{
  "mode": "dense",
  "variant": "pilot_gru_lowrank_d256_r16",
  "transition": "gru",
  "decoder_type": "low_rank",
  "args": {
    "variant_name": "pilot_gru_lowrank_d256_r16",
    "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": 256,
    "instr_dim": 128,
    "transition": "gru",
    "decoder_type": "low_rank",
    "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_lowrank_d256_r16",
    "checkpoint_dir": "large_artifacts/dense_teacher_distillation/checkpoints/pilot_gru_lowrank_d256_r16"
  },
  "env": {
    "python": "3.12.3",
    "torch": "2.8.0+cu128",
    "device": "cuda",
    "cuda_name": "NVIDIA RTX 6000 Ada Generation"
  },
  "train": [
    {
      "step": 1,
      "loss": 4.7960944175720215,
      "grad_norm": 0.1365613341331482,
      "query_loss": 0.0,
      "belief_loss": 4.7960944175720215,
      "decoder_query_loss": 0.0,
      "query_target_mass": 0.4239775240421295,
      "query_target_nll": 2.410278081893921,
      "query_top1_on_support": 0.4140625,
      "mean_query_support_size": 4.65625,
      "decoder_belief_target_mass": 0.050744444131851196,
      "decoder_belief_target_nll": 4.689797878265381,
      "decoder_belief_top1_on_support": 0.125,
      "mean_decoder_belief_support_size": 5.6953125,
      "decoder_query_target_mass": 0.42687535285949707,
      "decoder_query_target_nll": 2.3739848136901855,
      "decoder_query_top1_on_support": 0.51953125,
      "mean_decoder_query_support_size": 4.65625
    },
    {
      "step": 100,
      "loss": 3.2369368076324463,
      "grad_norm": 0.33899933099746704,
      "query_loss": 0.0,
      "belief_loss": 3.2369368076324463,
      "decoder_query_loss": 0.0,
      "query_target_mass": 0.43101632595062256,
      "query_target_nll": 2.4141016006469727,
      "query_top1_on_support": 0.44921875,
      "mean_query_support_size": 4.73046875,
      "decoder_belief_target_mass": 0.2500874400138855,
      "decoder_belief_target_nll": 3.67159104347229,
      "decoder_belief_top1_on_support": 0.2890625,
      "mean_decoder_belief_support_size": 5.9453125,
      "decoder_query_target_mass": 0.5732617378234863,
      "decoder_query_target_nll": 2.0627856254577637,
      "decoder_query_top1_on_support": 0.58203125,
      "mean_decoder_query_support_size": 4.73046875
    },
    {
      "step": 200,
      "loss": 3.262704372406006,
      "grad_norm": 0.3188554048538208,
      "query_loss": 0.0,
      "belief_loss": 3.262704372406006,
      "decoder_query_loss": 0.0,
      "query_target_mass": 0.47486239671707153,
      "query_target_nll": 2.408295154571533,
      "query_top1_on_support": 0.5,
      "mean_query_support_size": 5.20703125,
      "decoder_belief_target_mass": 0.2689853310585022,
      "decoder_belief_target_nll": 3.5410382747650146,
      "decoder_belief_top1_on_support": 0.3046875,
      "mean_decoder_belief_support_size": 5.75390625,
      "decoder_query_target_mass": 0.636425793170929,
      "decoder_query_target_nll": 2.0251142978668213,
      "decoder_query_top1_on_support": 0.6875,
      "mean_decoder_query_support_size": 5.20703125
    },
    {
      "step": 300,
      "loss": 3.1189277172088623,
      "grad_norm": 0.38044220209121704,
      "query_loss": 0.0,
      "belief_loss": 3.1189277172088623,
      "decoder_query_loss": 0.0,
      "query_target_mass": 0.4498646855354309,
      "query_target_nll": 2.416898727416992,
      "query_top1_on_support": 0.41796875,
      "mean_query_support_size": 4.94921875,
      "decoder_belief_target_mass": 0.3029497563838959,
      "decoder_belief_target_nll": 3.3816559314727783,
      "decoder_belief_top1_on_support": 0.3671875,
      "mean_decoder_belief_support_size": 5.81640625,
      "decoder_query_target_mass": 0.6558192372322083,
      "decoder_query_target_nll": 1.914000153541565,
      "decoder_query_top1_on_support": 0.71484375,
      "mean_decoder_query_support_size": 4.94921875
    },
    {
      "step": 400,
      "loss": 2.989154100418091,
      "grad_norm": 0.4023982584476471,
      "query_loss": 0.0,
      "belief_loss": 2.989154100418091,
      "decoder_query_loss": 0.0,
      "query_target_mass": 0.43072670698165894,
      "query_target_nll": 2.41259765625,
      "query_top1_on_support": 0.42578125,
      "mean_query_support_size": 4.734375,
      "decoder_belief_target_mass": 0.2959853410720825,
      "decoder_belief_target_nll": 3.344778299331665,
      "decoder_belief_top1_on_support": 0.40625,
      "mean_decoder_belief_support_size": 5.515625,
      "decoder_query_target_mass": 0.6198982000350952,
      "decoder_query_target_nll": 1.9123270511627197,
      "decoder_query_top1_on_support": 0.69921875,
      "mean_decoder_query_support_size": 4.734375
    },
    {
      "step": 500,
      "loss": 2.7321181297302246,
      "grad_norm": 0.534044623374939,
      "query_loss": 0.0,
      "belief_loss": 2.7321181297302246,
      "decoder_query_loss": 0.0,
      "query_target_mass": 0.4301447868347168,
      "query_target_nll": 2.4160640239715576,
      "query_top1_on_support": 0.4296875,
      "mean_query_support_size": 4.73828125,
      "decoder_belief_target_mass": 0.3766371011734009,
      "decoder_belief_target_nll": 3.0666048526763916,
      "decoder_belief_top1_on_support": 0.50390625,
      "mean_decoder_belief_support_size": 5.671875,
      "decoder_query_target_mass": 0.657516598701477,
      "decoder_query_target_nll": 1.8291822671890259,
      "decoder_query_top1_on_support": 0.72265625,
      "mean_decoder_query_support_size": 4.73828125
    },
    {
      "step": 600,
      "loss": 2.727497100830078,
      "grad_norm": 0.504955530166626,
      "query_loss": 0.0,
      "belief_loss": 2.727497100830078,
      "decoder_query_loss": 0.0,
      "query_target_mass": 0.4408799409866333,
      "query_target_nll": 2.425121307373047,
      "query_top1_on_support": 0.4609375,
      "mean_query_support_size": 4.85546875,
      "decoder_belief_target_mass": 0.4230186343193054,
      "decoder_belief_target_nll": 2.936278820037842,
      "decoder_belief_top1_on_support": 0.54296875,
      "mean_decoder_belief_support_size": 6.046875,
      "decoder_query_target_mass": 0.7125891447067261,
      "decoder_query_target_nll": 1.7136824131011963,
      "decoder_query_top1_on_support": 0.8125,
      "mean_decoder_query_support_size": 4.85546875
    },
    {
      "step": 700,
      "loss": 2.602102756500244,
      "grad_norm": 0.5746042728424072,
      "query_loss": 0.0,
      "belief_loss": 2.602102756500244,
      "decoder_query_loss": 0.0,
      "query_target_mass": 0.46020859479904175,
      "query_target_nll": 2.4151999950408936,
      "query_top1_on_support": 0.453125,
      "mean_query_support_size": 5.0625,
      "decoder_belief_target_mass": 0.47161656618118286,
      "decoder_belief_target_nll": 2.807579517364502,
      "decoder_belief_top1_on_support": 0.5859375,
      "mean_decoder_belief_support_size": 5.8671875,
      "decoder_query_target_mass": 0.7296038866043091,
      "decoder_query_target_nll": 1.7385830879211426,
      "decoder_query_top1_on_support": 0.8125,
      "mean_decoder_query_support_size": 5.0625
    },
    {
      "step": 800,
      "loss": 2.4212207794189453,
      "grad_norm": 0.5411489605903625,
      "query_loss": 0.0,
      "belief_loss": 2.4212207794189453,
      "decoder_query_loss": 0.0,
      "query_target_mass": 0.4645044505596161,
      "query_target_nll": 2.4186959266662598,
      "query_top1_on_support": 0.46875,
      "mean_query_support_size": 5.1171875,
      "decoder_belief_target_mass": 0.5350708365440369,
      "decoder_belief_target_nll": 2.6018762588500977,
      "decoder_belief_top1_on_support": 0.67578125,
      "mean_decoder_belief_support_size": 5.98046875,
      "decoder_query_target_mass": 0.76194828748703,
      "decoder_query_target_nll": 1.6672639846801758,
      "decoder_query_top1_on_support": 0.83203125,
      "mean_decoder_query_support_size": 5.1171875
    }
  ],
  "probe_train": [
    {
      "step": 1,
      "loss": 4.813169479370117,
      "grad_norm": 0.3901177644729614,
      "probe_belief_target_mass": 0.06059134379029274,
      "probe_belief_target_nll": 4.813169479370117,
      "probe_belief_top1_on_support": 0.0502232164144516,
      "mean_belief_support_size": 7.3058037757873535
    },
    {
      "step": 100,
      "loss": 2.7114036083221436,
      "grad_norm": 0.4647204577922821,
      "probe_belief_target_mass": 0.5862030386924744,
      "probe_belief_target_nll": 2.7114036083221436,
      "probe_belief_top1_on_support": 0.7159598469734192,
      "mean_belief_support_size": 7.245535850524902
    },
    {
      "step": 200,
      "loss": 2.506173849105835,
      "grad_norm": 0.46660247445106506,
      "probe_belief_target_mass": 0.6776570081710815,
      "probe_belief_target_nll": 2.506173849105835,
      "probe_belief_top1_on_support": 0.7767857313156128,
      "mean_belief_support_size": 7.470424652099609
    },
    {
      "step": 300,
      "loss": 2.3816916942596436,
      "grad_norm": 0.4514665901660919,
      "probe_belief_target_mass": 0.7249149680137634,
      "probe_belief_target_nll": 2.3816916942596436,
      "probe_belief_top1_on_support": 0.813058078289032,
      "mean_belief_support_size": 7.313058376312256
    },
    {
      "step": 400,
      "loss": 2.3798418045043945,
      "grad_norm": 0.500474750995636,
      "probe_belief_target_mass": 0.7229106426239014,
      "probe_belief_target_nll": 2.3798418045043945,
      "probe_belief_top1_on_support": 0.8074777126312256,
      "mean_belief_support_size": 7.136160850524902
    }
  ],
  "eval": [
    {
      "step": 800,
      "rows": [
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 3,
          "query_type": "A",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.4724474847316742,
          "query_target_nll": 2.4081382751464844,
          "query_top1_on_support": 0.46484375,
          "mean_query_support_size": 5.18359375,
          "probe_belief_target_mass": 0.06165938451886177,
          "probe_belief_target_nll": 20.526649475097656,
          "probe_belief_top1_on_support": 0.05078125,
          "mean_belief_support_size": 5.84375,
          "decoder_belief_target_mass": 0.06192343682050705,
          "decoder_belief_target_nll": 17.777679443359375,
          "decoder_belief_top1_on_support": 0.046875,
          "mean_decoder_belief_support_size": 5.84375,
          "decoder_query_target_mass": 0.4712299406528473,
          "decoder_query_target_nll": 2.410451889038086,
          "decoder_query_top1_on_support": 0.44921875,
          "mean_decoder_query_support_size": 5.18359375
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 3,
          "query_type": "A",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.4722716808319092,
          "query_target_nll": 2.404696464538574,
          "query_top1_on_support": 0.49609375,
          "mean_query_support_size": 5.18359375,
