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 4,
          "query_type": "A",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.5401357412338257,
          "query_target_nll": 4.575626373291016,
          "query_top1_on_support": 0.54296875,
          "mean_query_support_size": 12.01171875,
          "probe_belief_target_mass": 0.12435920536518097,
          "probe_belief_target_nll": 17.67121124267578,
          "probe_belief_top1_on_support": 0.1328125,
          "mean_belief_support_size": 13.84765625,
          "decoder_belief_target_mass": 0.12699714303016663,
          "decoder_belief_target_nll": 20.283519744873047,
          "decoder_belief_top1_on_support": 0.1328125,
          "mean_decoder_belief_support_size": 13.84765625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 4,
          "query_type": "A",
          "k": 4,
          "n": 256,
          "query_target_mass": 0.7716895341873169,
          "query_target_nll": 2.367856025695801,
          "query_top1_on_support": 0.828125,
          "mean_query_support_size": 12.01171875,
          "probe_belief_target_mass": 0.45342904329299927,
          "probe_belief_target_nll": 3.737516403198242,
          "probe_belief_top1_on_support": 0.6015625,
          "mean_belief_support_size": 13.84765625,
          "decoder_belief_target_mass": 0.4737215042114258,
          "decoder_belief_target_nll": 3.6338882446289062,
          "decoder_belief_top1_on_support": 0.609375,
          "mean_decoder_belief_support_size": 13.84765625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 4,
          "query_type": "A",
          "k": 8,
          "n": 256,
          "query_target_mass": 0.7716895341873169,
          "query_target_nll": 2.367856025695801,
          "query_top1_on_support": 0.828125,
          "mean_query_support_size": 12.01171875,
          "probe_belief_target_mass": 0.45342904329299927,
          "probe_belief_target_nll": 3.737516403198242,
          "probe_belief_top1_on_support": 0.6015625,
          "mean_belief_support_size": 13.84765625,
          "decoder_belief_target_mass": 0.4737215042114258,
          "decoder_belief_target_nll": 3.6338882446289062,
          "decoder_belief_top1_on_support": 0.609375,
          "mean_decoder_belief_support_size": 13.84765625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 4,
          "query_type": "A",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.7716895341873169,
          "query_target_nll": 2.367856025695801,
          "query_top1_on_support": 0.828125,
          "mean_query_support_size": 12.01171875,
          "probe_belief_target_mass": 0.45342904329299927,
          "probe_belief_target_nll": 3.737516403198242,
          "probe_belief_top1_on_support": 0.6015625,
          "mean_belief_support_size": 13.84765625,
          "decoder_belief_target_mass": 0.4737215042114258,
          "decoder_belief_target_nll": 3.6338882446289062,
          "decoder_belief_top1_on_support": 0.609375,
          "mean_decoder_belief_support_size": 13.84765625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 4,
          "query_type": "A",
          "k": 16,
          "n": 256,
          "query_target_mass": 0.7716895341873169,
          "query_target_nll": 2.367856025695801,
          "query_top1_on_support": 0.828125,
          "mean_query_support_size": 12.01171875,
          "probe_belief_target_mass": 0.45342904329299927,
          "probe_belief_target_nll": 3.737516403198242,
          "probe_belief_top1_on_support": 0.6015625,
          "mean_belief_support_size": 13.84765625,
          "decoder_belief_target_mass": 0.4737215042114258,
          "decoder_belief_target_nll": 3.6338882446289062,
          "decoder_belief_top1_on_support": 0.609375,
          "mean_decoder_belief_support_size": 13.84765625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 4,
          "query_type": "A",
          "k": 24,
          "n": 256,
          "query_target_mass": 0.7716895341873169,
          "query_target_nll": 2.367856025695801,
          "query_top1_on_support": 0.828125,
          "mean_query_support_size": 12.01171875,
          "probe_belief_target_mass": 0.45342904329299927,
          "probe_belief_target_nll": 3.737516403198242,
          "probe_belief_top1_on_support": 0.6015625,
          "mean_belief_support_size": 13.84765625,
          "decoder_belief_target_mass": 0.4737215042114258,
          "decoder_belief_target_nll": 3.6338882446289062,
          "decoder_belief_top1_on_support": 0.609375,
          "mean_decoder_belief_support_size": 13.84765625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 4,
          "query_type": "B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.35738152265548706,
          "query_target_nll": 3.45137619972229,
          "query_top1_on_support": 0.36328125,
          "mean_query_support_size": 11.109375,
          "probe_belief_target_mass": 0.024719536304473877,
          "probe_belief_target_nll": 19.975366592407227,
          "probe_belief_top1_on_support": 0.0234375,
          "mean_belief_support_size": 13.328125,
          "decoder_belief_target_mass": 0.024495981633663177,
          "decoder_belief_target_nll": 24.361562728881836,
          "decoder_belief_top1_on_support": 0.03515625,
          "mean_decoder_belief_support_size": 13.328125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 4,
          "query_type": "B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.42011135816574097,
          "query_target_nll": 5.680281639099121,
          "query_top1_on_support": 0.421875,
          "mean_query_support_size": 11.109375,
          "probe_belief_target_mass": 0.07697584480047226,
          "probe_belief_target_nll": 21.60219955444336,
          "probe_belief_top1_on_support": 0.078125,
          "mean_belief_support_size": 13.328125,
          "decoder_belief_target_mass": 0.07650911062955856,
          "decoder_belief_target_nll": 25.33791732788086,
          "decoder_belief_top1_on_support": 0.0703125,
          "mean_decoder_belief_support_size": 13.328125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 4,
          "query_type": "B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.525906503200531,
          "query_target_nll": 5.008063316345215,
          "query_top1_on_support": 0.546875,
          "mean_query_support_size": 11.109375,
          "probe_belief_target_mass": 0.13864076137542725,
          "probe_belief_target_nll": 18.22135353088379,
          "probe_belief_top1_on_support": 0.16796875,
          "mean_belief_support_size": 13.328125,
          "decoder_belief_target_mass": 0.14213040471076965,
          "decoder_belief_target_nll": 20.887447357177734,
          "decoder_belief_top1_on_support": 0.15234375,
          "mean_decoder_belief_support_size": 13.328125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 4,
          "query_type": "B",
          "k": 4,
          "n": 256,
          "query_target_mass": 0.7970174551010132,
          "query_target_nll": 2.1896252632141113,
          "query_top1_on_support": 0.8984375,
          "mean_query_support_size": 11.109375,
          "probe_belief_target_mass": 0.5176267623901367,
          "probe_belief_target_nll": 3.4372167587280273,
          "probe_belief_top1_on_support": 0.65625,
          "mean_belief_support_size": 13.328125,
          "decoder_belief_target_mass": 0.5278192758560181,
          "decoder_belief_target_nll": 3.3620972633361816,
          "decoder_belief_top1_on_support": 0.66015625,
          "mean_decoder_belief_support_size": 13.328125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 4,
          "query_type": "B",
          "k": 8,
          "n": 256,
          "query_target_mass": 0.7970174551010132,
          "query_target_nll": 2.1896252632141113,
          "query_top1_on_support": 0.8984375,
          "mean_query_support_size": 11.109375,
          "probe_belief_target_mass": 0.5176267623901367,
          "probe_belief_target_nll": 3.4372167587280273,
          "probe_belief_top1_on_support": 0.65625,
          "mean_belief_support_size": 13.328125,
          "decoder_belief_target_mass": 0.5278192758560181,
          "decoder_belief_target_nll": 3.3620972633361816,
          "decoder_belief_top1_on_support": 0.66015625,
          "mean_decoder_belief_support_size": 13.328125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 4,
          "query_type": "B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.7970174551010132,
          "query_target_nll": 2.1896252632141113,
          "query_top1_on_support": 0.8984375,
          "mean_query_support_size": 11.109375,
          "probe_belief_target_mass": 0.5176267623901367,
          "probe_belief_target_nll": 3.4372167587280273,
          "probe_belief_top1_on_support": 0.65625,
          "mean_belief_support_size": 13.328125,
          "decoder_belief_target_mass": 0.5278192758560181,
          "decoder_belief_target_nll": 3.3620972633361816,
          "decoder_belief_top1_on_support": 0.66015625,
          "mean_decoder_belief_support_size": 13.328125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 4,
          "query_type": "B",
          "k": 16,
          "n": 256,
          "query_target_mass": 0.7970174551010132,
          "query_target_nll": 2.1896252632141113,
          "query_top1_on_support": 0.8984375,
          "mean_query_support_size": 11.109375,
          "probe_belief_target_mass": 0.5176267623901367,
          "probe_belief_target_nll": 3.4372167587280273,
          "probe_belief_top1_on_support": 0.65625,
          "mean_belief_support_size": 13.328125,
          "decoder_belief_target_mass": 0.5278192758560181,
          "decoder_belief_target_nll": 3.3620972633361816,
          "decoder_belief_top1_on_support": 0.66015625,
          "mean_decoder_belief_support_size": 13.328125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 4,
          "query_type": "B",
          "k": 24,
          "n": 256,
          "query_target_mass": 0.7970174551010132,
          "query_target_nll": 2.1896252632141113,
          "query_top1_on_support": 0.8984375,
          "mean_query_support_size": 11.109375,
          "probe_belief_target_mass": 0.5176267623901367,
          "probe_belief_target_nll": 3.4372167587280273,
          "probe_belief_top1_on_support": 0.65625,
          "mean_belief_support_size": 13.328125,
          "decoder_belief_target_mass": 0.5278192758560181,
          "decoder_belief_target_nll": 3.3620972633361816,
          "decoder_belief_top1_on_support": 0.66015625,
          "mean_decoder_belief_support_size": 13.328125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 4,
          "query_type": "A_PLUS_B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.36923345923423767,
          "query_target_nll": 3.444960355758667,
          "query_top1_on_support": 0.36328125,
          "mean_query_support_size": 11.453125,
          "probe_belief_target_mass": 0.031481001526117325,
          "probe_belief_target_nll": 18.765033721923828,
          "probe_belief_top1_on_support": 0.03125,
          "mean_belief_support_size": 12.0859375,
          "decoder_belief_target_mass": 0.03191293776035309,
          "decoder_belief_target_nll": 22.737995147705078,
          "decoder_belief_top1_on_support": 0.02734375,
          "mean_decoder_belief_support_size": 12.0859375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 4,
          "query_type": "A_PLUS_B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.38201090693473816,
          "query_target_nll": 4.168905735015869,
          "query_top1_on_support": 0.3828125,
          "mean_query_support_size": 11.453125,
          "probe_belief_target_mass": 0.08882996439933777,
          "probe_belief_target_nll": 20.670331954956055,
          "probe_belief_top1_on_support": 0.08203125,
          "mean_belief_support_size": 12.0859375,
          "decoder_belief_target_mass": 0.0886441096663475,
          "decoder_belief_target_nll": 24.00691795349121,
          "decoder_belief_top1_on_support": 0.07421875,
          "mean_decoder_belief_support_size": 12.0859375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 4,
          "query_type": "A_PLUS_B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.40092310309410095,
          "query_target_nll": 4.293844223022461,
          "query_top1_on_support": 0.37890625,
          "mean_query_support_size": 11.453125,
          "probe_belief_target_mass": 0.1409604847431183,
          "probe_belief_target_nll": 17.613082885742188,
          "probe_belief_top1_on_support": 0.15234375,
          "mean_belief_support_size": 12.0859375,
          "decoder_belief_target_mass": 0.14299729466438293,
          "decoder_belief_target_nll": 19.569908142089844,
          "decoder_belief_top1_on_support": 0.1484375,
          "mean_decoder_belief_support_size": 12.0859375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 4,
          "query_type": "A_PLUS_B",
          "k": 4,
          "n": 256,
          "query_target_mass": 0.6381945610046387,
          "query_target_nll": 2.696251392364502,
          "query_top1_on_support": 0.75390625,
          "mean_query_support_size": 11.453125,
          "probe_belief_target_mass": 0.491230845451355,
          "probe_belief_target_nll": 3.499908924102783,
          "probe_belief_top1_on_support": 0.640625,
          "mean_belief_support_size": 12.0859375,
          "decoder_belief_target_mass": 0.510640025138855,
          "decoder_belief_target_nll": 3.363720178604126,
          "decoder_belief_top1_on_support": 0.62890625,
          "mean_decoder_belief_support_size": 12.0859375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 4,
          "query_type": "A_PLUS_B",
          "k": 8,
          "n": 256,
          "query_target_mass": 0.6381945610046387,
          "query_target_nll": 2.696251392364502,
          "query_top1_on_support": 0.75390625,
          "mean_query_support_size": 11.453125,
          "probe_belief_target_mass": 0.491230845451355,
          "probe_belief_target_nll": 3.499908924102783,
          "probe_belief_top1_on_support": 0.640625,
          "mean_belief_support_size": 12.0859375,
          "decoder_belief_target_mass": 0.510640025138855,
          "decoder_belief_target_nll": 3.363720178604126,
          "decoder_belief_top1_on_support": 0.62890625,
          "mean_decoder_belief_support_size": 12.0859375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 4,
