{
  "mode": "dense",
  "args": {
    "mode": "dense",
    "supervision": "sampled_final",
    "seed": 53,
    "modulus": 11,
    "observe_mod": 4,
    "observe_prob": 0.3,
    "op_family": "full",
    "query_family": "all",
    "train_min_len": 1,
    "train_max_len": 6,
    "eval_lengths": "3,6,9,12",
    "eval_k": "0,1,2,3,6,9,12",
    "eval_query_types": "all",
    "state_dim": 256,
    "instr_dim": 128,
    "heads": 4,
    "compiler_layers": 2,
    "dropout": 0.0,
    "train_steps": 1200,
    "batch_size": 256,
    "eval_batch_size": 256,
    "eval_examples": 256,
    "lr": 0.001,
    "weight_decay": 0.0001,
    "query_loss_weight": 1.0,
    "belief_loss_weight": 1.0,
    "belief_distill_prob": 0.1,
    "probe_steps": 600,
    "probe_batch_size": 256,
    "probe_lr": 0.001,
    "probe_weight_decay": 0.0001,
    "probe_dropout": 0.0,
    "max_grad_norm": 1.0,
    "log_every": 200,
    "probe_log_every": 100,
    "output_dir": "experiments/dense_supervision_ladder/runs/pilot_sampled_final_mod11",
    "checkpoint_dir": "large_artifacts/dense_supervision_ladder/checkpoints/pilot_sampled_final_mod11"
  },
  "env": {
    "python": "3.12.3",
    "torch": "2.8.0+cu128",
    "device": "cuda",
    "cuda_name": "NVIDIA RTX 6000 Ada Generation"
  },
  "train": [
    {
      "step": 1,
      "loss": 2.4312825202941895,
      "grad_norm": 0.7630031108856201,
      "query_loss": 2.4312825202941895,
      "belief_loss": 0.0,
      "query_target_mass": 0.44672441482543945,
      "query_target_nll": 2.2827816009521484,
      "query_top1_on_support": 0.703125,
      "mean_query_support_size": 4.65625
    },
    {
      "step": 200,
      "loss": 2.2287709712982178,
      "grad_norm": 0.5225602388381958,
      "query_loss": 2.2287709712982178,
      "belief_loss": 0.0,
      "query_target_mass": 0.5549721717834473,
      "query_target_nll": 2.2109317779541016,
      "query_top1_on_support": 0.62109375,
      "mean_query_support_size": 5.20703125
    },
    {
      "step": 400,
      "loss": 2.1657779216766357,
      "grad_norm": 0.5784071087837219,
      "query_loss": 2.1657779216766357,
      "belief_loss": 0.0,
      "query_target_mass": 0.5284178853034973,
      "query_target_nll": 2.1478846073150635,
      "query_top1_on_support": 0.578125,
      "mean_query_support_size": 4.734375
    },
    {
      "step": 600,
      "loss": 2.068990707397461,
      "grad_norm": 0.5910953283309937,
      "query_loss": 2.068990707397461,
      "belief_loss": 0.0,
      "query_target_mass": 0.5884122848510742,
      "query_target_nll": 2.046751022338867,
      "query_top1_on_support": 0.64453125,
      "mean_query_support_size": 4.85546875
    },
    {
      "step": 800,
      "loss": 2.021799325942993,
      "grad_norm": 0.7011949419975281,
      "query_loss": 2.021799325942993,
      "belief_loss": 0.0,
      "query_target_mass": 0.6280412077903748,
      "query_target_nll": 1.9977585077285767,
      "query_top1_on_support": 0.69140625,
      "mean_query_support_size": 5.1171875
    },
    {
      "step": 1000,
      "loss": 1.9364254474639893,
      "grad_norm": 0.8004977703094482,
      "query_loss": 1.9364254474639893,
      "belief_loss": 0.0,
      "query_target_mass": 0.6337892413139343,
      "query_target_nll": 1.9124162197113037,
      "query_top1_on_support": 0.703125,
      "mean_query_support_size": 4.8125
    },
    {
      "step": 1200,
      "loss": 1.8125019073486328,
      "grad_norm": 0.8936113119125366,
      "query_loss": 1.8125019073486328,
      "belief_loss": 0.0,
      "query_target_mass": 0.6747220158576965,
      "query_target_nll": 1.7701756954193115,
      "query_top1_on_support": 0.75,
      "mean_query_support_size": 4.62890625
    }
  ],
  "probe_train": [
    {
      "step": 1,
      "loss": 4.809880256652832,
      "grad_norm": 0.4655808210372925,
      "probe_belief_target_mass": 0.0605374351143837,
      "probe_belief_target_nll": 4.809880256652832,
      "probe_belief_top1_on_support": 0.0736607164144516,
      "mean_belief_support_size": 7.3058037757873535
    },
    {
      "step": 100,
      "loss": 2.8663430213928223,
      "grad_norm": 0.45194435119628906,
      "probe_belief_target_mass": 0.5232309699058533,
      "probe_belief_target_nll": 2.8663430213928223,
      "probe_belief_top1_on_support": 0.6796875596046448,
      "mean_belief_support_size": 7.245535850524902
    },
    {
      "step": 200,
      "loss": 2.637101888656616,
      "grad_norm": 0.39242589473724365,
      "probe_belief_target_mass": 0.6095945239067078,
      "probe_belief_target_nll": 2.637101888656616,
      "probe_belief_top1_on_support": 0.7254464626312256,
      "mean_belief_support_size": 7.470424652099609
    },
    {
      "step": 300,
      "loss": 2.5833334922790527,
      "grad_norm": 0.3754080832004547,
      "probe_belief_target_mass": 0.6324476003646851,
      "probe_belief_target_nll": 2.5833334922790527,
      "probe_belief_top1_on_support": 0.7393973469734192,
      "mean_belief_support_size": 7.313058376312256
    },
    {
      "step": 400,
      "loss": 2.4860472679138184,
      "grad_norm": 0.39385730028152466,
      "probe_belief_target_mass": 0.6510132551193237,
      "probe_belief_target_nll": 2.4860472679138184,
      "probe_belief_top1_on_support": 0.738839328289032,
      "mean_belief_support_size": 7.136160850524902
    },
    {
      "step": 500,
      "loss": 2.477740526199341,
      "grad_norm": 0.3865666389465332,
      "probe_belief_target_mass": 0.6635373830795288,
      "probe_belief_target_nll": 2.477740526199341,
      "probe_belief_top1_on_support": 0.7633928656578064,
      "mean_belief_support_size": 7.076451301574707
    },
    {
      "step": 600,
      "loss": 2.5707485675811768,
      "grad_norm": 0.380866140127182,
      "probe_belief_target_mass": 0.6514299511909485,
      "probe_belief_target_nll": 2.5707485675811768,
      "probe_belief_top1_on_support": 0.7360491156578064,
      "mean_belief_support_size": 7.327567100524902
    }
  ],
  "eval": [
    {
      "step": 1200,
      "rows": [
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 3,
          "query_type": "A",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.46988222002983093,
          "query_target_nll": 2.587775707244873,
          "query_top1_on_support": 0.4921875,
          "mean_query_support_size": 5.18359375,
          "probe_belief_target_mass": 0.06256759166717529,
          "probe_belief_target_nll": 17.43675994873047,
          "probe_belief_top1_on_support": 0.078125,
          "mean_belief_support_size": 5.84375,
          "decoder_belief_target_mass": 0.04832921177148819,
          "decoder_belief_target_nll": 4.811384201049805,
          "decoder_belief_top1_on_support": 0.04296875,
          "mean_decoder_belief_support_size": 5.84375
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 3,
          "query_type": "A",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.5492140054702759,
          "query_target_nll": 3.2111055850982666,
          "query_top1_on_support": 0.5390625,
          "mean_query_support_size": 5.18359375,
          "probe_belief_target_mass": 0.11901172995567322,
          "probe_belief_target_nll": 12.743096351623535,
          "probe_belief_top1_on_support": 0.109375,
          "mean_belief_support_size": 5.84375,
          "decoder_belief_target_mass": 0.04857911914587021,
          "decoder_belief_target_nll": 4.808718204498291,
          "decoder_belief_top1_on_support": 0.05859375,
          "mean_decoder_belief_support_size": 5.84375
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 3,
          "query_type": "A",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.6328621506690979,
          "query_target_nll": 2.683037042617798,
          "query_top1_on_support": 0.6640625,
          "mean_query_support_size": 5.18359375,
          "probe_belief_target_mass": 0.2013295590877533,
          "probe_belief_target_nll": 7.394757270812988,
          "probe_belief_top1_on_support": 0.21875,
          "mean_belief_support_size": 5.84375,
          "decoder_belief_target_mass": 0.048399850726127625,
          "decoder_belief_target_nll": 4.8151116371154785,
          "decoder_belief_top1_on_support": 0.046875,
          "mean_decoder_belief_support_size": 5.84375
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 3,
          "query_type": "A",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.7689529657363892,
          "query_target_nll": 1.7409001588821411,
          "query_top1_on_support": 0.82421875,
          "mean_query_support_size": 5.18359375,
          "probe_belief_target_mass": 0.4041691720485687,
          "probe_belief_target_nll": 2.852841377258301,
          "probe_belief_top1_on_support": 0.546875,
          "mean_belief_support_size": 5.84375,
          "decoder_belief_target_mass": 0.04843185842037201,
          "decoder_belief_target_nll": 4.818957805633545,
          "decoder_belief_top1_on_support": 0.05078125,
          "mean_decoder_belief_support_size": 5.84375
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 3,
          "query_type": "A",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.7689529657363892,
          "query_target_nll": 1.7409001588821411,
          "query_top1_on_support": 0.82421875,
          "mean_query_support_size": 5.18359375,
          "probe_belief_target_mass": 0.4041691720485687,
          "probe_belief_target_nll": 2.852841377258301,
          "probe_belief_top1_on_support": 0.546875,
          "mean_belief_support_size": 5.84375,
          "decoder_belief_target_mass": 0.04843185842037201,
          "decoder_belief_target_nll": 4.818957805633545,
          "decoder_belief_top1_on_support": 0.05078125,
          "mean_decoder_belief_support_size": 5.84375
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 3,
          "query_type": "A",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.7689529657363892,
          "query_target_nll": 1.7409001588821411,
          "query_top1_on_support": 0.82421875,
          "mean_query_support_size": 5.18359375,
          "probe_belief_target_mass": 0.4041691720485687,
          "probe_belief_target_nll": 2.852841377258301,
          "probe_belief_top1_on_support": 0.546875,
          "mean_belief_support_size": 5.84375,
          "decoder_belief_target_mass": 0.04843185842037201,
          "decoder_belief_target_nll": 4.818957805633545,
          "decoder_belief_top1_on_support": 0.05078125,
          "mean_decoder_belief_support_size": 5.84375
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 3,
          "query_type": "A",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.7689529657363892,
          "query_target_nll": 1.7409001588821411,
          "query_top1_on_support": 0.82421875,
          "mean_query_support_size": 5.18359375,
          "probe_belief_target_mass": 0.4041691720485687,
          "probe_belief_target_nll": 2.852841377258301,
          "probe_belief_top1_on_support": 0.546875,
          "mean_belief_support_size": 5.84375,
          "decoder_belief_target_mass": 0.04843185842037201,
          "decoder_belief_target_nll": 4.818957805633545,
          "decoder_belief_top1_on_support": 0.05078125,
          "mean_decoder_belief_support_size": 5.84375
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 3,