          "probe_belief_target_mass": 0.1212184727191925,
          "probe_belief_target_nll": 15.806905746459961,
          "probe_belief_top1_on_support": 0.1015625,
          "mean_belief_support_size": 5.84375,
          "decoder_belief_target_mass": 0.12106430530548096,
          "decoder_belief_target_nll": 11.892531394958496,
          "decoder_belief_top1_on_support": 0.12109375,
          "mean_decoder_belief_support_size": 5.84375,
          "decoder_query_target_mass": 0.5586518049240112,
          "decoder_query_target_nll": 3.7695326805114746,
          "decoder_query_top1_on_support": 0.5625,
          "mean_decoder_query_support_size": 5.18359375
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 3,
          "query_type": "A",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.47315165400505066,
          "query_target_nll": 2.4001095294952393,
          "query_top1_on_support": 0.50390625,
          "mean_query_support_size": 5.18359375,
          "probe_belief_target_mass": 0.23057694733142853,
          "probe_belief_target_nll": 9.011104583740234,
          "probe_belief_top1_on_support": 0.28125,
          "mean_belief_support_size": 5.84375,
          "decoder_belief_target_mass": 0.22554384171962738,
          "decoder_belief_target_nll": 7.2977399826049805,
          "decoder_belief_top1_on_support": 0.26171875,
          "mean_decoder_belief_support_size": 5.84375,
          "decoder_query_target_mass": 0.6807413101196289,
          "decoder_query_target_nll": 3.087006092071533,
          "decoder_query_top1_on_support": 0.6953125,
          "mean_decoder_query_support_size": 5.18359375
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 3,
          "query_type": "A",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.47115907073020935,
          "query_target_nll": 2.410308837890625,
          "query_top1_on_support": 0.46875,
          "mean_query_support_size": 5.18359375,
          "probe_belief_target_mass": 0.5273072719573975,
          "probe_belief_target_nll": 2.527426242828369,
          "probe_belief_top1_on_support": 0.65625,
          "mean_belief_support_size": 5.84375,
          "decoder_belief_target_mass": 0.4884257912635803,
          "decoder_belief_target_nll": 2.629747152328491,
          "decoder_belief_top1_on_support": 0.66015625,
          "mean_decoder_belief_support_size": 5.84375,
          "decoder_query_target_mass": 0.8538691997528076,
          "decoder_query_target_nll": 1.5624598264694214,
          "decoder_query_top1_on_support": 0.93359375,
          "mean_decoder_query_support_size": 5.18359375
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 3,
          "query_type": "A",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.47115907073020935,
          "query_target_nll": 2.410308837890625,
          "query_top1_on_support": 0.46875,
          "mean_query_support_size": 5.18359375,
          "probe_belief_target_mass": 0.5273072719573975,
          "probe_belief_target_nll": 2.527426242828369,
          "probe_belief_top1_on_support": 0.65625,
          "mean_belief_support_size": 5.84375,
          "decoder_belief_target_mass": 0.4884257912635803,
          "decoder_belief_target_nll": 2.629747152328491,
          "decoder_belief_top1_on_support": 0.66015625,
          "mean_decoder_belief_support_size": 5.84375,
          "decoder_query_target_mass": 0.8538691997528076,
          "decoder_query_target_nll": 1.5624598264694214,
          "decoder_query_top1_on_support": 0.93359375,
          "mean_decoder_query_support_size": 5.18359375
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 3,
          "query_type": "A",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.47115907073020935,
          "query_target_nll": 2.410308837890625,
          "query_top1_on_support": 0.46875,
          "mean_query_support_size": 5.18359375,
          "probe_belief_target_mass": 0.5273072719573975,
          "probe_belief_target_nll": 2.527426242828369,
          "probe_belief_top1_on_support": 0.65625,
          "mean_belief_support_size": 5.84375,
          "decoder_belief_target_mass": 0.4884257912635803,
          "decoder_belief_target_nll": 2.629747152328491,
          "decoder_belief_top1_on_support": 0.66015625,
          "mean_decoder_belief_support_size": 5.84375,
          "decoder_query_target_mass": 0.8538691997528076,
          "decoder_query_target_nll": 1.5624598264694214,
          "decoder_query_top1_on_support": 0.93359375,
          "mean_decoder_query_support_size": 5.18359375
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 3,
          "query_type": "A",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.47115907073020935,
          "query_target_nll": 2.410308837890625,
          "query_top1_on_support": 0.46875,
          "mean_query_support_size": 5.18359375,
          "probe_belief_target_mass": 0.5273072719573975,
          "probe_belief_target_nll": 2.527426242828369,
          "probe_belief_top1_on_support": 0.65625,
          "mean_belief_support_size": 5.84375,
          "decoder_belief_target_mass": 0.4884257912635803,
          "decoder_belief_target_nll": 2.629747152328491,
          "decoder_belief_top1_on_support": 0.66015625,
          "mean_decoder_belief_support_size": 5.84375,
          "decoder_query_target_mass": 0.8538691997528076,
          "decoder_query_target_nll": 1.5624598264694214,
          "decoder_query_top1_on_support": 0.93359375,
          "mean_decoder_query_support_size": 5.18359375
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 3,
          "query_type": "B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.4675944447517395,
          "query_target_nll": 2.4239730834960938,
          "query_top1_on_support": 0.484375,
          "mean_query_support_size": 5.15234375,
          "probe_belief_target_mass": 0.05307570844888687,
          "probe_belief_target_nll": 21.33190155029297,
          "probe_belief_top1_on_support": 0.03515625,
          "mean_belief_support_size": 5.77734375,
          "decoder_belief_target_mass": 0.05358303710818291,
          "decoder_belief_target_nll": 17.87875747680664,
          "decoder_belief_top1_on_support": 0.0625,
          "mean_decoder_belief_support_size": 5.77734375,
          "decoder_query_target_mass": 0.46897751092910767,
          "decoder_query_target_nll": 2.4010908603668213,
          "decoder_query_top1_on_support": 0.48046875,
          "mean_decoder_query_support_size": 5.15234375
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 3,
          "query_type": "B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.46796751022338867,
          "query_target_nll": 2.41627836227417,
          "query_top1_on_support": 0.46484375,
          "mean_query_support_size": 5.15234375,
          "probe_belief_target_mass": 0.13605894148349762,
          "probe_belief_target_nll": 14.90018081665039,
          "probe_belief_top1_on_support": 0.171875,
          "mean_belief_support_size": 5.77734375,
          "decoder_belief_target_mass": 0.13233178853988647,
          "decoder_belief_target_nll": 11.716575622558594,
          "decoder_belief_top1_on_support": 0.140625,
          "mean_decoder_belief_support_size": 5.77734375,
          "decoder_query_target_mass": 0.5399776697158813,
          "decoder_query_target_nll": 3.811605215072632,
          "decoder_query_top1_on_support": 0.5546875,
          "mean_decoder_query_support_size": 5.15234375
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 3,
          "query_type": "B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.466520220041275,
          "query_target_nll": 2.4203433990478516,
          "query_top1_on_support": 0.48046875,
          "mean_query_support_size": 5.15234375,
          "probe_belief_target_mass": 0.2425018846988678,
          "probe_belief_target_nll": 8.961723327636719,
          "probe_belief_top1_on_support": 0.265625,
          "mean_belief_support_size": 5.77734375,
          "decoder_belief_target_mass": 0.23340226709842682,
          "decoder_belief_target_nll": 7.321136951446533,
          "decoder_belief_top1_on_support": 0.2578125,
          "mean_decoder_belief_support_size": 5.77734375,
          "decoder_query_target_mass": 0.6714152693748474,
          "decoder_query_target_nll": 3.2281429767608643,
          "decoder_query_top1_on_support": 0.6953125,
          "mean_decoder_query_support_size": 5.15234375
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 3,
          "query_type": "B",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.4671507775783539,
          "query_target_nll": 2.418943405151367,
          "query_top1_on_support": 0.47265625,
          "mean_query_support_size": 5.15234375,
          "probe_belief_target_mass": 0.5340644121170044,
          "probe_belief_target_nll": 2.4800710678100586,
          "probe_belief_top1_on_support": 0.703125,
          "mean_belief_support_size": 5.77734375,
          "decoder_belief_target_mass": 0.49311432242393494,
          "decoder_belief_target_nll": 2.5936031341552734,
          "decoder_belief_top1_on_support": 0.66796875,
          "mean_decoder_belief_support_size": 5.77734375,
          "decoder_query_target_mass": 0.8429068922996521,
          "decoder_query_target_nll": 1.566602110862732,
          "decoder_query_top1_on_support": 0.88671875,
          "mean_decoder_query_support_size": 5.15234375
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 3,
          "query_type": "B",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.4671507775783539,
          "query_target_nll": 2.418943405151367,
          "query_top1_on_support": 0.47265625,
          "mean_query_support_size": 5.15234375,
          "probe_belief_target_mass": 0.5340644121170044,
          "probe_belief_target_nll": 2.4800710678100586,
          "probe_belief_top1_on_support": 0.703125,
          "mean_belief_support_size": 5.77734375,
          "decoder_belief_target_mass": 0.49311432242393494,
          "decoder_belief_target_nll": 2.5936031341552734,
          "decoder_belief_top1_on_support": 0.66796875,
          "mean_decoder_belief_support_size": 5.77734375,
          "decoder_query_target_mass": 0.8429068922996521,
          "decoder_query_target_nll": 1.566602110862732,
          "decoder_query_top1_on_support": 0.88671875,
          "mean_decoder_query_support_size": 5.15234375
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 3,
          "query_type": "B",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.4671507775783539,
          "query_target_nll": 2.418943405151367,
          "query_top1_on_support": 0.47265625,
          "mean_query_support_size": 5.15234375,
          "probe_belief_target_mass": 0.5340644121170044,
          "probe_belief_target_nll": 2.4800710678100586,
          "probe_belief_top1_on_support": 0.703125,
