{
  "mode": "dense",
  "variant": "pilot_gru_mlp_d128",
  "transition": "gru",
  "decoder_type": "mlp",
  "args": {
    "variant_name": "pilot_gru_mlp_d128",
    "mode": "dense",
    "supervision": "teacher_belief",
    "seed": 53,
    "modulus": 11,
    "observe_mod": 4,
    "observe_prob": 0.3,
    "op_family": "full",
    "query_family": "all",
    "train_min_len": 1,
    "train_max_len": 6,
    "eval_lengths": "3,6,9,12",
    "eval_k": "0,1,2,3,6,9,12",
    "eval_query_types": "all",
    "state_dim": 128,
    "instr_dim": 96,
    "transition": "gru",
    "decoder_type": "mlp",
    "decoder_rank": 16,
    "heads": 4,
    "compiler_layers": 2,
    "dropout": 0.0,
    "train_steps": 800,
    "batch_size": 256,
    "eval_batch_size": 256,
    "eval_examples": 256,
    "lr": 0.001,
    "weight_decay": 0.0001,
    "query_loss_weight": 0.0,
    "belief_loss_weight": 1.0,
    "decoder_query_loss_weight": 0.0,
    "belief_distill_prob": 1.0,
    "probe_steps": 400,
    "probe_batch_size": 256,
    "probe_lr": 0.001,
    "probe_weight_decay": 0.0001,
    "probe_dropout": 0.0,
    "max_grad_norm": 1.0,
    "log_every": 100,
    "probe_log_every": 100,
    "output_dir": "experiments/dense_teacher_distillation/runs/pilot_gru_mlp_d128",
    "checkpoint_dir": "large_artifacts/dense_teacher_distillation/checkpoints/pilot_gru_mlp_d128"
  },
  "env": {
    "python": "3.12.3",
    "torch": "2.8.0+cu128",
    "device": "cuda",
    "cuda_name": "NVIDIA RTX 6000 Ada Generation"
  },
  "train": [
    {
      "step": 1,
      "loss": 4.824020862579346,
      "grad_norm": 0.2694411277770996,
      "query_loss": 0.0,
      "belief_loss": 4.824020862579346,
      "decoder_query_loss": 0.0,
      "query_target_mass": 0.42348405718803406,
      "query_target_nll": 2.4182846546173096,
      "query_top1_on_support": 0.45703125,
      "mean_query_support_size": 4.65625,
      "decoder_belief_target_mass": 0.04856359213590622,
      "decoder_belief_target_nll": 4.74271297454834,
      "decoder_belief_top1_on_support": 0.08984375,
      "mean_decoder_belief_support_size": 5.6953125,
      "decoder_query_target_mass": 0.4252757430076599,
      "decoder_query_target_nll": 2.3862953186035156,
      "decoder_query_top1_on_support": 0.48046875,
      "mean_decoder_query_support_size": 4.65625
    },
    {
      "step": 100,
      "loss": 3.4188308715820312,
      "grad_norm": 0.3595808446407318,
      "query_loss": 0.0,
      "belief_loss": 3.4188308715820312,
      "decoder_query_loss": 0.0,
      "query_target_mass": 0.4310794472694397,
      "query_target_nll": 2.4121885299682617,
      "query_top1_on_support": 0.44140625,
      "mean_query_support_size": 4.73046875,
      "decoder_belief_target_mass": 0.20258992910385132,
      "decoder_belief_target_nll": 3.8955183029174805,
      "decoder_belief_top1_on_support": 0.25390625,
      "mean_decoder_belief_support_size": 5.9453125,
      "decoder_query_target_mass": 0.5555323362350464,
      "decoder_query_target_nll": 2.102229356765747,
      "decoder_query_top1_on_support": 0.61328125,
      "mean_decoder_query_support_size": 4.73046875
    },
    {
      "step": 200,
      "loss": 3.1687874794006348,
      "grad_norm": 0.5588650107383728,
      "query_loss": 0.0,
      "belief_loss": 3.1687874794006348,
      "decoder_query_loss": 0.0,
      "query_target_mass": 0.4703977704048157,
      "query_target_nll": 2.4346961975097656,
      "query_top1_on_support": 0.4453125,
      "mean_query_support_size": 5.20703125,
      "decoder_belief_target_mass": 0.26875996589660645,
      "decoder_belief_target_nll": 3.4790453910827637,
      "decoder_belief_top1_on_support": 0.34765625,
      "mean_decoder_belief_support_size": 5.75390625,
      "decoder_query_target_mass": 0.6457288265228271,
      "decoder_query_target_nll": 1.9907749891281128,
      "decoder_query_top1_on_support": 0.703125,
      "mean_decoder_query_support_size": 5.20703125
    },
    {
      "step": 300,
      "loss": 2.7569162845611572,
      "grad_norm": 0.7534188032150269,
      "query_loss": 0.0,
      "belief_loss": 2.7569162845611572,
      "decoder_query_loss": 0.0,
      "query_target_mass": 0.4488672912120819,
      "query_target_nll": 2.4259567260742188,
      "query_top1_on_support": 0.44921875,
      "mean_query_support_size": 4.94921875,
      "decoder_belief_target_mass": 0.4028274714946747,
      "decoder_belief_target_nll": 3.0358669757843018,
      "decoder_belief_top1_on_support": 0.5078125,
      "mean_decoder_belief_support_size": 5.81640625,
      "decoder_query_target_mass": 0.7049400806427002,
      "decoder_query_target_nll": 1.7874131202697754,
      "decoder_query_top1_on_support": 0.78515625,
      "mean_decoder_query_support_size": 4.94921875
    },
    {
      "step": 400,
      "loss": 2.4260783195495605,
      "grad_norm": 0.5603106617927551,
      "query_loss": 0.0,
      "belief_loss": 2.4260783195495605,
      "decoder_query_loss": 0.0,
      "query_target_mass": 0.4300532937049866,
      "query_target_nll": 2.417053699493408,
      "query_top1_on_support": 0.4140625,
      "mean_query_support_size": 4.734375,
      "decoder_belief_target_mass": 0.5300537347793579,
      "decoder_belief_target_nll": 2.6423802375793457,
      "decoder_belief_top1_on_support": 0.640625,
      "mean_decoder_belief_support_size": 5.515625,
      "decoder_query_target_mass": 0.7448897361755371,
      "decoder_query_target_nll": 1.6549811363220215,
      "decoder_query_top1_on_support": 0.80078125,
      "mean_decoder_query_support_size": 4.734375
    },
    {
      "step": 500,
      "loss": 2.3198914527893066,
      "grad_norm": 0.59250807762146,
      "query_loss": 0.0,
      "belief_loss": 2.3198914527893066,
      "decoder_query_loss": 0.0,
      "query_target_mass": 0.4313085079193115,
      "query_target_nll": 2.413430690765381,
      "query_top1_on_support": 0.44140625,
      "mean_query_support_size": 4.73828125,
      "decoder_belief_target_mass": 0.5911177396774292,
      "decoder_belief_target_nll": 2.4511148929595947,
      "decoder_belief_top1_on_support": 0.6796875,
      "mean_decoder_belief_support_size": 5.671875,
      "decoder_query_target_mass": 0.777594268321991,
      "decoder_query_target_nll": 1.592857837677002,
      "decoder_query_top1_on_support": 0.8203125,
      "mean_decoder_query_support_size": 4.73828125
    },
    {
      "step": 600,
      "loss": 2.3072166442871094,
      "grad_norm": 0.5800983309745789,
      "query_loss": 0.0,
      "belief_loss": 2.3072166442871094,
      "decoder_query_loss": 0.0,
      "query_target_mass": 0.4406753182411194,
      "query_target_nll": 2.418057918548584,
      "query_top1_on_support": 0.42578125,
      "mean_query_support_size": 4.85546875,
      "decoder_belief_target_mass": 0.656280517578125,
      "decoder_belief_target_nll": 2.3443398475646973,
      "decoder_belief_top1_on_support": 0.75390625,
      "mean_decoder_belief_support_size": 6.046875,
      "decoder_query_target_mass": 0.8384872674942017,
      "decoder_query_target_nll": 1.4889997243881226,
      "decoder_query_top1_on_support": 0.89453125,
      "mean_decoder_query_support_size": 4.85546875
    },
    {
      "step": 700,
      "loss": 2.2870142459869385,
      "grad_norm": 0.526483952999115,
      "query_loss": 0.0,
      "belief_loss": 2.2870142459869385,
      "decoder_query_loss": 0.0,
      "query_target_mass": 0.4602026343345642,
      "query_target_nll": 2.416264057159424,
      "query_top1_on_support": 0.4453125,
      "mean_query_support_size": 5.0625,
      "decoder_belief_target_mass": 0.6613610982894897,
      "decoder_belief_target_nll": 2.2985265254974365,
      "decoder_belief_top1_on_support": 0.75390625,
      "mean_decoder_belief_support_size": 5.8671875,
      "decoder_query_target_mass": 0.8426998853683472,
      "decoder_query_target_nll": 1.5242233276367188,
      "decoder_query_top1_on_support": 0.91015625,
      "mean_decoder_query_support_size": 5.0625
    },
    {
      "step": 800,
      "loss": 2.1457135677337646,
      "grad_norm": 0.5137642025947571,
      "query_loss": 0.0,
      "belief_loss": 2.1457135677337646,
      "decoder_query_loss": 0.0,
      "query_target_mass": 0.4629114866256714,
      "query_target_nll": 2.431854724884033,
      "query_top1_on_support": 0.45703125,
      "mean_query_support_size": 5.1171875,
      "decoder_belief_target_mass": 0.7119957804679871,
      "decoder_belief_target_nll": 2.1823768615722656,
      "decoder_belief_top1_on_support": 0.80859375,
      "mean_decoder_belief_support_size": 5.98046875,
      "decoder_query_target_mass": 0.8737783432006836,
      "decoder_query_target_nll": 1.4675037860870361,
      "decoder_query_top1_on_support": 0.9140625,
      "mean_decoder_query_support_size": 5.1171875
    }
  ],
  "probe_train": [
    {
      "step": 1,
      "loss": 4.814298629760742,
      "grad_norm": 0.36396947503089905,
      "probe_belief_target_mass": 0.06040330231189728,
      "probe_belief_target_nll": 4.814298629760742,
      "probe_belief_top1_on_support": 0.0703125,
      "mean_belief_support_size": 7.3058037757873535
    },
    {
      "step": 100,
      "loss": 2.475865602493286,
      "grad_norm": 0.27428001165390015,
      "probe_belief_target_mass": 0.6870558857917786,
      "probe_belief_target_nll": 2.475865602493286,
      "probe_belief_top1_on_support": 0.8203125596046448,
      "mean_belief_support_size": 7.245535850524902
    },
    {
      "step": 200,
      "loss": 2.301560163497925,
      "grad_norm": 0.27124059200286865,
      "probe_belief_target_mass": 0.7774045467376709,
      "probe_belief_target_nll": 2.301560163497925,
      "probe_belief_top1_on_support": 0.8610491752624512,
      "mean_belief_support_size": 7.470424652099609
    },
    {
      "step": 300,
      "loss": 2.169306516647339,
      "grad_norm": 0.2587418556213379,
      "probe_belief_target_mass": 0.8185343146324158,
      "probe_belief_target_nll": 2.169306516647339,
      "probe_belief_top1_on_support": 0.8939732313156128,
      "mean_belief_support_size": 7.313058376312256
    },
    {
      "step": 400,
      "loss": 2.2206857204437256,
      "grad_norm": 0.3195890188217163,
      "probe_belief_target_mass": 0.7955979704856873,
      "probe_belief_target_nll": 2.2206857204437256,
      "probe_belief_top1_on_support": 0.8571429252624512,
      "mean_belief_support_size": 7.136160850524902
    }
  ],
  "eval": [
    {
      "step": 800,
      "rows": [
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 3,
          "query_type": "A",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.47110384702682495,
          "query_target_nll": 2.4191734790802,
          "query_top1_on_support": 0.484375,
          "mean_query_support_size": 5.18359375,
          "probe_belief_target_mass": 0.0635276734828949,
          "probe_belief_target_nll": 15.048038482666016,
          "probe_belief_top1_on_support": 0.06640625,
          "mean_belief_support_size": 5.84375,
          "decoder_belief_target_mass": 0.0627020001411438,
          "decoder_belief_target_nll": 19.17376708984375,
          "decoder_belief_top1_on_support": 0.0625,
          "mean_decoder_belief_support_size": 5.84375,
          "decoder_query_target_mass": 0.47120803594589233,
          "decoder_query_target_nll": 2.4069528579711914,
          "decoder_query_top1_on_support": 0.46875,
          "mean_decoder_query_support_size": 5.18359375
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 3,
          "query_type": "A",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.47013863921165466,
          "query_target_nll": 2.4231321811676025,
          "query_top1_on_support": 0.484375,
          "mean_query_support_size": 5.18359375,
          "probe_belief_target_mass": 0.12043257057666779,
          "probe_belief_target_nll": 11.911369323730469,
          "probe_belief_top1_on_support": 0.0859375,
          "mean_belief_support_size": 5.84375,
          "decoder_belief_target_mass": 0.11983495950698853,
          "decoder_belief_target_nll": 13.87894058227539,
          "decoder_belief_top1_on_support": 0.1015625,
          "mean_decoder_belief_support_size": 5.84375,
          "decoder_query_target_mass": 0.5571562051773071,
          "decoder_query_target_nll": 3.8757054805755615,
          "decoder_query_top1_on_support": 0.5546875,
          "mean_decoder_query_support_size": 5.18359375
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 3,
          "query_type": "A",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.46795493364334106,
          "query_target_nll": 2.432373285293579,
          "query_top1_on_support": 0.4453125,
          "mean_query_support_size": 5.18359375,
          "probe_belief_target_mass": 0.2446410059928894,
          "probe_belief_target_nll": 7.722655296325684,
          "probe_belief_top1_on_support": 0.25390625,
          "mean_belief_support_size": 5.84375,
          "decoder_belief_target_mass": 0.24801595509052277,
          "decoder_belief_target_nll": 8.62706470489502,
          "decoder_belief_top1_on_support": 0.2890625,
          "mean_decoder_belief_support_size": 5.84375,
          "decoder_query_target_mass": 0.6931164264678955,
          "decoder_query_target_nll": 3.1799755096435547,
          "decoder_query_top1_on_support": 0.71484375,
          "mean_decoder_query_support_size": 5.18359375
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 3,
          "query_type": "A",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.4675104320049286,
          "query_target_nll": 2.4348554611206055,
          "query_top1_on_support": 0.41796875,
          "mean_query_support_size": 5.18359375,
          "probe_belief_target_mass": 0.6617501974105835,
          "probe_belief_target_nll": 2.315791130065918,
          "probe_belief_top1_on_support": 0.76171875,
          "mean_belief_support_size": 5.84375,
          "decoder_belief_target_mass": 0.6806540489196777,
          "decoder_belief_target_nll": 2.2753846645355225,
          "decoder_belief_top1_on_support": 0.7734375,
          "mean_decoder_belief_support_size": 5.84375,
          "decoder_query_target_mass": 0.921330451965332,
          "decoder_query_target_nll": 1.4591572284698486,
          "decoder_query_top1_on_support": 0.96875,
          "mean_decoder_query_support_size": 5.18359375
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 3,
          "query_type": "A",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.4675104320049286,
          "query_target_nll": 2.4348554611206055,
          "query_top1_on_support": 0.41796875,
          "mean_query_support_size": 5.18359375,
          "probe_belief_target_mass": 0.6617501974105835,
          "probe_belief_target_nll": 2.315791130065918,
          "probe_belief_top1_on_support": 0.76171875,
          "mean_belief_support_size": 5.84375,
          "decoder_belief_target_mass": 0.6806540489196777,
          "decoder_belief_target_nll": 2.2753846645355225,
          "decoder_belief_top1_on_support": 0.7734375,