          "query_type": "B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.46652406454086304,
          "query_target_nll": 2.5809335708618164,
          "query_top1_on_support": 0.4609375,
          "mean_query_support_size": 5.15234375,
          "probe_belief_target_mass": 0.05247144028544426,
          "probe_belief_target_nll": 17.75397491455078,
          "probe_belief_top1_on_support": 0.0390625,
          "mean_belief_support_size": 5.77734375,
          "decoder_belief_target_mass": 0.04789862036705017,
          "decoder_belief_target_nll": 4.807101249694824,
          "decoder_belief_top1_on_support": 0.0390625,
          "mean_decoder_belief_support_size": 5.77734375
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 3,
          "query_type": "B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.5322158336639404,
          "query_target_nll": 3.2302422523498535,
          "query_top1_on_support": 0.546875,
          "mean_query_support_size": 5.15234375,
          "probe_belief_target_mass": 0.13236385583877563,
          "probe_belief_target_nll": 12.143587112426758,
          "probe_belief_top1_on_support": 0.13671875,
          "mean_belief_support_size": 5.77734375,
          "decoder_belief_target_mass": 0.04766403138637543,
          "decoder_belief_target_nll": 4.819805145263672,
          "decoder_belief_top1_on_support": 0.0546875,
          "mean_decoder_belief_support_size": 5.77734375
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 3,
          "query_type": "B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.6230270862579346,
          "query_target_nll": 2.7199273109436035,
          "query_top1_on_support": 0.65625,
          "mean_query_support_size": 5.15234375,
          "probe_belief_target_mass": 0.2177877277135849,
          "probe_belief_target_nll": 7.474302291870117,
          "probe_belief_top1_on_support": 0.23828125,
          "mean_belief_support_size": 5.77734375,
          "decoder_belief_target_mass": 0.04812295734882355,
          "decoder_belief_target_nll": 4.808781623840332,
          "decoder_belief_top1_on_support": 0.04296875,
          "mean_decoder_belief_support_size": 5.77734375
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 3,
          "query_type": "B",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.7569008469581604,
          "query_target_nll": 1.7450308799743652,
          "query_top1_on_support": 0.828125,
          "mean_query_support_size": 5.15234375,
          "probe_belief_target_mass": 0.42490994930267334,
          "probe_belief_target_nll": 2.7252378463745117,
          "probe_belief_top1_on_support": 0.57421875,
          "mean_belief_support_size": 5.77734375,
          "decoder_belief_target_mass": 0.047950379550457,
          "decoder_belief_target_nll": 4.818295001983643,
          "decoder_belief_top1_on_support": 0.0703125,
          "mean_decoder_belief_support_size": 5.77734375
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 3,
          "query_type": "B",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.7569008469581604,
          "query_target_nll": 1.7450308799743652,
          "query_top1_on_support": 0.828125,
          "mean_query_support_size": 5.15234375,
          "probe_belief_target_mass": 0.42490994930267334,
          "probe_belief_target_nll": 2.7252378463745117,
          "probe_belief_top1_on_support": 0.57421875,
          "mean_belief_support_size": 5.77734375,
          "decoder_belief_target_mass": 0.047950379550457,
          "decoder_belief_target_nll": 4.818295001983643,
          "decoder_belief_top1_on_support": 0.0703125,
          "mean_decoder_belief_support_size": 5.77734375
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 3,
          "query_type": "B",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.7569008469581604,
          "query_target_nll": 1.7450308799743652,
          "query_top1_on_support": 0.828125,
          "mean_query_support_size": 5.15234375,
          "probe_belief_target_mass": 0.42490994930267334,
          "probe_belief_target_nll": 2.7252378463745117,
          "probe_belief_top1_on_support": 0.57421875,
          "mean_belief_support_size": 5.77734375,
          "decoder_belief_target_mass": 0.047950379550457,
          "decoder_belief_target_nll": 4.818295001983643,
          "decoder_belief_top1_on_support": 0.0703125,
          "mean_decoder_belief_support_size": 5.77734375
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 3,
          "query_type": "B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.7569008469581604,
          "query_target_nll": 1.7450308799743652,
          "query_top1_on_support": 0.828125,
          "mean_query_support_size": 5.15234375,
          "probe_belief_target_mass": 0.42490994930267334,
          "probe_belief_target_nll": 2.7252378463745117,
          "probe_belief_top1_on_support": 0.57421875,
          "mean_belief_support_size": 5.77734375,
          "decoder_belief_target_mass": 0.047950379550457,
          "decoder_belief_target_nll": 4.818295001983643,
          "decoder_belief_top1_on_support": 0.0703125,
          "mean_decoder_belief_support_size": 5.77734375
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 3,
          "query_type": "A_PLUS_B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.4829876124858856,
          "query_target_nll": 2.6273255348205566,
          "query_top1_on_support": 0.453125,
          "mean_query_support_size": 5.3671875,
          "probe_belief_target_mass": 0.06535881757736206,
          "probe_belief_target_nll": 17.34501838684082,
          "probe_belief_top1_on_support": 0.07421875,
          "mean_belief_support_size": 5.484375,
          "decoder_belief_target_mass": 0.04538842290639877,
          "decoder_belief_target_nll": 4.810123443603516,
          "decoder_belief_top1_on_support": 0.04296875,
          "mean_decoder_belief_support_size": 5.484375
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 3,
          "query_type": "A_PLUS_B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.48866790533065796,
          "query_target_nll": 2.9164552688598633,
          "query_top1_on_support": 0.49609375,
          "mean_query_support_size": 5.3671875,
          "probe_belief_target_mass": 0.15209926664829254,
          "probe_belief_target_nll": 13.029583930969238,
          "probe_belief_top1_on_support": 0.15234375,
          "mean_belief_support_size": 5.484375,
          "decoder_belief_target_mass": 0.04525510221719742,
          "decoder_belief_target_nll": 4.815186500549316,
          "decoder_belief_top1_on_support": 0.03515625,
          "mean_decoder_belief_support_size": 5.484375
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 3,
          "query_type": "A_PLUS_B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.5105758905410767,
          "query_target_nll": 2.738287925720215,
          "query_top1_on_support": 0.52734375,
          "mean_query_support_size": 5.3671875,
          "probe_belief_target_mass": 0.20042377710342407,
          "probe_belief_target_nll": 8.304998397827148,
          "probe_belief_top1_on_support": 0.203125,
          "mean_belief_support_size": 5.484375,
          "decoder_belief_target_mass": 0.044779591262340546,
          "decoder_belief_target_nll": 4.816498756408691,
          "decoder_belief_top1_on_support": 0.0390625,
          "mean_decoder_belief_support_size": 5.484375
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 3,
          "query_type": "A_PLUS_B",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.6115679740905762,
          "query_target_nll": 2.1287007331848145,
          "query_top1_on_support": 0.6953125,
          "mean_query_support_size": 5.3671875,
          "probe_belief_target_mass": 0.44858959317207336,
          "probe_belief_target_nll": 2.600292205810547,
          "probe_belief_top1_on_support": 0.61328125,
          "mean_belief_support_size": 5.484375,
          "decoder_belief_target_mass": 0.04476335644721985,
          "decoder_belief_target_nll": 4.822205543518066,
          "decoder_belief_top1_on_support": 0.03125,
          "mean_decoder_belief_support_size": 5.484375
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 3,
          "query_type": "A_PLUS_B",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.6115679740905762,
          "query_target_nll": 2.1287007331848145,
          "query_top1_on_support": 0.6953125,
          "mean_query_support_size": 5.3671875,
          "probe_belief_target_mass": 0.44858959317207336,
          "probe_belief_target_nll": 2.600292205810547,
          "probe_belief_top1_on_support": 0.61328125,
          "mean_belief_support_size": 5.484375,
          "decoder_belief_target_mass": 0.04476335644721985,
          "decoder_belief_target_nll": 4.822205543518066,
          "decoder_belief_top1_on_support": 0.03125,
          "mean_decoder_belief_support_size": 5.484375
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 3,
          "query_type": "A_PLUS_B",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.6115679740905762,
          "query_target_nll": 2.1287007331848145,
          "query_top1_on_support": 0.6953125,
          "mean_query_support_size": 5.3671875,
          "probe_belief_target_mass": 0.44858959317207336,
          "probe_belief_target_nll": 2.600292205810547,
          "probe_belief_top1_on_support": 0.61328125,
          "mean_belief_support_size": 5.484375,
          "decoder_belief_target_mass": 0.04476335644721985,
          "decoder_belief_target_nll": 4.822205543518066,
          "decoder_belief_top1_on_support": 0.03125,
          "mean_decoder_belief_support_size": 5.484375
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 3,
          "query_type": "A_PLUS_B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.6115679740905762,
          "query_target_nll": 2.1287007331848145,
          "query_top1_on_support": 0.6953125,
          "mean_query_support_size": 5.3671875,
          "probe_belief_target_mass": 0.44858959317207336,
          "probe_belief_target_nll": 2.600292205810547,
          "probe_belief_top1_on_support": 0.61328125,
          "mean_belief_support_size": 5.484375,
          "decoder_belief_target_mass": 0.04476335644721985,
          "decoder_belief_target_nll": 4.822205543518066,
          "decoder_belief_top1_on_support": 0.03125,
          "mean_decoder_belief_support_size": 5.484375
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 3,
          "query_type": "A_MINUS_B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.47254133224487305,
          "query_target_nll": 9.417624473571777,
          "query_top1_on_support": 0.47265625,
          "mean_query_support_size": 4.671875,
          "probe_belief_target_mass": 0.05917912721633911,
          "probe_belief_target_nll": 16.807941436767578,
          "probe_belief_top1_on_support": 0.05078125,
          "mean_belief_support_size": 5.6015625,
          "decoder_belief_target_mass": 0.0467354878783226,
          "decoder_belief_target_nll": 4.798651695251465,
          "decoder_belief_top1_on_support": 0.05078125,
          "mean_decoder_belief_support_size": 5.6015625
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 3,