          "mean_belief_support_size": 5.77734375,
          "decoder_belief_target_mass": 0.49311432242393494,
          "decoder_belief_target_nll": 2.5936031341552734,
          "decoder_belief_top1_on_support": 0.66796875,
          "mean_decoder_belief_support_size": 5.77734375,
          "decoder_query_target_mass": 0.8429068922996521,
          "decoder_query_target_nll": 1.566602110862732,
          "decoder_query_top1_on_support": 0.88671875,
          "mean_decoder_query_support_size": 5.15234375
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 3,
          "query_type": "B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.4671507775783539,
          "query_target_nll": 2.418943405151367,
          "query_top1_on_support": 0.47265625,
          "mean_query_support_size": 5.15234375,
          "probe_belief_target_mass": 0.5340644121170044,
          "probe_belief_target_nll": 2.4800710678100586,
          "probe_belief_top1_on_support": 0.703125,
          "mean_belief_support_size": 5.77734375,
          "decoder_belief_target_mass": 0.49311432242393494,
          "decoder_belief_target_nll": 2.5936031341552734,
          "decoder_belief_top1_on_support": 0.66796875,
          "mean_decoder_belief_support_size": 5.77734375,
          "decoder_query_target_mass": 0.8429068922996521,
          "decoder_query_target_nll": 1.566602110862732,
          "decoder_query_top1_on_support": 0.88671875,
          "mean_decoder_query_support_size": 5.15234375
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 3,
          "query_type": "A_PLUS_B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.4882711172103882,
          "query_target_nll": 2.415499210357666,
          "query_top1_on_support": 0.47265625,
          "mean_query_support_size": 5.3671875,
          "probe_belief_target_mass": 0.06361061334609985,
          "probe_belief_target_nll": 20.348066329956055,
          "probe_belief_top1_on_support": 0.0546875,
          "mean_belief_support_size": 5.484375,
          "decoder_belief_target_mass": 0.0659191682934761,
          "decoder_belief_target_nll": 17.22665786743164,
          "decoder_belief_top1_on_support": 0.0703125,
          "mean_decoder_belief_support_size": 5.484375,
          "decoder_query_target_mass": 0.48912355303764343,
          "decoder_query_target_nll": 2.416881561279297,
          "decoder_query_top1_on_support": 0.4921875,
          "mean_decoder_query_support_size": 5.3671875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 3,
          "query_type": "A_PLUS_B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.48894649744033813,
          "query_target_nll": 2.4137043952941895,
          "query_top1_on_support": 0.5078125,
          "mean_query_support_size": 5.3671875,
          "probe_belief_target_mass": 0.15090887248516083,
          "probe_belief_target_nll": 16.143169403076172,
          "probe_belief_top1_on_support": 0.14453125,
          "mean_belief_support_size": 5.484375,
          "decoder_belief_target_mass": 0.1507212221622467,
          "decoder_belief_target_nll": 12.017534255981445,
          "decoder_belief_top1_on_support": 0.13671875,
          "mean_decoder_belief_support_size": 5.484375,
          "decoder_query_target_mass": 0.4908345639705658,
          "decoder_query_target_nll": 3.85103178024292,
          "decoder_query_top1_on_support": 0.4921875,
          "mean_decoder_query_support_size": 5.3671875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 3,
          "query_type": "A_PLUS_B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.4902377128601074,
          "query_target_nll": 2.4059019088745117,
          "query_top1_on_support": 0.51171875,
          "mean_query_support_size": 5.3671875,
          "probe_belief_target_mass": 0.2237325757741928,
          "probe_belief_target_nll": 10.160980224609375,
          "probe_belief_top1_on_support": 0.23046875,
          "mean_belief_support_size": 5.484375,
          "decoder_belief_target_mass": 0.2166166752576828,
          "decoder_belief_target_nll": 7.8598432540893555,
          "decoder_belief_top1_on_support": 0.23828125,
          "mean_decoder_belief_support_size": 5.484375,
          "decoder_query_target_mass": 0.5369327068328857,
          "decoder_query_target_nll": 3.566835880279541,
          "decoder_query_top1_on_support": 0.56640625,
          "mean_decoder_query_support_size": 5.3671875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 3,
          "query_type": "A_PLUS_B",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.48888179659843445,
          "query_target_nll": 2.413473606109619,
          "query_top1_on_support": 0.5,
          "mean_query_support_size": 5.3671875,
          "probe_belief_target_mass": 0.5419090986251831,
          "probe_belief_target_nll": 2.4002978801727295,
          "probe_belief_top1_on_support": 0.71484375,
          "mean_belief_support_size": 5.484375,
          "decoder_belief_target_mass": 0.495158314704895,
          "decoder_belief_target_nll": 2.503220319747925,
          "decoder_belief_top1_on_support": 0.67578125,
          "mean_decoder_belief_support_size": 5.484375,
          "decoder_query_target_mass": 0.7121630907058716,
          "decoder_query_target_nll": 1.9072167873382568,
          "decoder_query_top1_on_support": 0.80859375,
          "mean_decoder_query_support_size": 5.3671875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 3,
          "query_type": "A_PLUS_B",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.48888179659843445,
          "query_target_nll": 2.413473606109619,
          "query_top1_on_support": 0.5,
          "mean_query_support_size": 5.3671875,
          "probe_belief_target_mass": 0.5419090986251831,
          "probe_belief_target_nll": 2.4002978801727295,
          "probe_belief_top1_on_support": 0.71484375,
          "mean_belief_support_size": 5.484375,
          "decoder_belief_target_mass": 0.495158314704895,
          "decoder_belief_target_nll": 2.503220319747925,
          "decoder_belief_top1_on_support": 0.67578125,
          "mean_decoder_belief_support_size": 5.484375,
          "decoder_query_target_mass": 0.7121630907058716,
          "decoder_query_target_nll": 1.9072167873382568,
          "decoder_query_top1_on_support": 0.80859375,
          "mean_decoder_query_support_size": 5.3671875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 3,
          "query_type": "A_PLUS_B",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.48888179659843445,
          "query_target_nll": 2.413473606109619,
          "query_top1_on_support": 0.5,
          "mean_query_support_size": 5.3671875,
          "probe_belief_target_mass": 0.5419090986251831,
          "probe_belief_target_nll": 2.4002978801727295,
          "probe_belief_top1_on_support": 0.71484375,
          "mean_belief_support_size": 5.484375,
          "decoder_belief_target_mass": 0.495158314704895,
          "decoder_belief_target_nll": 2.503220319747925,
          "decoder_belief_top1_on_support": 0.67578125,
          "mean_decoder_belief_support_size": 5.484375,
          "decoder_query_target_mass": 0.7121630907058716,
          "decoder_query_target_nll": 1.9072167873382568,
          "decoder_query_top1_on_support": 0.80859375,
          "mean_decoder_query_support_size": 5.3671875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 3,
          "query_type": "A_PLUS_B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.48888179659843445,
          "query_target_nll": 2.413473606109619,
          "query_top1_on_support": 0.5,
          "mean_query_support_size": 5.3671875,
          "probe_belief_target_mass": 0.5419090986251831,
          "probe_belief_target_nll": 2.4002978801727295,
          "probe_belief_top1_on_support": 0.71484375,
          "mean_belief_support_size": 5.484375,
          "decoder_belief_target_mass": 0.495158314704895,
          "decoder_belief_target_nll": 2.503220319747925,
          "decoder_belief_top1_on_support": 0.67578125,
          "mean_decoder_belief_support_size": 5.484375,
          "decoder_query_target_mass": 0.7121630907058716,
          "decoder_query_target_nll": 1.9072167873382568,
          "decoder_query_top1_on_support": 0.80859375,
          "mean_decoder_query_support_size": 5.3671875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 3,
          "query_type": "A_MINUS_B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.4238370656967163,
          "query_target_nll": 2.422173023223877,
          "query_top1_on_support": 0.40234375,
          "mean_query_support_size": 4.671875,
          "probe_belief_target_mass": 0.05753325670957565,
          "probe_belief_target_nll": 20.184864044189453,
          "probe_belief_top1_on_support": 0.0546875,
          "mean_belief_support_size": 5.6015625,
          "decoder_belief_target_mass": 0.05780297517776489,
          "decoder_belief_target_nll": 17.33003807067871,
          "decoder_belief_top1_on_support": 0.0390625,
          "mean_decoder_belief_support_size": 5.6015625,
          "decoder_query_target_mass": 0.4711865186691284,
          "decoder_query_target_nll": 9.25128173828125,
          "decoder_query_top1_on_support": 0.47265625,
          "mean_decoder_query_support_size": 4.671875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 3,
          "query_type": "A_MINUS_B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.4244525730609894,
          "query_target_nll": 2.415550947189331,
          "query_top1_on_support": 0.44140625,
          "mean_query_support_size": 4.671875,
          "probe_belief_target_mass": 0.15689273178577423,
          "probe_belief_target_nll": 14.931629180908203,
          "probe_belief_top1_on_support": 0.1484375,
          "mean_belief_support_size": 5.6015625,
          "decoder_belief_target_mass": 0.15901203453540802,
          "decoder_belief_target_nll": 11.361149787902832,
          "decoder_belief_top1_on_support": 0.171875,
          "mean_decoder_belief_support_size": 5.6015625,
          "decoder_query_target_mass": 0.5216417908668518,
          "decoder_query_target_nll": 5.044273853302002,
          "decoder_query_top1_on_support": 0.52734375,
          "mean_decoder_query_support_size": 4.671875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 3,
          "query_type": "A_MINUS_B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.4238811731338501,
          "query_target_nll": 2.4174931049346924,
          "query_top1_on_support": 0.4375,
          "mean_query_support_size": 4.671875,
          "probe_belief_target_mass": 0.2649506628513336,
          "probe_belief_target_nll": 8.894973754882812,
          "probe_belief_top1_on_support": 0.2890625,
          "mean_belief_support_size": 5.6015625,
          "decoder_belief_target_mass": 0.26069605350494385,
          "decoder_belief_target_nll": 6.629952430725098,