          "mean_decoder_belief_support_size": 5.84375,
          "decoder_query_target_mass": 0.921330451965332,
          "decoder_query_target_nll": 1.4591572284698486,
          "decoder_query_top1_on_support": 0.96875,
          "mean_decoder_query_support_size": 5.18359375
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 3,
          "query_type": "A",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.4675104320049286,
          "query_target_nll": 2.4348554611206055,
          "query_top1_on_support": 0.41796875,
          "mean_query_support_size": 5.18359375,
          "probe_belief_target_mass": 0.6617501974105835,
          "probe_belief_target_nll": 2.315791130065918,
          "probe_belief_top1_on_support": 0.76171875,
          "mean_belief_support_size": 5.84375,
          "decoder_belief_target_mass": 0.6806540489196777,
          "decoder_belief_target_nll": 2.2753846645355225,
          "decoder_belief_top1_on_support": 0.7734375,
          "mean_decoder_belief_support_size": 5.84375,
          "decoder_query_target_mass": 0.921330451965332,
          "decoder_query_target_nll": 1.4591572284698486,
          "decoder_query_top1_on_support": 0.96875,
          "mean_decoder_query_support_size": 5.18359375
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 3,
          "query_type": "A",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.4675104320049286,
          "query_target_nll": 2.4348554611206055,
          "query_top1_on_support": 0.41796875,
          "mean_query_support_size": 5.18359375,
          "probe_belief_target_mass": 0.6617501974105835,
          "probe_belief_target_nll": 2.315791130065918,
          "probe_belief_top1_on_support": 0.76171875,
          "mean_belief_support_size": 5.84375,
          "decoder_belief_target_mass": 0.6806540489196777,
          "decoder_belief_target_nll": 2.2753846645355225,
          "decoder_belief_top1_on_support": 0.7734375,
          "mean_decoder_belief_support_size": 5.84375,
          "decoder_query_target_mass": 0.921330451965332,
          "decoder_query_target_nll": 1.4591572284698486,
          "decoder_query_top1_on_support": 0.96875,
          "mean_decoder_query_support_size": 5.18359375
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 3,
          "query_type": "B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.46950191259384155,
          "query_target_nll": 2.406689405441284,
          "query_top1_on_support": 0.48046875,
          "mean_query_support_size": 5.15234375,
          "probe_belief_target_mass": 0.053820233792066574,
          "probe_belief_target_nll": 15.327560424804688,
          "probe_belief_top1_on_support": 0.0625,
          "mean_belief_support_size": 5.77734375,
          "decoder_belief_target_mass": 0.05332862585783005,
          "decoder_belief_target_nll": 19.403514862060547,
          "decoder_belief_top1_on_support": 0.04296875,
          "mean_decoder_belief_support_size": 5.77734375,
          "decoder_query_target_mass": 0.46912282705307007,
          "decoder_query_target_nll": 2.4038898944854736,
          "decoder_query_top1_on_support": 0.48828125,
          "mean_decoder_query_support_size": 5.15234375
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 3,
          "query_type": "B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.46953466534614563,
          "query_target_nll": 2.4139366149902344,
          "query_top1_on_support": 0.48046875,
          "mean_query_support_size": 5.15234375,
          "probe_belief_target_mass": 0.1357375979423523,
          "probe_belief_target_nll": 11.216041564941406,
          "probe_belief_top1_on_support": 0.13671875,
          "mean_belief_support_size": 5.77734375,
          "decoder_belief_target_mass": 0.13652178645133972,
          "decoder_belief_target_nll": 12.914661407470703,
          "decoder_belief_top1_on_support": 0.15234375,
          "mean_decoder_belief_support_size": 5.77734375,
          "decoder_query_target_mass": 0.5400673747062683,
          "decoder_query_target_nll": 4.013266563415527,
          "decoder_query_top1_on_support": 0.55859375,
          "mean_decoder_query_support_size": 5.15234375
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 3,
          "query_type": "B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.46769338846206665,
          "query_target_nll": 2.4227287769317627,
          "query_top1_on_support": 0.4921875,
          "mean_query_support_size": 5.15234375,
          "probe_belief_target_mass": 0.25874558091163635,
          "probe_belief_target_nll": 7.787363052368164,
          "probe_belief_top1_on_support": 0.28125,
          "mean_belief_support_size": 5.77734375,
          "decoder_belief_target_mass": 0.2589128017425537,
          "decoder_belief_target_nll": 8.542932510375977,
          "decoder_belief_top1_on_support": 0.2734375,
          "mean_decoder_belief_support_size": 5.77734375,
          "decoder_query_target_mass": 0.6802631616592407,
          "decoder_query_target_nll": 3.3026087284088135,
          "decoder_query_top1_on_support": 0.6796875,
          "mean_decoder_query_support_size": 5.15234375
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 3,
          "query_type": "B",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.46563589572906494,
          "query_target_nll": 2.4348487854003906,
          "query_top1_on_support": 0.4453125,
          "mean_query_support_size": 5.15234375,
          "probe_belief_target_mass": 0.7044081687927246,
          "probe_belief_target_nll": 2.1846587657928467,
          "probe_belief_top1_on_support": 0.81640625,
          "mean_belief_support_size": 5.77734375,
          "decoder_belief_target_mass": 0.7160639762878418,
          "decoder_belief_target_nll": 2.176726818084717,
          "decoder_belief_top1_on_support": 0.8046875,
          "mean_decoder_belief_support_size": 5.77734375,
          "decoder_query_target_mass": 0.9146192669868469,
          "decoder_query_target_nll": 1.4522721767425537,
          "decoder_query_top1_on_support": 0.95703125,
          "mean_decoder_query_support_size": 5.15234375
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 3,
          "query_type": "B",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.46563589572906494,
          "query_target_nll": 2.4348487854003906,
          "query_top1_on_support": 0.4453125,
          "mean_query_support_size": 5.15234375,
          "probe_belief_target_mass": 0.7044081687927246,
          "probe_belief_target_nll": 2.1846587657928467,
          "probe_belief_top1_on_support": 0.81640625,
          "mean_belief_support_size": 5.77734375,
          "decoder_belief_target_mass": 0.7160639762878418,
          "decoder_belief_target_nll": 2.176726818084717,
          "decoder_belief_top1_on_support": 0.8046875,
          "mean_decoder_belief_support_size": 5.77734375,
          "decoder_query_target_mass": 0.9146192669868469,
          "decoder_query_target_nll": 1.4522721767425537,
          "decoder_query_top1_on_support": 0.95703125,
          "mean_decoder_query_support_size": 5.15234375
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 3,
          "query_type": "B",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.46563589572906494,
          "query_target_nll": 2.4348487854003906,
          "query_top1_on_support": 0.4453125,
          "mean_query_support_size": 5.15234375,
          "probe_belief_target_mass": 0.7044081687927246,
          "probe_belief_target_nll": 2.1846587657928467,
          "probe_belief_top1_on_support": 0.81640625,
          "mean_belief_support_size": 5.77734375,
          "decoder_belief_target_mass": 0.7160639762878418,
          "decoder_belief_target_nll": 2.176726818084717,
          "decoder_belief_top1_on_support": 0.8046875,
          "mean_decoder_belief_support_size": 5.77734375,
          "decoder_query_target_mass": 0.9146192669868469,
          "decoder_query_target_nll": 1.4522721767425537,
          "decoder_query_top1_on_support": 0.95703125,
          "mean_decoder_query_support_size": 5.15234375
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 3,
          "query_type": "B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.46563589572906494,
          "query_target_nll": 2.4348487854003906,
          "query_top1_on_support": 0.4453125,
          "mean_query_support_size": 5.15234375,
          "probe_belief_target_mass": 0.7044081687927246,
          "probe_belief_target_nll": 2.1846587657928467,
          "probe_belief_top1_on_support": 0.81640625,
          "mean_belief_support_size": 5.77734375,
          "decoder_belief_target_mass": 0.7160639762878418,
          "decoder_belief_target_nll": 2.176726818084717,
          "decoder_belief_top1_on_support": 0.8046875,
          "mean_decoder_belief_support_size": 5.77734375,
          "decoder_query_target_mass": 0.9146192669868469,
          "decoder_query_target_nll": 1.4522721767425537,
          "decoder_query_top1_on_support": 0.95703125,
          "mean_decoder_query_support_size": 5.15234375
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 3,
          "query_type": "A_PLUS_B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.4883984923362732,
          "query_target_nll": 2.4130635261535645,
          "query_top1_on_support": 0.4765625,
          "mean_query_support_size": 5.3671875,
          "probe_belief_target_mass": 0.06478855013847351,
          "probe_belief_target_nll": 14.707942962646484,
          "probe_belief_top1_on_support": 0.05859375,
          "mean_belief_support_size": 5.484375,
          "decoder_belief_target_mass": 0.06581249088048935,
          "decoder_belief_target_nll": 18.736860275268555,
          "decoder_belief_top1_on_support": 0.06640625,
          "mean_decoder_belief_support_size": 5.484375,
          "decoder_query_target_mass": 0.4889189302921295,
          "decoder_query_target_nll": 2.4004502296447754,
          "decoder_query_top1_on_support": 0.5078125,
          "mean_decoder_query_support_size": 5.3671875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 3,
          "query_type": "A_PLUS_B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.4884129762649536,
          "query_target_nll": 2.4136266708374023,
          "query_top1_on_support": 0.50390625,
          "mean_query_support_size": 5.3671875,
          "probe_belief_target_mass": 0.15452393889427185,
          "probe_belief_target_nll": 11.97087287902832,
          "probe_belief_top1_on_support": 0.16796875,
          "mean_belief_support_size": 5.484375,
          "decoder_belief_target_mass": 0.1541256606578827,
          "decoder_belief_target_nll": 13.789974212646484,
          "decoder_belief_top1_on_support": 0.16015625,
          "mean_decoder_belief_support_size": 5.484375,
          "decoder_query_target_mass": 0.4921332597732544,
          "decoder_query_target_nll": 4.381202220916748,
          "decoder_query_top1_on_support": 0.515625,
          "mean_decoder_query_support_size": 5.3671875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 3,
          "query_type": "A_PLUS_B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.48764052987098694,
          "query_target_nll": 2.416640520095825,
          "query_top1_on_support": 0.48828125,
          "mean_query_support_size": 5.3671875,
          "probe_belief_target_mass": 0.23751813173294067,
          "probe_belief_target_nll": 8.699502944946289,
          "probe_belief_top1_on_support": 0.25390625,
          "mean_belief_support_size": 5.484375,
          "decoder_belief_target_mass": 0.24312759935855865,
          "decoder_belief_target_nll": 9.655357360839844,
          "decoder_belief_top1_on_support": 0.25390625,
          "mean_decoder_belief_support_size": 5.484375,
          "decoder_query_target_mass": 0.5469808578491211,
          "decoder_query_target_nll": 4.179184436798096,
          "decoder_query_top1_on_support": 0.56640625,
          "mean_decoder_query_support_size": 5.3671875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 3,
          "query_type": "A_PLUS_B",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.48798930644989014,
          "query_target_nll": 2.415081739425659,
          "query_top1_on_support": 0.48828125,
          "mean_query_support_size": 5.3671875,
          "probe_belief_target_mass": 0.7098647356033325,
          "probe_belief_target_nll": 2.100762367248535,
          "probe_belief_top1_on_support": 0.828125,
          "mean_belief_support_size": 5.484375,
          "decoder_belief_target_mass": 0.7332721948623657,
          "decoder_belief_target_nll": 2.0470166206359863,
          "decoder_belief_top1_on_support": 0.83984375,
          "mean_decoder_belief_support_size": 5.484375,
          "decoder_query_target_mass": 0.862246036529541,
          "decoder_query_target_nll": 1.6586408615112305,
          "decoder_query_top1_on_support": 0.9375,
          "mean_decoder_query_support_size": 5.3671875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 3,
          "query_type": "A_PLUS_B",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.48798930644989014,
          "query_target_nll": 2.415081739425659,
          "query_top1_on_support": 0.48828125,
          "mean_query_support_size": 5.3671875,
          "probe_belief_target_mass": 0.7098647356033325,
          "probe_belief_target_nll": 2.100762367248535,
          "probe_belief_top1_on_support": 0.828125,
          "mean_belief_support_size": 5.484375,
          "decoder_belief_target_mass": 0.7332721948623657,
          "decoder_belief_target_nll": 2.0470166206359863,
          "decoder_belief_top1_on_support": 0.83984375,
          "mean_decoder_belief_support_size": 5.484375,
          "decoder_query_target_mass": 0.862246036529541,
          "decoder_query_target_nll": 1.6586408615112305,
          "decoder_query_top1_on_support": 0.9375,
          "mean_decoder_query_support_size": 5.3671875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 3,
          "query_type": "A_PLUS_B",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.48798930644989014,
          "query_target_nll": 2.415081739425659,
          "query_top1_on_support": 0.48828125,
          "mean_query_support_size": 5.3671875,
          "probe_belief_target_mass": 0.7098647356033325,
          "probe_belief_target_nll": 2.100762367248535,
          "probe_belief_top1_on_support": 0.828125,
          "mean_belief_support_size": 5.484375,
          "decoder_belief_target_mass": 0.7332721948623657,
          "decoder_belief_target_nll": 2.0470166206359863,
          "decoder_belief_top1_on_support": 0.83984375,
          "mean_decoder_belief_support_size": 5.484375,
          "decoder_query_target_mass": 0.862246036529541,
          "decoder_query_target_nll": 1.6586408615112305,
          "decoder_query_top1_on_support": 0.9375,
          "mean_decoder_query_support_size": 5.3671875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 3,