          "query_type": "A_PLUS_B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.6381945610046387,
          "query_target_nll": 2.696251392364502,
          "query_top1_on_support": 0.75390625,
          "mean_query_support_size": 11.453125,
          "probe_belief_target_mass": 0.491230845451355,
          "probe_belief_target_nll": 3.499908924102783,
          "probe_belief_top1_on_support": 0.640625,
          "mean_belief_support_size": 12.0859375,
          "decoder_belief_target_mass": 0.510640025138855,
          "decoder_belief_target_nll": 3.363720178604126,
          "decoder_belief_top1_on_support": 0.62890625,
          "mean_decoder_belief_support_size": 12.0859375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 4,
          "query_type": "A_PLUS_B",
          "k": 16,
          "n": 256,
          "query_target_mass": 0.6381945610046387,
          "query_target_nll": 2.696251392364502,
          "query_top1_on_support": 0.75390625,
          "mean_query_support_size": 11.453125,
          "probe_belief_target_mass": 0.491230845451355,
          "probe_belief_target_nll": 3.499908924102783,
          "probe_belief_top1_on_support": 0.640625,
          "mean_belief_support_size": 12.0859375,
          "decoder_belief_target_mass": 0.510640025138855,
          "decoder_belief_target_nll": 3.363720178604126,
          "decoder_belief_top1_on_support": 0.62890625,
          "mean_decoder_belief_support_size": 12.0859375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 4,
          "query_type": "A_PLUS_B",
          "k": 24,
          "n": 256,
          "query_target_mass": 0.6381945610046387,
          "query_target_nll": 2.696251392364502,
          "query_top1_on_support": 0.75390625,
          "mean_query_support_size": 11.453125,
          "probe_belief_target_mass": 0.491230845451355,
          "probe_belief_target_nll": 3.499908924102783,
          "probe_belief_top1_on_support": 0.640625,
          "mean_belief_support_size": 12.0859375,
          "decoder_belief_target_mass": 0.510640025138855,
          "decoder_belief_target_nll": 3.363720178604126,
          "decoder_belief_top1_on_support": 0.62890625,
          "mean_decoder_belief_support_size": 12.0859375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 4,
          "query_type": "A_MINUS_B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.4296283423900604,
          "query_target_nll": 17.803972244262695,
          "query_top1_on_support": 0.4296875,
          "mean_query_support_size": 11.60546875,
          "probe_belief_target_mass": 0.03448978066444397,
          "probe_belief_target_nll": 19.863502502441406,
          "probe_belief_top1_on_support": 0.0390625,
          "mean_belief_support_size": 14.40625,
          "decoder_belief_target_mass": 0.035107143223285675,
          "decoder_belief_target_nll": 23.936626434326172,
          "decoder_belief_top1_on_support": 0.03125,
          "mean_decoder_belief_support_size": 14.40625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 4,
          "query_type": "A_MINUS_B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.42343324422836304,
          "query_target_nll": 18.641918182373047,
          "query_top1_on_support": 0.421875,
          "mean_query_support_size": 11.60546875,
          "probe_belief_target_mass": 0.06758037209510803,
          "probe_belief_target_nll": 22.970455169677734,
          "probe_belief_top1_on_support": 0.0703125,
          "mean_belief_support_size": 14.40625,
          "decoder_belief_target_mass": 0.06942355632781982,
          "decoder_belief_target_nll": 27.041967391967773,
          "decoder_belief_top1_on_support": 0.078125,
          "mean_decoder_belief_support_size": 14.40625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 4,
          "query_type": "A_MINUS_B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.449995219707489,
          "query_target_nll": 12.995741844177246,
          "query_top1_on_support": 0.453125,
          "mean_query_support_size": 11.60546875,
          "probe_belief_target_mass": 0.1469222754240036,
          "probe_belief_target_nll": 17.560333251953125,
          "probe_belief_top1_on_support": 0.1640625,
          "mean_belief_support_size": 14.40625,
          "decoder_belief_target_mass": 0.15168844163417816,
          "decoder_belief_target_nll": 20.01897621154785,
          "decoder_belief_top1_on_support": 0.16796875,
          "mean_decoder_belief_support_size": 14.40625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 4,
          "query_type": "A_MINUS_B",
          "k": 4,
          "n": 256,
          "query_target_mass": 0.7567803859710693,
          "query_target_nll": 2.185957908630371,
          "query_top1_on_support": 0.84375,
          "mean_query_support_size": 11.60546875,
          "probe_belief_target_mass": 0.5119541883468628,
          "probe_belief_target_nll": 3.5899887084960938,
          "probe_belief_top1_on_support": 0.6328125,
          "mean_belief_support_size": 14.40625,
          "decoder_belief_target_mass": 0.544764518737793,
          "decoder_belief_target_nll": 3.4640579223632812,
          "decoder_belief_top1_on_support": 0.67578125,
          "mean_decoder_belief_support_size": 14.40625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 4,
          "query_type": "A_MINUS_B",
          "k": 8,
          "n": 256,
          "query_target_mass": 0.7567803859710693,
          "query_target_nll": 2.185957908630371,
          "query_top1_on_support": 0.84375,
          "mean_query_support_size": 11.60546875,
          "probe_belief_target_mass": 0.5119541883468628,
          "probe_belief_target_nll": 3.5899887084960938,
          "probe_belief_top1_on_support": 0.6328125,
          "mean_belief_support_size": 14.40625,
          "decoder_belief_target_mass": 0.544764518737793,
          "decoder_belief_target_nll": 3.4640579223632812,
          "decoder_belief_top1_on_support": 0.67578125,
          "mean_decoder_belief_support_size": 14.40625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 4,
          "query_type": "A_MINUS_B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.7567803859710693,
          "query_target_nll": 2.185957908630371,
          "query_top1_on_support": 0.84375,
          "mean_query_support_size": 11.60546875,
          "probe_belief_target_mass": 0.5119541883468628,
          "probe_belief_target_nll": 3.5899887084960938,
          "probe_belief_top1_on_support": 0.6328125,
          "mean_belief_support_size": 14.40625,
          "decoder_belief_target_mass": 0.544764518737793,
          "decoder_belief_target_nll": 3.4640579223632812,
          "decoder_belief_top1_on_support": 0.67578125,
          "mean_decoder_belief_support_size": 14.40625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 4,
          "query_type": "A_MINUS_B",
          "k": 16,
          "n": 256,
          "query_target_mass": 0.7567803859710693,
          "query_target_nll": 2.185957908630371,
          "query_top1_on_support": 0.84375,
          "mean_query_support_size": 11.60546875,
          "probe_belief_target_mass": 0.5119541883468628,
          "probe_belief_target_nll": 3.5899887084960938,
          "probe_belief_top1_on_support": 0.6328125,
          "mean_belief_support_size": 14.40625,
          "decoder_belief_target_mass": 0.544764518737793,
          "decoder_belief_target_nll": 3.4640579223632812,
          "decoder_belief_top1_on_support": 0.67578125,
          "mean_decoder_belief_support_size": 14.40625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 4,
          "query_type": "A_MINUS_B",
          "k": 24,
          "n": 256,
          "query_target_mass": 0.7567803859710693,
          "query_target_nll": 2.185957908630371,
          "query_top1_on_support": 0.84375,
          "mean_query_support_size": 11.60546875,
          "probe_belief_target_mass": 0.5119541883468628,
          "probe_belief_target_nll": 3.5899887084960938,
          "probe_belief_top1_on_support": 0.6328125,
          "mean_belief_support_size": 14.40625,
          "decoder_belief_target_mass": 0.544764518737793,
          "decoder_belief_target_nll": 3.4640579223632812,
          "decoder_belief_top1_on_support": 0.67578125,
          "mean_decoder_belief_support_size": 14.40625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 8,
          "query_type": "A",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.17062892019748688,
          "query_target_nll": 3.43699049949646,
          "query_top1_on_support": 0.171875,
          "mean_query_support_size": 5.2890625,
          "probe_belief_target_mass": 0.007689051330089569,
          "probe_belief_target_nll": 20.730754852294922,
          "probe_belief_top1_on_support": 0.00390625,
          "mean_belief_support_size": 7.140625,
          "decoder_belief_target_mass": 0.007714299950748682,
          "decoder_belief_target_nll": 25.11417007446289,
          "decoder_belief_top1_on_support": 0.0078125,
          "mean_decoder_belief_support_size": 7.140625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 8,
          "query_type": "A",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.1960752308368683,
          "query_target_nll": 7.096009254455566,
          "query_top1_on_support": 0.1875,
          "mean_query_support_size": 5.2890625,
          "probe_belief_target_mass": 0.012637019157409668,
          "probe_belief_target_nll": 25.788055419921875,
          "probe_belief_top1_on_support": 0.0078125,
          "mean_belief_support_size": 7.140625,
          "decoder_belief_target_mass": 0.012698015198111534,
          "decoder_belief_target_nll": 30.0733642578125,
          "decoder_belief_top1_on_support": 0.01953125,
          "mean_decoder_belief_support_size": 7.140625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 8,
          "query_type": "A",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.2123914361000061,
          "query_target_nll": 7.5457916259765625,
          "query_top1_on_support": 0.23046875,
          "mean_query_support_size": 5.2890625,
          "probe_belief_target_mass": 0.01767536997795105,
          "probe_belief_target_nll": 23.277751922607422,
          "probe_belief_top1_on_support": 0.015625,
          "mean_belief_support_size": 7.140625,
          "decoder_belief_target_mass": 0.019800476729869843,
          "decoder_belief_target_nll": 26.786800384521484,
          "decoder_belief_top1_on_support": 0.02734375,
          "mean_decoder_belief_support_size": 7.140625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 8,
          "query_type": "A",
          "k": 4,
          "n": 256,
          "query_target_mass": 0.2434786558151245,
          "query_target_nll": 8.739091873168945,
          "query_top1_on_support": 0.25,
          "mean_query_support_size": 5.2890625,
          "probe_belief_target_mass": 0.034201569855213165,
          "probe_belief_target_nll": 18.376543045043945,
          "probe_belief_top1_on_support": 0.05078125,
          "mean_belief_support_size": 7.140625,
          "decoder_belief_target_mass": 0.036072444170713425,
          "decoder_belief_target_nll": 20.660919189453125,
          "decoder_belief_top1_on_support": 0.046875,
          "mean_decoder_belief_support_size": 7.140625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 8,
          "query_type": "A",
          "k": 8,
          "n": 256,
          "query_target_mass": 0.5141621232032776,
          "query_target_nll": 2.1218762397766113,
          "query_top1_on_support": 0.609375,
          "mean_query_support_size": 5.2890625,
          "probe_belief_target_mass": 0.2132147252559662,
          "probe_belief_target_nll": 3.7970974445343018,
          "probe_belief_top1_on_support": 0.30078125,
          "mean_belief_support_size": 7.140625,
          "decoder_belief_target_mass": 0.22018641233444214,
          "decoder_belief_target_nll": 3.781383991241455,
          "decoder_belief_top1_on_support": 0.29296875,
          "mean_decoder_belief_support_size": 7.140625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 8,
          "query_type": "A",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.5141621232032776,
          "query_target_nll": 2.1218762397766113,
          "query_top1_on_support": 0.609375,
          "mean_query_support_size": 5.2890625,
          "probe_belief_target_mass": 0.2132147252559662,
          "probe_belief_target_nll": 3.7970974445343018,
          "probe_belief_top1_on_support": 0.30078125,
          "mean_belief_support_size": 7.140625,
          "decoder_belief_target_mass": 0.22018641233444214,
          "decoder_belief_target_nll": 3.781383991241455,
          "decoder_belief_top1_on_support": 0.29296875,
          "mean_decoder_belief_support_size": 7.140625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 8,
          "query_type": "A",
          "k": 16,
          "n": 256,
          "query_target_mass": 0.5141621232032776,
          "query_target_nll": 2.1218762397766113,
          "query_top1_on_support": 0.609375,
          "mean_query_support_size": 5.2890625,
          "probe_belief_target_mass": 0.2132147252559662,
          "probe_belief_target_nll": 3.7970974445343018,
          "probe_belief_top1_on_support": 0.30078125,
          "mean_belief_support_size": 7.140625,
          "decoder_belief_target_mass": 0.22018641233444214,
          "decoder_belief_target_nll": 3.781383991241455,
          "decoder_belief_top1_on_support": 0.29296875,
          "mean_decoder_belief_support_size": 7.140625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 8,
          "query_type": "A",
          "k": 24,
          "n": 256,
          "query_target_mass": 0.5141621232032776,
          "query_target_nll": 2.1218762397766113,
          "query_top1_on_support": 0.609375,
          "mean_query_support_size": 5.2890625,
          "probe_belief_target_mass": 0.2132147252559662,
          "probe_belief_target_nll": 3.7970974445343018,
          "probe_belief_top1_on_support": 0.30078125,
          "mean_belief_support_size": 7.140625,
          "decoder_belief_target_mass": 0.22018641233444214,
          "decoder_belief_target_nll": 3.781383991241455,
          "decoder_belief_top1_on_support": 0.29296875,
          "mean_decoder_belief_support_size": 7.140625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 8,