          "query_type": "A_MINUS_B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.49754825234413147,
          "query_target_nll": 4.484367847442627,
          "query_top1_on_support": 0.5234375,
          "mean_query_support_size": 4.671875,
          "probe_belief_target_mass": 0.15483994781970978,
          "probe_belief_target_nll": 11.964420318603516,
          "probe_belief_top1_on_support": 0.13671875,
          "mean_belief_support_size": 5.6015625,
          "decoder_belief_target_mass": 0.046453312039375305,
          "decoder_belief_target_nll": 4.814164638519287,
          "decoder_belief_top1_on_support": 0.03125,
          "mean_decoder_belief_support_size": 5.6015625
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 3,
          "query_type": "A_MINUS_B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.5074940919876099,
          "query_target_nll": 3.1024210453033447,
          "query_top1_on_support": 0.52734375,
          "mean_query_support_size": 4.671875,
          "probe_belief_target_mass": 0.24015921354293823,
          "probe_belief_target_nll": 7.043479919433594,
          "probe_belief_top1_on_support": 0.2734375,
          "mean_belief_support_size": 5.6015625,
          "decoder_belief_target_mass": 0.046161308884620667,
          "decoder_belief_target_nll": 4.826233863830566,
          "decoder_belief_top1_on_support": 0.05078125,
          "mean_decoder_belief_support_size": 5.6015625
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 3,
          "query_type": "A_MINUS_B",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.6278619170188904,
          "query_target_nll": 1.788727045059204,
          "query_top1_on_support": 0.70703125,
          "mean_query_support_size": 4.671875,
          "probe_belief_target_mass": 0.4301525950431824,
          "probe_belief_target_nll": 2.703197956085205,
          "probe_belief_top1_on_support": 0.55859375,
          "mean_belief_support_size": 5.6015625,
          "decoder_belief_target_mass": 0.04623959958553314,
          "decoder_belief_target_nll": 4.81721305847168,
          "decoder_belief_top1_on_support": 0.0234375,
          "mean_decoder_belief_support_size": 5.6015625
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 3,
          "query_type": "A_MINUS_B",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.6278619170188904,
          "query_target_nll": 1.788727045059204,
          "query_top1_on_support": 0.70703125,
          "mean_query_support_size": 4.671875,
          "probe_belief_target_mass": 0.4301525950431824,
          "probe_belief_target_nll": 2.703197956085205,
          "probe_belief_top1_on_support": 0.55859375,
          "mean_belief_support_size": 5.6015625,
          "decoder_belief_target_mass": 0.04623959958553314,
          "decoder_belief_target_nll": 4.81721305847168,
          "decoder_belief_top1_on_support": 0.0234375,
          "mean_decoder_belief_support_size": 5.6015625
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 3,
          "query_type": "A_MINUS_B",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.6278619170188904,
          "query_target_nll": 1.788727045059204,
          "query_top1_on_support": 0.70703125,
          "mean_query_support_size": 4.671875,
          "probe_belief_target_mass": 0.4301525950431824,
          "probe_belief_target_nll": 2.703197956085205,
          "probe_belief_top1_on_support": 0.55859375,
          "mean_belief_support_size": 5.6015625,
          "decoder_belief_target_mass": 0.04623959958553314,
          "decoder_belief_target_nll": 4.81721305847168,
          "decoder_belief_top1_on_support": 0.0234375,
          "mean_decoder_belief_support_size": 5.6015625
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 3,
          "query_type": "A_MINUS_B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.6278619170188904,
          "query_target_nll": 1.788727045059204,
          "query_top1_on_support": 0.70703125,
          "mean_query_support_size": 4.671875,
          "probe_belief_target_mass": 0.4301525950431824,
          "probe_belief_target_nll": 2.703197956085205,
          "probe_belief_top1_on_support": 0.55859375,
          "mean_belief_support_size": 5.6015625,
          "decoder_belief_target_mass": 0.04623959958553314,
          "decoder_belief_target_nll": 4.81721305847168,
          "decoder_belief_top1_on_support": 0.0234375,
          "mean_decoder_belief_support_size": 5.6015625
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 6,
          "query_type": "A",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.2473030984401703,
          "query_target_nll": 2.6040053367614746,
          "query_top1_on_support": 0.2265625,
          "mean_query_support_size": 2.78125,
          "probe_belief_target_mass": 0.029890164732933044,
          "probe_belief_target_nll": 17.779258728027344,
          "probe_belief_top1_on_support": 0.02734375,
          "mean_belief_support_size": 3.45703125,
          "decoder_belief_target_mass": 0.028695974498987198,
          "decoder_belief_target_nll": 4.814629554748535,
          "decoder_belief_top1_on_support": 0.0234375,
          "mean_decoder_belief_support_size": 3.45703125
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 6,
          "query_type": "A",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.26927804946899414,
          "query_target_nll": 3.7467775344848633,
          "query_top1_on_support": 0.25390625,
          "mean_query_support_size": 2.78125,
          "probe_belief_target_mass": 0.07114387303590775,
          "probe_belief_target_nll": 13.927730560302734,
          "probe_belief_top1_on_support": 0.0703125,
          "mean_belief_support_size": 3.45703125,
          "decoder_belief_target_mass": 0.02847321704030037,
          "decoder_belief_target_nll": 4.816279411315918,
          "decoder_belief_top1_on_support": 0.01953125,
          "mean_decoder_belief_support_size": 3.45703125
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 6,
          "query_type": "A",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.31720104813575745,
          "query_target_nll": 3.4597418308258057,
          "query_top1_on_support": 0.31640625,
          "mean_query_support_size": 2.78125,
          "probe_belief_target_mass": 0.07384228706359863,
          "probe_belief_target_nll": 10.85362434387207,
          "probe_belief_top1_on_support": 0.09375,
          "mean_belief_support_size": 3.45703125,
          "decoder_belief_target_mass": 0.02835235744714737,
          "decoder_belief_target_nll": 4.822488784790039,
          "decoder_belief_top1_on_support": 0.015625,
          "mean_decoder_belief_support_size": 3.45703125
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 6,
          "query_type": "A",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.3411954343318939,
          "query_target_nll": 3.309664249420166,
          "query_top1_on_support": 0.3515625,
          "mean_query_support_size": 2.78125,
          "probe_belief_target_mass": 0.07494892925024033,
          "probe_belief_target_nll": 9.429313659667969,
          "probe_belief_top1_on_support": 0.07421875,
          "mean_belief_support_size": 3.45703125,
          "decoder_belief_target_mass": 0.028358137235045433,
          "decoder_belief_target_nll": 4.815634250640869,
          "decoder_belief_top1_on_support": 0.03125,
          "mean_decoder_belief_support_size": 3.45703125
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 6,
          "query_type": "A",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.48593881726264954,
          "query_target_nll": 1.784972071647644,
          "query_top1_on_support": 0.55078125,
          "mean_query_support_size": 2.78125,
          "probe_belief_target_mass": 0.18819698691368103,
          "probe_belief_target_nll": 3.167755126953125,
          "probe_belief_top1_on_support": 0.27734375,
          "mean_belief_support_size": 3.45703125,
          "decoder_belief_target_mass": 0.02833785116672516,
          "decoder_belief_target_nll": 4.822076797485352,
          "decoder_belief_top1_on_support": 0.01171875,
          "mean_decoder_belief_support_size": 3.45703125
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 6,
          "query_type": "A",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.48593881726264954,
          "query_target_nll": 1.784972071647644,
          "query_top1_on_support": 0.55078125,
          "mean_query_support_size": 2.78125,
          "probe_belief_target_mass": 0.18819698691368103,
          "probe_belief_target_nll": 3.167755126953125,
          "probe_belief_top1_on_support": 0.27734375,
          "mean_belief_support_size": 3.45703125,
          "decoder_belief_target_mass": 0.02833785116672516,
          "decoder_belief_target_nll": 4.822076797485352,
          "decoder_belief_top1_on_support": 0.01171875,
          "mean_decoder_belief_support_size": 3.45703125
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 6,
          "query_type": "A",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.48593881726264954,
          "query_target_nll": 1.784972071647644,
          "query_top1_on_support": 0.55078125,
          "mean_query_support_size": 2.78125,
          "probe_belief_target_mass": 0.18819698691368103,
          "probe_belief_target_nll": 3.167755126953125,
          "probe_belief_top1_on_support": 0.27734375,
          "mean_belief_support_size": 3.45703125,
          "decoder_belief_target_mass": 0.02833785116672516,
          "decoder_belief_target_nll": 4.822076797485352,
          "decoder_belief_top1_on_support": 0.01171875,
          "mean_decoder_belief_support_size": 3.45703125
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 6,
          "query_type": "B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.30441606044769287,
          "query_target_nll": 2.643301486968994,
          "query_top1_on_support": 0.3046875,
          "mean_query_support_size": 3.4140625,
          "probe_belief_target_mass": 0.03445829078555107,
          "probe_belief_target_nll": 18.42621612548828,
          "probe_belief_top1_on_support": 0.0234375,
          "mean_belief_support_size": 4.03125,
          "decoder_belief_target_mass": 0.03332225978374481,
          "decoder_belief_target_nll": 4.810358047485352,
          "decoder_belief_top1_on_support": 0.0390625,
          "mean_decoder_belief_support_size": 4.03125
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 6,
          "query_type": "B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.3363608717918396,
          "query_target_nll": 3.9921748638153076,
          "query_top1_on_support": 0.35546875,
          "mean_query_support_size": 3.4140625,
          "probe_belief_target_mass": 0.04518283158540726,
          "probe_belief_target_nll": 15.449813842773438,
          "probe_belief_top1_on_support": 0.046875,
          "mean_belief_support_size": 4.03125,
          "decoder_belief_target_mass": 0.03297550976276398,
          "decoder_belief_target_nll": 4.826999187469482,
          "decoder_belief_top1_on_support": 0.03515625,
          "mean_decoder_belief_support_size": 4.03125
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 6,
          "query_type": "B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.3674909174442291,