          "query_type": "A_PLUS_B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.48798930644989014,
          "query_target_nll": 2.415081739425659,
          "query_top1_on_support": 0.48828125,
          "mean_query_support_size": 5.3671875,
          "probe_belief_target_mass": 0.7098647356033325,
          "probe_belief_target_nll": 2.100762367248535,
          "probe_belief_top1_on_support": 0.828125,
          "mean_belief_support_size": 5.484375,
          "decoder_belief_target_mass": 0.7332721948623657,
          "decoder_belief_target_nll": 2.0470166206359863,
          "decoder_belief_top1_on_support": 0.83984375,
          "mean_decoder_belief_support_size": 5.484375,
          "decoder_query_target_mass": 0.862246036529541,
          "decoder_query_target_nll": 1.6586408615112305,
          "decoder_query_top1_on_support": 0.9375,
          "mean_decoder_query_support_size": 5.3671875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 3,
          "query_type": "A_MINUS_B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.42282915115356445,
          "query_target_nll": 2.424910068511963,
          "query_top1_on_support": 0.40625,
          "mean_query_support_size": 4.671875,
          "probe_belief_target_mass": 0.059614941477775574,
          "probe_belief_target_nll": 14.775103569030762,
          "probe_belief_top1_on_support": 0.07421875,
          "mean_belief_support_size": 5.6015625,
          "decoder_belief_target_mass": 0.0589413121342659,
          "decoder_belief_target_nll": 18.82172393798828,
          "decoder_belief_top1_on_support": 0.05859375,
          "mean_decoder_belief_support_size": 5.6015625,
          "decoder_query_target_mass": 0.47265586256980896,
          "decoder_query_target_nll": 14.964567184448242,
          "decoder_query_top1_on_support": 0.47265625,
          "mean_decoder_query_support_size": 4.671875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 3,
          "query_type": "A_MINUS_B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.42443594336509705,
          "query_target_nll": 2.4166154861450195,
          "query_top1_on_support": 0.421875,
          "mean_query_support_size": 4.671875,
          "probe_belief_target_mass": 0.1571921408176422,
          "probe_belief_target_nll": 11.11817741394043,
          "probe_belief_top1_on_support": 0.16796875,
          "mean_belief_support_size": 5.6015625,
          "decoder_belief_target_mass": 0.1574312150478363,
          "decoder_belief_target_nll": 12.834342956542969,
          "decoder_belief_top1_on_support": 0.15234375,
          "mean_decoder_belief_support_size": 5.6015625,
          "decoder_query_target_mass": 0.5295737385749817,
          "decoder_query_target_nll": 8.14964485168457,
          "decoder_query_top1_on_support": 0.53125,
          "mean_decoder_query_support_size": 4.671875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 3,
          "query_type": "A_MINUS_B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.4249086081981659,
          "query_target_nll": 2.4161853790283203,
          "query_top1_on_support": 0.45703125,
          "mean_query_support_size": 4.671875,
          "probe_belief_target_mass": 0.2777051031589508,
          "probe_belief_target_nll": 7.496294975280762,
          "probe_belief_top1_on_support": 0.29296875,
          "mean_belief_support_size": 5.6015625,
          "decoder_belief_target_mass": 0.28368306159973145,
          "decoder_belief_target_nll": 8.341564178466797,
          "decoder_belief_top1_on_support": 0.3203125,
          "mean_decoder_belief_support_size": 5.6015625,
          "decoder_query_target_mass": 0.5874843001365662,
          "decoder_query_target_nll": 5.337933540344238,
          "decoder_query_top1_on_support": 0.57421875,
          "mean_decoder_query_support_size": 4.671875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 3,
          "query_type": "A_MINUS_B",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.4247581362724304,
          "query_target_nll": 2.416616916656494,
          "query_top1_on_support": 0.4375,
          "mean_query_support_size": 4.671875,
          "probe_belief_target_mass": 0.673723578453064,
          "probe_belief_target_nll": 2.200747489929199,
          "probe_belief_top1_on_support": 0.79296875,
          "mean_belief_support_size": 5.6015625,
          "decoder_belief_target_mass": 0.6928728818893433,
          "decoder_belief_target_nll": 2.166213035583496,
          "decoder_belief_top1_on_support": 0.796875,
          "mean_decoder_belief_support_size": 5.6015625,
          "decoder_query_target_mass": 0.9221807718276978,
          "decoder_query_target_nll": 1.243563175201416,
          "decoder_query_top1_on_support": 0.9609375,
          "mean_decoder_query_support_size": 4.671875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 3,
          "query_type": "A_MINUS_B",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.4247581362724304,
          "query_target_nll": 2.416616916656494,
          "query_top1_on_support": 0.4375,
          "mean_query_support_size": 4.671875,
          "probe_belief_target_mass": 0.673723578453064,
          "probe_belief_target_nll": 2.200747489929199,
          "probe_belief_top1_on_support": 0.79296875,
          "mean_belief_support_size": 5.6015625,
          "decoder_belief_target_mass": 0.6928728818893433,
          "decoder_belief_target_nll": 2.166213035583496,
          "decoder_belief_top1_on_support": 0.796875,
          "mean_decoder_belief_support_size": 5.6015625,
          "decoder_query_target_mass": 0.9221807718276978,
          "decoder_query_target_nll": 1.243563175201416,
          "decoder_query_top1_on_support": 0.9609375,
          "mean_decoder_query_support_size": 4.671875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 3,
          "query_type": "A_MINUS_B",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.4247581362724304,
          "query_target_nll": 2.416616916656494,
          "query_top1_on_support": 0.4375,
          "mean_query_support_size": 4.671875,
          "probe_belief_target_mass": 0.673723578453064,
          "probe_belief_target_nll": 2.200747489929199,
          "probe_belief_top1_on_support": 0.79296875,
          "mean_belief_support_size": 5.6015625,
          "decoder_belief_target_mass": 0.6928728818893433,
          "decoder_belief_target_nll": 2.166213035583496,
          "decoder_belief_top1_on_support": 0.796875,
          "mean_decoder_belief_support_size": 5.6015625,
          "decoder_query_target_mass": 0.9221807718276978,
          "decoder_query_target_nll": 1.243563175201416,
          "decoder_query_top1_on_support": 0.9609375,
          "mean_decoder_query_support_size": 4.671875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 3,
          "query_type": "A_MINUS_B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.4247581362724304,
          "query_target_nll": 2.416616916656494,
          "query_top1_on_support": 0.4375,
          "mean_query_support_size": 4.671875,
          "probe_belief_target_mass": 0.673723578453064,
          "probe_belief_target_nll": 2.200747489929199,
          "probe_belief_top1_on_support": 0.79296875,
          "mean_belief_support_size": 5.6015625,
          "decoder_belief_target_mass": 0.6928728818893433,
          "decoder_belief_target_nll": 2.166213035583496,
          "decoder_belief_top1_on_support": 0.796875,
          "mean_decoder_belief_support_size": 5.6015625,
          "decoder_query_target_mass": 0.9221807718276978,
          "decoder_query_target_nll": 1.243563175201416,
          "decoder_query_top1_on_support": 0.9609375,
          "mean_decoder_query_support_size": 4.671875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 6,
          "query_type": "A",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.25166404247283936,
          "query_target_nll": 2.4238812923431396,
          "query_top1_on_support": 0.2421875,
          "mean_query_support_size": 2.78125,
          "probe_belief_target_mass": 0.029168590903282166,
          "probe_belief_target_nll": 15.291852951049805,
          "probe_belief_top1_on_support": 0.01171875,
          "mean_belief_support_size": 3.45703125,
          "decoder_belief_target_mass": 0.02950304001569748,
          "decoder_belief_target_nll": 19.36859703063965,
          "decoder_belief_top1_on_support": 0.0234375,
          "mean_decoder_belief_support_size": 3.45703125,
          "decoder_query_target_mass": 0.25274208188056946,
          "decoder_query_target_nll": 2.4025540351867676,
          "decoder_query_top1_on_support": 0.234375,
          "mean_decoder_query_support_size": 2.78125
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 6,
          "query_type": "A",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.2518327534198761,
          "query_target_nll": 2.426527976989746,
          "query_top1_on_support": 0.2734375,
          "mean_query_support_size": 2.78125,
          "probe_belief_target_mass": 0.07259135693311691,
          "probe_belief_target_nll": 12.755531311035156,
          "probe_belief_top1_on_support": 0.0859375,
          "mean_belief_support_size": 3.45703125,
          "decoder_belief_target_mass": 0.0700208991765976,
          "decoder_belief_target_nll": 14.70085620880127,
          "decoder_belief_top1_on_support": 0.06640625,
          "mean_decoder_belief_support_size": 3.45703125,
          "decoder_query_target_mass": 0.28065940737724304,
          "decoder_query_target_nll": 4.701322078704834,
          "decoder_query_top1_on_support": 0.2734375,
          "mean_decoder_query_support_size": 2.78125
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 6,
          "query_type": "A",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.25447583198547363,
          "query_target_nll": 2.4077672958374023,
          "query_top1_on_support": 0.265625,
          "mean_query_support_size": 2.78125,
          "probe_belief_target_mass": 0.08253698796033859,
          "probe_belief_target_nll": 11.309873580932617,
          "probe_belief_top1_on_support": 0.1015625,
          "mean_belief_support_size": 3.45703125,
          "decoder_belief_target_mass": 0.08264337480068207,
          "decoder_belief_target_nll": 12.729463577270508,
          "decoder_belief_top1_on_support": 0.0859375,
          "mean_decoder_belief_support_size": 3.45703125,
          "decoder_query_target_mass": 0.3470911383628845,
          "decoder_query_target_nll": 4.540765762329102,
          "decoder_query_top1_on_support": 0.36328125,
          "mean_decoder_query_support_size": 2.78125
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 6,
          "query_type": "A",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.25459587574005127,
          "query_target_nll": 2.40433931350708,
          "query_top1_on_support": 0.25390625,
          "mean_query_support_size": 2.78125,
          "probe_belief_target_mass": 0.09405414015054703,
          "probe_belief_target_nll": 10.37038803100586,
          "probe_belief_top1_on_support": 0.1015625,
          "mean_belief_support_size": 3.45703125,
          "decoder_belief_target_mass": 0.09792867302894592,
          "decoder_belief_target_nll": 11.445039749145508,
          "decoder_belief_top1_on_support": 0.109375,
          "mean_decoder_belief_support_size": 3.45703125,
          "decoder_query_target_mass": 0.39327359199523926,
          "decoder_query_target_nll": 4.398672580718994,
          "decoder_query_top1_on_support": 0.3828125,
          "mean_decoder_query_support_size": 2.78125
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 6,
          "query_type": "A",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.25147804617881775,
          "query_target_nll": 2.424131393432617,
          "query_top1_on_support": 0.22265625,
          "mean_query_support_size": 2.78125,
          "probe_belief_target_mass": 0.3735561966896057,
          "probe_belief_target_nll": 2.551007032394409,
          "probe_belief_top1_on_support": 0.52734375,
          "mean_belief_support_size": 3.45703125,
          "decoder_belief_target_mass": 0.38643890619277954,
          "decoder_belief_target_nll": 2.5020406246185303,
          "decoder_belief_top1_on_support": 0.53515625,
          "mean_decoder_belief_support_size": 3.45703125,
          "decoder_query_target_mass": 0.6263805627822876,
          "decoder_query_target_nll": 1.4433948993682861,
          "decoder_query_top1_on_support": 0.73828125,
          "mean_decoder_query_support_size": 2.78125
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 6,
          "query_type": "A",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.25147804617881775,
          "query_target_nll": 2.424131393432617,
          "query_top1_on_support": 0.22265625,
          "mean_query_support_size": 2.78125,
          "probe_belief_target_mass": 0.3735561966896057,
          "probe_belief_target_nll": 2.551007032394409,
          "probe_belief_top1_on_support": 0.52734375,
          "mean_belief_support_size": 3.45703125,
          "decoder_belief_target_mass": 0.38643890619277954,
          "decoder_belief_target_nll": 2.5020406246185303,
          "decoder_belief_top1_on_support": 0.53515625,
          "mean_decoder_belief_support_size": 3.45703125,
          "decoder_query_target_mass": 0.6263805627822876,
          "decoder_query_target_nll": 1.4433948993682861,
          "decoder_query_top1_on_support": 0.73828125,
          "mean_decoder_query_support_size": 2.78125
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 6,
          "query_type": "A",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.25147804617881775,
          "query_target_nll": 2.424131393432617,
          "query_top1_on_support": 0.22265625,
          "mean_query_support_size": 2.78125,
          "probe_belief_target_mass": 0.3735561966896057,
          "probe_belief_target_nll": 2.551007032394409,
          "probe_belief_top1_on_support": 0.52734375,
          "mean_belief_support_size": 3.45703125,
          "decoder_belief_target_mass": 0.38643890619277954,
          "decoder_belief_target_nll": 2.5020406246185303,
          "decoder_belief_top1_on_support": 0.53515625,
          "mean_decoder_belief_support_size": 3.45703125,
          "decoder_query_target_mass": 0.6263805627822876,
          "decoder_query_target_nll": 1.4433948993682861,
          "decoder_query_top1_on_support": 0.73828125,
          "mean_decoder_query_support_size": 2.78125
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 6,
          "query_type": "B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.31201937794685364,
          "query_target_nll": 2.402050495147705,
          "query_top1_on_support": 0.375,
          "mean_query_support_size": 3.4140625,
          "probe_belief_target_mass": 0.034444015473127365,
          "probe_belief_target_nll": 15.774734497070312,
          "probe_belief_top1_on_support": 0.03515625,
          "mean_belief_support_size": 4.03125,