          "decoder_belief_top1_on_support": 0.29296875,
          "mean_decoder_belief_support_size": 5.6015625,
          "decoder_query_target_mass": 0.5539066791534424,
          "decoder_query_target_nll": 3.348081588745117,
          "decoder_query_top1_on_support": 0.5703125,
          "mean_decoder_query_support_size": 4.671875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 3,
          "query_type": "A_MINUS_B",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.4248504340648651,
          "query_target_nll": 2.4096314907073975,
          "query_top1_on_support": 0.44921875,
          "mean_query_support_size": 4.671875,
          "probe_belief_target_mass": 0.5420727729797363,
          "probe_belief_target_nll": 2.4268851280212402,
          "probe_belief_top1_on_support": 0.69140625,
          "mean_belief_support_size": 5.6015625,
          "decoder_belief_target_mass": 0.49906858801841736,
          "decoder_belief_target_nll": 2.5301876068115234,
          "decoder_belief_top1_on_support": 0.72265625,
          "mean_decoder_belief_support_size": 5.6015625,
          "decoder_query_target_mass": 0.7436374425888062,
          "decoder_query_target_nll": 1.5115315914154053,
          "decoder_query_top1_on_support": 0.90234375,
          "mean_decoder_query_support_size": 4.671875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 3,
          "query_type": "A_MINUS_B",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.4248504340648651,
          "query_target_nll": 2.4096314907073975,
          "query_top1_on_support": 0.44921875,
          "mean_query_support_size": 4.671875,
          "probe_belief_target_mass": 0.5420727729797363,
          "probe_belief_target_nll": 2.4268851280212402,
          "probe_belief_top1_on_support": 0.69140625,
          "mean_belief_support_size": 5.6015625,
          "decoder_belief_target_mass": 0.49906858801841736,
          "decoder_belief_target_nll": 2.5301876068115234,
          "decoder_belief_top1_on_support": 0.72265625,
          "mean_decoder_belief_support_size": 5.6015625,
          "decoder_query_target_mass": 0.7436374425888062,
          "decoder_query_target_nll": 1.5115315914154053,
          "decoder_query_top1_on_support": 0.90234375,
          "mean_decoder_query_support_size": 4.671875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 3,
          "query_type": "A_MINUS_B",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.4248504340648651,
          "query_target_nll": 2.4096314907073975,
          "query_top1_on_support": 0.44921875,
          "mean_query_support_size": 4.671875,
          "probe_belief_target_mass": 0.5420727729797363,
          "probe_belief_target_nll": 2.4268851280212402,
          "probe_belief_top1_on_support": 0.69140625,
          "mean_belief_support_size": 5.6015625,
          "decoder_belief_target_mass": 0.49906858801841736,
          "decoder_belief_target_nll": 2.5301876068115234,
          "decoder_belief_top1_on_support": 0.72265625,
          "mean_decoder_belief_support_size": 5.6015625,
          "decoder_query_target_mass": 0.7436374425888062,
          "decoder_query_target_nll": 1.5115315914154053,
          "decoder_query_top1_on_support": 0.90234375,
          "mean_decoder_query_support_size": 4.671875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 3,
          "query_type": "A_MINUS_B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.4248504340648651,
          "query_target_nll": 2.4096314907073975,
          "query_top1_on_support": 0.44921875,
          "mean_query_support_size": 4.671875,
          "probe_belief_target_mass": 0.5420727729797363,
          "probe_belief_target_nll": 2.4268851280212402,
          "probe_belief_top1_on_support": 0.69140625,
          "mean_belief_support_size": 5.6015625,
          "decoder_belief_target_mass": 0.49906858801841736,
          "decoder_belief_target_nll": 2.5301876068115234,
          "decoder_belief_top1_on_support": 0.72265625,
          "mean_decoder_belief_support_size": 5.6015625,
          "decoder_query_target_mass": 0.7436374425888062,
          "decoder_query_target_nll": 1.5115315914154053,
          "decoder_query_top1_on_support": 0.90234375,
          "mean_decoder_query_support_size": 4.671875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 6,
          "query_type": "A",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.2527785003185272,
          "query_target_nll": 2.4310302734375,
          "query_top1_on_support": 0.26171875,
          "mean_query_support_size": 2.78125,
          "probe_belief_target_mass": 0.029864490032196045,
          "probe_belief_target_nll": 20.87744140625,
          "probe_belief_top1_on_support": 0.03515625,
          "mean_belief_support_size": 3.45703125,
          "decoder_belief_target_mass": 0.030595283955335617,
          "decoder_belief_target_nll": 17.954526901245117,
          "decoder_belief_top1_on_support": 0.03125,
          "mean_decoder_belief_support_size": 3.45703125,
          "decoder_query_target_mass": 0.25339582562446594,
          "decoder_query_target_nll": 2.399840831756592,
          "decoder_query_top1_on_support": 0.26953125,
          "mean_decoder_query_support_size": 2.78125
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 6,
          "query_type": "A",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.2501792907714844,
          "query_target_nll": 2.437351703643799,
          "query_top1_on_support": 0.24609375,
          "mean_query_support_size": 2.78125,
          "probe_belief_target_mass": 0.07261766493320465,
          "probe_belief_target_nll": 16.628080368041992,
          "probe_belief_top1_on_support": 0.08203125,
          "mean_belief_support_size": 3.45703125,
          "decoder_belief_target_mass": 0.07108329236507416,
          "decoder_belief_target_nll": 13.611248016357422,
          "decoder_belief_top1_on_support": 0.0703125,
          "mean_decoder_belief_support_size": 3.45703125,
          "decoder_query_target_mass": 0.2848970293998718,
          "decoder_query_target_nll": 4.745427131652832,
          "decoder_query_top1_on_support": 0.2890625,
          "mean_decoder_query_support_size": 2.78125
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 6,
          "query_type": "A",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.2521877586841583,
          "query_target_nll": 2.4193620681762695,
          "query_top1_on_support": 0.2421875,
          "mean_query_support_size": 2.78125,
          "probe_belief_target_mass": 0.07945874333381653,
          "probe_belief_target_nll": 13.323448181152344,
          "probe_belief_top1_on_support": 0.08984375,
          "mean_belief_support_size": 3.45703125,
          "decoder_belief_target_mass": 0.07840122282505035,
          "decoder_belief_target_nll": 10.774110794067383,
          "decoder_belief_top1_on_support": 0.0859375,
          "mean_decoder_belief_support_size": 3.45703125,
          "decoder_query_target_mass": 0.3417888283729553,
          "decoder_query_target_nll": 4.563794136047363,
          "decoder_query_top1_on_support": 0.359375,
          "mean_decoder_query_support_size": 2.78125
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 6,
          "query_type": "A",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.2522081732749939,
          "query_target_nll": 2.415283203125,
          "query_top1_on_support": 0.25390625,
          "mean_query_support_size": 2.78125,
          "probe_belief_target_mass": 0.08944875001907349,
          "probe_belief_target_nll": 11.369555473327637,
          "probe_belief_top1_on_support": 0.09765625,
          "mean_belief_support_size": 3.45703125,
          "decoder_belief_target_mass": 0.08697383105754852,
          "decoder_belief_target_nll": 9.36297607421875,
          "decoder_belief_top1_on_support": 0.09765625,
          "mean_decoder_belief_support_size": 3.45703125,
          "decoder_query_target_mass": 0.3815232813358307,
          "decoder_query_target_nll": 4.043517589569092,
          "decoder_query_top1_on_support": 0.4140625,
          "mean_decoder_query_support_size": 2.78125
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 6,
          "query_type": "A",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.25394439697265625,
          "query_target_nll": 2.4073872566223145,
          "query_top1_on_support": 0.25,
          "mean_query_support_size": 2.78125,
          "probe_belief_target_mass": 0.22399313747882843,
          "probe_belief_target_nll": 3.0703678131103516,
          "probe_belief_top1_on_support": 0.359375,
          "mean_belief_support_size": 3.45703125,
          "decoder_belief_target_mass": 0.19754676520824432,
          "decoder_belief_target_nll": 3.1409478187561035,
          "decoder_belief_top1_on_support": 0.28515625,
          "mean_decoder_belief_support_size": 3.45703125,
          "decoder_query_target_mass": 0.5186177492141724,
          "decoder_query_target_nll": 1.6696866750717163,
          "decoder_query_top1_on_support": 0.5859375,
          "mean_decoder_query_support_size": 2.78125
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 6,
          "query_type": "A",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.25394439697265625,
          "query_target_nll": 2.4073872566223145,
          "query_top1_on_support": 0.25,
          "mean_query_support_size": 2.78125,
          "probe_belief_target_mass": 0.22399313747882843,
          "probe_belief_target_nll": 3.0703678131103516,
          "probe_belief_top1_on_support": 0.359375,
          "mean_belief_support_size": 3.45703125,
          "decoder_belief_target_mass": 0.19754676520824432,
          "decoder_belief_target_nll": 3.1409478187561035,
          "decoder_belief_top1_on_support": 0.28515625,
          "mean_decoder_belief_support_size": 3.45703125,
          "decoder_query_target_mass": 0.5186177492141724,
          "decoder_query_target_nll": 1.6696865558624268,
          "decoder_query_top1_on_support": 0.5859375,
          "mean_decoder_query_support_size": 2.78125
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 6,
          "query_type": "A",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.25394439697265625,
          "query_target_nll": 2.4073872566223145,
          "query_top1_on_support": 0.25,
          "mean_query_support_size": 2.78125,
          "probe_belief_target_mass": 0.22399313747882843,
          "probe_belief_target_nll": 3.0703678131103516,
          "probe_belief_top1_on_support": 0.359375,
          "mean_belief_support_size": 3.45703125,
          "decoder_belief_target_mass": 0.19754676520824432,
          "decoder_belief_target_nll": 3.1409478187561035,
          "decoder_belief_top1_on_support": 0.28515625,
          "mean_decoder_belief_support_size": 3.45703125,