          "query_type": "B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.18891799449920654,
          "query_target_nll": 3.438258409500122,
          "query_top1_on_support": 0.19140625,
          "mean_query_support_size": 5.84765625,
          "probe_belief_target_mass": 0.0075687747448682785,
          "probe_belief_target_nll": 20.60618782043457,
          "probe_belief_top1_on_support": 0.015625,
          "mean_belief_support_size": 7.09765625,
          "decoder_belief_target_mass": 0.00785771943628788,
          "decoder_belief_target_nll": 24.89812469482422,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 7.09765625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 8,
          "query_type": "B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.200783371925354,
          "query_target_nll": 7.038937568664551,
          "query_top1_on_support": 0.1796875,
          "mean_query_support_size": 5.84765625,
          "probe_belief_target_mass": 0.014147687703371048,
          "probe_belief_target_nll": 25.203067779541016,
          "probe_belief_top1_on_support": 0.01171875,
          "mean_belief_support_size": 7.09765625,
          "decoder_belief_target_mass": 0.014486076310276985,
          "decoder_belief_target_nll": 29.23245620727539,
          "decoder_belief_top1_on_support": 0.015625,
          "mean_decoder_belief_support_size": 7.09765625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 8,
          "query_type": "B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.22321107983589172,
          "query_target_nll": 7.762014865875244,
          "query_top1_on_support": 0.22265625,
          "mean_query_support_size": 5.84765625,
          "probe_belief_target_mass": 0.021793434396386147,
          "probe_belief_target_nll": 22.33222198486328,
          "probe_belief_top1_on_support": 0.01953125,
          "mean_belief_support_size": 7.09765625,
          "decoder_belief_target_mass": 0.022100120782852173,
          "decoder_belief_target_nll": 25.767730712890625,
          "decoder_belief_top1_on_support": 0.0234375,
          "mean_decoder_belief_support_size": 7.09765625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 8,
          "query_type": "B",
          "k": 4,
          "n": 256,
          "query_target_mass": 0.2867579460144043,
          "query_target_nll": 7.653135776519775,
          "query_top1_on_support": 0.30859375,
          "mean_query_support_size": 5.84765625,
          "probe_belief_target_mass": 0.03244370222091675,
          "probe_belief_target_nll": 18.660110473632812,
          "probe_belief_top1_on_support": 0.0390625,
          "mean_belief_support_size": 7.09765625,
          "decoder_belief_target_mass": 0.03327341377735138,
          "decoder_belief_target_nll": 21.015329360961914,
          "decoder_belief_top1_on_support": 0.03125,
          "mean_decoder_belief_support_size": 7.09765625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 8,
          "query_type": "B",
          "k": 8,
          "n": 256,
          "query_target_mass": 0.5706254243850708,
          "query_target_nll": 2.1008524894714355,
          "query_top1_on_support": 0.65625,
          "mean_query_support_size": 5.84765625,
          "probe_belief_target_mass": 0.22567038238048553,
          "probe_belief_target_nll": 3.9217960834503174,
          "probe_belief_top1_on_support": 0.3125,
          "mean_belief_support_size": 7.09765625,
          "decoder_belief_target_mass": 0.2327306866645813,
          "decoder_belief_target_nll": 3.8376524448394775,
          "decoder_belief_top1_on_support": 0.3203125,
          "mean_decoder_belief_support_size": 7.09765625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 8,
          "query_type": "B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.5706254243850708,
          "query_target_nll": 2.1008524894714355,
          "query_top1_on_support": 0.65625,
          "mean_query_support_size": 5.84765625,
          "probe_belief_target_mass": 0.22567038238048553,
          "probe_belief_target_nll": 3.9217960834503174,
          "probe_belief_top1_on_support": 0.3125,
          "mean_belief_support_size": 7.09765625,
          "decoder_belief_target_mass": 0.2327306866645813,
          "decoder_belief_target_nll": 3.8376524448394775,
          "decoder_belief_top1_on_support": 0.3203125,
          "mean_decoder_belief_support_size": 7.09765625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 8,
          "query_type": "B",
          "k": 16,
          "n": 256,
          "query_target_mass": 0.5706254243850708,
          "query_target_nll": 2.1008524894714355,
          "query_top1_on_support": 0.65625,
          "mean_query_support_size": 5.84765625,
          "probe_belief_target_mass": 0.22567038238048553,
          "probe_belief_target_nll": 3.9217960834503174,
          "probe_belief_top1_on_support": 0.3125,
          "mean_belief_support_size": 7.09765625,
          "decoder_belief_target_mass": 0.2327306866645813,
          "decoder_belief_target_nll": 3.8376524448394775,
          "decoder_belief_top1_on_support": 0.3203125,
          "mean_decoder_belief_support_size": 7.09765625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 8,
          "query_type": "B",
          "k": 24,
          "n": 256,
          "query_target_mass": 0.5706254243850708,
          "query_target_nll": 2.1008524894714355,
          "query_top1_on_support": 0.65625,
          "mean_query_support_size": 5.84765625,
          "probe_belief_target_mass": 0.22567038238048553,
          "probe_belief_target_nll": 3.9217960834503174,
          "probe_belief_top1_on_support": 0.3125,
          "mean_belief_support_size": 7.09765625,
          "decoder_belief_target_mass": 0.2327306866645813,
          "decoder_belief_target_nll": 3.8376524448394775,
          "decoder_belief_top1_on_support": 0.3203125,
          "mean_decoder_belief_support_size": 7.09765625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 8,
          "query_type": "A_PLUS_B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.17336010932922363,
          "query_target_nll": 3.436509132385254,
          "query_top1_on_support": 0.1953125,
          "mean_query_support_size": 5.3515625,
          "probe_belief_target_mass": 0.009554926306009293,
          "probe_belief_target_nll": 20.989295959472656,
          "probe_belief_top1_on_support": 0.0078125,
          "mean_belief_support_size": 5.765625,
          "decoder_belief_target_mass": 0.009639071300625801,
          "decoder_belief_target_nll": 25.743999481201172,
          "decoder_belief_top1_on_support": 0.0078125,
          "mean_decoder_belief_support_size": 5.765625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 8,
          "query_type": "A_PLUS_B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.17745478451251984,
          "query_target_nll": 4.386720657348633,
          "query_top1_on_support": 0.16015625,
          "mean_query_support_size": 5.3515625,
          "probe_belief_target_mass": 0.008424295112490654,
          "probe_belief_target_nll": 26.559988021850586,
          "probe_belief_top1_on_support": 0.01171875,
          "mean_belief_support_size": 5.765625,
          "decoder_belief_target_mass": 0.007505240384489298,
          "decoder_belief_target_nll": 30.9483699798584,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 5.765625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 8,
          "query_type": "A_PLUS_B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.16887956857681274,
          "query_target_nll": 4.68864107131958,
          "query_top1_on_support": 0.18359375,
          "mean_query_support_size": 5.3515625,
          "probe_belief_target_mass": 0.013062138110399246,
          "probe_belief_target_nll": 23.698143005371094,
          "probe_belief_top1_on_support": 0.015625,
          "mean_belief_support_size": 5.765625,
          "decoder_belief_target_mass": 0.010924098081886768,
          "decoder_belief_target_nll": 27.215694427490234,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 5.765625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 8,
          "query_type": "A_PLUS_B",
          "k": 4,
          "n": 256,
          "query_target_mass": 0.17949600517749786,
          "query_target_nll": 5.037837982177734,
          "query_top1_on_support": 0.16796875,
          "mean_query_support_size": 5.3515625,
          "probe_belief_target_mass": 0.03017348051071167,
          "probe_belief_target_nll": 18.525224685668945,
          "probe_belief_top1_on_support": 0.03515625,
          "mean_belief_support_size": 5.765625,
          "decoder_belief_target_mass": 0.028242740780115128,
          "decoder_belief_target_nll": 20.779735565185547,
          "decoder_belief_top1_on_support": 0.0390625,
          "mean_decoder_belief_support_size": 5.765625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 8,
          "query_type": "A_PLUS_B",
          "k": 8,
          "n": 256,
          "query_target_mass": 0.3895808458328247,
          "query_target_nll": 2.5520036220550537,
          "query_top1_on_support": 0.49609375,
          "mean_query_support_size": 5.3515625,
          "probe_belief_target_mass": 0.2055542916059494,
          "probe_belief_target_nll": 3.8428430557250977,
          "probe_belief_top1_on_support": 0.33203125,
          "mean_belief_support_size": 5.765625,
          "decoder_belief_target_mass": 0.19136548042297363,
          "decoder_belief_target_nll": 3.8604912757873535,
          "decoder_belief_top1_on_support": 0.296875,
          "mean_decoder_belief_support_size": 5.765625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 8,
          "query_type": "A_PLUS_B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.3895808458328247,
          "query_target_nll": 2.5520036220550537,
          "query_top1_on_support": 0.49609375,
          "mean_query_support_size": 5.3515625,
          "probe_belief_target_mass": 0.2055542916059494,
          "probe_belief_target_nll": 3.8428430557250977,
          "probe_belief_top1_on_support": 0.33203125,
          "mean_belief_support_size": 5.765625,
          "decoder_belief_target_mass": 0.19136548042297363,
          "decoder_belief_target_nll": 3.8604912757873535,
          "decoder_belief_top1_on_support": 0.296875,
          "mean_decoder_belief_support_size": 5.765625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 8,
          "query_type": "A_PLUS_B",
          "k": 16,
          "n": 256,
          "query_target_mass": 0.3895808458328247,
          "query_target_nll": 2.5520036220550537,
          "query_top1_on_support": 0.49609375,
          "mean_query_support_size": 5.3515625,
          "probe_belief_target_mass": 0.2055542916059494,
          "probe_belief_target_nll": 3.8428430557250977,
          "probe_belief_top1_on_support": 0.33203125,
          "mean_belief_support_size": 5.765625,
          "decoder_belief_target_mass": 0.19136548042297363,
          "decoder_belief_target_nll": 3.8604912757873535,
          "decoder_belief_top1_on_support": 0.296875,
          "mean_decoder_belief_support_size": 5.765625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 8,
          "query_type": "A_PLUS_B",
          "k": 24,
          "n": 256,
          "query_target_mass": 0.3895808458328247,
          "query_target_nll": 2.5520036220550537,
          "query_top1_on_support": 0.49609375,
          "mean_query_support_size": 5.3515625,
          "probe_belief_target_mass": 0.2055542916059494,
          "probe_belief_target_nll": 3.8428430557250977,
          "probe_belief_top1_on_support": 0.33203125,
          "mean_belief_support_size": 5.765625,
          "decoder_belief_target_mass": 0.19136548042297363,
          "decoder_belief_target_nll": 3.8604912757873535,
          "decoder_belief_top1_on_support": 0.296875,
          "mean_decoder_belief_support_size": 5.765625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 8,
          "query_type": "A_MINUS_B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.17969729006290436,
          "query_target_nll": 19.268463134765625,
          "query_top1_on_support": 0.1796875,
          "mean_query_support_size": 6.21875,
          "probe_belief_target_mass": 0.0060858167707920074,
          "probe_belief_target_nll": 21.666210174560547,
          "probe_belief_top1_on_support": 0.00390625,
          "mean_belief_support_size": 6.9609375,
          "decoder_belief_target_mass": 0.005891532637178898,
          "decoder_belief_target_nll": 26.076976776123047,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 6.9609375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 8,
          "query_type": "A_MINUS_B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.19136767089366913,
          "query_target_nll": 20.429170608520508,
          "query_top1_on_support": 0.19140625,
          "mean_query_support_size": 6.21875,
          "probe_belief_target_mass": 0.016381122171878815,
          "probe_belief_target_nll": 25.71109390258789,
          "probe_belief_top1_on_support": 0.0234375,
          "mean_belief_support_size": 6.9609375,
          "decoder_belief_target_mass": 0.01631227880716324,
          "decoder_belief_target_nll": 29.57242202758789,
          "decoder_belief_top1_on_support": 0.01953125,
          "mean_decoder_belief_support_size": 6.9609375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 8,
          "query_type": "A_MINUS_B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.218773752450943,
          "query_target_nll": 17.554080963134766,
          "query_top1_on_support": 0.21875,
          "mean_query_support_size": 6.21875,
          "probe_belief_target_mass": 0.020375410094857216,
          "probe_belief_target_nll": 23.912511825561523,
          "probe_belief_top1_on_support": 0.01953125,
          "mean_belief_support_size": 6.9609375,
          "decoder_belief_target_mass": 0.02027694508433342,
          "decoder_belief_target_nll": 27.383989334106445,
          "decoder_belief_top1_on_support": 0.02734375,
          "mean_decoder_belief_support_size": 6.9609375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 8,
          "query_type": "A_MINUS_B",
          "k": 4,
          "n": 256,
          "query_target_mass": 0.22679273784160614,
          "query_target_nll": 11.51268196105957,
          "query_top1_on_support": 0.21875,
          "mean_query_support_size": 6.21875,