          "decoder_belief_target_mass": 0.03434707224369049,
          "decoder_belief_target_nll": 20.240074157714844,
          "decoder_belief_top1_on_support": 0.0234375,
          "mean_decoder_belief_support_size": 4.03125,
          "decoder_query_target_mass": 0.31116628646850586,
          "decoder_query_target_nll": 2.4003424644470215,
          "decoder_query_top1_on_support": 0.3125,
          "mean_decoder_query_support_size": 3.4140625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 6,
          "query_type": "B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.3105176091194153,
          "query_target_nll": 2.414137840270996,
          "query_top1_on_support": 0.3203125,
          "mean_query_support_size": 3.4140625,
          "probe_belief_target_mass": 0.04466042295098305,
          "probe_belief_target_nll": 14.183162689208984,
          "probe_belief_top1_on_support": 0.05859375,
          "mean_belief_support_size": 4.03125,
          "decoder_belief_target_mass": 0.04474988952279091,
          "decoder_belief_target_nll": 16.460283279418945,
          "decoder_belief_top1_on_support": 0.05078125,
          "mean_decoder_belief_support_size": 4.03125,
          "decoder_query_target_mass": 0.34040409326553345,
          "decoder_query_target_nll": 5.309167861938477,
          "decoder_query_top1_on_support": 0.35546875,
          "mean_decoder_query_support_size": 3.4140625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 6,
          "query_type": "B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.31084638833999634,
          "query_target_nll": 2.410517454147339,
          "query_top1_on_support": 0.296875,
          "mean_query_support_size": 3.4140625,
          "probe_belief_target_mass": 0.0691615417599678,
          "probe_belief_target_nll": 11.800688743591309,
          "probe_belief_top1_on_support": 0.07421875,
          "mean_belief_support_size": 4.03125,
          "decoder_belief_target_mass": 0.06966784596443176,
          "decoder_belief_target_nll": 13.367157936096191,
          "decoder_belief_top1_on_support": 0.078125,
          "mean_decoder_belief_support_size": 4.03125,
          "decoder_query_target_mass": 0.38453638553619385,
          "decoder_query_target_nll": 4.844334602355957,
          "decoder_query_top1_on_support": 0.41015625,
          "mean_decoder_query_support_size": 3.4140625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 6,
          "query_type": "B",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.3106018304824829,
          "query_target_nll": 2.413517713546753,
          "query_top1_on_support": 0.3046875,
          "mean_query_support_size": 3.4140625,
          "probe_belief_target_mass": 0.09033704549074173,
          "probe_belief_target_nll": 10.471715927124023,
          "probe_belief_top1_on_support": 0.11328125,
          "mean_belief_support_size": 4.03125,
          "decoder_belief_target_mass": 0.09146375954151154,
          "decoder_belief_target_nll": 11.677251815795898,
          "decoder_belief_top1_on_support": 0.11328125,
          "mean_decoder_belief_support_size": 4.03125,
          "decoder_query_target_mass": 0.42143094539642334,
          "decoder_query_target_nll": 4.87171745300293,
          "decoder_query_top1_on_support": 0.44921875,
          "mean_decoder_query_support_size": 3.4140625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 6,
          "query_type": "B",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.30933240056037903,
          "query_target_nll": 2.431696653366089,
          "query_top1_on_support": 0.30078125,
          "mean_query_support_size": 3.4140625,
          "probe_belief_target_mass": 0.3785366415977478,
          "probe_belief_target_nll": 2.736133575439453,
          "probe_belief_top1_on_support": 0.4921875,
          "mean_belief_support_size": 4.03125,
          "decoder_belief_target_mass": 0.393803209066391,
          "decoder_belief_target_nll": 2.674856424331665,
          "decoder_belief_top1_on_support": 0.5078125,
          "mean_decoder_belief_support_size": 4.03125,
          "decoder_query_target_mass": 0.6696572303771973,
          "decoder_query_target_nll": 1.5436899662017822,
          "decoder_query_top1_on_support": 0.7421875,
          "mean_decoder_query_support_size": 3.4140625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 6,
          "query_type": "B",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.30933240056037903,
          "query_target_nll": 2.431696653366089,
          "query_top1_on_support": 0.30078125,
          "mean_query_support_size": 3.4140625,
          "probe_belief_target_mass": 0.3785366415977478,
          "probe_belief_target_nll": 2.736133575439453,
          "probe_belief_top1_on_support": 0.4921875,
          "mean_belief_support_size": 4.03125,
          "decoder_belief_target_mass": 0.393803209066391,
          "decoder_belief_target_nll": 2.674856424331665,
          "decoder_belief_top1_on_support": 0.5078125,
          "mean_decoder_belief_support_size": 4.03125,
          "decoder_query_target_mass": 0.6696572303771973,
          "decoder_query_target_nll": 1.5436899662017822,
          "decoder_query_top1_on_support": 0.7421875,
          "mean_decoder_query_support_size": 3.4140625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 6,
          "query_type": "B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.30933240056037903,
          "query_target_nll": 2.431696653366089,
          "query_top1_on_support": 0.30078125,
          "mean_query_support_size": 3.4140625,
          "probe_belief_target_mass": 0.3785366415977478,
          "probe_belief_target_nll": 2.736133575439453,
          "probe_belief_top1_on_support": 0.4921875,
          "mean_belief_support_size": 4.03125,
          "decoder_belief_target_mass": 0.393803209066391,
          "decoder_belief_target_nll": 2.674856424331665,
          "decoder_belief_top1_on_support": 0.5078125,
          "mean_decoder_belief_support_size": 4.03125,
          "decoder_query_target_mass": 0.6696572303771973,
          "decoder_query_target_nll": 1.5436899662017822,
          "decoder_query_top1_on_support": 0.7421875,
          "mean_decoder_query_support_size": 3.4140625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 6,
          "query_type": "A_PLUS_B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.32671499252319336,
          "query_target_nll": 2.4264111518859863,
          "query_top1_on_support": 0.296875,
          "mean_query_support_size": 3.6171875,
          "probe_belief_target_mass": 0.03292188420891762,
          "probe_belief_target_nll": 15.617737770080566,
          "probe_belief_top1_on_support": 0.0234375,
          "mean_belief_support_size": 3.97265625,
          "decoder_belief_target_mass": 0.03354159742593765,
          "decoder_belief_target_nll": 19.653993606567383,
          "decoder_belief_top1_on_support": 0.04296875,
          "mean_decoder_belief_support_size": 3.97265625,
          "decoder_query_target_mass": 0.32998067140579224,
          "decoder_query_target_nll": 2.3992886543273926,
          "decoder_query_top1_on_support": 0.3359375,
          "mean_decoder_query_support_size": 3.6171875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 6,
          "query_type": "A_PLUS_B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.32788729667663574,
          "query_target_nll": 2.4189891815185547,
          "query_top1_on_support": 0.30859375,
          "mean_query_support_size": 3.6171875,
          "probe_belief_target_mass": 0.04918728768825531,
          "probe_belief_target_nll": 13.338727951049805,
          "probe_belief_top1_on_support": 0.03125,
          "mean_belief_support_size": 3.97265625,
          "decoder_belief_target_mass": 0.052063342183828354,
          "decoder_belief_target_nll": 15.434245109558105,
          "decoder_belief_top1_on_support": 0.05859375,
          "mean_decoder_belief_support_size": 3.97265625,
          "decoder_query_target_mass": 0.33848750591278076,
          "decoder_query_target_nll": 3.815155506134033,
          "decoder_query_top1_on_support": 0.3671875,
          "mean_decoder_query_support_size": 3.6171875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 6,
          "query_type": "A_PLUS_B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.32635897397994995,
          "query_target_nll": 2.4271860122680664,
          "query_top1_on_support": 0.28125,
          "mean_query_support_size": 3.6171875,
          "probe_belief_target_mass": 0.06887058913707733,
          "probe_belief_target_nll": 11.400657653808594,
          "probe_belief_top1_on_support": 0.07421875,
          "mean_belief_support_size": 3.97265625,
          "decoder_belief_target_mass": 0.06791515648365021,
          "decoder_belief_target_nll": 12.77479076385498,
          "decoder_belief_top1_on_support": 0.06640625,
          "mean_decoder_belief_support_size": 3.97265625,
          "decoder_query_target_mass": 0.33099204301834106,
          "decoder_query_target_nll": 4.331418514251709,
          "decoder_query_top1_on_support": 0.32421875,
          "mean_decoder_query_support_size": 3.6171875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 6,
          "query_type": "A_PLUS_B",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.3263581693172455,
          "query_target_nll": 2.4292309284210205,
          "query_top1_on_support": 0.296875,
          "mean_query_support_size": 3.6171875,
          "probe_belief_target_mass": 0.09682142734527588,
          "probe_belief_target_nll": 9.60102367401123,
          "probe_belief_top1_on_support": 0.09765625,
          "mean_belief_support_size": 3.97265625,
          "decoder_belief_target_mass": 0.09690393507480621,
          "decoder_belief_target_nll": 10.657740592956543,
          "decoder_belief_top1_on_support": 0.09765625,
          "mean_decoder_belief_support_size": 3.97265625,
          "decoder_query_target_mass": 0.3264571726322174,
          "decoder_query_target_nll": 4.1781907081604,
          "decoder_query_top1_on_support": 0.3125,
          "mean_decoder_query_support_size": 3.6171875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 6,
          "query_type": "A_PLUS_B",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.3270219564437866,
          "query_target_nll": 2.428271770477295,
          "query_top1_on_support": 0.32421875,
          "mean_query_support_size": 3.6171875,
          "probe_belief_target_mass": 0.34989112615585327,
          "probe_belief_target_nll": 2.734478235244751,
          "probe_belief_top1_on_support": 0.46875,
          "mean_belief_support_size": 3.97265625,
          "decoder_belief_target_mass": 0.3539012670516968,
          "decoder_belief_target_nll": 2.734510660171509,
          "decoder_belief_top1_on_support": 0.46484375,
          "mean_decoder_belief_support_size": 3.97265625,
          "decoder_query_target_mass": 0.55483478307724,
          "decoder_query_target_nll": 1.8413000106811523,
          "decoder_query_top1_on_support": 0.65625,
          "mean_decoder_query_support_size": 3.6171875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 6,
          "query_type": "A_PLUS_B",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.3270219564437866,
          "query_target_nll": 2.428271770477295,
          "query_top1_on_support": 0.32421875,
          "mean_query_support_size": 3.6171875,
          "probe_belief_target_mass": 0.34989112615585327,
          "probe_belief_target_nll": 2.734478235244751,
          "probe_belief_top1_on_support": 0.46875,
          "mean_belief_support_size": 3.97265625,
          "decoder_belief_target_mass": 0.3539012670516968,
          "decoder_belief_target_nll": 2.734510660171509,
          "decoder_belief_top1_on_support": 0.46484375,
          "mean_decoder_belief_support_size": 3.97265625,
          "decoder_query_target_mass": 0.55483478307724,
          "decoder_query_target_nll": 1.8413000106811523,
          "decoder_query_top1_on_support": 0.65625,
          "mean_decoder_query_support_size": 3.6171875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 6,
          "query_type": "A_PLUS_B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.3270219564437866,
          "query_target_nll": 2.428271770477295,
          "query_top1_on_support": 0.32421875,
          "mean_query_support_size": 3.6171875,
          "probe_belief_target_mass": 0.34989112615585327,
          "probe_belief_target_nll": 2.734478235244751,
          "probe_belief_top1_on_support": 0.46875,
          "mean_belief_support_size": 3.97265625,
          "decoder_belief_target_mass": 0.3539012670516968,
          "decoder_belief_target_nll": 2.734510660171509,
          "decoder_belief_top1_on_support": 0.46484375,
          "mean_decoder_belief_support_size": 3.97265625,
          "decoder_query_target_mass": 0.55483478307724,
          "decoder_query_target_nll": 1.8413000106811523,
          "decoder_query_top1_on_support": 0.65625,
          "mean_decoder_query_support_size": 3.6171875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 6,
          "query_type": "A_MINUS_B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.2941204309463501,
          "query_target_nll": 2.4192824363708496,
          "query_top1_on_support": 0.3046875,
          "mean_query_support_size": 3.2265625,
          "probe_belief_target_mass": 0.030512649565935135,
          "probe_belief_target_nll": 15.441719055175781,
          "probe_belief_top1_on_support": 0.015625,
          "mean_belief_support_size": 3.76171875,
          "decoder_belief_target_mass": 0.031184181571006775,
          "decoder_belief_target_nll": 19.51642417907715,
          "decoder_belief_top1_on_support": 0.01953125,
          "mean_decoder_belief_support_size": 3.76171875,
          "decoder_query_target_mass": 0.3125,
          "decoder_query_target_nll": 15.431215286254883,
          "decoder_query_top1_on_support": 0.3125,
          "mean_decoder_query_support_size": 3.2265625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 6,
          "query_type": "A_MINUS_B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.2948492169380188,
          "query_target_nll": 2.410552501678467,
          "query_top1_on_support": 0.296875,
          "mean_query_support_size": 3.2265625,
          "probe_belief_target_mass": 0.04365333169698715,
          "probe_belief_target_nll": 13.526325225830078,
          "probe_belief_top1_on_support": 0.0390625,
          "mean_belief_support_size": 3.76171875,
          "decoder_belief_target_mass": 0.04493442177772522,
          "decoder_belief_target_nll": 15.432703018188477,
          "decoder_belief_top1_on_support": 0.03125,
          "mean_decoder_belief_support_size": 3.76171875,
          "decoder_query_target_mass": 0.3033238649368286,
          "decoder_query_target_nll": 9.189085960388184,
          "decoder_query_top1_on_support": 0.3125,