          "decoder_query_target_mass": 0.5186177492141724,
          "decoder_query_target_nll": 1.6696866750717163,
          "decoder_query_top1_on_support": 0.5859375,
          "mean_decoder_query_support_size": 2.78125
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 6,
          "query_type": "B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.31076592206954956,
          "query_target_nll": 2.4202215671539307,
          "query_top1_on_support": 0.32421875,
          "mean_query_support_size": 3.4140625,
          "probe_belief_target_mass": 0.0354711152613163,
          "probe_belief_target_nll": 21.617019653320312,
          "probe_belief_top1_on_support": 0.05078125,
          "mean_belief_support_size": 4.03125,
          "decoder_belief_target_mass": 0.03475470095872879,
          "decoder_belief_target_nll": 18.41317367553711,
          "decoder_belief_top1_on_support": 0.046875,
          "mean_decoder_belief_support_size": 4.03125,
          "decoder_query_target_mass": 0.31161123514175415,
          "decoder_query_target_nll": 2.399221658706665,
          "decoder_query_top1_on_support": 0.328125,
          "mean_decoder_query_support_size": 3.4140625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 6,
          "query_type": "B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.31249427795410156,
          "query_target_nll": 2.4078803062438965,
          "query_top1_on_support": 0.32421875,
          "mean_query_support_size": 3.4140625,
          "probe_belief_target_mass": 0.04426170885562897,
          "probe_belief_target_nll": 18.533336639404297,
          "probe_belief_top1_on_support": 0.046875,
          "mean_belief_support_size": 4.03125,
          "decoder_belief_target_mass": 0.04376210272312164,
          "decoder_belief_target_nll": 14.738940238952637,
          "decoder_belief_top1_on_support": 0.02734375,
          "mean_decoder_belief_support_size": 4.03125,
          "decoder_query_target_mass": 0.3391587436199188,
          "decoder_query_target_nll": 4.945141792297363,
          "decoder_query_top1_on_support": 0.35546875,
          "mean_decoder_query_support_size": 3.4140625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 6,
          "query_type": "B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.3123280107975006,
          "query_target_nll": 2.407947540283203,
          "query_top1_on_support": 0.328125,
          "mean_query_support_size": 3.4140625,
          "probe_belief_target_mass": 0.06703880429267883,
          "probe_belief_target_nll": 14.26873779296875,
          "probe_belief_top1_on_support": 0.07421875,
          "mean_belief_support_size": 4.03125,
          "decoder_belief_target_mass": 0.06860530376434326,
          "decoder_belief_target_nll": 11.4982328414917,
          "decoder_belief_top1_on_support": 0.078125,
          "mean_decoder_belief_support_size": 4.03125,
          "decoder_query_target_mass": 0.3763154149055481,
          "decoder_query_target_nll": 4.586565971374512,
          "decoder_query_top1_on_support": 0.37890625,
          "mean_decoder_query_support_size": 3.4140625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 6,
          "query_type": "B",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.3131853938102722,
          "query_target_nll": 2.40191388130188,
          "query_top1_on_support": 0.35546875,
          "mean_query_support_size": 3.4140625,
          "probe_belief_target_mass": 0.07775022089481354,
          "probe_belief_target_nll": 11.640929222106934,
          "probe_belief_top1_on_support": 0.0859375,
          "mean_belief_support_size": 4.03125,
          "decoder_belief_target_mass": 0.0747174471616745,
          "decoder_belief_target_nll": 9.92082405090332,
          "decoder_belief_top1_on_support": 0.09375,
          "mean_decoder_belief_support_size": 4.03125,
          "decoder_query_target_mass": 0.4091891646385193,
          "decoder_query_target_nll": 4.417701721191406,
          "decoder_query_top1_on_support": 0.4296875,
          "mean_decoder_query_support_size": 3.4140625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 6,
          "query_type": "B",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.31235018372535706,
          "query_target_nll": 2.4005610942840576,
          "query_top1_on_support": 0.34375,
          "mean_query_support_size": 3.4140625,
          "probe_belief_target_mass": 0.24165722727775574,
          "probe_belief_target_nll": 3.187602996826172,
          "probe_belief_top1_on_support": 0.3671875,
          "mean_belief_support_size": 4.03125,
          "decoder_belief_target_mass": 0.21859267354011536,
          "decoder_belief_target_nll": 3.2231884002685547,
          "decoder_belief_top1_on_support": 0.31640625,
          "mean_decoder_belief_support_size": 4.03125,
          "decoder_query_target_mass": 0.5825638175010681,
          "decoder_query_target_nll": 1.7054595947265625,
          "decoder_query_top1_on_support": 0.64453125,
          "mean_decoder_query_support_size": 3.4140625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 6,
          "query_type": "B",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.31235018372535706,
          "query_target_nll": 2.4005610942840576,
          "query_top1_on_support": 0.34375,
          "mean_query_support_size": 3.4140625,
          "probe_belief_target_mass": 0.24165722727775574,
          "probe_belief_target_nll": 3.187602996826172,
          "probe_belief_top1_on_support": 0.3671875,
          "mean_belief_support_size": 4.03125,
          "decoder_belief_target_mass": 0.21859267354011536,
          "decoder_belief_target_nll": 3.2231884002685547,
          "decoder_belief_top1_on_support": 0.31640625,
          "mean_decoder_belief_support_size": 4.03125,
          "decoder_query_target_mass": 0.5825638175010681,
          "decoder_query_target_nll": 1.7054595947265625,
          "decoder_query_top1_on_support": 0.64453125,
          "mean_decoder_query_support_size": 3.4140625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 6,
          "query_type": "B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.31235018372535706,
          "query_target_nll": 2.4005610942840576,
          "query_top1_on_support": 0.34375,
          "mean_query_support_size": 3.4140625,
          "probe_belief_target_mass": 0.24165722727775574,
          "probe_belief_target_nll": 3.187602996826172,
          "probe_belief_top1_on_support": 0.3671875,
          "mean_belief_support_size": 4.03125,
          "decoder_belief_target_mass": 0.21859267354011536,
          "decoder_belief_target_nll": 3.2231884002685547,
          "decoder_belief_top1_on_support": 0.31640625,
          "mean_decoder_belief_support_size": 4.03125,
          "decoder_query_target_mass": 0.5825638175010681,
          "decoder_query_target_nll": 1.7054595947265625,
          "decoder_query_top1_on_support": 0.64453125,
          "mean_decoder_query_support_size": 3.4140625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 6,
          "query_type": "A_PLUS_B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.3297267556190491,
          "query_target_nll": 2.4129292964935303,
          "query_top1_on_support": 0.3125,
          "mean_query_support_size": 3.6171875,
          "probe_belief_target_mass": 0.0332917645573616,
          "probe_belief_target_nll": 21.338430404663086,
          "probe_belief_top1_on_support": 0.03515625,
          "mean_belief_support_size": 3.97265625,
          "decoder_belief_target_mass": 0.033563874661922455,
          "decoder_belief_target_nll": 18.113880157470703,
          "decoder_belief_top1_on_support": 0.0234375,
          "mean_decoder_belief_support_size": 3.97265625,
          "decoder_query_target_mass": 0.3303530812263489,
          "decoder_query_target_nll": 2.418466567993164,
          "decoder_query_top1_on_support": 0.3203125,
          "mean_decoder_query_support_size": 3.6171875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 6,
          "query_type": "A_PLUS_B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.32968491315841675,
          "query_target_nll": 2.4147982597351074,
          "query_top1_on_support": 0.32421875,
          "mean_query_support_size": 3.6171875,
          "probe_belief_target_mass": 0.05039028078317642,
          "probe_belief_target_nll": 17.547555923461914,
          "probe_belief_top1_on_support": 0.05078125,
          "mean_belief_support_size": 3.97265625,
          "decoder_belief_target_mass": 0.049543410539627075,
          "decoder_belief_target_nll": 13.470821380615234,
          "decoder_belief_top1_on_support": 0.04296875,
          "mean_decoder_belief_support_size": 3.97265625,
          "decoder_query_target_mass": 0.3392288386821747,
          "decoder_query_target_nll": 3.377131223678589,
          "decoder_query_top1_on_support": 0.3515625,
          "mean_decoder_query_support_size": 3.6171875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 6,
          "query_type": "A_PLUS_B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.32910996675491333,
          "query_target_nll": 2.4137871265411377,
          "query_top1_on_support": 0.35546875,
          "mean_query_support_size": 3.6171875,
          "probe_belief_target_mass": 0.06798690557479858,
          "probe_belief_target_nll": 13.230849266052246,
          "probe_belief_top1_on_support": 0.0625,
          "mean_belief_support_size": 3.97265625,
          "decoder_belief_target_mass": 0.06733609735965729,
          "decoder_belief_target_nll": 10.876144409179688,
          "decoder_belief_top1_on_support": 0.078125,
          "mean_decoder_belief_support_size": 3.97265625,
          "decoder_query_target_mass": 0.3308259844779968,
          "decoder_query_target_nll": 3.7987756729125977,
          "decoder_query_top1_on_support": 0.3203125,
          "mean_decoder_query_support_size": 3.6171875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 6,
          "query_type": "A_PLUS_B",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.3287602663040161,
          "query_target_nll": 2.4147071838378906,
          "query_top1_on_support": 0.3203125,
          "mean_query_support_size": 3.6171875,
          "probe_belief_target_mass": 0.09349982440471649,
          "probe_belief_target_nll": 10.491569519042969,
          "probe_belief_top1_on_support": 0.11328125,
          "mean_belief_support_size": 3.97265625,
          "decoder_belief_target_mass": 0.08170415461063385,
          "decoder_belief_target_nll": 9.262191772460938,
          "decoder_belief_top1_on_support": 0.09375,
          "mean_decoder_belief_support_size": 3.97265625,
          "decoder_query_target_mass": 0.31904494762420654,
          "decoder_query_target_nll": 3.521233081817627,
          "decoder_query_top1_on_support": 0.3125,