          "probe_belief_target_mass": 0.0253602247685194,
          "probe_belief_target_nll": 19.484867095947266,
          "probe_belief_top1_on_support": 0.03125,
          "mean_belief_support_size": 6.9609375,
          "decoder_belief_target_mass": 0.028305385261774063,
          "decoder_belief_target_nll": 21.935440063476562,
          "decoder_belief_top1_on_support": 0.0390625,
          "mean_decoder_belief_support_size": 6.9609375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 8,
          "query_type": "A_MINUS_B",
          "k": 8,
          "n": 256,
          "query_target_mass": 0.5136630535125732,
          "query_target_nll": 2.3046205043792725,
          "query_top1_on_support": 0.5859375,
          "mean_query_support_size": 6.21875,
          "probe_belief_target_mass": 0.2171868085861206,
          "probe_belief_target_nll": 3.8453800678253174,
          "probe_belief_top1_on_support": 0.32421875,
          "mean_belief_support_size": 6.9609375,
          "decoder_belief_target_mass": 0.22240647673606873,
          "decoder_belief_target_nll": 3.7892463207244873,
          "decoder_belief_top1_on_support": 0.31640625,
          "mean_decoder_belief_support_size": 6.9609375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 8,
          "query_type": "A_MINUS_B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.5136630535125732,
          "query_target_nll": 2.3046205043792725,
          "query_top1_on_support": 0.5859375,
          "mean_query_support_size": 6.21875,
          "probe_belief_target_mass": 0.2171868085861206,
          "probe_belief_target_nll": 3.8453800678253174,
          "probe_belief_top1_on_support": 0.32421875,
          "mean_belief_support_size": 6.9609375,
          "decoder_belief_target_mass": 0.22240647673606873,
          "decoder_belief_target_nll": 3.7892463207244873,
          "decoder_belief_top1_on_support": 0.31640625,
          "mean_decoder_belief_support_size": 6.9609375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 8,
          "query_type": "A_MINUS_B",
          "k": 16,
          "n": 256,
          "query_target_mass": 0.5136630535125732,
          "query_target_nll": 2.3046205043792725,
          "query_top1_on_support": 0.5859375,
          "mean_query_support_size": 6.21875,
          "probe_belief_target_mass": 0.2171868085861206,
          "probe_belief_target_nll": 3.8453800678253174,
          "probe_belief_top1_on_support": 0.32421875,
          "mean_belief_support_size": 6.9609375,
          "decoder_belief_target_mass": 0.22240647673606873,
          "decoder_belief_target_nll": 3.7892463207244873,
          "decoder_belief_top1_on_support": 0.31640625,
          "mean_decoder_belief_support_size": 6.9609375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 8,
          "query_type": "A_MINUS_B",
          "k": 24,
          "n": 256,
          "query_target_mass": 0.5136630535125732,
          "query_target_nll": 2.3046205043792725,
          "query_top1_on_support": 0.5859375,
          "mean_query_support_size": 6.21875,
          "probe_belief_target_mass": 0.2171868085861206,
          "probe_belief_target_nll": 3.8453800678253174,
          "probe_belief_top1_on_support": 0.32421875,
          "mean_belief_support_size": 6.9609375,
          "decoder_belief_target_mass": 0.22240647673606873,
          "decoder_belief_target_nll": 3.7892463207244873,
          "decoder_belief_top1_on_support": 0.31640625,
          "mean_decoder_belief_support_size": 6.9609375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "A",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.11347337067127228,
          "query_target_nll": 3.441542387008667,
          "query_top1_on_support": 0.125,
          "mean_query_support_size": 3.5078125,
          "probe_belief_target_mass": 0.005247227847576141,
          "probe_belief_target_nll": 20.691295623779297,
          "probe_belief_top1_on_support": 0.0078125,
          "mean_belief_support_size": 3.8125,
          "decoder_belief_target_mass": 0.005186660215258598,
          "decoder_belief_target_nll": 25.08948516845703,
          "decoder_belief_top1_on_support": 0.01171875,
          "mean_decoder_belief_support_size": 3.8125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "A",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.11596766859292984,
          "query_target_nll": 7.061614513397217,
          "query_top1_on_support": 0.1328125,
          "mean_query_support_size": 3.5078125,
          "probe_belief_target_mass": 0.004456790164113045,
          "probe_belief_target_nll": 25.19416046142578,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 3.8125,
          "decoder_belief_target_mass": 0.004365209955722094,
          "decoder_belief_target_nll": 28.94913673400879,
          "decoder_belief_top1_on_support": 0.01171875,
          "mean_decoder_belief_support_size": 3.8125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "A",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.13747292757034302,
          "query_target_nll": 7.489340305328369,
          "query_top1_on_support": 0.19140625,
          "mean_query_support_size": 3.5078125,
          "probe_belief_target_mass": 0.00907878391444683,
          "probe_belief_target_nll": 22.19698715209961,
          "probe_belief_top1_on_support": 0.01171875,
          "mean_belief_support_size": 3.8125,
          "decoder_belief_target_mass": 0.00831284187734127,
          "decoder_belief_target_nll": 25.122608184814453,
          "decoder_belief_top1_on_support": 0.0078125,
          "mean_decoder_belief_support_size": 3.8125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "A",
          "k": 4,
          "n": 256,
          "query_target_mass": 0.12902364134788513,
          "query_target_nll": 8.232946395874023,
          "query_top1_on_support": 0.1171875,
          "mean_query_support_size": 3.5078125,
          "probe_belief_target_mass": 0.006108233705163002,
          "probe_belief_target_nll": 19.441394805908203,
          "probe_belief_top1_on_support": 0.01171875,
          "mean_belief_support_size": 3.8125,
          "decoder_belief_target_mass": 0.005677476525306702,
          "decoder_belief_target_nll": 21.499149322509766,
          "decoder_belief_top1_on_support": 0.01171875,
          "mean_decoder_belief_support_size": 3.8125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "A",
          "k": 8,
          "n": 256,
          "query_target_mass": 0.1580829918384552,
          "query_target_nll": 7.742947578430176,
          "query_top1_on_support": 0.1796875,
          "mean_query_support_size": 3.5078125,
          "probe_belief_target_mass": 0.014927646145224571,
          "probe_belief_target_nll": 14.984375953674316,
          "probe_belief_top1_on_support": 0.01953125,
          "mean_belief_support_size": 3.8125,
          "decoder_belief_target_mass": 0.012747695669531822,
          "decoder_belief_target_nll": 16.358110427856445,
          "decoder_belief_top1_on_support": 0.0234375,
          "mean_decoder_belief_support_size": 3.8125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "A",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.37379252910614014,
          "query_target_nll": 2.229926586151123,
          "query_top1_on_support": 0.42578125,
          "mean_query_support_size": 3.5078125,
          "probe_belief_target_mass": 0.1040026992559433,
          "probe_belief_target_nll": 4.089742660522461,
          "probe_belief_top1_on_support": 0.15625,
          "mean_belief_support_size": 3.8125,
          "decoder_belief_target_mass": 0.09536483138799667,
          "decoder_belief_target_nll": 4.1067352294921875,
          "decoder_belief_top1_on_support": 0.15234375,
          "mean_decoder_belief_support_size": 3.8125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "A",
          "k": 16,
          "n": 256,
          "query_target_mass": 0.37379252910614014,
          "query_target_nll": 2.229926586151123,
          "query_top1_on_support": 0.42578125,
          "mean_query_support_size": 3.5078125,
          "probe_belief_target_mass": 0.1040026992559433,
          "probe_belief_target_nll": 4.089742660522461,
          "probe_belief_top1_on_support": 0.15625,
          "mean_belief_support_size": 3.8125,
          "decoder_belief_target_mass": 0.09536483138799667,
          "decoder_belief_target_nll": 4.1067352294921875,
          "decoder_belief_top1_on_support": 0.15234375,
          "mean_decoder_belief_support_size": 3.8125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "A",
          "k": 24,
          "n": 256,
          "query_target_mass": 0.37379252910614014,
          "query_target_nll": 2.229926586151123,
          "query_top1_on_support": 0.42578125,
          "mean_query_support_size": 3.5078125,
          "probe_belief_target_mass": 0.1040026992559433,
          "probe_belief_target_nll": 4.089742660522461,
          "probe_belief_top1_on_support": 0.15625,
          "mean_belief_support_size": 3.8125,
          "decoder_belief_target_mass": 0.09536483138799667,
          "decoder_belief_target_nll": 4.1067352294921875,
          "decoder_belief_top1_on_support": 0.15234375,
          "mean_decoder_belief_support_size": 3.8125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.11945437639951706,
          "query_target_nll": 3.4536824226379395,
          "query_top1_on_support": 0.09375,
          "mean_query_support_size": 3.7265625,
          "probe_belief_target_mass": 0.005461900494992733,
          "probe_belief_target_nll": 21.24953269958496,
          "probe_belief_top1_on_support": 0.015625,
          "mean_belief_support_size": 4.64453125,
          "decoder_belief_target_mass": 0.005614397116005421,
          "decoder_belief_target_nll": 25.432769775390625,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 4.64453125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.12414209544658661,
          "query_target_nll": 6.964838981628418,
          "query_top1_on_support": 0.12890625,
          "mean_query_support_size": 3.7265625,
          "probe_belief_target_mass": 0.005325024481862783,
          "probe_belief_target_nll": 24.72874641418457,
          "probe_belief_top1_on_support": 0.00390625,
          "mean_belief_support_size": 4.64453125,
          "decoder_belief_target_mass": 0.005316514056175947,
          "decoder_belief_target_nll": 28.374549865722656,
          "decoder_belief_top1_on_support": 0.0078125,
          "mean_decoder_belief_support_size": 4.64453125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.10852386802434921,
          "query_target_nll": 8.08963394165039,
          "query_top1_on_support": 0.11328125,
          "mean_query_support_size": 3.7265625,
          "probe_belief_target_mass": 0.009666250087320805,
          "probe_belief_target_nll": 23.333702087402344,
          "probe_belief_top1_on_support": 0.0078125,
          "mean_belief_support_size": 4.64453125,
          "decoder_belief_target_mass": 0.009396908804774284,
          "decoder_belief_target_nll": 26.830379486083984,
          "decoder_belief_top1_on_support": 0.0078125,
          "mean_decoder_belief_support_size": 4.64453125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "B",
          "k": 4,
          "n": 256,
          "query_target_mass": 0.1373930126428604,
          "query_target_nll": 8.472413063049316,
          "query_top1_on_support": 0.12890625,
          "mean_query_support_size": 3.7265625,
          "probe_belief_target_mass": 0.013189887627959251,
          "probe_belief_target_nll": 19.91069793701172,
          "probe_belief_top1_on_support": 0.01953125,
          "mean_belief_support_size": 4.64453125,
          "decoder_belief_target_mass": 0.010777302086353302,
          "decoder_belief_target_nll": 22.075424194335938,
          "decoder_belief_top1_on_support": 0.0078125,
          "mean_decoder_belief_support_size": 4.64453125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "B",
          "k": 8,
          "n": 256,
          "query_target_mass": 0.18769514560699463,
          "query_target_nll": 7.027449607849121,
          "query_top1_on_support": 0.1796875,
          "mean_query_support_size": 3.7265625,
          "probe_belief_target_mass": 0.020231515169143677,
          "probe_belief_target_nll": 15.140985488891602,
          "probe_belief_top1_on_support": 0.02734375,
          "mean_belief_support_size": 4.64453125,
          "decoder_belief_target_mass": 0.02119268849492073,
          "decoder_belief_target_nll": 16.409854888916016,
          "decoder_belief_top1_on_support": 0.046875,
          "mean_decoder_belief_support_size": 4.64453125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.41324055194854736,
          "query_target_nll": 2.083564281463623,
          "query_top1_on_support": 0.5234375,
          "mean_query_support_size": 3.7265625,
          "probe_belief_target_mass": 0.12079696357250214,
          "probe_belief_target_nll": 3.94431209564209,
          "probe_belief_top1_on_support": 0.1875,
          "mean_belief_support_size": 4.64453125,
          "decoder_belief_target_mass": 0.11509508639574051,
          "decoder_belief_target_nll": 3.908050537109375,
          "decoder_belief_top1_on_support": 0.17578125,
          "mean_decoder_belief_support_size": 4.64453125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "B",
          "k": 16,
          "n": 256,
          "query_target_mass": 0.41324055194854736,
          "query_target_nll": 2.083564281463623,
          "query_top1_on_support": 0.5234375,
          "mean_query_support_size": 3.7265625,
          "probe_belief_target_mass": 0.12079696357250214,
          "probe_belief_target_nll": 3.94431209564209,
          "probe_belief_top1_on_support": 0.1875,
          "mean_belief_support_size": 4.64453125,
          "decoder_belief_target_mass": 0.11509508639574051,
          "decoder_belief_target_nll": 3.908050537109375,
          "decoder_belief_top1_on_support": 0.17578125,
          "mean_decoder_belief_support_size": 4.64453125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "B",
          "k": 24,
          "n": 256,
          "query_target_mass": 0.41324055194854736,
          "query_target_nll": 2.083564281463623,
          "query_top1_on_support": 0.5234375,
          "mean_query_support_size": 3.7265625,
          "probe_belief_target_mass": 0.12079696357250214,