          "mean_decoder_query_support_size": 3.2265625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 6,
          "query_type": "A_MINUS_B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.2938900589942932,
          "query_target_nll": 2.416015148162842,
          "query_top1_on_support": 0.328125,
          "mean_query_support_size": 3.2265625,
          "probe_belief_target_mass": 0.06927317380905151,
          "probe_belief_target_nll": 11.581522941589355,
          "probe_belief_top1_on_support": 0.078125,
          "mean_belief_support_size": 3.76171875,
          "decoder_belief_target_mass": 0.06976403295993805,
          "decoder_belief_target_nll": 12.89130973815918,
          "decoder_belief_top1_on_support": 0.08203125,
          "mean_decoder_belief_support_size": 3.76171875,
          "decoder_query_target_mass": 0.33412355184555054,
          "decoder_query_target_nll": 7.2140212059021,
          "decoder_query_top1_on_support": 0.33203125,
          "mean_decoder_query_support_size": 3.2265625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 6,
          "query_type": "A_MINUS_B",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.293886661529541,
          "query_target_nll": 2.415499687194824,
          "query_top1_on_support": 0.27734375,
          "mean_query_support_size": 3.2265625,
          "probe_belief_target_mass": 0.07779116928577423,
          "probe_belief_target_nll": 9.994932174682617,
          "probe_belief_top1_on_support": 0.0859375,
          "mean_belief_support_size": 3.76171875,
          "decoder_belief_target_mass": 0.07860880345106125,
          "decoder_belief_target_nll": 11.129804611206055,
          "decoder_belief_top1_on_support": 0.0703125,
          "mean_decoder_belief_support_size": 3.76171875,
          "decoder_query_target_mass": 0.3181135952472687,
          "decoder_query_target_nll": 5.997439384460449,
          "decoder_query_top1_on_support": 0.34765625,
          "mean_decoder_query_support_size": 3.2265625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 6,
          "query_type": "A_MINUS_B",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.2950218915939331,
          "query_target_nll": 2.4091601371765137,
          "query_top1_on_support": 0.328125,
          "mean_query_support_size": 3.2265625,
          "probe_belief_target_mass": 0.35932230949401855,
          "probe_belief_target_nll": 2.6827821731567383,
          "probe_belief_top1_on_support": 0.5078125,
          "mean_belief_support_size": 3.76171875,
          "decoder_belief_target_mass": 0.37872231006622314,
          "decoder_belief_target_nll": 2.632678985595703,
          "decoder_belief_top1_on_support": 0.5078125,
          "mean_decoder_belief_support_size": 3.76171875,
          "decoder_query_target_mass": 0.6266967058181763,
          "decoder_query_target_nll": 1.5649724006652832,
          "decoder_query_top1_on_support": 0.69140625,
          "mean_decoder_query_support_size": 3.2265625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 6,
          "query_type": "A_MINUS_B",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.2950218915939331,
          "query_target_nll": 2.4091601371765137,
          "query_top1_on_support": 0.328125,
          "mean_query_support_size": 3.2265625,
          "probe_belief_target_mass": 0.35932230949401855,
          "probe_belief_target_nll": 2.6827821731567383,
          "probe_belief_top1_on_support": 0.5078125,
          "mean_belief_support_size": 3.76171875,
          "decoder_belief_target_mass": 0.37872231006622314,
          "decoder_belief_target_nll": 2.632678985595703,
          "decoder_belief_top1_on_support": 0.5078125,
          "mean_decoder_belief_support_size": 3.76171875,
          "decoder_query_target_mass": 0.6266967058181763,
          "decoder_query_target_nll": 1.5649724006652832,
          "decoder_query_top1_on_support": 0.69140625,
          "mean_decoder_query_support_size": 3.2265625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 6,
          "query_type": "A_MINUS_B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.2950218915939331,
          "query_target_nll": 2.4091601371765137,
          "query_top1_on_support": 0.328125,
          "mean_query_support_size": 3.2265625,
          "probe_belief_target_mass": 0.35932230949401855,
          "probe_belief_target_nll": 2.6827821731567383,
          "probe_belief_top1_on_support": 0.5078125,
          "mean_belief_support_size": 3.76171875,
          "decoder_belief_target_mass": 0.37872231006622314,
          "decoder_belief_target_nll": 2.632678985595703,
          "decoder_belief_top1_on_support": 0.5078125,
          "mean_decoder_belief_support_size": 3.76171875,
          "decoder_query_target_mass": 0.6266967058181763,
          "decoder_query_target_nll": 1.5649724006652832,
          "decoder_query_top1_on_support": 0.69140625,
          "mean_decoder_query_support_size": 3.2265625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 9,
          "query_type": "A",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.2173953354358673,
          "query_target_nll": 2.418349027633667,
          "query_top1_on_support": 0.21875,
          "mean_query_support_size": 2.40625,
          "probe_belief_target_mass": 0.021447153761982918,
          "probe_belief_target_nll": 15.527957916259766,
          "probe_belief_top1_on_support": 0.0234375,
          "mean_belief_support_size": 2.73828125,
          "decoder_belief_target_mass": 0.021122723817825317,
          "decoder_belief_target_nll": 19.934385299682617,
          "decoder_belief_top1_on_support": 0.02734375,
          "mean_decoder_belief_support_size": 2.73828125,
          "decoder_query_target_mass": 0.21702948212623596,
          "decoder_query_target_nll": 2.41515851020813,
          "decoder_query_top1_on_support": 0.21875,
          "mean_decoder_query_support_size": 2.40625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 9,
          "query_type": "A",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.21920032799243927,
          "query_target_nll": 2.4128456115722656,
          "query_top1_on_support": 0.19140625,
          "mean_query_support_size": 2.40625,
          "probe_belief_target_mass": 0.0356263630092144,
          "probe_belief_target_nll": 13.26718807220459,
          "probe_belief_top1_on_support": 0.03515625,
          "mean_belief_support_size": 2.73828125,
          "decoder_belief_target_mass": 0.034599125385284424,
          "decoder_belief_target_nll": 15.442214965820312,
          "decoder_belief_top1_on_support": 0.0390625,
          "mean_decoder_belief_support_size": 2.73828125,
          "decoder_query_target_mass": 0.24070313572883606,
          "decoder_query_target_nll": 4.805291175842285,
          "decoder_query_top1_on_support": 0.25390625,
          "mean_decoder_query_support_size": 2.40625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 9,
          "query_type": "A",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.21938131749629974,
          "query_target_nll": 2.416624069213867,
          "query_top1_on_support": 0.21875,
          "mean_query_support_size": 2.40625,
          "probe_belief_target_mass": 0.035185862332582474,
          "probe_belief_target_nll": 12.486356735229492,
          "probe_belief_top1_on_support": 0.03125,
          "mean_belief_support_size": 2.73828125,
          "decoder_belief_target_mass": 0.03622954711318016,
          "decoder_belief_target_nll": 14.139609336853027,
          "decoder_belief_top1_on_support": 0.04296875,
          "mean_decoder_belief_support_size": 2.73828125,
          "decoder_query_target_mass": 0.26098161935806274,
          "decoder_query_target_nll": 5.417749404907227,
          "decoder_query_top1_on_support": 0.2734375,
          "mean_decoder_query_support_size": 2.40625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 9,
          "query_type": "A",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.22000497579574585,
          "query_target_nll": 2.406236171722412,
          "query_top1_on_support": 0.20703125,
          "mean_query_support_size": 2.40625,
          "probe_belief_target_mass": 0.037877678871154785,
          "probe_belief_target_nll": 11.436624526977539,
          "probe_belief_top1_on_support": 0.03515625,
          "mean_belief_support_size": 2.73828125,
          "decoder_belief_target_mass": 0.039632149040699005,
          "decoder_belief_target_nll": 12.742849349975586,
          "decoder_belief_top1_on_support": 0.03515625,
          "mean_decoder_belief_support_size": 2.73828125,
          "decoder_query_target_mass": 0.26378512382507324,
          "decoder_query_target_nll": 5.5594377517700195,
          "decoder_query_top1_on_support": 0.24609375,
          "mean_decoder_query_support_size": 2.40625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 9,
          "query_type": "A",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.21984978020191193,
          "query_target_nll": 2.4176414012908936,
          "query_top1_on_support": 0.2109375,
          "mean_query_support_size": 2.40625,
          "probe_belief_target_mass": 0.05249246582388878,
          "probe_belief_target_nll": 8.843938827514648,
          "probe_belief_top1_on_support": 0.046875,
          "mean_belief_support_size": 2.73828125,
          "decoder_belief_target_mass": 0.05634628236293793,
          "decoder_belief_target_nll": 9.750123977661133,
          "decoder_belief_top1_on_support": 0.0625,
          "mean_decoder_belief_support_size": 2.73828125,
          "decoder_query_target_mass": 0.31492865085601807,
          "decoder_query_target_nll": 4.624454498291016,
          "decoder_query_top1_on_support": 0.34765625,
          "mean_decoder_query_support_size": 2.40625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 9,
          "query_type": "A",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.21840865910053253,
          "query_target_nll": 2.4213051795959473,
          "query_top1_on_support": 0.20703125,
          "mean_query_support_size": 2.40625,
          "probe_belief_target_mass": 0.2072945237159729,
          "probe_belief_target_nll": 3.1388463973999023,
          "probe_belief_top1_on_support": 0.27734375,
          "mean_belief_support_size": 2.73828125,
          "decoder_belief_target_mass": 0.20926669239997864,
          "decoder_belief_target_nll": 3.1585135459899902,
          "decoder_belief_top1_on_support": 0.28125,
          "mean_decoder_belief_support_size": 2.73828125,
          "decoder_query_target_mass": 0.48488250374794006,
          "decoder_query_target_nll": 1.6657767295837402,
          "decoder_query_top1_on_support": 0.5703125,
          "mean_decoder_query_support_size": 2.40625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 9,
          "query_type": "A",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.21840865910053253,
          "query_target_nll": 2.4213051795959473,
          "query_top1_on_support": 0.20703125,
          "mean_query_support_size": 2.40625,
          "probe_belief_target_mass": 0.2072945237159729,
          "probe_belief_target_nll": 3.1388463973999023,
          "probe_belief_top1_on_support": 0.27734375,
          "mean_belief_support_size": 2.73828125,
          "decoder_belief_target_mass": 0.20926669239997864,
          "decoder_belief_target_nll": 3.1585135459899902,
          "decoder_belief_top1_on_support": 0.28125,
          "mean_decoder_belief_support_size": 2.73828125,
          "decoder_query_target_mass": 0.48488250374794006,
          "decoder_query_target_nll": 1.6657767295837402,
          "decoder_query_top1_on_support": 0.5703125,
          "mean_decoder_query_support_size": 2.40625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 9,
          "query_type": "B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.19217932224273682,
          "query_target_nll": 2.421501398086548,
          "query_top1_on_support": 0.20703125,
          "mean_query_support_size": 2.1171875,
          "probe_belief_target_mass": 0.025784967467188835,
          "probe_belief_target_nll": 15.507177352905273,
          "probe_belief_top1_on_support": 0.02734375,
          "mean_belief_support_size": 2.37109375,
          "decoder_belief_target_mass": 0.024145497009158134,
          "decoder_belief_target_nll": 19.79434585571289,
          "decoder_belief_top1_on_support": 0.015625,
          "mean_decoder_belief_support_size": 2.37109375,
          "decoder_query_target_mass": 0.193277508020401,
          "decoder_query_target_nll": 2.402801990509033,
          "decoder_query_top1_on_support": 0.203125,
          "mean_decoder_query_support_size": 2.1171875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 9,
          "query_type": "B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.1920354962348938,
          "query_target_nll": 2.4222402572631836,
          "query_top1_on_support": 0.203125,
          "mean_query_support_size": 2.1171875,
          "probe_belief_target_mass": 0.024483980610966682,
          "probe_belief_target_nll": 13.657074928283691,
          "probe_belief_top1_on_support": 0.0234375,
          "mean_belief_support_size": 2.37109375,
          "decoder_belief_target_mass": 0.025735095143318176,
          "decoder_belief_target_nll": 15.954259872436523,
          "decoder_belief_top1_on_support": 0.0390625,
          "mean_decoder_belief_support_size": 2.37109375,
          "decoder_query_target_mass": 0.19393709301948547,
          "decoder_query_target_nll": 5.217952728271484,
          "decoder_query_top1_on_support": 0.21484375,
          "mean_decoder_query_support_size": 2.1171875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 9,
          "query_type": "B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.19015023112297058,
          "query_target_nll": 2.434356689453125,
          "query_top1_on_support": 0.1953125,
          "mean_query_support_size": 2.1171875,
          "probe_belief_target_mass": 0.03351377695798874,
          "probe_belief_target_nll": 12.154568672180176,
          "probe_belief_top1_on_support": 0.03515625,
          "mean_belief_support_size": 2.37109375,
          "decoder_belief_target_mass": 0.028447071090340614,
          "decoder_belief_target_nll": 13.809832572937012,
          "decoder_belief_top1_on_support": 0.015625,
          "mean_decoder_belief_support_size": 2.37109375,
          "decoder_query_target_mass": 0.2084154486656189,
          "decoder_query_target_nll": 5.396471977233887,
          "decoder_query_top1_on_support": 0.203125,
          "mean_decoder_query_support_size": 2.1171875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 9,
          "query_type": "B",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.19125506281852722,
          "query_target_nll": 2.4265365600585938,
          "query_top1_on_support": 0.20703125,
          "mean_query_support_size": 2.1171875,