          "mean_decoder_query_support_size": 3.6171875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 6,
          "query_type": "A_PLUS_B",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.3298276960849762,
          "query_target_nll": 2.4162020683288574,
          "query_top1_on_support": 0.328125,
          "mean_query_support_size": 3.6171875,
          "probe_belief_target_mass": 0.22745844721794128,
          "probe_belief_target_nll": 3.1122798919677734,
          "probe_belief_top1_on_support": 0.33984375,
          "mean_belief_support_size": 3.97265625,
          "decoder_belief_target_mass": 0.20910391211509705,
          "decoder_belief_target_nll": 3.183773994445801,
          "decoder_belief_top1_on_support": 0.2890625,
          "mean_decoder_belief_support_size": 3.97265625,
          "decoder_query_target_mass": 0.4453887343406677,
          "decoder_query_target_nll": 2.0519604682922363,
          "decoder_query_top1_on_support": 0.515625,
          "mean_decoder_query_support_size": 3.6171875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 6,
          "query_type": "A_PLUS_B",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.3298276960849762,
          "query_target_nll": 2.4162020683288574,
          "query_top1_on_support": 0.328125,
          "mean_query_support_size": 3.6171875,
          "probe_belief_target_mass": 0.22745844721794128,
          "probe_belief_target_nll": 3.1122798919677734,
          "probe_belief_top1_on_support": 0.33984375,
          "mean_belief_support_size": 3.97265625,
          "decoder_belief_target_mass": 0.20910391211509705,
          "decoder_belief_target_nll": 3.183773994445801,
          "decoder_belief_top1_on_support": 0.2890625,
          "mean_decoder_belief_support_size": 3.97265625,
          "decoder_query_target_mass": 0.4453887343406677,
          "decoder_query_target_nll": 2.0519604682922363,
          "decoder_query_top1_on_support": 0.515625,
          "mean_decoder_query_support_size": 3.6171875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 6,
          "query_type": "A_PLUS_B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.3298276960849762,
          "query_target_nll": 2.4162020683288574,
          "query_top1_on_support": 0.328125,
          "mean_query_support_size": 3.6171875,
          "probe_belief_target_mass": 0.22745844721794128,
          "probe_belief_target_nll": 3.1122798919677734,
          "probe_belief_top1_on_support": 0.33984375,
          "mean_belief_support_size": 3.97265625,
          "decoder_belief_target_mass": 0.20910391211509705,
          "decoder_belief_target_nll": 3.183773994445801,
          "decoder_belief_top1_on_support": 0.2890625,
          "mean_decoder_belief_support_size": 3.97265625,
          "decoder_query_target_mass": 0.4453887343406677,
          "decoder_query_target_nll": 2.0519604682922363,
          "decoder_query_top1_on_support": 0.515625,
          "mean_decoder_query_support_size": 3.6171875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 6,
          "query_type": "A_MINUS_B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.2936553359031677,
          "query_target_nll": 2.4076223373413086,
          "query_top1_on_support": 0.30859375,
          "mean_query_support_size": 3.2265625,
          "probe_belief_target_mass": 0.029076505452394485,
          "probe_belief_target_nll": 20.63088607788086,
          "probe_belief_top1_on_support": 0.01953125,
          "mean_belief_support_size": 3.76171875,
          "decoder_belief_target_mass": 0.03165994584560394,
          "decoder_belief_target_nll": 18.1296443939209,
          "decoder_belief_top1_on_support": 0.03125,
          "mean_decoder_belief_support_size": 3.76171875,
          "decoder_query_target_mass": 0.3117271363735199,
          "decoder_query_target_nll": 9.684219360351562,
          "decoder_query_top1_on_support": 0.3125,
          "mean_decoder_query_support_size": 3.2265625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 6,
          "query_type": "A_MINUS_B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.29279598593711853,
          "query_target_nll": 2.416402578353882,
          "query_top1_on_support": 0.296875,
          "mean_query_support_size": 3.2265625,
          "probe_belief_target_mass": 0.043895743787288666,
          "probe_belief_target_nll": 18.049802780151367,
          "probe_belief_top1_on_support": 0.04296875,
          "mean_belief_support_size": 3.76171875,
          "decoder_belief_target_mass": 0.042697712779045105,
          "decoder_belief_target_nll": 13.999984741210938,
          "decoder_belief_top1_on_support": 0.02734375,
          "mean_decoder_belief_support_size": 3.76171875,
          "decoder_query_target_mass": 0.30055105686187744,
          "decoder_query_target_nll": 5.513845443725586,
          "decoder_query_top1_on_support": 0.29296875,
          "mean_decoder_query_support_size": 3.2265625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 6,
          "query_type": "A_MINUS_B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.291471004486084,
          "query_target_nll": 2.426262140274048,
          "query_top1_on_support": 0.2578125,
          "mean_query_support_size": 3.2265625,
          "probe_belief_target_mass": 0.06597362458705902,
          "probe_belief_target_nll": 14.213123321533203,
          "probe_belief_top1_on_support": 0.0703125,
          "mean_belief_support_size": 3.76171875,
          "decoder_belief_target_mass": 0.06308464705944061,
          "decoder_belief_target_nll": 11.237106323242188,
          "decoder_belief_top1_on_support": 0.05859375,
          "mean_decoder_belief_support_size": 3.76171875,
          "decoder_query_target_mass": 0.3183556795120239,
          "decoder_query_target_nll": 4.435617446899414,
          "decoder_query_top1_on_support": 0.31640625,
          "mean_decoder_query_support_size": 3.2265625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 6,
          "query_type": "A_MINUS_B",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.2922719120979309,
          "query_target_nll": 2.424588680267334,
          "query_top1_on_support": 0.26171875,
          "mean_query_support_size": 3.2265625,
          "probe_belief_target_mass": 0.07659319043159485,
          "probe_belief_target_nll": 11.128060340881348,
          "probe_belief_top1_on_support": 0.0859375,
          "mean_belief_support_size": 3.76171875,
          "decoder_belief_target_mass": 0.07384563982486725,
          "decoder_belief_target_nll": 9.449763298034668,
          "decoder_belief_top1_on_support": 0.0859375,
          "mean_decoder_belief_support_size": 3.76171875,
          "decoder_query_target_mass": 0.3140830099582672,
          "decoder_query_target_nll": 4.008111000061035,
          "decoder_query_top1_on_support": 0.32421875,
          "mean_decoder_query_support_size": 3.2265625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 6,
          "query_type": "A_MINUS_B",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.2924385070800781,
          "query_target_nll": 2.419478416442871,
          "query_top1_on_support": 0.27734375,
          "mean_query_support_size": 3.2265625,
          "probe_belief_target_mass": 0.22546625137329102,
          "probe_belief_target_nll": 3.1709983348846436,
          "probe_belief_top1_on_support": 0.28125,
          "mean_belief_support_size": 3.76171875,
          "decoder_belief_target_mass": 0.22148706018924713,
          "decoder_belief_target_nll": 3.1424717903137207,
          "decoder_belief_top1_on_support": 0.328125,
          "mean_decoder_belief_support_size": 3.76171875,
          "decoder_query_target_mass": 0.46611177921295166,
          "decoder_query_target_nll": 1.8931626081466675,
          "decoder_query_top1_on_support": 0.578125,
          "mean_decoder_query_support_size": 3.2265625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 6,
          "query_type": "A_MINUS_B",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.2924385070800781,
          "query_target_nll": 2.419478416442871,
          "query_top1_on_support": 0.27734375,
          "mean_query_support_size": 3.2265625,
          "probe_belief_target_mass": 0.22546625137329102,
          "probe_belief_target_nll": 3.1709983348846436,
          "probe_belief_top1_on_support": 0.28125,
          "mean_belief_support_size": 3.76171875,
          "decoder_belief_target_mass": 0.22148706018924713,
          "decoder_belief_target_nll": 3.1424717903137207,
          "decoder_belief_top1_on_support": 0.328125,
          "mean_decoder_belief_support_size": 3.76171875,
          "decoder_query_target_mass": 0.46611177921295166,
          "decoder_query_target_nll": 1.8931626081466675,
          "decoder_query_top1_on_support": 0.578125,
          "mean_decoder_query_support_size": 3.2265625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 6,
          "query_type": "A_MINUS_B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.2924385070800781,
          "query_target_nll": 2.419478416442871,
          "query_top1_on_support": 0.27734375,
          "mean_query_support_size": 3.2265625,
          "probe_belief_target_mass": 0.22546625137329102,
          "probe_belief_target_nll": 3.1709983348846436,
          "probe_belief_top1_on_support": 0.28125,
          "mean_belief_support_size": 3.76171875,
          "decoder_belief_target_mass": 0.22148706018924713,
          "decoder_belief_target_nll": 3.1424717903137207,
          "decoder_belief_top1_on_support": 0.328125,
          "mean_decoder_belief_support_size": 3.76171875,
          "decoder_query_target_mass": 0.46611177921295166,
          "decoder_query_target_nll": 1.8931626081466675,
          "decoder_query_top1_on_support": 0.578125,
          "mean_decoder_query_support_size": 3.2265625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 9,
          "query_type": "A",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.21900299191474915,
          "query_target_nll": 2.418043851852417,
          "query_top1_on_support": 0.1953125,
          "mean_query_support_size": 2.40625,
          "probe_belief_target_mass": 0.021631527692079544,
          "probe_belief_target_nll": 21.471817016601562,
          "probe_belief_top1_on_support": 0.01953125,
          "mean_belief_support_size": 2.73828125,
          "decoder_belief_target_mass": 0.021610047668218613,
          "decoder_belief_target_nll": 18.15215301513672,
          "decoder_belief_top1_on_support": 0.0234375,
          "mean_decoder_belief_support_size": 2.73828125,
          "decoder_query_target_mass": 0.21907417476177216,
          "decoder_query_target_nll": 2.4037961959838867,
          "decoder_query_top1_on_support": 0.22265625,
          "mean_decoder_query_support_size": 2.40625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 9,