          "probe_belief_target_nll": 3.94431209564209,
          "probe_belief_top1_on_support": 0.1875,
          "mean_belief_support_size": 4.64453125,
          "decoder_belief_target_mass": 0.11509508639574051,
          "decoder_belief_target_nll": 3.908050537109375,
          "decoder_belief_top1_on_support": 0.17578125,
          "mean_decoder_belief_support_size": 4.64453125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "A_PLUS_B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.1286548674106598,
          "query_target_nll": 3.4442973136901855,
          "query_top1_on_support": 0.13671875,
          "mean_query_support_size": 4.0,
          "probe_belief_target_mass": 0.0036242958158254623,
          "probe_belief_target_nll": 21.522563934326172,
          "probe_belief_top1_on_support": 0.0078125,
          "mean_belief_support_size": 4.4140625,
          "decoder_belief_target_mass": 0.003626459976658225,
          "decoder_belief_target_nll": 26.15912628173828,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 4.4140625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "A_PLUS_B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.12161271274089813,
          "query_target_nll": 4.473797798156738,
          "query_top1_on_support": 0.08984375,
          "mean_query_support_size": 4.0,
          "probe_belief_target_mass": 0.006020868197083473,
          "probe_belief_target_nll": 26.424976348876953,
          "probe_belief_top1_on_support": 0.00390625,
          "mean_belief_support_size": 4.4140625,
          "decoder_belief_target_mass": 0.006791967898607254,
          "decoder_belief_target_nll": 30.62468719482422,
          "decoder_belief_top1_on_support": 0.015625,
          "mean_decoder_belief_support_size": 4.4140625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "A_PLUS_B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.13399001955986023,
          "query_target_nll": 4.709643363952637,
          "query_top1_on_support": 0.140625,
          "mean_query_support_size": 4.0,
          "probe_belief_target_mass": 0.007279363926500082,
          "probe_belief_target_nll": 24.137855529785156,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 4.4140625,
          "decoder_belief_target_mass": 0.009777356870472431,
          "decoder_belief_target_nll": 27.463516235351562,
          "decoder_belief_top1_on_support": 0.01953125,
          "mean_decoder_belief_support_size": 4.4140625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "A_PLUS_B",
          "k": 4,
          "n": 256,
          "query_target_mass": 0.1283898651599884,
          "query_target_nll": 4.951449394226074,
          "query_top1_on_support": 0.12109375,
          "mean_query_support_size": 4.0,
          "probe_belief_target_mass": 0.006699924357235432,
          "probe_belief_target_nll": 20.692733764648438,
          "probe_belief_top1_on_support": 0.01171875,
          "mean_belief_support_size": 4.4140625,
          "decoder_belief_target_mass": 0.007758313324302435,
          "decoder_belief_target_nll": 23.081483840942383,
          "decoder_belief_top1_on_support": 0.0078125,
          "mean_decoder_belief_support_size": 4.4140625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "A_PLUS_B",
          "k": 8,
          "n": 256,
          "query_target_mass": 0.12810207903385162,
          "query_target_nll": 5.448395729064941,
          "query_top1_on_support": 0.12109375,
          "mean_query_support_size": 4.0,
          "probe_belief_target_mass": 0.022791430354118347,
          "probe_belief_target_nll": 16.30373764038086,
          "probe_belief_top1_on_support": 0.0390625,
          "mean_belief_support_size": 4.4140625,
          "decoder_belief_target_mass": 0.020109497010707855,
          "decoder_belief_target_nll": 17.70688819885254,
          "decoder_belief_top1_on_support": 0.02734375,
          "mean_decoder_belief_support_size": 4.4140625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "A_PLUS_B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.2978871464729309,
          "query_target_nll": 2.586127758026123,
          "query_top1_on_support": 0.359375,
          "mean_query_support_size": 4.0,
          "probe_belief_target_mass": 0.12410593032836914,
          "probe_belief_target_nll": 3.9624483585357666,
          "probe_belief_top1_on_support": 0.16015625,
          "mean_belief_support_size": 4.4140625,
          "decoder_belief_target_mass": 0.1240033507347107,
          "decoder_belief_target_nll": 3.983875274658203,
          "decoder_belief_top1_on_support": 0.19140625,
          "mean_decoder_belief_support_size": 4.4140625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "A_PLUS_B",
          "k": 16,
          "n": 256,
          "query_target_mass": 0.2978871464729309,
          "query_target_nll": 2.586127758026123,
          "query_top1_on_support": 0.359375,
          "mean_query_support_size": 4.0,
          "probe_belief_target_mass": 0.12410593032836914,
          "probe_belief_target_nll": 3.9624483585357666,
          "probe_belief_top1_on_support": 0.16015625,
          "mean_belief_support_size": 4.4140625,
          "decoder_belief_target_mass": 0.1240033507347107,
          "decoder_belief_target_nll": 3.983875274658203,
          "decoder_belief_top1_on_support": 0.19140625,
          "mean_decoder_belief_support_size": 4.4140625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "A_PLUS_B",
          "k": 24,
          "n": 256,
          "query_target_mass": 0.2978871464729309,
          "query_target_nll": 2.586127758026123,
          "query_top1_on_support": 0.359375,
          "mean_query_support_size": 4.0,
          "probe_belief_target_mass": 0.12410593032836914,
          "probe_belief_target_nll": 3.9624483585357666,
          "probe_belief_top1_on_support": 0.16015625,
          "mean_belief_support_size": 4.4140625,
          "decoder_belief_target_mass": 0.1240033507347107,
          "decoder_belief_target_nll": 3.983875274658203,
          "decoder_belief_top1_on_support": 0.19140625,
          "mean_decoder_belief_support_size": 4.4140625
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "A_MINUS_B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.15623024106025696,
          "query_target_nll": 18.600444793701172,
          "query_top1_on_support": 0.15625,
          "mean_query_support_size": 4.078125,
          "probe_belief_target_mass": 0.005357361398637295,
          "probe_belief_target_nll": 20.703350067138672,
          "probe_belief_top1_on_support": 0.0078125,
          "mean_belief_support_size": 4.32421875,
          "decoder_belief_target_mass": 0.005061738193035126,
          "decoder_belief_target_nll": 24.982328414916992,
          "decoder_belief_top1_on_support": 0.01171875,
          "mean_decoder_belief_support_size": 4.32421875
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "A_MINUS_B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.15257859230041504,
          "query_target_nll": 18.876205444335938,
          "query_top1_on_support": 0.1484375,
          "mean_query_support_size": 4.078125,
          "probe_belief_target_mass": 0.007461001630872488,
          "probe_belief_target_nll": 24.0849609375,
          "probe_belief_top1_on_support": 0.01953125,
          "mean_belief_support_size": 4.32421875,
          "decoder_belief_target_mass": 0.006707549095153809,
          "decoder_belief_target_nll": 27.941072463989258,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 4.32421875
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "A_MINUS_B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.13537821173667908,
          "query_target_nll": 15.46552848815918,
          "query_top1_on_support": 0.140625,
          "mean_query_support_size": 4.078125,
          "probe_belief_target_mass": 0.004985133185982704,
          "probe_belief_target_nll": 22.803070068359375,
          "probe_belief_top1_on_support": 0.00390625,
          "mean_belief_support_size": 4.32421875,
          "decoder_belief_target_mass": 0.004455901682376862,
          "decoder_belief_target_nll": 25.753612518310547,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 4.32421875
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "A_MINUS_B",
          "k": 4,
          "n": 256,
          "query_target_mass": 0.12535226345062256,
          "query_target_nll": 10.921182632446289,
          "query_top1_on_support": 0.1328125,
          "mean_query_support_size": 4.078125,
          "probe_belief_target_mass": 0.004252538550645113,
          "probe_belief_target_nll": 20.081645965576172,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 4.32421875,
          "decoder_belief_target_mass": 0.005639819428324699,
          "decoder_belief_target_nll": 22.401596069335938,
          "decoder_belief_top1_on_support": 0.01171875,
          "mean_decoder_belief_support_size": 4.32421875
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "A_MINUS_B",
          "k": 8,
          "n": 256,
          "query_target_mass": 0.15081295371055603,
          "query_target_nll": 7.453424453735352,
          "query_top1_on_support": 0.15625,
          "mean_query_support_size": 4.078125,
          "probe_belief_target_mass": 0.01800757832825184,
          "probe_belief_target_nll": 15.254117012023926,
          "probe_belief_top1_on_support": 0.0234375,
          "mean_belief_support_size": 4.32421875,
          "decoder_belief_target_mass": 0.016606628894805908,
          "decoder_belief_target_nll": 16.9791259765625,
          "decoder_belief_top1_on_support": 0.015625,
          "mean_decoder_belief_support_size": 4.32421875
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "A_MINUS_B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.3484933078289032,
          "query_target_nll": 2.4559500217437744,
          "query_top1_on_support": 0.39453125,
          "mean_query_support_size": 4.078125,
          "probe_belief_target_mass": 0.12170298397541046,
          "probe_belief_target_nll": 3.9552969932556152,
          "probe_belief_top1_on_support": 0.171875,
          "mean_belief_support_size": 4.32421875,
          "decoder_belief_target_mass": 0.11627607047557831,
          "decoder_belief_target_nll": 3.9489827156066895,
          "decoder_belief_top1_on_support": 0.17578125,
          "mean_decoder_belief_support_size": 4.32421875
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "A_MINUS_B",
          "k": 16,
          "n": 256,
          "query_target_mass": 0.3484933078289032,
          "query_target_nll": 2.4559500217437744,
          "query_top1_on_support": 0.39453125,
          "mean_query_support_size": 4.078125,
          "probe_belief_target_mass": 0.12170298397541046,
          "probe_belief_target_nll": 3.9552969932556152,
          "probe_belief_top1_on_support": 0.171875,
          "mean_belief_support_size": 4.32421875,
          "decoder_belief_target_mass": 0.11627607047557831,
          "decoder_belief_target_nll": 3.9489827156066895,
          "decoder_belief_top1_on_support": 0.17578125,
          "mean_decoder_belief_support_size": 4.32421875
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 12,
          "query_type": "A_MINUS_B",
          "k": 24,
          "n": 256,
          "query_target_mass": 0.3484933078289032,
          "query_target_nll": 2.4559500217437744,
          "query_top1_on_support": 0.39453125,
          "mean_query_support_size": 4.078125,
          "probe_belief_target_mass": 0.12170298397541046,
          "probe_belief_target_nll": 3.9552969932556152,
          "probe_belief_top1_on_support": 0.171875,
          "mean_belief_support_size": 4.32421875,
          "decoder_belief_target_mass": 0.11627607047557831,
          "decoder_belief_target_nll": 3.9489827156066895,
          "decoder_belief_top1_on_support": 0.17578125,
          "mean_decoder_belief_support_size": 4.32421875
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 16,
          "query_type": "A",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.07391535490751266,
          "query_target_nll": 3.4457290172576904,
          "query_top1_on_support": 0.0859375,
          "mean_query_support_size": 2.296875,
          "probe_belief_target_mass": 0.003135921899229288,
          "probe_belief_target_nll": 21.19908905029297,
          "probe_belief_top1_on_support": 0.0078125,
          "mean_belief_support_size": 2.58984375,
          "decoder_belief_target_mass": 0.0030500830616801977,
          "decoder_belief_target_nll": 25.86792755126953,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 2.58984375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 16,
          "query_type": "A",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.07713442295789719,
          "query_target_nll": 6.884642601013184,
          "query_top1_on_support": 0.0546875,
          "mean_query_support_size": 2.296875,
          "probe_belief_target_mass": 0.004460257478058338,
          "probe_belief_target_nll": 25.90934944152832,
          "probe_belief_top1_on_support": 0.0078125,
          "mean_belief_support_size": 2.58984375,
          "decoder_belief_target_mass": 0.003960292786359787,
          "decoder_belief_target_nll": 30.352188110351562,
          "decoder_belief_top1_on_support": 0.01171875,
          "mean_decoder_belief_support_size": 2.58984375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 16,
          "query_type": "A",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.07211387157440186,
          "query_target_nll": 7.988980770111084,
          "query_top1_on_support": 0.0859375,
          "mean_query_support_size": 2.296875,
          "probe_belief_target_mass": 0.002064925152808428,
          "probe_belief_target_nll": 22.980789184570312,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 2.58984375,
          "decoder_belief_target_mass": 0.002317200880497694,
          "decoder_belief_target_nll": 26.426895141601562,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 2.58984375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 16,
          "query_type": "A",
          "k": 4,
          "n": 256,
          "query_target_mass": 0.0729062408208847,
          "query_target_nll": 8.534212112426758,
          "query_top1_on_support": 0.08203125,
          "mean_query_support_size": 2.296875,
          "probe_belief_target_mass": 0.0018834545044228435,