          "query_target_nll": 3.5843424797058105,
          "query_top1_on_support": 0.375,
          "mean_query_support_size": 3.4140625,
          "probe_belief_target_mass": 0.06573092937469482,
          "probe_belief_target_nll": 11.43690299987793,
          "probe_belief_top1_on_support": 0.078125,
          "mean_belief_support_size": 4.03125,
          "decoder_belief_target_mass": 0.03312522917985916,
          "decoder_belief_target_nll": 4.816170692443848,
          "decoder_belief_top1_on_support": 0.01953125,
          "mean_decoder_belief_support_size": 4.03125
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 6,
          "query_type": "B",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.3772066533565521,
          "query_target_nll": 3.673529624938965,
          "query_top1_on_support": 0.41015625,
          "mean_query_support_size": 3.4140625,
          "probe_belief_target_mass": 0.06816184520721436,
          "probe_belief_target_nll": 9.836244583129883,
          "probe_belief_top1_on_support": 0.0859375,
          "mean_belief_support_size": 4.03125,
          "decoder_belief_target_mass": 0.032919954508543015,
          "decoder_belief_target_nll": 4.824705123901367,
          "decoder_belief_top1_on_support": 0.0546875,
          "mean_decoder_belief_support_size": 4.03125
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 6,
          "query_type": "B",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.5668624639511108,
          "query_target_nll": 1.7720649242401123,
          "query_top1_on_support": 0.6171875,
          "mean_query_support_size": 3.4140625,
          "probe_belief_target_mass": 0.2100953608751297,
          "probe_belief_target_nll": 3.19893479347229,
          "probe_belief_top1_on_support": 0.33984375,
          "mean_belief_support_size": 4.03125,
          "decoder_belief_target_mass": 0.03310411050915718,
          "decoder_belief_target_nll": 4.822439193725586,
          "decoder_belief_top1_on_support": 0.015625,
          "mean_decoder_belief_support_size": 4.03125
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 6,
          "query_type": "B",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.5668624639511108,
          "query_target_nll": 1.7720649242401123,
          "query_top1_on_support": 0.6171875,
          "mean_query_support_size": 3.4140625,
          "probe_belief_target_mass": 0.2100953608751297,
          "probe_belief_target_nll": 3.19893479347229,
          "probe_belief_top1_on_support": 0.33984375,
          "mean_belief_support_size": 4.03125,
          "decoder_belief_target_mass": 0.03310411050915718,
          "decoder_belief_target_nll": 4.822439193725586,
          "decoder_belief_top1_on_support": 0.015625,
          "mean_decoder_belief_support_size": 4.03125
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 6,
          "query_type": "B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.5668624639511108,
          "query_target_nll": 1.7720649242401123,
          "query_top1_on_support": 0.6171875,
          "mean_query_support_size": 3.4140625,
          "probe_belief_target_mass": 0.2100953608751297,
          "probe_belief_target_nll": 3.19893479347229,
          "probe_belief_top1_on_support": 0.33984375,
          "mean_belief_support_size": 4.03125,
          "decoder_belief_target_mass": 0.03310411050915718,
          "decoder_belief_target_nll": 4.822439193725586,
          "decoder_belief_top1_on_support": 0.015625,
          "mean_decoder_belief_support_size": 4.03125
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 6,
          "query_type": "A_PLUS_B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.33343398571014404,
          "query_target_nll": 2.6266047954559326,
          "query_top1_on_support": 0.34375,
          "mean_query_support_size": 3.6171875,
          "probe_belief_target_mass": 0.033798325806856155,
          "probe_belief_target_nll": 18.320701599121094,
          "probe_belief_top1_on_support": 0.03125,
          "mean_belief_support_size": 3.97265625,
          "decoder_belief_target_mass": 0.032552339136600494,
          "decoder_belief_target_nll": 4.830920219421387,
          "decoder_belief_top1_on_support": 0.046875,
          "mean_decoder_belief_support_size": 3.97265625
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 6,
          "query_type": "A_PLUS_B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.3292003870010376,
          "query_target_nll": 2.7928683757781982,
          "query_top1_on_support": 0.29296875,
          "mean_query_support_size": 3.6171875,
          "probe_belief_target_mass": 0.04969996213912964,
          "probe_belief_target_nll": 14.530515670776367,
          "probe_belief_top1_on_support": 0.0390625,
          "mean_belief_support_size": 3.97265625,
          "decoder_belief_target_mass": 0.03252776712179184,
          "decoder_belief_target_nll": 4.811452865600586,
          "decoder_belief_top1_on_support": 0.0078125,
          "mean_decoder_belief_support_size": 3.97265625
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 6,
          "query_type": "A_PLUS_B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.3318507671356201,
          "query_target_nll": 2.8440802097320557,
          "query_top1_on_support": 0.33203125,
          "mean_query_support_size": 3.6171875,
          "probe_belief_target_mass": 0.06376940757036209,
          "probe_belief_target_nll": 11.03817367553711,
          "probe_belief_top1_on_support": 0.0703125,
          "mean_belief_support_size": 3.97265625,
          "decoder_belief_target_mass": 0.03267209231853485,
          "decoder_belief_target_nll": 4.811470985412598,
          "decoder_belief_top1_on_support": 0.0390625,
          "mean_decoder_belief_support_size": 3.97265625
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 6,
          "query_type": "A_PLUS_B",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.32627248764038086,
          "query_target_nll": 2.8988637924194336,
          "query_top1_on_support": 0.3046875,
          "mean_query_support_size": 3.6171875,
          "probe_belief_target_mass": 0.07215237617492676,
          "probe_belief_target_nll": 8.917096138000488,
          "probe_belief_top1_on_support": 0.0625,
          "mean_belief_support_size": 3.97265625,
          "decoder_belief_target_mass": 0.03251643106341362,
          "decoder_belief_target_nll": 4.815543174743652,
          "decoder_belief_top1_on_support": 0.01953125,
          "mean_decoder_belief_support_size": 3.97265625
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 6,
          "query_type": "A_PLUS_B",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.4286966621875763,
          "query_target_nll": 2.0998599529266357,
          "query_top1_on_support": 0.51953125,
          "mean_query_support_size": 3.6171875,
          "probe_belief_target_mass": 0.20382089912891388,
          "probe_belief_target_nll": 3.199127197265625,
          "probe_belief_top1_on_support": 0.3203125,
          "mean_belief_support_size": 3.97265625,
          "decoder_belief_target_mass": 0.03282991051673889,
          "decoder_belief_target_nll": 4.818681240081787,
          "decoder_belief_top1_on_support": 0.01953125,
          "mean_decoder_belief_support_size": 3.97265625
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 6,
          "query_type": "A_PLUS_B",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.4286966621875763,
          "query_target_nll": 2.0998599529266357,
          "query_top1_on_support": 0.51953125,
          "mean_query_support_size": 3.6171875,
          "probe_belief_target_mass": 0.20382089912891388,
          "probe_belief_target_nll": 3.199127197265625,
          "probe_belief_top1_on_support": 0.3203125,
          "mean_belief_support_size": 3.97265625,
          "decoder_belief_target_mass": 0.03282991051673889,
          "decoder_belief_target_nll": 4.818681240081787,
          "decoder_belief_top1_on_support": 0.01953125,
          "mean_decoder_belief_support_size": 3.97265625
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 6,
          "query_type": "A_PLUS_B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.4286966621875763,
          "query_target_nll": 2.0998599529266357,
          "query_top1_on_support": 0.51953125,
          "mean_query_support_size": 3.6171875,
          "probe_belief_target_mass": 0.20382089912891388,
          "probe_belief_target_nll": 3.199127197265625,
          "probe_belief_top1_on_support": 0.3203125,
          "mean_belief_support_size": 3.97265625,
          "decoder_belief_target_mass": 0.03282991051673889,
          "decoder_belief_target_nll": 4.818681240081787,
          "decoder_belief_top1_on_support": 0.01953125,
          "mean_decoder_belief_support_size": 3.97265625
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 6,
          "query_type": "A_MINUS_B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.31243348121643066,
          "query_target_nll": 9.957746505737305,
          "query_top1_on_support": 0.3125,
          "mean_query_support_size": 3.2265625,
          "probe_belief_target_mass": 0.03048647940158844,
          "probe_belief_target_nll": 17.938594818115234,
          "probe_belief_top1_on_support": 0.015625,
          "mean_belief_support_size": 3.76171875,
          "decoder_belief_target_mass": 0.030961286276578903,
          "decoder_belief_target_nll": 4.817396640777588,
          "decoder_belief_top1_on_support": 0.015625,
          "mean_decoder_belief_support_size": 3.76171875
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 6,
          "query_type": "A_MINUS_B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.30359530448913574,
          "query_target_nll": 5.016068935394287,
          "query_top1_on_support": 0.29296875,
          "mean_query_support_size": 3.2265625,
          "probe_belief_target_mass": 0.044032271951436996,
          "probe_belief_target_nll": 14.731951713562012,
          "probe_belief_top1_on_support": 0.0390625,
          "mean_belief_support_size": 3.76171875,
          "decoder_belief_target_mass": 0.030929310247302055,
          "decoder_belief_target_nll": 4.8176798820495605,
          "decoder_belief_top1_on_support": 0.03125,
          "mean_decoder_belief_support_size": 3.76171875
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 6,
          "query_type": "A_MINUS_B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.30834197998046875,
          "query_target_nll": 3.7823240756988525,
          "query_top1_on_support": 0.33203125,
          "mean_query_support_size": 3.2265625,
          "probe_belief_target_mass": 0.06375878304243088,
          "probe_belief_target_nll": 10.926966667175293,
          "probe_belief_top1_on_support": 0.06640625,
          "mean_belief_support_size": 3.76171875,
          "decoder_belief_target_mass": 0.031005363911390305,
          "decoder_belief_target_nll": 4.814269065856934,
          "decoder_belief_top1_on_support": 0.03125,
          "mean_decoder_belief_support_size": 3.76171875
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 6,
          "query_type": "A_MINUS_B",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.30338114500045776,