          "query_type": "A",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.220951646566391,
          "query_target_nll": 2.4032392501831055,
          "query_top1_on_support": 0.2265625,
          "mean_query_support_size": 2.40625,
          "probe_belief_target_mass": 0.03730778396129608,
          "probe_belief_target_nll": 17.707809448242188,
          "probe_belief_top1_on_support": 0.046875,
          "mean_belief_support_size": 2.73828125,
          "decoder_belief_target_mass": 0.0348881371319294,
          "decoder_belief_target_nll": 13.364859580993652,
          "decoder_belief_top1_on_support": 0.04296875,
          "mean_decoder_belief_support_size": 2.73828125,
          "decoder_query_target_mass": 0.24321198463439941,
          "decoder_query_target_nll": 4.698854446411133,
          "decoder_query_top1_on_support": 0.23828125,
          "mean_decoder_query_support_size": 2.40625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 9,
          "query_type": "A",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.21828067302703857,
          "query_target_nll": 2.417978048324585,
          "query_top1_on_support": 0.22265625,
          "mean_query_support_size": 2.40625,
          "probe_belief_target_mass": 0.030084354802966118,
          "probe_belief_target_nll": 14.91852855682373,
          "probe_belief_top1_on_support": 0.03125,
          "mean_belief_support_size": 2.73828125,
          "decoder_belief_target_mass": 0.03421579301357269,
          "decoder_belief_target_nll": 12.027567863464355,
          "decoder_belief_top1_on_support": 0.03515625,
          "mean_decoder_belief_support_size": 2.73828125,
          "decoder_query_target_mass": 0.2588315010070801,
          "decoder_query_target_nll": 5.2525410652160645,
          "decoder_query_top1_on_support": 0.25390625,
          "mean_decoder_query_support_size": 2.40625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 9,
          "query_type": "A",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.22003233432769775,
          "query_target_nll": 2.405550241470337,
          "query_top1_on_support": 0.2421875,
          "mean_query_support_size": 2.40625,
          "probe_belief_target_mass": 0.034992437809705734,
          "probe_belief_target_nll": 12.834015846252441,
          "probe_belief_top1_on_support": 0.03125,
          "mean_belief_support_size": 2.73828125,
          "decoder_belief_target_mass": 0.03811271861195564,
          "decoder_belief_target_nll": 10.772750854492188,
          "decoder_belief_top1_on_support": 0.04296875,
          "mean_decoder_belief_support_size": 2.73828125,
          "decoder_query_target_mass": 0.25732773542404175,
          "decoder_query_target_nll": 5.489830017089844,
          "decoder_query_top1_on_support": 0.25390625,
          "mean_decoder_query_support_size": 2.40625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 9,
          "query_type": "A",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.21821913123130798,
          "query_target_nll": 2.419857978820801,
          "query_top1_on_support": 0.19921875,
          "mean_query_support_size": 2.40625,
          "probe_belief_target_mass": 0.05284116417169571,
          "probe_belief_target_nll": 8.153726577758789,
          "probe_belief_top1_on_support": 0.05078125,
          "mean_belief_support_size": 2.73828125,
          "decoder_belief_target_mass": 0.051214445382356644,
          "decoder_belief_target_nll": 8.093389511108398,
          "decoder_belief_top1_on_support": 0.0546875,
          "mean_decoder_belief_support_size": 2.73828125,
          "decoder_query_target_mass": 0.29781797528266907,
          "decoder_query_target_nll": 4.137447834014893,
          "decoder_query_top1_on_support": 0.34375,
          "mean_decoder_query_support_size": 2.40625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 9,
          "query_type": "A",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.21812798082828522,
          "query_target_nll": 2.4149169921875,
          "query_top1_on_support": 0.2109375,
          "mean_query_support_size": 2.40625,
          "probe_belief_target_mass": 0.14116206765174866,
          "probe_belief_target_nll": 3.3361663818359375,
          "probe_belief_top1_on_support": 0.18359375,
          "mean_belief_support_size": 2.73828125,
          "decoder_belief_target_mass": 0.13419456779956818,
          "decoder_belief_target_nll": 3.367373466491699,
          "decoder_belief_top1_on_support": 0.19921875,
          "mean_decoder_belief_support_size": 2.73828125,
          "decoder_query_target_mass": 0.43623101711273193,
          "decoder_query_target_nll": 1.7365283966064453,
          "decoder_query_top1_on_support": 0.51953125,
          "mean_decoder_query_support_size": 2.40625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 9,
          "query_type": "A",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.21812798082828522,
          "query_target_nll": 2.4149169921875,
          "query_top1_on_support": 0.2109375,
          "mean_query_support_size": 2.40625,
          "probe_belief_target_mass": 0.14116206765174866,
          "probe_belief_target_nll": 3.3361663818359375,
          "probe_belief_top1_on_support": 0.18359375,
          "mean_belief_support_size": 2.73828125,
          "decoder_belief_target_mass": 0.13419456779956818,
          "decoder_belief_target_nll": 3.367373466491699,
          "decoder_belief_top1_on_support": 0.19921875,
          "mean_decoder_belief_support_size": 2.73828125,
          "decoder_query_target_mass": 0.43623101711273193,
          "decoder_query_target_nll": 1.7365283966064453,
          "decoder_query_top1_on_support": 0.51953125,
          "mean_decoder_query_support_size": 2.40625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 9,
          "query_type": "B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.19027817249298096,
          "query_target_nll": 2.4296841621398926,
          "query_top1_on_support": 0.16796875,
          "mean_query_support_size": 2.1171875,
          "probe_belief_target_mass": 0.024481050670146942,
          "probe_belief_target_nll": 21.587594985961914,
          "probe_belief_top1_on_support": 0.01953125,
          "mean_belief_support_size": 2.37109375,
          "decoder_belief_target_mass": 0.025130566209554672,
          "decoder_belief_target_nll": 18.525238037109375,
          "decoder_belief_top1_on_support": 0.01171875,
          "mean_decoder_belief_support_size": 2.37109375,
          "decoder_query_target_mass": 0.19173559546470642,
          "decoder_query_target_nll": 2.4115076065063477,
          "decoder_query_top1_on_support": 0.1875,
          "mean_decoder_query_support_size": 2.1171875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 9,
          "query_type": "B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.193424791097641,
          "query_target_nll": 2.4025111198425293,
          "query_top1_on_support": 0.1953125,
          "mean_query_support_size": 2.1171875,
          "probe_belief_target_mass": 0.024728111922740936,
          "probe_belief_target_nll": 18.363985061645508,
          "probe_belief_top1_on_support": 0.03125,
          "mean_belief_support_size": 2.37109375,
          "decoder_belief_target_mass": 0.02415289357304573,
          "decoder_belief_target_nll": 14.266276359558105,
          "decoder_belief_top1_on_support": 0.02734375,
          "mean_decoder_belief_support_size": 2.37109375,
          "decoder_query_target_mass": 0.1911018341779709,
          "decoder_query_target_nll": 4.921765327453613,
          "decoder_query_top1_on_support": 0.1953125,
          "mean_decoder_query_support_size": 2.1171875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 9,
          "query_type": "B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.193841814994812,
          "query_target_nll": 2.4046072959899902,
          "query_top1_on_support": 0.171875,
          "mean_query_support_size": 2.1171875,
          "probe_belief_target_mass": 0.02658645063638687,
          "probe_belief_target_nll": 14.447484016418457,
          "probe_belief_top1_on_support": 0.01171875,
          "mean_belief_support_size": 2.37109375,
          "decoder_belief_target_mass": 0.029350396245718002,
          "decoder_belief_target_nll": 11.761978149414062,
          "decoder_belief_top1_on_support": 0.03515625,
          "mean_decoder_belief_support_size": 2.37109375,
          "decoder_query_target_mass": 0.20641447603702545,
          "decoder_query_target_nll": 4.871077537536621,
          "decoder_query_top1_on_support": 0.19921875,
          "mean_decoder_query_support_size": 2.1171875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 9,
          "query_type": "B",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.19323685765266418,
          "query_target_nll": 2.4134716987609863,
          "query_top1_on_support": 0.19921875,
          "mean_query_support_size": 2.1171875,
          "probe_belief_target_mass": 0.028761807829141617,
          "probe_belief_target_nll": 12.184160232543945,
          "probe_belief_top1_on_support": 0.02734375,
          "mean_belief_support_size": 2.37109375,
          "decoder_belief_target_mass": 0.029130494222044945,
          "decoder_belief_target_nll": 10.726828575134277,
          "decoder_belief_top1_on_support": 0.04296875,
          "mean_decoder_belief_support_size": 2.37109375,
          "decoder_query_target_mass": 0.22329413890838623,
          "decoder_query_target_nll": 4.89052152633667,
          "decoder_query_top1_on_support": 0.26171875,
          "mean_decoder_query_support_size": 2.1171875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 9,
          "query_type": "B",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.19124144315719604,
          "query_target_nll": 2.423872947692871,
          "query_top1_on_support": 0.14453125,
          "mean_query_support_size": 2.1171875,
          "probe_belief_target_mass": 0.03544319421052933,
          "probe_belief_target_nll": 9.027854919433594,
          "probe_belief_top1_on_support": 0.0390625,
          "mean_belief_support_size": 2.37109375,
          "decoder_belief_target_mass": 0.035628389567136765,
          "decoder_belief_target_nll": 8.955249786376953,
          "decoder_belief_top1_on_support": 0.0390625,
          "mean_decoder_belief_support_size": 2.37109375,
          "decoder_query_target_mass": 0.23094458878040314,
          "decoder_query_target_nll": 4.53026008605957,
          "decoder_query_top1_on_support": 0.2578125,
          "mean_decoder_query_support_size": 2.1171875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 9,
          "query_type": "B",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.19300776720046997,