          "probe_belief_target_nll": 19.988311767578125,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 2.58984375,
          "decoder_belief_target_mass": 0.0020824503153562546,
          "decoder_belief_target_nll": 22.281646728515625,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 2.58984375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 16,
          "query_type": "A",
          "k": 8,
          "n": 256,
          "query_target_mass": 0.08982579410076141,
          "query_target_nll": 8.346597671508789,
          "query_top1_on_support": 0.09765625,
          "mean_query_support_size": 2.296875,
          "probe_belief_target_mass": 0.0029836331959813833,
          "probe_belief_target_nll": 17.333436965942383,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 2.58984375,
          "decoder_belief_target_mass": 0.0028044297359883785,
          "decoder_belief_target_nll": 19.2227840423584,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 2.58984375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 16,
          "query_type": "A",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.12690018117427826,
          "query_target_nll": 7.944700717926025,
          "query_top1_on_support": 0.15625,
          "mean_query_support_size": 2.296875,
          "probe_belief_target_mass": 0.0077566723339259624,
          "probe_belief_target_nll": 15.202644348144531,
          "probe_belief_top1_on_support": 0.01953125,
          "mean_belief_support_size": 2.58984375,
          "decoder_belief_target_mass": 0.007489344105124474,
          "decoder_belief_target_nll": 16.467092514038086,
          "decoder_belief_top1_on_support": 0.0078125,
          "mean_decoder_belief_support_size": 2.58984375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 16,
          "query_type": "A",
          "k": 16,
          "n": 256,
          "query_target_mass": 0.3113999366760254,
          "query_target_nll": 2.163252830505371,
          "query_top1_on_support": 0.3828125,
          "mean_query_support_size": 2.296875,
          "probe_belief_target_mass": 0.08000831305980682,
          "probe_belief_target_nll": 4.0142693519592285,
          "probe_belief_top1_on_support": 0.13671875,
          "mean_belief_support_size": 2.58984375,
          "decoder_belief_target_mass": 0.07334187626838684,
          "decoder_belief_target_nll": 4.009781360626221,
          "decoder_belief_top1_on_support": 0.11328125,
          "mean_decoder_belief_support_size": 2.58984375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 16,
          "query_type": "A",
          "k": 24,
          "n": 256,
          "query_target_mass": 0.3113999366760254,
          "query_target_nll": 2.163252830505371,
          "query_top1_on_support": 0.3828125,
          "mean_query_support_size": 2.296875,
          "probe_belief_target_mass": 0.08000831305980682,
          "probe_belief_target_nll": 4.0142693519592285,
          "probe_belief_top1_on_support": 0.13671875,
          "mean_belief_support_size": 2.58984375,
          "decoder_belief_target_mass": 0.07334187626838684,
          "decoder_belief_target_nll": 4.009781360626221,
          "decoder_belief_top1_on_support": 0.11328125,
          "mean_decoder_belief_support_size": 2.58984375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 16,
          "query_type": "B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.07847846299409866,
          "query_target_nll": 3.447706460952759,
          "query_top1_on_support": 0.0546875,
          "mean_query_support_size": 2.44140625,
          "probe_belief_target_mass": 0.003271300345659256,
          "probe_belief_target_nll": 21.738178253173828,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 2.828125,
          "decoder_belief_target_mass": 0.003362104529514909,
          "decoder_belief_target_nll": 25.962533950805664,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 2.828125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 16,
          "query_type": "B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.07741744071245193,
          "query_target_nll": 7.133663654327393,
          "query_top1_on_support": 0.0859375,
          "mean_query_support_size": 2.44140625,
          "probe_belief_target_mass": 0.003556810785084963,
          "probe_belief_target_nll": 24.906503677368164,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 2.828125,
          "decoder_belief_target_mass": 0.004148405510932207,
          "decoder_belief_target_nll": 28.75959014892578,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 2.828125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 16,
          "query_type": "B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.07624329626560211,
          "query_target_nll": 8.175710678100586,
          "query_top1_on_support": 0.078125,
          "mean_query_support_size": 2.44140625,
          "probe_belief_target_mass": 0.0032038562931120396,
          "probe_belief_target_nll": 23.03062629699707,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 2.828125,
          "decoder_belief_target_mass": 0.004444648511707783,
          "decoder_belief_target_nll": 26.193517684936523,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 2.828125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 16,
          "query_type": "B",
          "k": 4,
          "n": 256,
          "query_target_mass": 0.08014118671417236,
          "query_target_nll": 8.288249015808105,
          "query_top1_on_support": 0.078125,
          "mean_query_support_size": 2.44140625,
          "probe_belief_target_mass": 0.0026019439101219177,
          "probe_belief_target_nll": 19.59066390991211,
          "probe_belief_top1_on_support": 0.0078125,
          "mean_belief_support_size": 2.828125,
          "decoder_belief_target_mass": 0.0021855635568499565,
          "decoder_belief_target_nll": 22.156518936157227,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 2.828125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 16,
          "query_type": "B",
          "k": 8,
          "n": 256,
          "query_target_mass": 0.0860026627779007,
          "query_target_nll": 8.681936264038086,
          "query_top1_on_support": 0.09765625,
          "mean_query_support_size": 2.44140625,
          "probe_belief_target_mass": 0.00420977920293808,
          "probe_belief_target_nll": 17.03221893310547,
          "probe_belief_top1_on_support": 0.01171875,
          "mean_belief_support_size": 2.828125,
          "decoder_belief_target_mass": 0.004138953983783722,
          "decoder_belief_target_nll": 18.77341079711914,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 2.828125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 16,
          "query_type": "B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.1265241801738739,
          "query_target_nll": 7.586702346801758,
          "query_top1_on_support": 0.1484375,
          "mean_query_support_size": 2.44140625,
          "probe_belief_target_mass": 0.008270961232483387,
          "probe_belief_target_nll": 15.346421241760254,
          "probe_belief_top1_on_support": 0.0078125,
          "mean_belief_support_size": 2.828125,
          "decoder_belief_target_mass": 0.007396100088953972,
          "decoder_belief_target_nll": 16.606998443603516,
          "decoder_belief_top1_on_support": 0.01171875,
          "mean_decoder_belief_support_size": 2.828125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 16,
          "query_type": "B",
          "k": 16,
          "n": 256,
          "query_target_mass": 0.3083140552043915,
          "query_target_nll": 2.1389336585998535,
          "query_top1_on_support": 0.40625,
          "mean_query_support_size": 2.44140625,
          "probe_belief_target_mass": 0.07336777448654175,
          "probe_belief_target_nll": 4.038908004760742,
          "probe_belief_top1_on_support": 0.10546875,
          "mean_belief_support_size": 2.828125,
          "decoder_belief_target_mass": 0.07468593120574951,
          "decoder_belief_target_nll": 4.0132222175598145,
          "decoder_belief_top1_on_support": 0.125,
          "mean_decoder_belief_support_size": 2.828125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 16,
          "query_type": "B",
          "k": 24,
          "n": 256,
          "query_target_mass": 0.3083140552043915,
          "query_target_nll": 2.1389336585998535,
          "query_top1_on_support": 0.40625,
          "mean_query_support_size": 2.44140625,
          "probe_belief_target_mass": 0.07336777448654175,
          "probe_belief_target_nll": 4.038908004760742,
          "probe_belief_top1_on_support": 0.10546875,
          "mean_belief_support_size": 2.828125,
          "decoder_belief_target_mass": 0.07468593120574951,
          "decoder_belief_target_nll": 4.0132222175598145,
          "decoder_belief_top1_on_support": 0.125,
          "mean_decoder_belief_support_size": 2.828125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 16,
          "query_type": "A_PLUS_B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.07516573369503021,
          "query_target_nll": 3.4352829456329346,
          "query_top1_on_support": 0.08984375,
          "mean_query_support_size": 2.31640625,
          "probe_belief_target_mass": 0.0022279047407209873,
          "probe_belief_target_nll": 21.58127784729004,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 2.46484375,
          "decoder_belief_target_mass": 0.0020222975872457027,
          "decoder_belief_target_nll": 26.223827362060547,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 2.46484375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 16,
          "query_type": "A_PLUS_B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.07038529217243195,
          "query_target_nll": 4.555548191070557,
          "query_top1_on_support": 0.0703125,
          "mean_query_support_size": 2.31640625,
          "probe_belief_target_mass": 0.0033472059294581413,
          "probe_belief_target_nll": 25.364112854003906,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 2.46484375,
          "decoder_belief_target_mass": 0.00380672886967659,
          "decoder_belief_target_nll": 29.422351837158203,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 2.46484375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 16,
          "query_type": "A_PLUS_B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.06486523151397705,
          "query_target_nll": 4.8046464920043945,
          "query_top1_on_support": 0.0546875,
          "mean_query_support_size": 2.31640625,
          "probe_belief_target_mass": 0.002668073633685708,
          "probe_belief_target_nll": 24.27609634399414,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 2.46484375,
          "decoder_belief_target_mass": 0.002655287506058812,
          "decoder_belief_target_nll": 27.770877838134766,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 2.46484375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 16,
          "query_type": "A_PLUS_B",
          "k": 4,
          "n": 256,
          "query_target_mass": 0.07827344536781311,
          "query_target_nll": 4.984312057495117,
          "query_top1_on_support": 0.0859375,
          "mean_query_support_size": 2.31640625,
          "probe_belief_target_mass": 0.003221767023205757,
          "probe_belief_target_nll": 21.101572036743164,
          "probe_belief_top1_on_support": 0.0078125,
          "mean_belief_support_size": 2.46484375,
          "decoder_belief_target_mass": 0.002215176122263074,
          "decoder_belief_target_nll": 23.65903091430664,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 2.46484375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 16,
          "query_type": "A_PLUS_B",
          "k": 8,
          "n": 256,
          "query_target_mass": 0.08187062293291092,
          "query_target_nll": 5.728529453277588,
          "query_top1_on_support": 0.0859375,
          "mean_query_support_size": 2.31640625,
          "probe_belief_target_mass": 0.003696782747283578,
          "probe_belief_target_nll": 17.303815841674805,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 2.46484375,
          "decoder_belief_target_mass": 0.003712737001478672,
          "decoder_belief_target_nll": 19.15384292602539,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 2.46484375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 16,
          "query_type": "A_PLUS_B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.07953639328479767,
          "query_target_nll": 6.025744438171387,
          "query_top1_on_support": 0.08984375,
          "mean_query_support_size": 2.31640625,
          "probe_belief_target_mass": 0.009525220841169357,
          "probe_belief_target_nll": 15.479944229125977,
          "probe_belief_top1_on_support": 0.01953125,
          "mean_belief_support_size": 2.46484375,
          "decoder_belief_target_mass": 0.009050844237208366,
          "decoder_belief_target_nll": 17.185285568237305,
          "decoder_belief_top1_on_support": 0.01171875,
          "mean_decoder_belief_support_size": 2.46484375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 16,
          "query_type": "A_PLUS_B",
          "k": 16,
          "n": 256,
          "query_target_mass": 0.2137909084558487,
          "query_target_nll": 2.5870933532714844,
          "query_top1_on_support": 0.24609375,
          "mean_query_support_size": 2.31640625,
          "probe_belief_target_mass": 0.08401121944189072,
          "probe_belief_target_nll": 4.076594352722168,
          "probe_belief_top1_on_support": 0.13671875,
          "mean_belief_support_size": 2.46484375,
          "decoder_belief_target_mass": 0.07899656891822815,
          "decoder_belief_target_nll": 3.9972455501556396,
          "decoder_belief_top1_on_support": 0.1015625,
          "mean_decoder_belief_support_size": 2.46484375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 16,
          "query_type": "A_PLUS_B",
          "k": 24,
          "n": 256,
          "query_target_mass": 0.2137909084558487,
          "query_target_nll": 2.5870933532714844,
          "query_top1_on_support": 0.24609375,
          "mean_query_support_size": 2.31640625,
          "probe_belief_target_mass": 0.08401121944189072,
          "probe_belief_target_nll": 4.076594352722168,
          "probe_belief_top1_on_support": 0.13671875,