          "query_target_nll": 3.390533208847046,
          "query_top1_on_support": 0.3046875,
          "mean_query_support_size": 3.2265625,
          "probe_belief_target_mass": 0.06701186299324036,
          "probe_belief_target_nll": 9.038179397583008,
          "probe_belief_top1_on_support": 0.07421875,
          "mean_belief_support_size": 3.76171875,
          "decoder_belief_target_mass": 0.030876310542225838,
          "decoder_belief_target_nll": 4.822976589202881,
          "decoder_belief_top1_on_support": 0.0234375,
          "mean_decoder_belief_support_size": 3.76171875
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 6,
          "query_type": "A_MINUS_B",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.4430456757545471,
          "query_target_nll": 1.98830246925354,
          "query_top1_on_support": 0.49609375,
          "mean_query_support_size": 3.2265625,
          "probe_belief_target_mass": 0.18825840950012207,
          "probe_belief_target_nll": 3.299530267715454,
          "probe_belief_top1_on_support": 0.2421875,
          "mean_belief_support_size": 3.76171875,
          "decoder_belief_target_mass": 0.03082234412431717,
          "decoder_belief_target_nll": 4.823643684387207,
          "decoder_belief_top1_on_support": 0.01171875,
          "mean_decoder_belief_support_size": 3.76171875
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 6,
          "query_type": "A_MINUS_B",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.4430456757545471,
          "query_target_nll": 1.98830246925354,
          "query_top1_on_support": 0.49609375,
          "mean_query_support_size": 3.2265625,
          "probe_belief_target_mass": 0.18825840950012207,
          "probe_belief_target_nll": 3.299530267715454,
          "probe_belief_top1_on_support": 0.2421875,
          "mean_belief_support_size": 3.76171875,
          "decoder_belief_target_mass": 0.03082234412431717,
          "decoder_belief_target_nll": 4.823643684387207,
          "decoder_belief_top1_on_support": 0.01171875,
          "mean_decoder_belief_support_size": 3.76171875
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 6,
          "query_type": "A_MINUS_B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.4430456757545471,
          "query_target_nll": 1.98830246925354,
          "query_top1_on_support": 0.49609375,
          "mean_query_support_size": 3.2265625,
          "probe_belief_target_mass": 0.18825840950012207,
          "probe_belief_target_nll": 3.299530267715454,
          "probe_belief_top1_on_support": 0.2421875,
          "mean_belief_support_size": 3.76171875,
          "decoder_belief_target_mass": 0.03082234412431717,
          "decoder_belief_target_nll": 4.823643684387207,
          "decoder_belief_top1_on_support": 0.01171875,
          "mean_decoder_belief_support_size": 3.76171875
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 9,
          "query_type": "A",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.22090023756027222,
          "query_target_nll": 2.571146011352539,
          "query_top1_on_support": 0.2109375,
          "mean_query_support_size": 2.40625,
          "probe_belief_target_mass": 0.02142459526658058,
          "probe_belief_target_nll": 18.408824920654297,
          "probe_belief_top1_on_support": 0.0234375,
          "mean_belief_support_size": 2.73828125,
          "decoder_belief_target_mass": 0.022584300488233566,
          "decoder_belief_target_nll": 4.813832759857178,
          "decoder_belief_top1_on_support": 0.0234375,
          "mean_decoder_belief_support_size": 2.73828125
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 9,
          "query_type": "A",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.2371445745229721,
          "query_target_nll": 3.849165439605713,
          "query_top1_on_support": 0.21875,
          "mean_query_support_size": 2.40625,
          "probe_belief_target_mass": 0.03368290513753891,
          "probe_belief_target_nll": 14.697354316711426,
          "probe_belief_top1_on_support": 0.0390625,
          "mean_belief_support_size": 2.73828125,
          "decoder_belief_target_mass": 0.022928662598133087,
          "decoder_belief_target_nll": 4.809299468994141,
          "decoder_belief_top1_on_support": 0.02734375,
          "mean_decoder_belief_support_size": 2.73828125
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 9,
          "query_type": "A",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.24382053315639496,
          "query_target_nll": 3.932662010192871,
          "query_top1_on_support": 0.24609375,
          "mean_query_support_size": 2.40625,
          "probe_belief_target_mass": 0.033069759607315063,
          "probe_belief_target_nll": 12.001561164855957,
          "probe_belief_top1_on_support": 0.046875,
          "mean_belief_support_size": 2.73828125,
          "decoder_belief_target_mass": 0.022842001169919968,
          "decoder_belief_target_nll": 4.810747146606445,
          "decoder_belief_top1_on_support": 0.01953125,
          "mean_decoder_belief_support_size": 2.73828125
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 9,
          "query_type": "A",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.24388451874256134,
          "query_target_nll": 4.01344108581543,
          "query_top1_on_support": 0.2578125,
          "mean_query_support_size": 2.40625,
          "probe_belief_target_mass": 0.0343589223921299,
          "probe_belief_target_nll": 10.343520164489746,
          "probe_belief_top1_on_support": 0.0234375,
          "mean_belief_support_size": 2.73828125,
          "decoder_belief_target_mass": 0.022807827219367027,
          "decoder_belief_target_nll": 4.811180114746094,
          "decoder_belief_top1_on_support": 0.0234375,
          "mean_decoder_belief_support_size": 2.73828125
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 9,
          "query_type": "A",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.2884332537651062,
          "query_target_nll": 3.3322834968566895,
          "query_top1_on_support": 0.27734375,
          "mean_query_support_size": 2.40625,
          "probe_belief_target_mass": 0.047394733875989914,
          "probe_belief_target_nll": 8.113056182861328,
          "probe_belief_top1_on_support": 0.05859375,
          "mean_belief_support_size": 2.73828125,
          "decoder_belief_target_mass": 0.02264590747654438,
          "decoder_belief_target_nll": 4.8244524002075195,
          "decoder_belief_top1_on_support": 0.0234375,
          "mean_decoder_belief_support_size": 2.73828125
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 9,
          "query_type": "A",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.4096178412437439,
          "query_target_nll": 1.7995884418487549,
          "query_top1_on_support": 0.4765625,
          "mean_query_support_size": 2.40625,
          "probe_belief_target_mass": 0.15215972065925598,
          "probe_belief_target_nll": 3.18841552734375,
          "probe_belief_top1_on_support": 0.2265625,
          "mean_belief_support_size": 2.73828125,
          "decoder_belief_target_mass": 0.022692177444696426,
          "decoder_belief_target_nll": 4.82077693939209,
          "decoder_belief_top1_on_support": 0.01953125,
          "mean_decoder_belief_support_size": 2.73828125
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 9,
          "query_type": "A",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.4096178412437439,
          "query_target_nll": 1.7995884418487549,
          "query_top1_on_support": 0.4765625,
          "mean_query_support_size": 2.40625,
          "probe_belief_target_mass": 0.15215972065925598,
          "probe_belief_target_nll": 3.18841552734375,
          "probe_belief_top1_on_support": 0.2265625,
          "mean_belief_support_size": 2.73828125,
          "decoder_belief_target_mass": 0.022692177444696426,
          "decoder_belief_target_nll": 4.82077693939209,
          "decoder_belief_top1_on_support": 0.01953125,
          "mean_decoder_belief_support_size": 2.73828125
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 9,
          "query_type": "B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.187042236328125,
          "query_target_nll": 2.643123149871826,
          "query_top1_on_support": 0.15625,
          "mean_query_support_size": 2.1171875,
          "probe_belief_target_mass": 0.02537365071475506,
          "probe_belief_target_nll": 18.19559097290039,
          "probe_belief_top1_on_support": 0.04296875,
          "mean_belief_support_size": 2.37109375,
          "decoder_belief_target_mass": 0.019422966986894608,
          "decoder_belief_target_nll": 4.821810722351074,
          "decoder_belief_top1_on_support": 0.0234375,
          "mean_decoder_belief_support_size": 2.37109375
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 9,
          "query_type": "B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.1920313984155655,
          "query_target_nll": 3.9965310096740723,
          "query_top1_on_support": 0.16015625,
          "mean_query_support_size": 2.1171875,
          "probe_belief_target_mass": 0.02388213574886322,
          "probe_belief_target_nll": 14.96171760559082,
          "probe_belief_top1_on_support": 0.01953125,
          "mean_belief_support_size": 2.37109375,
          "decoder_belief_target_mass": 0.019419748336076736,
          "decoder_belief_target_nll": 4.826468467712402,
          "decoder_belief_top1_on_support": 0.0234375,
          "mean_decoder_belief_support_size": 2.37109375
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 9,
          "query_type": "B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.20821303129196167,
          "query_target_nll": 3.8996198177337646,
          "query_top1_on_support": 0.20703125,
          "mean_query_support_size": 2.1171875,
          "probe_belief_target_mass": 0.025061389431357384,
          "probe_belief_target_nll": 11.451470375061035,
          "probe_belief_top1_on_support": 0.0234375,
          "mean_belief_support_size": 2.37109375,
          "decoder_belief_target_mass": 0.019575731828808784,
          "decoder_belief_target_nll": 4.816307067871094,
          "decoder_belief_top1_on_support": 0.01953125,
          "mean_decoder_belief_support_size": 2.37109375
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 9,
          "query_type": "B",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.2068260908126831,
          "query_target_nll": 4.03657341003418,
          "query_top1_on_support": 0.20703125,
          "mean_query_support_size": 2.1171875,
          "probe_belief_target_mass": 0.02668747678399086,
          "probe_belief_target_nll": 10.184585571289062,
          "probe_belief_top1_on_support": 0.03125,
          "mean_belief_support_size": 2.37109375,
          "decoder_belief_target_mass": 0.0195605531334877,
          "decoder_belief_target_nll": 4.81576681137085,
          "decoder_belief_top1_on_support": 0.01171875,
          "mean_decoder_belief_support_size": 2.37109375
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 9,
          "query_type": "B",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.24243319034576416,