          "probe_belief_target_mass": 0.03152240067720413,
          "probe_belief_target_nll": 11.085882186889648,
          "probe_belief_top1_on_support": 0.02734375,
          "mean_belief_support_size": 2.37109375,
          "decoder_belief_target_mass": 0.0323014035820961,
          "decoder_belief_target_nll": 12.456039428710938,
          "decoder_belief_top1_on_support": 0.0390625,
          "mean_decoder_belief_support_size": 2.37109375,
          "decoder_query_target_mass": 0.22293971478939056,
          "decoder_query_target_nll": 5.329303741455078,
          "decoder_query_top1_on_support": 0.2265625,
          "mean_decoder_query_support_size": 2.1171875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 9,
          "query_type": "B",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.1903984546661377,
          "query_target_nll": 2.4407505989074707,
          "query_top1_on_support": 0.203125,
          "mean_query_support_size": 2.1171875,
          "probe_belief_target_mass": 0.04283447563648224,
          "probe_belief_target_nll": 9.541810035705566,
          "probe_belief_top1_on_support": 0.046875,
          "mean_belief_support_size": 2.37109375,
          "decoder_belief_target_mass": 0.045041367411613464,
          "decoder_belief_target_nll": 10.542394638061523,
          "decoder_belief_top1_on_support": 0.05859375,
          "mean_decoder_belief_support_size": 2.37109375,
          "decoder_query_target_mass": 0.2574509084224701,
          "decoder_query_target_nll": 4.958747386932373,
          "decoder_query_top1_on_support": 0.26953125,
          "mean_decoder_query_support_size": 2.1171875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 9,
          "query_type": "B",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.1886242926120758,
          "query_target_nll": 2.4456357955932617,
          "query_top1_on_support": 0.18359375,
          "mean_query_support_size": 2.1171875,
          "probe_belief_target_mass": 0.20266272127628326,
          "probe_belief_target_nll": 2.9478683471679688,
          "probe_belief_top1_on_support": 0.28125,
          "mean_belief_support_size": 2.37109375,
          "decoder_belief_target_mass": 0.20748509466648102,
          "decoder_belief_target_nll": 2.9563560485839844,
          "decoder_belief_top1_on_support": 0.28125,
          "mean_decoder_belief_support_size": 2.37109375,
          "decoder_query_target_mass": 0.4631088376045227,
          "decoder_query_target_nll": 1.659258484840393,
          "decoder_query_top1_on_support": 0.55078125,
          "mean_decoder_query_support_size": 2.1171875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 9,
          "query_type": "B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.1886242926120758,
          "query_target_nll": 2.4456357955932617,
          "query_top1_on_support": 0.18359375,
          "mean_query_support_size": 2.1171875,
          "probe_belief_target_mass": 0.20266272127628326,
          "probe_belief_target_nll": 2.9478683471679688,
          "probe_belief_top1_on_support": 0.28125,
          "mean_belief_support_size": 2.37109375,
          "decoder_belief_target_mass": 0.20748509466648102,
          "decoder_belief_target_nll": 2.9563560485839844,
          "decoder_belief_top1_on_support": 0.28125,
          "mean_decoder_belief_support_size": 2.37109375,
          "decoder_query_target_mass": 0.4631088376045227,
          "decoder_query_target_nll": 1.659258484840393,
          "decoder_query_top1_on_support": 0.55078125,
          "mean_decoder_query_support_size": 2.1171875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 9,
          "query_type": "A_PLUS_B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.23008953034877777,
          "query_target_nll": 2.391150951385498,
          "query_top1_on_support": 0.23828125,
          "mean_query_support_size": 2.50390625,
          "probe_belief_target_mass": 0.022646905854344368,
          "probe_belief_target_nll": 15.650851249694824,
          "probe_belief_top1_on_support": 0.01953125,
          "mean_belief_support_size": 2.71875,
          "decoder_belief_target_mass": 0.022184669971466064,
          "decoder_belief_target_nll": 19.97473907470703,
          "decoder_belief_top1_on_support": 0.0234375,
          "mean_decoder_belief_support_size": 2.71875,
          "decoder_query_target_mass": 0.227195605635643,
          "decoder_query_target_nll": 2.4096341133117676,
          "decoder_query_top1_on_support": 0.22265625,
          "mean_decoder_query_support_size": 2.50390625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 9,
          "query_type": "A_PLUS_B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.2270626574754715,
          "query_target_nll": 2.4225237369537354,
          "query_top1_on_support": 0.21875,
          "mean_query_support_size": 2.50390625,
          "probe_belief_target_mass": 0.029886286705732346,
          "probe_belief_target_nll": 13.498055458068848,
          "probe_belief_top1_on_support": 0.01953125,
          "mean_belief_support_size": 2.71875,
          "decoder_belief_target_mass": 0.030882224440574646,
          "decoder_belief_target_nll": 15.74579906463623,
          "decoder_belief_top1_on_support": 0.0234375,
          "mean_decoder_belief_support_size": 2.71875,
          "decoder_query_target_mass": 0.23695239424705505,
          "decoder_query_target_nll": 3.8286538124084473,
          "decoder_query_top1_on_support": 0.24609375,
          "mean_decoder_query_support_size": 2.50390625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 9,
          "query_type": "A_PLUS_B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.22784513235092163,
          "query_target_nll": 2.4209632873535156,
          "query_top1_on_support": 0.25,
          "mean_query_support_size": 2.50390625,
          "probe_belief_target_mass": 0.02438339591026306,
          "probe_belief_target_nll": 12.10353946685791,
          "probe_belief_top1_on_support": 0.01953125,
          "mean_belief_support_size": 2.71875,
          "decoder_belief_target_mass": 0.026452062651515007,
          "decoder_belief_target_nll": 13.669113159179688,
          "decoder_belief_top1_on_support": 0.02734375,
          "mean_decoder_belief_support_size": 2.71875,
          "decoder_query_target_mass": 0.21507810056209564,
          "decoder_query_target_nll": 4.466488361358643,
          "decoder_query_top1_on_support": 0.22265625,
          "mean_decoder_query_support_size": 2.50390625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 9,
          "query_type": "A_PLUS_B",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.22836941480636597,
          "query_target_nll": 2.419605255126953,
          "query_top1_on_support": 0.25390625,
          "mean_query_support_size": 2.50390625,
          "probe_belief_target_mass": 0.04180474579334259,
          "probe_belief_target_nll": 10.577249526977539,
          "probe_belief_top1_on_support": 0.03125,
          "mean_belief_support_size": 2.71875,
          "decoder_belief_target_mass": 0.042642734944820404,
          "decoder_belief_target_nll": 11.806680679321289,
          "decoder_belief_top1_on_support": 0.03515625,
          "mean_decoder_belief_support_size": 2.71875,
          "decoder_query_target_mass": 0.2265668511390686,
          "decoder_query_target_nll": 4.114984035491943,
          "decoder_query_top1_on_support": 0.1875,
          "mean_decoder_query_support_size": 2.50390625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 9,
          "query_type": "A_PLUS_B",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.22875306010246277,
          "query_target_nll": 2.4164440631866455,
          "query_top1_on_support": 0.24609375,
          "mean_query_support_size": 2.50390625,
          "probe_belief_target_mass": 0.054439447820186615,
          "probe_belief_target_nll": 8.805171012878418,
          "probe_belief_top1_on_support": 0.06640625,
          "mean_belief_support_size": 2.71875,
          "decoder_belief_target_mass": 0.0525551363825798,
          "decoder_belief_target_nll": 9.628729820251465,
          "decoder_belief_top1_on_support": 0.06640625,
          "mean_decoder_belief_support_size": 2.71875,
          "decoder_query_target_mass": 0.21765679121017456,
          "decoder_query_target_nll": 3.629549741744995,
          "decoder_query_top1_on_support": 0.20703125,
          "mean_decoder_query_support_size": 2.50390625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 9,
          "query_type": "A_PLUS_B",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.2266707420349121,
          "query_target_nll": 2.424114942550659,
          "query_top1_on_support": 0.21484375,
          "mean_query_support_size": 2.50390625,
          "probe_belief_target_mass": 0.2175387144088745,
          "probe_belief_target_nll": 3.025674343109131,
          "probe_belief_top1_on_support": 0.2734375,
          "mean_belief_support_size": 2.71875,
          "decoder_belief_target_mass": 0.22546961903572083,
          "decoder_belief_target_nll": 2.989323854446411,
          "decoder_belief_top1_on_support": 0.3046875,
          "mean_decoder_belief_support_size": 2.71875,
          "decoder_query_target_mass": 0.4075146019458771,
          "decoder_query_target_nll": 1.8716799020767212,
          "decoder_query_top1_on_support": 0.4765625,
          "mean_decoder_query_support_size": 2.50390625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 9,
          "query_type": "A_PLUS_B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.2266707420349121,
          "query_target_nll": 2.424114942550659,
          "query_top1_on_support": 0.21484375,
          "mean_query_support_size": 2.50390625,
          "probe_belief_target_mass": 0.2175387144088745,
          "probe_belief_target_nll": 3.025674343109131,
          "probe_belief_top1_on_support": 0.2734375,
          "mean_belief_support_size": 2.71875,
          "decoder_belief_target_mass": 0.22546961903572083,
          "decoder_belief_target_nll": 2.989323854446411,
          "decoder_belief_top1_on_support": 0.3046875,
          "mean_decoder_belief_support_size": 2.71875,
          "decoder_query_target_mass": 0.4075146019458771,
          "decoder_query_target_nll": 1.8716799020767212,
          "decoder_query_top1_on_support": 0.4765625,
          "mean_decoder_query_support_size": 2.50390625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 9,
          "query_type": "A_MINUS_B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.2133101373910904,
          "query_target_nll": 2.431548595428467,
          "query_top1_on_support": 0.2109375,
          "mean_query_support_size": 2.35546875,
          "probe_belief_target_mass": 0.016749586910009384,
          "probe_belief_target_nll": 15.958711624145508,
          "probe_belief_top1_on_support": 0.0078125,
          "mean_belief_support_size": 2.48046875,
          "decoder_belief_target_mass": 0.01688450574874878,
          "decoder_belief_target_nll": 20.352447509765625,
          "decoder_belief_top1_on_support": 0.0078125,
          "mean_decoder_belief_support_size": 2.48046875,
          "decoder_query_target_mass": 0.1835939884185791,
          "decoder_query_target_nll": 15.857067108154297,
          "decoder_query_top1_on_support": 0.18359375,
          "mean_decoder_query_support_size": 2.35546875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 9,
          "query_type": "A_MINUS_B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.21514980494976044,
          "query_target_nll": 2.413130521774292,
          "query_top1_on_support": 0.203125,
          "mean_query_support_size": 2.35546875,
          "probe_belief_target_mass": 0.021334489807486534,
          "probe_belief_target_nll": 13.340267181396484,
          "probe_belief_top1_on_support": 0.0234375,
          "mean_belief_support_size": 2.48046875,
          "decoder_belief_target_mass": 0.021612316370010376,
          "decoder_belief_target_nll": 15.386428833007812,
          "decoder_belief_top1_on_support": 0.015625,
          "mean_decoder_belief_support_size": 2.48046875,
          "decoder_query_target_mass": 0.19958209991455078,
          "decoder_query_target_nll": 8.419122695922852,
          "decoder_query_top1_on_support": 0.19140625,
          "mean_decoder_query_support_size": 2.35546875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 9,
          "query_type": "A_MINUS_B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.2128385305404663,
          "query_target_nll": 2.4280152320861816,
          "query_top1_on_support": 0.20703125,
          "mean_query_support_size": 2.35546875,
          "probe_belief_target_mass": 0.02307635173201561,
          "probe_belief_target_nll": 11.973088264465332,
          "probe_belief_top1_on_support": 0.0234375,
          "mean_belief_support_size": 2.48046875,
          "decoder_belief_target_mass": 0.02115948125720024,
          "decoder_belief_target_nll": 13.465518951416016,
          "decoder_belief_top1_on_support": 0.01171875,
          "mean_decoder_belief_support_size": 2.48046875,
          "decoder_query_target_mass": 0.2365237921476364,
          "decoder_query_target_nll": 7.115877628326416,
          "decoder_query_top1_on_support": 0.2109375,
          "mean_decoder_query_support_size": 2.35546875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 9,
          "query_type": "A_MINUS_B",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.21350878477096558,
          "query_target_nll": 2.428291082382202,
          "query_top1_on_support": 0.20703125,
          "mean_query_support_size": 2.35546875,
          "probe_belief_target_mass": 0.02870083972811699,
          "probe_belief_target_nll": 11.459609985351562,
          "probe_belief_top1_on_support": 0.02734375,
          "mean_belief_support_size": 2.48046875,
          "decoder_belief_target_mass": 0.02880816161632538,
          "decoder_belief_target_nll": 12.7356538772583,
          "decoder_belief_top1_on_support": 0.03125,
          "mean_decoder_belief_support_size": 2.48046875,
          "decoder_query_target_mass": 0.24276131391525269,
          "decoder_query_target_nll": 6.211129665374756,
          "decoder_query_top1_on_support": 0.24609375,
          "mean_decoder_query_support_size": 2.35546875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 9,
          "query_type": "A_MINUS_B",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.2142428159713745,
          "query_target_nll": 2.422636032104492,
          "query_top1_on_support": 0.2109375,
          "mean_query_support_size": 2.35546875,
          "probe_belief_target_mass": 0.05013090744614601,
          "probe_belief_target_nll": 9.368474960327148,
          "probe_belief_top1_on_support": 0.06640625,
          "mean_belief_support_size": 2.48046875,
          "decoder_belief_target_mass": 0.051718905568122864,
          "decoder_belief_target_nll": 10.272544860839844,