          "mean_belief_support_size": 2.46484375,
          "decoder_belief_target_mass": 0.07899656891822815,
          "decoder_belief_target_nll": 3.9972455501556396,
          "decoder_belief_top1_on_support": 0.1015625,
          "mean_decoder_belief_support_size": 2.46484375
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 16,
          "query_type": "A_MINUS_B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.0976245105266571,
          "query_target_nll": 18.953824996948242,
          "query_top1_on_support": 0.09765625,
          "mean_query_support_size": 2.5546875,
          "probe_belief_target_mass": 0.0031704329885542393,
          "probe_belief_target_nll": 21.317867279052734,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 2.69921875,
          "decoder_belief_target_mass": 0.0031837269198149443,
          "decoder_belief_target_nll": 26.062448501586914,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 2.69921875
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 16,
          "query_type": "A_MINUS_B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.07475785166025162,
          "query_target_nll": 20.975862503051758,
          "query_top1_on_support": 0.07421875,
          "mean_query_support_size": 2.5546875,
          "probe_belief_target_mass": 0.0024690190330147743,
          "probe_belief_target_nll": 26.988656997680664,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 2.69921875,
          "decoder_belief_target_mass": 0.0026284505147486925,
          "decoder_belief_target_nll": 31.311927795410156,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 2.69921875
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 16,
          "query_type": "A_MINUS_B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.08856484293937683,
          "query_target_nll": 15.45101547241211,
          "query_top1_on_support": 0.0859375,
          "mean_query_support_size": 2.5546875,
          "probe_belief_target_mass": 0.002539386972784996,
          "probe_belief_target_nll": 24.336158752441406,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 2.69921875,
          "decoder_belief_target_mass": 0.0027859187684953213,
          "decoder_belief_target_nll": 28.063674926757812,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 2.69921875
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 16,
          "query_type": "A_MINUS_B",
          "k": 4,
          "n": 256,
          "query_target_mass": 0.06959381699562073,
          "query_target_nll": 11.705877304077148,
          "query_top1_on_support": 0.0703125,
          "mean_query_support_size": 2.5546875,
          "probe_belief_target_mass": 0.00122431805357337,
          "probe_belief_target_nll": 21.811771392822266,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 2.69921875,
          "decoder_belief_target_mass": 0.000909874914214015,
          "decoder_belief_target_nll": 24.22890853881836,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 2.69921875
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 16,
          "query_type": "A_MINUS_B",
          "k": 8,
          "n": 256,
          "query_target_mass": 0.08048954606056213,
          "query_target_nll": 8.604353904724121,
          "query_top1_on_support": 0.06640625,
          "mean_query_support_size": 2.5546875,
          "probe_belief_target_mass": 0.003854743205010891,
          "probe_belief_target_nll": 17.93354034423828,
          "probe_belief_top1_on_support": 0.01171875,
          "mean_belief_support_size": 2.69921875,
          "decoder_belief_target_mass": 0.0036020888946950436,
          "decoder_belief_target_nll": 19.75257110595703,
          "decoder_belief_top1_on_support": 0.0078125,
          "mean_decoder_belief_support_size": 2.69921875
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 16,
          "query_type": "A_MINUS_B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.08299977332353592,
          "query_target_nll": 7.4795403480529785,
          "query_top1_on_support": 0.08203125,
          "mean_query_support_size": 2.5546875,
          "probe_belief_target_mass": 0.005978097207844257,
          "probe_belief_target_nll": 15.404470443725586,
          "probe_belief_top1_on_support": 0.00390625,
          "mean_belief_support_size": 2.69921875,
          "decoder_belief_target_mass": 0.006487196311354637,
          "decoder_belief_target_nll": 16.872085571289062,
          "decoder_belief_top1_on_support": 0.0078125,
          "mean_decoder_belief_support_size": 2.69921875
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 16,
          "query_type": "A_MINUS_B",
          "k": 16,
          "n": 256,
          "query_target_mass": 0.258537232875824,
          "query_target_nll": 2.4821534156799316,
          "query_top1_on_support": 0.27734375,
          "mean_query_support_size": 2.5546875,
          "probe_belief_target_mass": 0.07685523480176926,
          "probe_belief_target_nll": 4.145617961883545,
          "probe_belief_top1_on_support": 0.14453125,
          "mean_belief_support_size": 2.69921875,
          "decoder_belief_target_mass": 0.07018118351697922,
          "decoder_belief_target_nll": 4.120429992675781,
          "decoder_belief_top1_on_support": 0.11328125,
          "mean_decoder_belief_support_size": 2.69921875
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 16,
          "query_type": "A_MINUS_B",
          "k": 24,
          "n": 256,
          "query_target_mass": 0.258537232875824,
          "query_target_nll": 2.4821534156799316,
          "query_top1_on_support": 0.27734375,
          "mean_query_support_size": 2.5546875,
          "probe_belief_target_mass": 0.07685523480176926,
          "probe_belief_target_nll": 4.145617961883545,
          "probe_belief_top1_on_support": 0.14453125,
          "mean_belief_support_size": 2.69921875,
          "decoder_belief_target_mass": 0.07018118351697922,
          "decoder_belief_target_nll": 4.120429992675781,
          "decoder_belief_top1_on_support": 0.11328125,
          "mean_decoder_belief_support_size": 2.69921875
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 24,
          "query_type": "A",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.0500708669424057,
          "query_target_nll": 3.443030834197998,
          "query_top1_on_support": 0.0390625,
          "mean_query_support_size": 1.56640625,
          "probe_belief_target_mass": 0.0014333580620586872,
          "probe_belief_target_nll": 21.39141273498535,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 1.64453125,
          "decoder_belief_target_mass": 0.0015374207869172096,
          "decoder_belief_target_nll": 25.672496795654297,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 1.64453125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 24,
          "query_type": "A",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.048929858952760696,
          "query_target_nll": 8.315622329711914,
          "query_top1_on_support": 0.04296875,
          "mean_query_support_size": 1.56640625,
          "probe_belief_target_mass": 0.0012763486010953784,
          "probe_belief_target_nll": 25.852506637573242,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 1.64453125,
          "decoder_belief_target_mass": 0.001409392338246107,
          "decoder_belief_target_nll": 29.785791397094727,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 1.64453125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 24,
          "query_type": "A",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.0480174794793129,
          "query_target_nll": 8.400054931640625,
          "query_top1_on_support": 0.05078125,
          "mean_query_support_size": 1.56640625,
          "probe_belief_target_mass": 0.0015382003039121628,
          "probe_belief_target_nll": 24.19517707824707,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 1.64453125,
          "decoder_belief_target_mass": 0.0014902468537911773,
          "decoder_belief_target_nll": 27.546905517578125,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 1.64453125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 24,
          "query_type": "A",
          "k": 4,
          "n": 256,
          "query_target_mass": 0.0542115643620491,
          "query_target_nll": 8.624311447143555,
          "query_top1_on_support": 0.0625,
          "mean_query_support_size": 1.56640625,
          "probe_belief_target_mass": 0.0026488034054636955,
          "probe_belief_target_nll": 19.95915985107422,
          "probe_belief_top1_on_support": 0.00390625,
          "mean_belief_support_size": 1.64453125,
          "decoder_belief_target_mass": 0.002854306250810623,
          "decoder_belief_target_nll": 22.03384017944336,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 1.64453125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 24,
          "query_type": "A",
          "k": 8,
          "n": 256,
          "query_target_mass": 0.05059289187192917,
          "query_target_nll": 9.398111343383789,
          "query_top1_on_support": 0.0703125,
          "mean_query_support_size": 1.56640625,
          "probe_belief_target_mass": 0.003162094159051776,
          "probe_belief_target_nll": 17.30068016052246,
          "probe_belief_top1_on_support": 0.00390625,
          "mean_belief_support_size": 1.64453125,
          "decoder_belief_target_mass": 0.0025477129966020584,
          "decoder_belief_target_nll": 19.119945526123047,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 1.64453125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 24,
          "query_type": "A",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.041648924350738525,
          "query_target_nll": 9.288469314575195,
          "query_top1_on_support": 0.03515625,
          "mean_query_support_size": 1.56640625,
          "probe_belief_target_mass": 0.0032740728929638863,
          "probe_belief_target_nll": 16.475181579589844,
          "probe_belief_top1_on_support": 0.00390625,
          "mean_belief_support_size": 1.64453125,
          "decoder_belief_target_mass": 0.0025211956817656755,
          "decoder_belief_target_nll": 18.19573402404785,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 1.64453125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 24,
          "query_type": "A",
          "k": 16,
          "n": 256,
          "query_target_mass": 0.05866536870598793,
          "query_target_nll": 9.180864334106445,
          "query_top1_on_support": 0.0625,
          "mean_query_support_size": 1.56640625,
          "probe_belief_target_mass": 0.0017496634973213077,
          "probe_belief_target_nll": 16.318572998046875,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 1.64453125,
          "decoder_belief_target_mass": 0.0016693209763616323,
          "decoder_belief_target_nll": 17.833721160888672,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 1.64453125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 24,
          "query_type": "A",
          "k": 24,
          "n": 256,
          "query_target_mass": 0.22858992218971252,
          "query_target_nll": 2.157907247543335,
          "query_top1_on_support": 0.2578125,
          "mean_query_support_size": 1.56640625,
          "probe_belief_target_mass": 0.05037049576640129,
          "probe_belief_target_nll": 4.149055004119873,
          "probe_belief_top1_on_support": 0.078125,
          "mean_belief_support_size": 1.64453125,
          "decoder_belief_target_mass": 0.04819731041789055,
          "decoder_belief_target_nll": 4.081865310668945,
          "decoder_belief_top1_on_support": 0.0703125,
          "mean_decoder_belief_support_size": 1.64453125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 24,
          "query_type": "B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.050803110003471375,
          "query_target_nll": 3.4441020488739014,
          "query_top1_on_support": 0.0625,
          "mean_query_support_size": 1.578125,
          "probe_belief_target_mass": 0.0026295860297977924,
          "probe_belief_target_nll": 20.9365291595459,
          "probe_belief_top1_on_support": 0.0078125,
          "mean_belief_support_size": 1.78125,
          "decoder_belief_target_mass": 0.00255156890489161,
          "decoder_belief_target_nll": 25.23039436340332,
          "decoder_belief_top1_on_support": 0.0078125,
          "mean_decoder_belief_support_size": 1.78125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 24,
          "query_type": "B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.04906778037548065,
          "query_target_nll": 6.210253715515137,
          "query_top1_on_support": 0.04296875,
          "mean_query_support_size": 1.578125,
          "probe_belief_target_mass": 0.0023997374810278416,
          "probe_belief_target_nll": 25.62588119506836,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 1.78125,
          "decoder_belief_target_mass": 0.002125219674780965,
          "decoder_belief_target_nll": 29.09856414794922,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 1.78125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 24,
          "query_type": "B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.04933788999915123,
          "query_target_nll": 6.9960103034973145,
          "query_top1_on_support": 0.05078125,
          "mean_query_support_size": 1.578125,
          "probe_belief_target_mass": 0.003437318839132786,
          "probe_belief_target_nll": 22.629135131835938,
          "probe_belief_top1_on_support": 0.00390625,
          "mean_belief_support_size": 1.78125,
          "decoder_belief_target_mass": 0.0037936940789222717,
          "decoder_belief_target_nll": 25.657859802246094,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 1.78125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 24,
          "query_type": "B",
          "k": 4,
          "n": 256,
          "query_target_mass": 0.04553280025720596,
          "query_target_nll": 8.428838729858398,
          "query_top1_on_support": 0.03515625,
          "mean_query_support_size": 1.578125,
          "probe_belief_target_mass": 0.0016946650575846434,
          "probe_belief_target_nll": 20.006221771240234,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 1.78125,
          "decoder_belief_target_mass": 0.0017761618364602327,