          "query_target_nll": 3.72904109954834,
          "query_top1_on_support": 0.296875,
          "mean_query_support_size": 2.1171875,
          "probe_belief_target_mass": 0.03555576875805855,
          "probe_belief_target_nll": 8.720361709594727,
          "probe_belief_top1_on_support": 0.046875,
          "mean_belief_support_size": 2.37109375,
          "decoder_belief_target_mass": 0.019475743174552917,
          "decoder_belief_target_nll": 4.817638874053955,
          "decoder_belief_top1_on_support": 0.015625,
          "mean_decoder_belief_support_size": 2.37109375
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 9,
          "query_type": "B",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.38550686836242676,
          "query_target_nll": 1.8169198036193848,
          "query_top1_on_support": 0.421875,
          "mean_query_support_size": 2.1171875,
          "probe_belief_target_mass": 0.12753993272781372,
          "probe_belief_target_nll": 3.3703103065490723,
          "probe_belief_top1_on_support": 0.16796875,
          "mean_belief_support_size": 2.37109375,
          "decoder_belief_target_mass": 0.019435234367847443,
          "decoder_belief_target_nll": 4.823925971984863,
          "decoder_belief_top1_on_support": 0.015625,
          "mean_decoder_belief_support_size": 2.37109375
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 9,
          "query_type": "B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.38550686836242676,
          "query_target_nll": 1.8169198036193848,
          "query_top1_on_support": 0.421875,
          "mean_query_support_size": 2.1171875,
          "probe_belief_target_mass": 0.12753993272781372,
          "probe_belief_target_nll": 3.3703103065490723,
          "probe_belief_top1_on_support": 0.16796875,
          "mean_belief_support_size": 2.37109375,
          "decoder_belief_target_mass": 0.019435234367847443,
          "decoder_belief_target_nll": 4.823925971984863,
          "decoder_belief_top1_on_support": 0.015625,
          "mean_decoder_belief_support_size": 2.37109375
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 9,
          "query_type": "A_PLUS_B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.2212672084569931,
          "query_target_nll": 2.6769633293151855,
          "query_top1_on_support": 0.19140625,
          "mean_query_support_size": 2.50390625,
          "probe_belief_target_mass": 0.022219959646463394,
          "probe_belief_target_nll": 18.709083557128906,
          "probe_belief_top1_on_support": 0.015625,
          "mean_belief_support_size": 2.71875,
          "decoder_belief_target_mass": 0.02258916199207306,
          "decoder_belief_target_nll": 4.803956031799316,
          "decoder_belief_top1_on_support": 0.01953125,
          "mean_decoder_belief_support_size": 2.71875
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 9,
          "query_type": "A_PLUS_B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.2334207445383072,
          "query_target_nll": 2.7948927879333496,
          "query_top1_on_support": 0.2578125,
          "mean_query_support_size": 2.50390625,
          "probe_belief_target_mass": 0.03061053715646267,
          "probe_belief_target_nll": 14.824993133544922,
          "probe_belief_top1_on_support": 0.04296875,
          "mean_belief_support_size": 2.71875,
          "decoder_belief_target_mass": 0.022648885846138,
          "decoder_belief_target_nll": 4.803682327270508,
          "decoder_belief_top1_on_support": 0.03515625,
          "mean_decoder_belief_support_size": 2.71875
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 9,
          "query_type": "A_PLUS_B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.21801359951496124,
          "query_target_nll": 2.9401350021362305,
          "query_top1_on_support": 0.2265625,
          "mean_query_support_size": 2.50390625,
          "probe_belief_target_mass": 0.028337232768535614,
          "probe_belief_target_nll": 11.448262214660645,
          "probe_belief_top1_on_support": 0.04296875,
          "mean_belief_support_size": 2.71875,
          "decoder_belief_target_mass": 0.022706113755702972,
          "decoder_belief_target_nll": 4.805651664733887,
          "decoder_belief_top1_on_support": 0.03125,
          "mean_decoder_belief_support_size": 2.71875
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 9,
          "query_type": "A_PLUS_B",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.21944785118103027,
          "query_target_nll": 2.976011276245117,
          "query_top1_on_support": 0.20703125,
          "mean_query_support_size": 2.50390625,
          "probe_belief_target_mass": 0.038333356380462646,
          "probe_belief_target_nll": 9.696935653686523,
          "probe_belief_top1_on_support": 0.04296875,
          "mean_belief_support_size": 2.71875,
          "decoder_belief_target_mass": 0.022302065044641495,
          "decoder_belief_target_nll": 4.826377868652344,
          "decoder_belief_top1_on_support": 0.01171875,
          "mean_decoder_belief_support_size": 2.71875
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 9,
          "query_type": "A_PLUS_B",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.22663217782974243,
          "query_target_nll": 2.807781219482422,
          "query_top1_on_support": 0.21875,
          "mean_query_support_size": 2.50390625,
          "probe_belief_target_mass": 0.041166823357343674,
          "probe_belief_target_nll": 7.926517486572266,
          "probe_belief_top1_on_support": 0.05078125,
          "mean_belief_support_size": 2.71875,
          "decoder_belief_target_mass": 0.02256845124065876,
          "decoder_belief_target_nll": 4.812149524688721,
          "decoder_belief_top1_on_support": 0.01171875,
          "mean_decoder_belief_support_size": 2.71875
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 9,
          "query_type": "A_PLUS_B",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.30634236335754395,
          "query_target_nll": 2.1553287506103516,
          "query_top1_on_support": 0.37109375,
          "mean_query_support_size": 2.50390625,
          "probe_belief_target_mass": 0.14063134789466858,
          "probe_belief_target_nll": 3.3321306705474854,
          "probe_belief_top1_on_support": 0.16796875,
          "mean_belief_support_size": 2.71875,
          "decoder_belief_target_mass": 0.022284358739852905,
          "decoder_belief_target_nll": 4.82647705078125,
          "decoder_belief_top1_on_support": 0.015625,
          "mean_decoder_belief_support_size": 2.71875
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 9,
          "query_type": "A_PLUS_B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.30634236335754395,
          "query_target_nll": 2.1553287506103516,
          "query_top1_on_support": 0.37109375,
          "mean_query_support_size": 2.50390625,
          "probe_belief_target_mass": 0.14063134789466858,
          "probe_belief_target_nll": 3.3321306705474854,
          "probe_belief_top1_on_support": 0.16796875,
          "mean_belief_support_size": 2.71875,
          "decoder_belief_target_mass": 0.022284358739852905,
          "decoder_belief_target_nll": 4.82647705078125,
          "decoder_belief_top1_on_support": 0.015625,
          "mean_decoder_belief_support_size": 2.71875
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 9,
          "query_type": "A_MINUS_B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.1837502419948578,
          "query_target_nll": 10.42789077758789,
          "query_top1_on_support": 0.18359375,
          "mean_query_support_size": 2.35546875,
          "probe_belief_target_mass": 0.0176092516630888,
          "probe_belief_target_nll": 18.891162872314453,
          "probe_belief_top1_on_support": 0.0234375,
          "mean_belief_support_size": 2.48046875,
          "decoder_belief_target_mass": 0.020732248201966286,
          "decoder_belief_target_nll": 4.8164825439453125,
          "decoder_belief_top1_on_support": 0.0625,
          "mean_decoder_belief_support_size": 2.48046875
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 9,
          "query_type": "A_MINUS_B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.21106088161468506,
          "query_target_nll": 4.678603172302246,
          "query_top1_on_support": 0.19921875,
          "mean_query_support_size": 2.35546875,
          "probe_belief_target_mass": 0.021550804376602173,
          "probe_belief_target_nll": 14.448516845703125,
          "probe_belief_top1_on_support": 0.02734375,
          "mean_belief_support_size": 2.48046875,
          "decoder_belief_target_mass": 0.020411673933267593,
          "decoder_belief_target_nll": 4.828613758087158,
          "decoder_belief_top1_on_support": 0.015625,
          "mean_decoder_belief_support_size": 2.48046875
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 9,
          "query_type": "A_MINUS_B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.2094467729330063,
          "query_target_nll": 3.778968334197998,
          "query_top1_on_support": 0.20703125,
          "mean_query_support_size": 2.35546875,
          "probe_belief_target_mass": 0.021304331719875336,
          "probe_belief_target_nll": 11.594578742980957,
          "probe_belief_top1_on_support": 0.01171875,
          "mean_belief_support_size": 2.48046875,
          "decoder_belief_target_mass": 0.020438559353351593,
          "decoder_belief_target_nll": 4.819063186645508,
          "decoder_belief_top1_on_support": 0.02734375,
          "mean_decoder_belief_support_size": 2.48046875
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 9,
          "query_type": "A_MINUS_B",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.21160031855106354,
          "query_target_nll": 3.6183226108551025,
          "query_top1_on_support": 0.1953125,
          "mean_query_support_size": 2.35546875,
          "probe_belief_target_mass": 0.02290727011859417,
          "probe_belief_target_nll": 10.617734909057617,
          "probe_belief_top1_on_support": 0.0234375,
          "mean_belief_support_size": 2.48046875,
          "decoder_belief_target_mass": 0.020567700266838074,
          "decoder_belief_target_nll": 4.813740253448486,
          "decoder_belief_top1_on_support": 0.01953125,
          "mean_decoder_belief_support_size": 2.48046875
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 9,
          "query_type": "A_MINUS_B",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.22157636284828186,
          "query_target_nll": 3.1633872985839844,
          "query_top1_on_support": 0.21875,
          "mean_query_support_size": 2.35546875,
          "probe_belief_target_mass": 0.03514540567994118,
          "probe_belief_target_nll": 8.608607292175293,
          "probe_belief_top1_on_support": 0.0390625,
          "mean_belief_support_size": 2.48046875,
          "decoder_belief_target_mass": 0.020478615537285805,
          "decoder_belief_target_nll": 4.810596466064453,
          "decoder_belief_top1_on_support": 0.015625,
          "mean_decoder_belief_support_size": 2.48046875
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 9,
          "query_type": "A_MINUS_B",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.3293148875236511,