          "decoder_belief_top1_on_support": 0.05078125,
          "mean_decoder_belief_support_size": 2.48046875,
          "decoder_query_target_mass": 0.23347535729408264,
          "decoder_query_target_nll": 4.853728294372559,
          "decoder_query_top1_on_support": 0.2421875,
          "mean_decoder_query_support_size": 2.35546875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 9,
          "query_type": "A_MINUS_B",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.21663126349449158,
          "query_target_nll": 2.406384229660034,
          "query_top1_on_support": 0.23046875,
          "mean_query_support_size": 2.35546875,
          "probe_belief_target_mass": 0.20806412398815155,
          "probe_belief_target_nll": 3.0457448959350586,
          "probe_belief_top1_on_support": 0.2578125,
          "mean_belief_support_size": 2.48046875,
          "decoder_belief_target_mass": 0.22238636016845703,
          "decoder_belief_target_nll": 3.016629219055176,
          "decoder_belief_top1_on_support": 0.28515625,
          "mean_decoder_belief_support_size": 2.48046875,
          "decoder_query_target_mass": 0.4353267550468445,
          "decoder_query_target_nll": 1.847685694694519,
          "decoder_query_top1_on_support": 0.4765625,
          "mean_decoder_query_support_size": 2.35546875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 9,
          "query_type": "A_MINUS_B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.21663126349449158,
          "query_target_nll": 2.406384229660034,
          "query_top1_on_support": 0.23046875,
          "mean_query_support_size": 2.35546875,
          "probe_belief_target_mass": 0.20806412398815155,
          "probe_belief_target_nll": 3.0457448959350586,
          "probe_belief_top1_on_support": 0.2578125,
          "mean_belief_support_size": 2.48046875,
          "decoder_belief_target_mass": 0.22238636016845703,
          "decoder_belief_target_nll": 3.016629219055176,
          "decoder_belief_top1_on_support": 0.28515625,
          "mean_decoder_belief_support_size": 2.48046875,
          "decoder_query_target_mass": 0.4353267550468445,
          "decoder_query_target_nll": 1.847685694694519,
          "decoder_query_top1_on_support": 0.4765625,
          "mean_decoder_query_support_size": 2.35546875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 12,
          "query_type": "A",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.17462894320487976,
          "query_target_nll": 2.4141242504119873,
          "query_top1_on_support": 0.203125,
          "mean_query_support_size": 1.90625,
          "probe_belief_target_mass": 0.01718699000775814,
          "probe_belief_target_nll": 15.629461288452148,
          "probe_belief_top1_on_support": 0.0078125,
          "mean_belief_support_size": 2.21484375,
          "decoder_belief_target_mass": 0.01738426275551319,
          "decoder_belief_target_nll": 19.74510955810547,
          "decoder_belief_top1_on_support": 0.0078125,
          "mean_decoder_belief_support_size": 2.21484375,
          "decoder_query_target_mass": 0.17225682735443115,
          "decoder_query_target_nll": 2.4133760929107666,
          "decoder_query_top1_on_support": 0.1171875,
          "mean_decoder_query_support_size": 1.90625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 12,
          "query_type": "A",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.17175087332725525,
          "query_target_nll": 2.4284017086029053,
          "query_top1_on_support": 0.12890625,
          "mean_query_support_size": 1.90625,
          "probe_belief_target_mass": 0.028218481689691544,
          "probe_belief_target_nll": 13.541804313659668,
          "probe_belief_top1_on_support": 0.02734375,
          "mean_belief_support_size": 2.21484375,
          "decoder_belief_target_mass": 0.02816130220890045,
          "decoder_belief_target_nll": 15.761367797851562,
          "decoder_belief_top1_on_support": 0.015625,
          "mean_decoder_belief_support_size": 2.21484375,
          "decoder_query_target_mass": 0.18960072100162506,
          "decoder_query_target_nll": 4.74046516418457,
          "decoder_query_top1_on_support": 0.20703125,
          "mean_decoder_query_support_size": 1.90625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 12,
          "query_type": "A",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.17126703262329102,
          "query_target_nll": 2.430234432220459,
          "query_top1_on_support": 0.140625,
          "mean_query_support_size": 1.90625,
          "probe_belief_target_mass": 0.01791740208864212,
          "probe_belief_target_nll": 12.737646102905273,
          "probe_belief_top1_on_support": 0.015625,
          "mean_belief_support_size": 2.21484375,
          "decoder_belief_target_mass": 0.017902381718158722,
          "decoder_belief_target_nll": 14.403749465942383,
          "decoder_belief_top1_on_support": 0.03125,
          "mean_decoder_belief_support_size": 2.21484375,
          "decoder_query_target_mass": 0.18418067693710327,
          "decoder_query_target_nll": 5.650548458099365,
          "decoder_query_top1_on_support": 0.19921875,
          "mean_decoder_query_support_size": 1.90625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 12,
          "query_type": "A",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.1699678897857666,
          "query_target_nll": 2.443767547607422,
          "query_top1_on_support": 0.125,
          "mean_query_support_size": 1.90625,
          "probe_belief_target_mass": 0.02477056346833706,
          "probe_belief_target_nll": 11.207929611206055,
          "probe_belief_top1_on_support": 0.0234375,
          "mean_belief_support_size": 2.21484375,
          "decoder_belief_target_mass": 0.022814761847257614,
          "decoder_belief_target_nll": 12.496767044067383,
          "decoder_belief_top1_on_support": 0.01953125,
          "mean_decoder_belief_support_size": 2.21484375,
          "decoder_query_target_mass": 0.1798866242170334,
          "decoder_query_target_nll": 5.742835998535156,
          "decoder_query_top1_on_support": 0.16015625,
          "mean_decoder_query_support_size": 1.90625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 12,
          "query_type": "A",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.1710362732410431,
          "query_target_nll": 2.4335575103759766,
          "query_top1_on_support": 0.15625,
          "mean_query_support_size": 1.90625,
          "probe_belief_target_mass": 0.025061335414648056,
          "probe_belief_target_nll": 9.854846954345703,
          "probe_belief_top1_on_support": 0.03125,
          "mean_belief_support_size": 2.21484375,
          "decoder_belief_target_mass": 0.026679355651140213,
          "decoder_belief_target_nll": 10.762558937072754,
          "decoder_belief_top1_on_support": 0.0234375,
          "mean_decoder_belief_support_size": 2.21484375,
          "decoder_query_target_mass": 0.18973511457443237,
          "decoder_query_target_nll": 5.0829854011535645,
          "decoder_query_top1_on_support": 0.18359375,
          "mean_decoder_query_support_size": 1.90625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 12,
          "query_type": "A",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.17106115818023682,
          "query_target_nll": 2.4290361404418945,
          "query_top1_on_support": 0.1484375,
          "mean_query_support_size": 1.90625,
          "probe_belief_target_mass": 0.05106022581458092,
          "probe_belief_target_nll": 8.627975463867188,
          "probe_belief_top1_on_support": 0.05078125,
          "mean_belief_support_size": 2.21484375,
          "decoder_belief_target_mass": 0.05080382525920868,
          "decoder_belief_target_nll": 9.393651962280273,
          "decoder_belief_top1_on_support": 0.05859375,
          "mean_decoder_belief_support_size": 2.21484375,
          "decoder_query_target_mass": 0.24550870060920715,
          "decoder_query_target_nll": 4.214076995849609,
          "decoder_query_top1_on_support": 0.2421875,
          "mean_decoder_query_support_size": 1.90625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 12,
          "query_type": "A",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.17062002420425415,
          "query_target_nll": 2.4339184761047363,
          "query_top1_on_support": 0.125,
          "mean_query_support_size": 1.90625,
          "probe_belief_target_mass": 0.16080564260482788,
          "probe_belief_target_nll": 3.1984262466430664,
          "probe_belief_top1_on_support": 0.234375,
          "mean_belief_support_size": 2.21484375,
          "decoder_belief_target_mass": 0.1608790159225464,
          "decoder_belief_target_nll": 3.2072556018829346,
          "decoder_belief_top1_on_support": 0.1875,
          "mean_decoder_belief_support_size": 2.21484375,
          "decoder_query_target_mass": 0.4143306016921997,
          "decoder_query_target_nll": 1.6439239978790283,
          "decoder_query_top1_on_support": 0.48046875,
          "mean_decoder_query_support_size": 1.90625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 12,
          "query_type": "B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.15211495757102966,
          "query_target_nll": 2.4137580394744873,
          "query_top1_on_support": 0.15625,
          "mean_query_support_size": 1.6796875,
          "probe_belief_target_mass": 0.01312563568353653,
          "probe_belief_target_nll": 15.887737274169922,
          "probe_belief_top1_on_support": 0.0078125,
          "mean_belief_support_size": 1.84765625,
          "decoder_belief_target_mass": 0.012936179526150227,
          "decoder_belief_target_nll": 20.22970199584961,
          "decoder_belief_top1_on_support": 0.01171875,
          "mean_decoder_belief_support_size": 1.84765625,
          "decoder_query_target_mass": 0.1515640914440155,
          "decoder_query_target_nll": 2.412230968475342,
          "decoder_query_top1_on_support": 0.140625,
          "mean_decoder_query_support_size": 1.6796875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 12,
          "query_type": "B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.15305885672569275,
          "query_target_nll": 2.4115123748779297,
          "query_top1_on_support": 0.16015625,
          "mean_query_support_size": 1.6796875,
          "probe_belief_target_mass": 0.014185957610607147,
          "probe_belief_target_nll": 13.538520812988281,
          "probe_belief_top1_on_support": 0.015625,
          "mean_belief_support_size": 1.84765625,
          "decoder_belief_target_mass": 0.01437797024846077,
          "decoder_belief_target_nll": 15.643942832946777,
          "decoder_belief_top1_on_support": 0.0078125,
          "mean_decoder_belief_support_size": 1.84765625,
          "decoder_query_target_mass": 0.15523579716682434,
          "decoder_query_target_nll": 4.612979888916016,
          "decoder_query_top1_on_support": 0.171875,
          "mean_decoder_query_support_size": 1.6796875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 12,
          "query_type": "B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.15364877879619598,
          "query_target_nll": 2.4057326316833496,
          "query_top1_on_support": 0.13671875,
          "mean_query_support_size": 1.6796875,
          "probe_belief_target_mass": 0.013956759124994278,
          "probe_belief_target_nll": 12.04821491241455,
          "probe_belief_top1_on_support": 0.015625,
          "mean_belief_support_size": 1.84765625,
          "decoder_belief_target_mass": 0.01612776890397072,
          "decoder_belief_target_nll": 13.342222213745117,
          "decoder_belief_top1_on_support": 0.0234375,
          "mean_decoder_belief_support_size": 1.84765625,
          "decoder_query_target_mass": 0.170750692486763,
          "decoder_query_target_nll": 5.520190238952637,
          "decoder_query_top1_on_support": 0.16015625,
          "mean_decoder_query_support_size": 1.6796875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 12,
          "query_type": "B",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.15608935058116913,
          "query_target_nll": 2.388587474822998,
          "query_top1_on_support": 0.16796875,
          "mean_query_support_size": 1.6796875,
          "probe_belief_target_mass": 0.011102870106697083,
          "probe_belief_target_nll": 11.288517951965332,
          "probe_belief_top1_on_support": 0.0078125,
          "mean_belief_support_size": 1.84765625,
          "decoder_belief_target_mass": 0.013298971578478813,
          "decoder_belief_target_nll": 12.488784790039062,
          "decoder_belief_top1_on_support": 0.01953125,
          "mean_decoder_belief_support_size": 1.84765625,
          "decoder_query_target_mass": 0.1658865511417389,
          "decoder_query_target_nll": 5.4214701652526855,
          "decoder_query_top1_on_support": 0.16015625,
          "mean_decoder_query_support_size": 1.6796875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 12,
          "query_type": "B",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.1539210081100464,
          "query_target_nll": 2.4050779342651367,
          "query_top1_on_support": 0.1484375,
          "mean_query_support_size": 1.6796875,
          "probe_belief_target_mass": 0.019355760887265205,
          "probe_belief_target_nll": 10.15687370300293,
          "probe_belief_top1_on_support": 0.01953125,
          "mean_belief_support_size": 1.84765625,
          "decoder_belief_target_mass": 0.020197562873363495,
          "decoder_belief_target_nll": 11.241415023803711,
          "decoder_belief_top1_on_support": 0.03515625,
          "mean_decoder_belief_support_size": 1.84765625,
          "decoder_query_target_mass": 0.15970349311828613,
          "decoder_query_target_nll": 5.446552276611328,
          "decoder_query_top1_on_support": 0.16015625,
          "mean_decoder_query_support_size": 1.6796875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 12,
          "query_type": "B",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.1533793807029724,
          "query_target_nll": 2.407517433166504,
          "query_top1_on_support": 0.13671875,
          "mean_query_support_size": 1.6796875,
          "probe_belief_target_mass": 0.04069221019744873,
          "probe_belief_target_nll": 8.2049560546875,
          "probe_belief_top1_on_support": 0.0546875,
          "mean_belief_support_size": 1.84765625,
          "decoder_belief_target_mass": 0.03996644914150238,
          "decoder_belief_target_nll": 9.096324920654297,
          "decoder_belief_top1_on_support": 0.05078125,
          "mean_decoder_belief_support_size": 1.84765625,
          "decoder_query_target_mass": 0.22389180958271027,
          "decoder_query_target_nll": 4.094874858856201,
          "decoder_query_top1_on_support": 0.23828125,
          "mean_decoder_query_support_size": 1.6796875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 12,