          "decoder_belief_target_nll": 22.642963409423828,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 1.78125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 24,
          "query_type": "B",
          "k": 8,
          "n": 256,
          "query_target_mass": 0.06439875811338425,
          "query_target_nll": 8.782342910766602,
          "query_top1_on_support": 0.0703125,
          "mean_query_support_size": 1.578125,
          "probe_belief_target_mass": 0.0010900229681283236,
          "probe_belief_target_nll": 18.400650024414062,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 1.78125,
          "decoder_belief_target_mass": 0.0012461475562304258,
          "decoder_belief_target_nll": 20.463138580322266,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 1.78125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 24,
          "query_type": "B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.053781427443027496,
          "query_target_nll": 8.645956039428711,
          "query_top1_on_support": 0.046875,
          "mean_query_support_size": 1.578125,
          "probe_belief_target_mass": 0.0011208694195374846,
          "probe_belief_target_nll": 17.130252838134766,
          "probe_belief_top1_on_support": 0.00390625,
          "mean_belief_support_size": 1.78125,
          "decoder_belief_target_mass": 0.0011311201378703117,
          "decoder_belief_target_nll": 19.044368743896484,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 1.78125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 24,
          "query_type": "B",
          "k": 16,
          "n": 256,
          "query_target_mass": 0.05228535085916519,
          "query_target_nll": 8.515451431274414,
          "query_top1_on_support": 0.0625,
          "mean_query_support_size": 1.578125,
          "probe_belief_target_mass": 0.002429573331028223,
          "probe_belief_target_nll": 16.387802124023438,
          "probe_belief_top1_on_support": 0.00390625,
          "mean_belief_support_size": 1.78125,
          "decoder_belief_target_mass": 0.002493852050974965,
          "decoder_belief_target_nll": 17.85413932800293,
          "decoder_belief_top1_on_support": 0.0078125,
          "mean_decoder_belief_support_size": 1.78125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 24,
          "query_type": "B",
          "k": 24,
          "n": 256,
          "query_target_mass": 0.2143578827381134,
          "query_target_nll": 2.1785173416137695,
          "query_top1_on_support": 0.28125,
          "mean_query_support_size": 1.578125,
          "probe_belief_target_mass": 0.04732155054807663,
          "probe_belief_target_nll": 4.157291412353516,
          "probe_belief_top1_on_support": 0.078125,
          "mean_belief_support_size": 1.78125,
          "decoder_belief_target_mass": 0.04515793174505234,
          "decoder_belief_target_nll": 4.11705207824707,
          "decoder_belief_top1_on_support": 0.06640625,
          "mean_decoder_belief_support_size": 1.78125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 24,
          "query_type": "A_PLUS_B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.04825025796890259,
          "query_target_nll": 3.455349922180176,
          "query_top1_on_support": 0.046875,
          "mean_query_support_size": 1.515625,
          "probe_belief_target_mass": 0.0014574886299669743,
          "probe_belief_target_nll": 21.78738784790039,
          "probe_belief_top1_on_support": 0.00390625,
          "mean_belief_support_size": 1.578125,
          "decoder_belief_target_mass": 0.0012936179991811514,
          "decoder_belief_target_nll": 26.572322845458984,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 1.578125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 24,
          "query_type": "A_PLUS_B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.051915526390075684,
          "query_target_nll": 4.261604309082031,
          "query_top1_on_support": 0.0625,
          "mean_query_support_size": 1.515625,
          "probe_belief_target_mass": 0.0008325342205353081,
          "probe_belief_target_nll": 27.19382667541504,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 1.578125,
          "decoder_belief_target_mass": 0.0007742843590676785,
          "decoder_belief_target_nll": 31.595884323120117,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 1.578125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 24,
          "query_type": "A_PLUS_B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.04606422781944275,
          "query_target_nll": 4.5742645263671875,
          "query_top1_on_support": 0.046875,
          "mean_query_support_size": 1.515625,
          "probe_belief_target_mass": 0.00043543544597923756,
          "probe_belief_target_nll": 25.67795181274414,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 1.578125,
          "decoder_belief_target_mass": 0.0003183918306604028,
          "decoder_belief_target_nll": 29.210277557373047,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 1.578125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 24,
          "query_type": "A_PLUS_B",
          "k": 4,
          "n": 256,
          "query_target_mass": 0.049174923449754715,
          "query_target_nll": 5.333070755004883,
          "query_top1_on_support": 0.0546875,
          "mean_query_support_size": 1.515625,
          "probe_belief_target_mass": 0.0017524807481095195,
          "probe_belief_target_nll": 21.464237213134766,
          "probe_belief_top1_on_support": 0.00390625,
          "mean_belief_support_size": 1.578125,
          "decoder_belief_target_mass": 0.0016906794626265764,
          "decoder_belief_target_nll": 23.739160537719727,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 1.578125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 24,
          "query_type": "A_PLUS_B",
          "k": 8,
          "n": 256,
          "query_target_mass": 0.044418275356292725,
          "query_target_nll": 5.5819993019104,
          "query_top1_on_support": 0.02734375,
          "mean_query_support_size": 1.515625,
          "probe_belief_target_mass": 0.001116985804401338,
          "probe_belief_target_nll": 18.283687591552734,
          "probe_belief_top1_on_support": 0.00390625,
          "mean_belief_support_size": 1.578125,
          "decoder_belief_target_mass": 0.001171605079434812,
          "decoder_belief_target_nll": 19.93109893798828,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 1.578125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 24,
          "query_type": "A_PLUS_B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.046447791159152985,
          "query_target_nll": 5.369767189025879,
          "query_top1_on_support": 0.03515625,
          "mean_query_support_size": 1.515625,
          "probe_belief_target_mass": 0.0015214894665405154,
          "probe_belief_target_nll": 16.878101348876953,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 1.578125,
          "decoder_belief_target_mass": 0.0015334230847656727,
          "decoder_belief_target_nll": 18.397428512573242,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 1.578125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 24,
          "query_type": "A_PLUS_B",
          "k": 16,
          "n": 256,
          "query_target_mass": 0.054639678448438644,
          "query_target_nll": 5.611838340759277,
          "query_top1_on_support": 0.0546875,
          "mean_query_support_size": 1.515625,
          "probe_belief_target_mass": 0.0024306599516421556,
          "probe_belief_target_nll": 16.163352966308594,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 1.578125,
          "decoder_belief_target_mass": 0.002916657831519842,
          "decoder_belief_target_nll": 17.624862670898438,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 1.578125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 24,
          "query_type": "A_PLUS_B",
          "k": 24,
          "n": 256,
          "query_target_mass": 0.15765362977981567,
          "query_target_nll": 2.5235753059387207,
          "query_top1_on_support": 0.21484375,
          "mean_query_support_size": 1.515625,
          "probe_belief_target_mass": 0.04541557654738426,
          "probe_belief_target_nll": 4.14217472076416,
          "probe_belief_top1_on_support": 0.08984375,
          "mean_belief_support_size": 1.578125,
          "decoder_belief_target_mass": 0.04625755548477173,
          "decoder_belief_target_nll": 4.092135906219482,
          "decoder_belief_top1_on_support": 0.05078125,
          "mean_decoder_belief_support_size": 1.578125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 24,
          "query_type": "A_MINUS_B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.05857759714126587,
          "query_target_nll": 18.397899627685547,
          "query_top1_on_support": 0.05859375,
          "mean_query_support_size": 1.51953125,
          "probe_belief_target_mass": 0.0017375817988067865,
          "probe_belief_target_nll": 20.80959701538086,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 1.55078125,
          "decoder_belief_target_mass": 0.0018338833469897509,
          "decoder_belief_target_nll": 25.233264923095703,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 1.55078125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 24,
          "query_type": "A_MINUS_B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.050973080098629,
          "query_target_nll": 20.139265060424805,
          "query_top1_on_support": 0.05078125,
          "mean_query_support_size": 1.51953125,
          "probe_belief_target_mass": 0.0021660507190972567,
          "probe_belief_target_nll": 25.837284088134766,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 1.55078125,
          "decoder_belief_target_mass": 0.0022280856501311064,
          "decoder_belief_target_nll": 30.113082885742188,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 1.55078125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 24,
          "query_type": "A_MINUS_B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.03839056193828583,
          "query_target_nll": 18.102521896362305,
          "query_top1_on_support": 0.0546875,
          "mean_query_support_size": 1.51953125,
          "probe_belief_target_mass": 0.00032926321728155017,
          "probe_belief_target_nll": 25.495298385620117,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 1.55078125,
          "decoder_belief_target_mass": 0.00041789968963712454,
          "decoder_belief_target_nll": 29.30652618408203,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 1.55078125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 24,
          "query_type": "A_MINUS_B",
          "k": 4,
          "n": 256,
          "query_target_mass": 0.044963181018829346,
          "query_target_nll": 11.422538757324219,
          "query_top1_on_support": 0.03515625,
          "mean_query_support_size": 1.51953125,
          "probe_belief_target_mass": 0.0017436982598155737,
          "probe_belief_target_nll": 20.618303298950195,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 1.55078125,
          "decoder_belief_target_mass": 0.002405658597126603,
          "decoder_belief_target_nll": 23.12558937072754,
          "decoder_belief_top1_on_support": 0.0078125,
          "mean_decoder_belief_support_size": 1.55078125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 24,
          "query_type": "A_MINUS_B",
          "k": 8,
          "n": 256,
          "query_target_mass": 0.05068668723106384,
          "query_target_nll": 8.097721099853516,
          "query_top1_on_support": 0.046875,
          "mean_query_support_size": 1.51953125,
          "probe_belief_target_mass": 0.001732110744342208,
          "probe_belief_target_nll": 17.26290512084961,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 1.55078125,
          "decoder_belief_target_mass": 0.002041223691776395,
          "decoder_belief_target_nll": 19.014232635498047,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 1.55078125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 24,
          "query_type": "A_MINUS_B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.05499601364135742,
          "query_target_nll": 7.470462799072266,
          "query_top1_on_support": 0.07421875,
          "mean_query_support_size": 1.51953125,
          "probe_belief_target_mass": 0.0011450279271230102,
          "probe_belief_target_nll": 16.880266189575195,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 1.55078125,
          "decoder_belief_target_mass": 0.0014779148623347282,
          "decoder_belief_target_nll": 18.46977996826172,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 1.55078125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 24,
          "query_type": "A_MINUS_B",
          "k": 16,
          "n": 256,
          "query_target_mass": 0.05751677230000496,
          "query_target_nll": 6.755842685699463,
          "query_top1_on_support": 0.0546875,
          "mean_query_support_size": 1.51953125,
          "probe_belief_target_mass": 0.0022029196843504906,
          "probe_belief_target_nll": 16.02764892578125,
          "probe_belief_top1_on_support": 0.0,
          "mean_belief_support_size": 1.55078125,
          "decoder_belief_target_mass": 0.002189175458624959,
          "decoder_belief_target_nll": 17.455123901367188,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 1.55078125
        },
        {
          "model": "dense",
          "supervision": "full_belief",
          "length": 24,
          "query_type": "A_MINUS_B",
          "k": 24,
          "n": 256,
          "query_target_mass": 0.17393182218074799,
          "query_target_nll": 2.515559196472168,
          "query_top1_on_support": 0.23828125,
          "mean_query_support_size": 1.51953125,
          "probe_belief_target_mass": 0.049228738993406296,
          "probe_belief_target_nll": 4.104497909545898,
          "probe_belief_top1_on_support": 0.07421875,
          "mean_belief_support_size": 1.55078125,
          "decoder_belief_target_mass": 0.04544755816459656,
          "decoder_belief_target_nll": 4.09470796585083,
          "decoder_belief_top1_on_support": 0.09765625,
          "mean_decoder_belief_support_size": 1.55078125
        }
      ]
    }
  ],
  "checkpoints": [
    "large_artifacts/dense_supervision_ladder/checkpoints/main_full_belief_mod31/checkpoint_final.pt"
  ]
}