          "query_target_nll": 2.089789867401123,
          "query_top1_on_support": 0.40625,
          "mean_query_support_size": 2.35546875,
          "probe_belief_target_mass": 0.1276468187570572,
          "probe_belief_target_nll": 3.457775354385376,
          "probe_belief_top1_on_support": 0.171875,
          "mean_belief_support_size": 2.48046875,
          "decoder_belief_target_mass": 0.020315252244472504,
          "decoder_belief_target_nll": 4.822732925415039,
          "decoder_belief_top1_on_support": 0.01171875,
          "mean_decoder_belief_support_size": 2.48046875
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 9,
          "query_type": "A_MINUS_B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.3293148875236511,
          "query_target_nll": 2.089789867401123,
          "query_top1_on_support": 0.40625,
          "mean_query_support_size": 2.35546875,
          "probe_belief_target_mass": 0.1276468187570572,
          "probe_belief_target_nll": 3.457775354385376,
          "probe_belief_top1_on_support": 0.171875,
          "mean_belief_support_size": 2.48046875,
          "decoder_belief_target_mass": 0.020315252244472504,
          "decoder_belief_target_nll": 4.822732925415039,
          "decoder_belief_top1_on_support": 0.01171875,
          "mean_decoder_belief_support_size": 2.48046875
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 12,
          "query_type": "A",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.17213335633277893,
          "query_target_nll": 2.6157050132751465,
          "query_top1_on_support": 0.1484375,
          "mean_query_support_size": 1.90625,
          "probe_belief_target_mass": 0.01814807951450348,
          "probe_belief_target_nll": 18.293428421020508,
          "probe_belief_top1_on_support": 0.01171875,
          "mean_belief_support_size": 2.21484375,
          "decoder_belief_target_mass": 0.018473032861948013,
          "decoder_belief_target_nll": 4.794489860534668,
          "decoder_belief_top1_on_support": 0.015625,
          "mean_decoder_belief_support_size": 2.21484375
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 12,
          "query_type": "A",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.19660162925720215,
          "query_target_nll": 3.6950762271881104,
          "query_top1_on_support": 0.2421875,
          "mean_query_support_size": 1.90625,
          "probe_belief_target_mass": 0.029951199889183044,
          "probe_belief_target_nll": 14.598289489746094,
          "probe_belief_top1_on_support": 0.03515625,
          "mean_belief_support_size": 2.21484375,
          "decoder_belief_target_mass": 0.018585417419672012,
          "decoder_belief_target_nll": 4.79763126373291,
          "decoder_belief_top1_on_support": 0.01953125,
          "mean_decoder_belief_support_size": 2.21484375
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 12,
          "query_type": "A",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.18879713118076324,
          "query_target_nll": 4.030126094818115,
          "query_top1_on_support": 0.20703125,
          "mean_query_support_size": 1.90625,
          "probe_belief_target_mass": 0.021227166056632996,
          "probe_belief_target_nll": 12.403993606567383,
          "probe_belief_top1_on_support": 0.03515625,
          "mean_belief_support_size": 2.21484375,
          "decoder_belief_target_mass": 0.018348654732108116,
          "decoder_belief_target_nll": 4.796041011810303,
          "decoder_belief_top1_on_support": 0.015625,
          "mean_decoder_belief_support_size": 2.21484375
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 12,
          "query_type": "A",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.1864052712917328,
          "query_target_nll": 4.046446800231934,
          "query_top1_on_support": 0.23046875,
          "mean_query_support_size": 1.90625,
          "probe_belief_target_mass": 0.022054610773921013,
          "probe_belief_target_nll": 10.551101684570312,
          "probe_belief_top1_on_support": 0.03515625,
          "mean_belief_support_size": 2.21484375,
          "decoder_belief_target_mass": 0.018369581550359726,
          "decoder_belief_target_nll": 4.804891109466553,
          "decoder_belief_top1_on_support": 0.03125,
          "mean_decoder_belief_support_size": 2.21484375
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 12,
          "query_type": "A",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.18097689747810364,
          "query_target_nll": 3.809791088104248,
          "query_top1_on_support": 0.1796875,
          "mean_query_support_size": 1.90625,
          "probe_belief_target_mass": 0.02383875660598278,
          "probe_belief_target_nll": 9.013718605041504,
          "probe_belief_top1_on_support": 0.02734375,
          "mean_belief_support_size": 2.21484375,
          "decoder_belief_target_mass": 0.018426213413476944,
          "decoder_belief_target_nll": 4.801424980163574,
          "decoder_belief_top1_on_support": 0.015625,
          "mean_decoder_belief_support_size": 2.21484375
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 12,
          "query_type": "A",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.21209770441055298,
          "query_target_nll": 3.2429018020629883,
          "query_top1_on_support": 0.20703125,
          "mean_query_support_size": 1.90625,
          "probe_belief_target_mass": 0.032761428505182266,
          "probe_belief_target_nll": 8.01764965057373,
          "probe_belief_top1_on_support": 0.03125,
          "mean_belief_support_size": 2.21484375,
          "decoder_belief_target_mass": 0.018467506393790245,
          "decoder_belief_target_nll": 4.800258636474609,
          "decoder_belief_top1_on_support": 0.015625,
          "mean_decoder_belief_support_size": 2.21484375
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 12,
          "query_type": "A",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.33341777324676514,
          "query_target_nll": 1.8395557403564453,
          "query_top1_on_support": 0.3984375,
          "mean_query_support_size": 1.90625,
          "probe_belief_target_mass": 0.10366392880678177,
          "probe_belief_target_nll": 3.570664405822754,
          "probe_belief_top1_on_support": 0.12890625,
          "mean_belief_support_size": 2.21484375,
          "decoder_belief_target_mass": 0.018307846039533615,
          "decoder_belief_target_nll": 4.808688163757324,
          "decoder_belief_top1_on_support": 0.01171875,
          "mean_decoder_belief_support_size": 2.21484375
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 12,
          "query_type": "B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.14917227625846863,
          "query_target_nll": 2.588888168334961,
          "query_top1_on_support": 0.14453125,
          "mean_query_support_size": 1.6796875,
          "probe_belief_target_mass": 0.01332557201385498,
          "probe_belief_target_nll": 18.4035587310791,
          "probe_belief_top1_on_support": 0.015625,
          "mean_belief_support_size": 1.84765625,
          "decoder_belief_target_mass": 0.01537800207734108,
          "decoder_belief_target_nll": 4.807626724243164,
          "decoder_belief_top1_on_support": 0.0078125,
          "mean_decoder_belief_support_size": 1.84765625
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 12,
          "query_type": "B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.15009969472885132,
          "query_target_nll": 3.7110280990600586,
          "query_top1_on_support": 0.15234375,
          "mean_query_support_size": 1.6796875,
          "probe_belief_target_mass": 0.015110809355974197,
          "probe_belief_target_nll": 14.8231201171875,
          "probe_belief_top1_on_support": 0.01171875,
          "mean_belief_support_size": 1.84765625,
          "decoder_belief_target_mass": 0.015284418128430843,
          "decoder_belief_target_nll": 4.8119072914123535,
          "decoder_belief_top1_on_support": 0.01953125,
          "mean_decoder_belief_support_size": 1.84765625
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 12,
          "query_type": "B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.17008331418037415,
          "query_target_nll": 3.9272727966308594,
          "query_top1_on_support": 0.1796875,
          "mean_query_support_size": 1.6796875,
          "probe_belief_target_mass": 0.019535906612873077,
          "probe_belief_target_nll": 11.380084037780762,
          "probe_belief_top1_on_support": 0.01953125,
          "mean_belief_support_size": 1.84765625,
          "decoder_belief_target_mass": 0.015215076506137848,
          "decoder_belief_target_nll": 4.824094295501709,
          "decoder_belief_top1_on_support": 0.01953125,
          "mean_decoder_belief_support_size": 1.84765625
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 12,
          "query_type": "B",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.16309255361557007,
          "query_target_nll": 4.017511367797852,
          "query_top1_on_support": 0.18359375,
          "mean_query_support_size": 1.6796875,
          "probe_belief_target_mass": 0.012974375858902931,
          "probe_belief_target_nll": 10.471226692199707,
          "probe_belief_top1_on_support": 0.0078125,
          "mean_belief_support_size": 1.84765625,
          "decoder_belief_target_mass": 0.015122308395802975,
          "decoder_belief_target_nll": 4.820509910583496,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 1.84765625
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 12,
          "query_type": "B",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.15170177817344666,
          "query_target_nll": 3.9471468925476074,
          "query_top1_on_support": 0.16015625,
          "mean_query_support_size": 1.6796875,
          "probe_belief_target_mass": 0.016546104103326797,
          "probe_belief_target_nll": 9.030381202697754,
          "probe_belief_top1_on_support": 0.015625,
          "mean_belief_support_size": 1.84765625,
          "decoder_belief_target_mass": 0.015201719477772713,
          "decoder_belief_target_nll": 4.813813209533691,
          "decoder_belief_top1_on_support": 0.0078125,
          "mean_decoder_belief_support_size": 1.84765625
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 12,
          "query_type": "B",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.1966700404882431,
          "query_target_nll": 3.2246105670928955,
          "query_top1_on_support": 0.19140625,
          "mean_query_support_size": 1.6796875,
          "probe_belief_target_mass": 0.029360651969909668,
          "probe_belief_target_nll": 7.613982677459717,
          "probe_belief_top1_on_support": 0.03125,
          "mean_belief_support_size": 1.84765625,
          "decoder_belief_target_mass": 0.015323900617659092,
          "decoder_belief_target_nll": 4.811687469482422,
          "decoder_belief_top1_on_support": 0.015625,
          "mean_decoder_belief_support_size": 1.84765625
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 12,
          "query_type": "B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.3197818994522095,