          "query_type": "B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.15350399911403656,
          "query_target_nll": 2.4109764099121094,
          "query_top1_on_support": 0.17578125,
          "mean_query_support_size": 1.6796875,
          "probe_belief_target_mass": 0.15458296239376068,
          "probe_belief_target_nll": 3.237107753753662,
          "probe_belief_top1_on_support": 0.2265625,
          "mean_belief_support_size": 1.84765625,
          "decoder_belief_target_mass": 0.14958572387695312,
          "decoder_belief_target_nll": 3.3036465644836426,
          "decoder_belief_top1_on_support": 0.22265625,
          "mean_decoder_belief_support_size": 1.84765625,
          "decoder_query_target_mass": 0.37103337049484253,
          "decoder_query_target_nll": 1.757891058921814,
          "decoder_query_top1_on_support": 0.421875,
          "mean_decoder_query_support_size": 1.6796875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 12,
          "query_type": "A_PLUS_B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.17687882483005524,
          "query_target_nll": 2.4070777893066406,
          "query_top1_on_support": 0.19921875,
          "mean_query_support_size": 1.93359375,
          "probe_belief_target_mass": 0.024815086275339127,
          "probe_belief_target_nll": 15.313039779663086,
          "probe_belief_top1_on_support": 0.0234375,
          "mean_belief_support_size": 2.0,
          "decoder_belief_target_mass": 0.024964656680822372,
          "decoder_belief_target_nll": 19.472396850585938,
          "decoder_belief_top1_on_support": 0.0234375,
          "mean_decoder_belief_support_size": 2.0,
          "decoder_query_target_mass": 0.17563292384147644,
          "decoder_query_target_nll": 2.4035120010375977,
          "decoder_query_top1_on_support": 0.15234375,
          "mean_decoder_query_support_size": 1.93359375
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 12,
          "query_type": "A_PLUS_B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.17644010484218597,
          "query_target_nll": 2.4085464477539062,
          "query_top1_on_support": 0.1953125,
          "mean_query_support_size": 1.93359375,
          "probe_belief_target_mass": 0.01613939180970192,
          "probe_belief_target_nll": 13.914175033569336,
          "probe_belief_top1_on_support": 0.0234375,
          "mean_belief_support_size": 2.0,
          "decoder_belief_target_mass": 0.015676848590373993,
          "decoder_belief_target_nll": 16.09649658203125,
          "decoder_belief_top1_on_support": 0.01171875,
          "mean_decoder_belief_support_size": 2.0,
          "decoder_query_target_mass": 0.18647733330726624,
          "decoder_query_target_nll": 4.109813690185547,
          "decoder_query_top1_on_support": 0.19921875,
          "mean_decoder_query_support_size": 1.93359375
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 12,
          "query_type": "A_PLUS_B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.17787159979343414,
          "query_target_nll": 2.3936166763305664,
          "query_top1_on_support": 0.2109375,
          "mean_query_support_size": 1.93359375,
          "probe_belief_target_mass": 0.019857686012983322,
          "probe_belief_target_nll": 12.800413131713867,
          "probe_belief_top1_on_support": 0.015625,
          "mean_belief_support_size": 2.0,
          "decoder_belief_target_mass": 0.0199945829808712,
          "decoder_belief_target_nll": 14.33509635925293,
          "decoder_belief_top1_on_support": 0.01953125,
          "mean_decoder_belief_support_size": 2.0,
          "decoder_query_target_mass": 0.17168056964874268,
          "decoder_query_target_nll": 4.359709739685059,
          "decoder_query_top1_on_support": 0.1484375,
          "mean_decoder_query_support_size": 1.93359375
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 12,
          "query_type": "A_PLUS_B",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.17695339024066925,
          "query_target_nll": 2.396679401397705,
          "query_top1_on_support": 0.1953125,
          "mean_query_support_size": 1.93359375,
          "probe_belief_target_mass": 0.015294811688363552,
          "probe_belief_target_nll": 12.327058792114258,
          "probe_belief_top1_on_support": 0.01171875,
          "mean_belief_support_size": 2.0,
          "decoder_belief_target_mass": 0.016918513923883438,
          "decoder_belief_target_nll": 13.63517951965332,
          "decoder_belief_top1_on_support": 0.01171875,
          "mean_decoder_belief_support_size": 2.0,
          "decoder_query_target_mass": 0.17465835809707642,
          "decoder_query_target_nll": 4.3768744468688965,
          "decoder_query_top1_on_support": 0.1953125,
          "mean_decoder_query_support_size": 1.93359375
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 12,
          "query_type": "A_PLUS_B",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.1769425868988037,
          "query_target_nll": 2.3987388610839844,
          "query_top1_on_support": 0.140625,
          "mean_query_support_size": 1.93359375,
          "probe_belief_target_mass": 0.022439219057559967,
          "probe_belief_target_nll": 10.24063491821289,
          "probe_belief_top1_on_support": 0.05078125,
          "mean_belief_support_size": 2.0,
          "decoder_belief_target_mass": 0.0255202017724514,
          "decoder_belief_target_nll": 11.264364242553711,
          "decoder_belief_top1_on_support": 0.03515625,
          "mean_decoder_belief_support_size": 2.0,
          "decoder_query_target_mass": 0.1726289987564087,
          "decoder_query_target_nll": 3.7814781665802,
          "decoder_query_top1_on_support": 0.15234375,
          "mean_decoder_query_support_size": 1.93359375
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 12,
          "query_type": "A_PLUS_B",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.17639392614364624,
          "query_target_nll": 2.406798839569092,
          "query_top1_on_support": 0.16796875,
          "mean_query_support_size": 1.93359375,
          "probe_belief_target_mass": 0.03654772415757179,
          "probe_belief_target_nll": 8.653221130371094,
          "probe_belief_top1_on_support": 0.0390625,
          "mean_belief_support_size": 2.0,
          "decoder_belief_target_mass": 0.03433029353618622,
          "decoder_belief_target_nll": 9.442667961120605,
          "decoder_belief_top1_on_support": 0.03515625,
          "mean_decoder_belief_support_size": 2.0,
          "decoder_query_target_mass": 0.17917048931121826,
          "decoder_query_target_nll": 3.6488418579101562,
          "decoder_query_top1_on_support": 0.17578125,
          "mean_decoder_query_support_size": 1.93359375
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 12,
          "query_type": "A_PLUS_B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.1748615950345993,
          "query_target_nll": 2.418363571166992,
          "query_top1_on_support": 0.1796875,
          "mean_query_support_size": 1.93359375,
          "probe_belief_target_mass": 0.15657415986061096,
          "probe_belief_target_nll": 3.1982884407043457,
          "probe_belief_top1_on_support": 0.2109375,
          "mean_belief_support_size": 2.0,
          "decoder_belief_target_mass": 0.1539027839899063,
          "decoder_belief_target_nll": 3.191218137741089,
          "decoder_belief_top1_on_support": 0.21484375,
          "mean_decoder_belief_support_size": 2.0,
          "decoder_query_target_mass": 0.2877967953681946,
          "decoder_query_target_nll": 2.057342052459717,
          "decoder_query_top1_on_support": 0.3125,
          "mean_decoder_query_support_size": 1.93359375
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 12,
          "query_type": "A_MINUS_B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.1778937578201294,
          "query_target_nll": 2.4168577194213867,
          "query_top1_on_support": 0.17578125,
          "mean_query_support_size": 1.95703125,
          "probe_belief_target_mass": 0.020947713404893875,
          "probe_belief_target_nll": 15.386316299438477,
          "probe_belief_top1_on_support": 0.0234375,
          "mean_belief_support_size": 2.17578125,
          "decoder_belief_target_mass": 0.02102092280983925,
          "decoder_belief_target_nll": 19.528385162353516,
          "decoder_belief_top1_on_support": 0.0234375,
          "mean_decoder_belief_support_size": 2.17578125,
          "decoder_query_target_mass": 0.21484358608722687,
          "decoder_query_target_nll": 15.421475410461426,
          "decoder_query_top1_on_support": 0.21484375,
          "mean_decoder_query_support_size": 1.95703125
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 12,
          "query_type": "A_MINUS_B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.17894107103347778,
          "query_target_nll": 2.408451795578003,
          "query_top1_on_support": 0.1875,
          "mean_query_support_size": 1.95703125,
          "probe_belief_target_mass": 0.016421005129814148,
          "probe_belief_target_nll": 13.984004020690918,
          "probe_belief_top1_on_support": 0.00390625,
          "mean_belief_support_size": 2.17578125,
          "decoder_belief_target_mass": 0.01659698784351349,
          "decoder_belief_target_nll": 16.389606475830078,
          "decoder_belief_top1_on_support": 0.01171875,
          "mean_decoder_belief_support_size": 2.17578125,
          "decoder_query_target_mass": 0.18329191207885742,
          "decoder_query_target_nll": 8.765880584716797,
          "decoder_query_top1_on_support": 0.171875,
          "mean_decoder_query_support_size": 1.95703125
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 12,
          "query_type": "A_MINUS_B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.17970244586467743,
          "query_target_nll": 2.409942626953125,
          "query_top1_on_support": 0.2109375,
          "mean_query_support_size": 1.95703125,
          "probe_belief_target_mass": 0.015474754385650158,
          "probe_belief_target_nll": 12.669325828552246,
          "probe_belief_top1_on_support": 0.01171875,
          "mean_belief_support_size": 2.17578125,
          "decoder_belief_target_mass": 0.015183680690824986,
          "decoder_belief_target_nll": 14.287564277648926,
          "decoder_belief_top1_on_support": 0.0,
          "mean_decoder_belief_support_size": 2.17578125,
          "decoder_query_target_mass": 0.16078846156597137,
          "decoder_query_target_nll": 7.105837821960449,
          "decoder_query_top1_on_support": 0.1640625,
          "mean_decoder_query_support_size": 1.95703125
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 12,
          "query_type": "A_MINUS_B",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.1791113018989563,
          "query_target_nll": 2.411658763885498,
          "query_top1_on_support": 0.1953125,
          "mean_query_support_size": 1.95703125,
          "probe_belief_target_mass": 0.01534990780055523,
          "probe_belief_target_nll": 11.412936210632324,
          "probe_belief_top1_on_support": 0.0078125,
          "mean_belief_support_size": 2.17578125,
          "decoder_belief_target_mass": 0.01555049791932106,
          "decoder_belief_target_nll": 12.743518829345703,
          "decoder_belief_top1_on_support": 0.015625,
          "mean_decoder_belief_support_size": 2.17578125,
          "decoder_query_target_mass": 0.16706688702106476,
          "decoder_query_target_nll": 6.054551124572754,
          "decoder_query_top1_on_support": 0.1796875,
          "mean_decoder_query_support_size": 1.95703125
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 12,
          "query_type": "A_MINUS_B",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.1786377876996994,
          "query_target_nll": 2.407041549682617,
          "query_top1_on_support": 0.16796875,
          "mean_query_support_size": 1.95703125,
          "probe_belief_target_mass": 0.018628336489200592,
          "probe_belief_target_nll": 10.152790069580078,
          "probe_belief_top1_on_support": 0.01171875,
          "mean_belief_support_size": 2.17578125,
          "decoder_belief_target_mass": 0.019317857921123505,
          "decoder_belief_target_nll": 11.259057998657227,
          "decoder_belief_top1_on_support": 0.015625,
          "mean_decoder_belief_support_size": 2.17578125,
          "decoder_query_target_mass": 0.18633073568344116,
          "decoder_query_target_nll": 4.581335544586182,
          "decoder_query_top1_on_support": 0.2109375,
          "mean_decoder_query_support_size": 1.95703125
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 12,
          "query_type": "A_MINUS_B",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.17947810888290405,
          "query_target_nll": 2.4123356342315674,
          "query_top1_on_support": 0.203125,
          "mean_query_support_size": 1.95703125,
          "probe_belief_target_mass": 0.040985602885484695,
          "probe_belief_target_nll": 8.058320045471191,
          "probe_belief_top1_on_support": 0.0390625,
          "mean_belief_support_size": 2.17578125,
          "decoder_belief_target_mass": 0.04201223701238632,
          "decoder_belief_target_nll": 8.921529769897461,
          "decoder_belief_top1_on_support": 0.0546875,
          "mean_decoder_belief_support_size": 2.17578125,
          "decoder_query_target_mass": 0.1971958726644516,
          "decoder_query_target_nll": 4.247076034545898,
          "decoder_query_top1_on_support": 0.19921875,
          "mean_decoder_query_support_size": 1.95703125
        },
        {
          "model": "dense",
          "variant": "pilot_gru_mlp_d128",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "mlp",
          "state_dim": 128,
          "instr_dim": 96,
          "decoder_rank": 16,
          "length": 12,
          "query_type": "A_MINUS_B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.17856577038764954,
          "query_target_nll": 2.407665252685547,
          "query_top1_on_support": 0.1796875,
          "mean_query_support_size": 1.95703125,
          "probe_belief_target_mass": 0.13653075695037842,
          "probe_belief_target_nll": 3.266841411590576,
          "probe_belief_top1_on_support": 0.17578125,
          "mean_belief_support_size": 2.17578125,
          "decoder_belief_target_mass": 0.134201318025589,
          "decoder_belief_target_nll": 3.312962055206299,
          "decoder_belief_top1_on_support": 0.1640625,
          "mean_decoder_belief_support_size": 2.17578125,
          "decoder_query_target_mass": 0.33567726612091064,
          "decoder_query_target_nll": 1.9736098051071167,
          "decoder_query_top1_on_support": 0.37890625,
          "mean_decoder_query_support_size": 1.95703125
        }
      ]
    }
  ],
  "checkpoints": [
    "large_artifacts/dense_teacher_distillation/checkpoints/pilot_gru_mlp_d128/checkpoint_final.pt"
  ]
}