          "query_target_nll": 1.8510003089904785,
          "query_top1_on_support": 0.375,
          "mean_query_support_size": 1.6796875,
          "probe_belief_target_mass": 0.09407847374677658,
          "probe_belief_target_nll": 3.4600493907928467,
          "probe_belief_top1_on_support": 0.1328125,
          "mean_belief_support_size": 1.84765625,
          "decoder_belief_target_mass": 0.015031401067972183,
          "decoder_belief_target_nll": 4.83796501159668,
          "decoder_belief_top1_on_support": 0.015625,
          "mean_decoder_belief_support_size": 1.84765625
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 12,
          "query_type": "A_PLUS_B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.1775628924369812,
          "query_target_nll": 2.6373372077941895,
          "query_top1_on_support": 0.19140625,
          "mean_query_support_size": 1.93359375,
          "probe_belief_target_mass": 0.02555187977850437,
          "probe_belief_target_nll": 17.792404174804688,
          "probe_belief_top1_on_support": 0.03125,
          "mean_belief_support_size": 2.0,
          "decoder_belief_target_mass": 0.016588382422924042,
          "decoder_belief_target_nll": 4.809507369995117,
          "decoder_belief_top1_on_support": 0.015625,
          "mean_decoder_belief_support_size": 2.0
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 12,
          "query_type": "A_PLUS_B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.1850622147321701,
          "query_target_nll": 2.8068935871124268,
          "query_top1_on_support": 0.1953125,
          "mean_query_support_size": 1.93359375,
          "probe_belief_target_mass": 0.015222523361444473,
          "probe_belief_target_nll": 15.046089172363281,
          "probe_belief_top1_on_support": 0.015625,
          "mean_belief_support_size": 2.0,
          "decoder_belief_target_mass": 0.016497455537319183,
          "decoder_belief_target_nll": 4.803901672363281,
          "decoder_belief_top1_on_support": 0.01171875,
          "mean_decoder_belief_support_size": 2.0
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 12,
          "query_type": "A_PLUS_B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.17088115215301514,
          "query_target_nll": 2.8475587368011475,
          "query_top1_on_support": 0.14453125,
          "mean_query_support_size": 1.93359375,
          "probe_belief_target_mass": 0.021190352737903595,
          "probe_belief_target_nll": 12.095610618591309,
          "probe_belief_top1_on_support": 0.02734375,
          "mean_belief_support_size": 2.0,
          "decoder_belief_target_mass": 0.01668158918619156,
          "decoder_belief_target_nll": 4.797701358795166,
          "decoder_belief_top1_on_support": 0.015625,
          "mean_decoder_belief_support_size": 2.0
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 12,
          "query_type": "A_PLUS_B",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.16595637798309326,
          "query_target_nll": 2.87034273147583,
          "query_top1_on_support": 0.13671875,
          "mean_query_support_size": 1.93359375,
          "probe_belief_target_mass": 0.01451747864484787,
          "probe_belief_target_nll": 10.972213745117188,
          "probe_belief_top1_on_support": 0.01171875,
          "mean_belief_support_size": 2.0,
          "decoder_belief_target_mass": 0.01663852483034134,
          "decoder_belief_target_nll": 4.7952046394348145,
          "decoder_belief_top1_on_support": 0.03515625,
          "mean_decoder_belief_support_size": 2.0
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 12,
          "query_type": "A_PLUS_B",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.17773905396461487,
          "query_target_nll": 2.8428449630737305,
          "query_top1_on_support": 0.15625,
          "mean_query_support_size": 1.93359375,
          "probe_belief_target_mass": 0.01902991533279419,
          "probe_belief_target_nll": 9.12874984741211,
          "probe_belief_top1_on_support": 0.015625,
          "mean_belief_support_size": 2.0,
          "decoder_belief_target_mass": 0.0165302325040102,
          "decoder_belief_target_nll": 4.802526473999023,
          "decoder_belief_top1_on_support": 0.01171875,
          "mean_decoder_belief_support_size": 2.0
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 12,
          "query_type": "A_PLUS_B",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.18126492202281952,
          "query_target_nll": 2.7917003631591797,
          "query_top1_on_support": 0.21875,
          "mean_query_support_size": 1.93359375,
          "probe_belief_target_mass": 0.029317542910575867,
          "probe_belief_target_nll": 7.914249897003174,
          "probe_belief_top1_on_support": 0.01953125,
          "mean_belief_support_size": 2.0,
          "decoder_belief_target_mass": 0.016492949798703194,
          "decoder_belief_target_nll": 4.811379432678223,
          "decoder_belief_top1_on_support": 0.02734375,
          "mean_decoder_belief_support_size": 2.0
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 12,
          "query_type": "A_PLUS_B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.23579445481300354,
          "query_target_nll": 2.1747233867645264,
          "query_top1_on_support": 0.25390625,
          "mean_query_support_size": 1.93359375,
          "probe_belief_target_mass": 0.10347418487071991,
          "probe_belief_target_nll": 3.4609148502349854,
          "probe_belief_top1_on_support": 0.1640625,
          "mean_belief_support_size": 2.0,
          "decoder_belief_target_mass": 0.016607286408543587,
          "decoder_belief_target_nll": 4.7909111976623535,
          "decoder_belief_top1_on_support": 0.015625,
          "mean_decoder_belief_support_size": 2.0
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 12,
          "query_type": "A_MINUS_B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.2146395444869995,
          "query_target_nll": 10.113341331481934,
          "query_top1_on_support": 0.21484375,
          "mean_query_support_size": 1.95703125,
          "probe_belief_target_mass": 0.02035471796989441,
          "probe_belief_target_nll": 17.853694915771484,
          "probe_belief_top1_on_support": 0.015625,
          "mean_belief_support_size": 2.17578125,
          "decoder_belief_target_mass": 0.018026262521743774,
          "decoder_belief_target_nll": 4.821809768676758,
          "decoder_belief_top1_on_support": 0.01171875,
          "mean_decoder_belief_support_size": 2.17578125
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 12,
          "query_type": "A_MINUS_B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.17071972787380219,
          "query_target_nll": 5.023070335388184,
          "query_top1_on_support": 0.1640625,
          "mean_query_support_size": 1.95703125,
          "probe_belief_target_mass": 0.015676062554121017,
          "probe_belief_target_nll": 15.495765686035156,
          "probe_belief_top1_on_support": 0.01953125,
          "mean_belief_support_size": 2.17578125,
          "decoder_belief_target_mass": 0.017709707841277122,
          "decoder_belief_target_nll": 4.827645301818848,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 2.17578125
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 12,
          "query_type": "A_MINUS_B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.16195760667324066,
          "query_target_nll": 3.8412365913391113,
          "query_top1_on_support": 0.12109375,
          "mean_query_support_size": 1.95703125,
          "probe_belief_target_mass": 0.017613016068935394,
          "probe_belief_target_nll": 12.247017860412598,
          "probe_belief_top1_on_support": 0.0078125,
          "mean_belief_support_size": 2.17578125,
          "decoder_belief_target_mass": 0.017834849655628204,
          "decoder_belief_target_nll": 4.827598571777344,
          "decoder_belief_top1_on_support": 0.00390625,
          "mean_decoder_belief_support_size": 2.17578125
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 12,
          "query_type": "A_MINUS_B",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.1692376732826233,
          "query_target_nll": 3.522169351577759,
          "query_top1_on_support": 0.140625,
          "mean_query_support_size": 1.95703125,
          "probe_belief_target_mass": 0.015027519315481186,
          "probe_belief_target_nll": 10.596442222595215,
          "probe_belief_top1_on_support": 0.01953125,
          "mean_belief_support_size": 2.17578125,
          "decoder_belief_target_mass": 0.018133491277694702,
          "decoder_belief_target_nll": 4.813499450683594,
          "decoder_belief_top1_on_support": 0.01171875,
          "mean_decoder_belief_support_size": 2.17578125
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 12,
          "query_type": "A_MINUS_B",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.18105213344097137,
          "query_target_nll": 3.3539137840270996,
          "query_top1_on_support": 0.14453125,
          "mean_query_support_size": 1.95703125,
          "probe_belief_target_mass": 0.015564126893877983,
          "probe_belief_target_nll": 9.673836708068848,
          "probe_belief_top1_on_support": 0.01171875,
          "mean_belief_support_size": 2.17578125,
          "decoder_belief_target_mass": 0.01780867390334606,
          "decoder_belief_target_nll": 4.828852653503418,
          "decoder_belief_top1_on_support": 0.01171875,
          "mean_decoder_belief_support_size": 2.17578125
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 12,
          "query_type": "A_MINUS_B",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.1804182231426239,
          "query_target_nll": 3.2168960571289062,
          "query_top1_on_support": 0.1875,
          "mean_query_support_size": 1.95703125,
          "probe_belief_target_mass": 0.029245322570204735,
          "probe_belief_target_nll": 7.815713882446289,
          "probe_belief_top1_on_support": 0.02734375,
          "mean_belief_support_size": 2.17578125,
          "decoder_belief_target_mass": 0.018030963838100433,
          "decoder_belief_target_nll": 4.806764602661133,
          "decoder_belief_top1_on_support": 0.015625,
          "mean_decoder_belief_support_size": 2.17578125
        },
        {
          "model": "dense",
          "supervision": "sampled_final",
          "length": 12,
          "query_type": "A_MINUS_B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.26854419708251953,
          "query_target_nll": 2.1574888229370117,
          "query_top1_on_support": 0.296875,
          "mean_query_support_size": 1.95703125,
          "probe_belief_target_mass": 0.09589309990406036,
          "probe_belief_target_nll": 3.5952014923095703,
          "probe_belief_top1_on_support": 0.14453125,
          "mean_belief_support_size": 2.17578125,
          "decoder_belief_target_mass": 0.01805247738957405,
          "decoder_belief_target_nll": 4.80674934387207,
          "decoder_belief_top1_on_support": 0.0234375,
          "mean_decoder_belief_support_size": 2.17578125
        }
      ]
    }
  ],
  "checkpoints": [
    "large_artifacts/dense_supervision_ladder/checkpoints/pilot_sampled_final_mod11/checkpoint_final.pt"
  ]
}