          "query_target_nll": 2.4142212867736816,
          "query_top1_on_support": 0.21484375,
          "mean_query_support_size": 2.1171875,
          "probe_belief_target_mass": 0.13250242173671722,
          "probe_belief_target_nll": 3.2959628105163574,
          "probe_belief_top1_on_support": 0.2109375,
          "mean_belief_support_size": 2.37109375,
          "decoder_belief_target_mass": 0.1198950931429863,
          "decoder_belief_target_nll": 3.3351516723632812,
          "decoder_belief_top1_on_support": 0.203125,
          "mean_decoder_belief_support_size": 2.37109375,
          "decoder_query_target_mass": 0.3969704210758209,
          "decoder_query_target_nll": 1.7439463138580322,
          "decoder_query_top1_on_support": 0.47265625,
          "mean_decoder_query_support_size": 2.1171875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 9,
          "query_type": "B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.19300776720046997,
          "query_target_nll": 2.4142212867736816,
          "query_top1_on_support": 0.21484375,
          "mean_query_support_size": 2.1171875,
          "probe_belief_target_mass": 0.13250242173671722,
          "probe_belief_target_nll": 3.2959628105163574,
          "probe_belief_top1_on_support": 0.2109375,
          "mean_belief_support_size": 2.37109375,
          "decoder_belief_target_mass": 0.1198950931429863,
          "decoder_belief_target_nll": 3.3351516723632812,
          "decoder_belief_top1_on_support": 0.203125,
          "mean_decoder_belief_support_size": 2.37109375,
          "decoder_query_target_mass": 0.3969704210758209,
          "decoder_query_target_nll": 1.7439463138580322,
          "decoder_query_top1_on_support": 0.47265625,
          "mean_decoder_query_support_size": 2.1171875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 9,
          "query_type": "A_PLUS_B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.22787417471408844,
          "query_target_nll": 2.410775661468506,
          "query_top1_on_support": 0.20703125,
          "mean_query_support_size": 2.50390625,
          "probe_belief_target_mass": 0.0229348111897707,
          "probe_belief_target_nll": 21.128795623779297,
          "probe_belief_top1_on_support": 0.015625,
          "mean_belief_support_size": 2.71875,
          "decoder_belief_target_mass": 0.022537898272275925,
          "decoder_belief_target_nll": 18.29022979736328,
          "decoder_belief_top1_on_support": 0.02734375,
          "mean_decoder_belief_support_size": 2.71875,
          "decoder_query_target_mass": 0.22865945100784302,
          "decoder_query_target_nll": 2.431025505065918,
          "decoder_query_top1_on_support": 0.21875,
          "mean_decoder_query_support_size": 2.50390625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 9,
          "query_type": "A_PLUS_B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.22788622975349426,
          "query_target_nll": 2.411581039428711,
          "query_top1_on_support": 0.21875,
          "mean_query_support_size": 2.50390625,
          "probe_belief_target_mass": 0.030033374205231667,
          "probe_belief_target_nll": 18.39821434020996,
          "probe_belief_top1_on_support": 0.03515625,
          "mean_belief_support_size": 2.71875,
          "decoder_belief_target_mass": 0.030992813408374786,
          "decoder_belief_target_nll": 14.01161003112793,
          "decoder_belief_top1_on_support": 0.0234375,
          "mean_decoder_belief_support_size": 2.71875,
          "decoder_query_target_mass": 0.23625625669956207,
          "decoder_query_target_nll": 3.4130187034606934,
          "decoder_query_top1_on_support": 0.234375,
          "mean_decoder_query_support_size": 2.50390625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 9,
          "query_type": "A_PLUS_B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.22846558690071106,
          "query_target_nll": 2.4112064838409424,
          "query_top1_on_support": 0.1953125,
          "mean_query_support_size": 2.50390625,
          "probe_belief_target_mass": 0.02726122736930847,
          "probe_belief_target_nll": 14.480352401733398,
          "probe_belief_top1_on_support": 0.03515625,
          "mean_belief_support_size": 2.71875,
          "decoder_belief_target_mass": 0.03118482045829296,
          "decoder_belief_target_nll": 11.543022155761719,
          "decoder_belief_top1_on_support": 0.03125,
          "mean_decoder_belief_support_size": 2.71875,
          "decoder_query_target_mass": 0.22212493419647217,
          "decoder_query_target_nll": 3.8233230113983154,
          "decoder_query_top1_on_support": 0.23828125,
          "mean_decoder_query_support_size": 2.50390625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 9,
          "query_type": "A_PLUS_B",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.22874441742897034,
          "query_target_nll": 2.4081692695617676,
          "query_top1_on_support": 0.2109375,
          "mean_query_support_size": 2.50390625,
          "probe_belief_target_mass": 0.03716862574219704,
          "probe_belief_target_nll": 11.501039505004883,
          "probe_belief_top1_on_support": 0.046875,
          "mean_belief_support_size": 2.71875,
          "decoder_belief_target_mass": 0.03873613104224205,
          "decoder_belief_target_nll": 9.704584121704102,
          "decoder_belief_top1_on_support": 0.04296875,
          "mean_decoder_belief_support_size": 2.71875,
          "decoder_query_target_mass": 0.22540532052516937,
          "decoder_query_target_nll": 3.495692253112793,
          "decoder_query_top1_on_support": 0.23046875,
          "mean_decoder_query_support_size": 2.50390625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 9,
          "query_type": "A_PLUS_B",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.229184091091156,
          "query_target_nll": 2.405803680419922,
          "query_top1_on_support": 0.2578125,
          "mean_query_support_size": 2.50390625,
          "probe_belief_target_mass": 0.04993891716003418,
          "probe_belief_target_nll": 8.325678825378418,
          "probe_belief_top1_on_support": 0.10546875,
          "mean_belief_support_size": 2.71875,
          "decoder_belief_target_mass": 0.045595645904541016,
          "decoder_belief_target_nll": 8.1626558303833,
          "decoder_belief_top1_on_support": 0.046875,
          "mean_decoder_belief_support_size": 2.71875,
          "decoder_query_target_mass": 0.22751814126968384,
          "decoder_query_target_nll": 3.188206195831299,
          "decoder_query_top1_on_support": 0.2109375,
          "mean_decoder_query_support_size": 2.50390625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 9,
          "query_type": "A_PLUS_B",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.22921918332576752,
          "query_target_nll": 2.3998942375183105,
          "query_top1_on_support": 0.2578125,
          "mean_query_support_size": 2.50390625,
          "probe_belief_target_mass": 0.11991897225379944,
          "probe_belief_target_nll": 3.4781723022460938,
          "probe_belief_top1_on_support": 0.1640625,
          "mean_belief_support_size": 2.71875,
          "decoder_belief_target_mass": 0.11113123595714569,
          "decoder_belief_target_nll": 3.462674379348755,
          "decoder_belief_top1_on_support": 0.1328125,
          "mean_decoder_belief_support_size": 2.71875,
          "decoder_query_target_mass": 0.29932236671447754,
          "decoder_query_target_nll": 2.137068748474121,
          "decoder_query_top1_on_support": 0.3671875,
          "mean_decoder_query_support_size": 2.50390625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 9,
          "query_type": "A_PLUS_B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.22921918332576752,
          "query_target_nll": 2.3998942375183105,
          "query_top1_on_support": 0.2578125,
          "mean_query_support_size": 2.50390625,
          "probe_belief_target_mass": 0.11991897225379944,
          "probe_belief_target_nll": 3.4781723022460938,
          "probe_belief_top1_on_support": 0.1640625,
          "mean_belief_support_size": 2.71875,
          "decoder_belief_target_mass": 0.11113123595714569,
          "decoder_belief_target_nll": 3.462674379348755,
          "decoder_belief_top1_on_support": 0.1328125,
          "mean_decoder_belief_support_size": 2.71875,
          "decoder_query_target_mass": 0.29932236671447754,
          "decoder_query_target_nll": 2.137068748474121,
          "decoder_query_top1_on_support": 0.3671875,
          "mean_decoder_query_support_size": 2.50390625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 9,
          "query_type": "A_MINUS_B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.21431730687618256,
          "query_target_nll": 2.413658618927002,
          "query_top1_on_support": 0.19921875,
          "mean_query_support_size": 2.35546875,
          "probe_belief_target_mass": 0.016899891197681427,
          "probe_belief_target_nll": 21.84446144104004,
          "probe_belief_top1_on_support": 0.01953125,
          "mean_belief_support_size": 2.48046875,
          "decoder_belief_target_mass": 0.016230084002017975,
          "decoder_belief_target_nll": 18.86080551147461,
          "decoder_belief_top1_on_support": 0.02734375,
          "mean_decoder_belief_support_size": 2.48046875,
          "decoder_query_target_mass": 0.18403001129627228,
          "decoder_query_target_nll": 10.019756317138672,
          "decoder_query_top1_on_support": 0.18359375,
          "mean_decoder_query_support_size": 2.35546875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 9,
          "query_type": "A_MINUS_B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.21390725672245026,
          "query_target_nll": 2.4152607917785645,
          "query_top1_on_support": 0.2109375,
          "mean_query_support_size": 2.35546875,
          "probe_belief_target_mass": 0.020990032702684402,
          "probe_belief_target_nll": 17.612686157226562,
          "probe_belief_top1_on_support": 0.02734375,
          "mean_belief_support_size": 2.48046875,
          "decoder_belief_target_mass": 0.021069491282105446,
          "decoder_belief_target_nll": 14.314237594604492,
          "decoder_belief_top1_on_support": 0.0234375,
          "mean_decoder_belief_support_size": 2.48046875,
          "decoder_query_target_mass": 0.2003161907196045,
          "decoder_query_target_nll": 5.225517749786377,
          "decoder_query_top1_on_support": 0.1953125,
          "mean_decoder_query_support_size": 2.35546875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 9,
          "query_type": "A_MINUS_B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.21356059610843658,
          "query_target_nll": 2.4138660430908203,
          "query_top1_on_support": 0.203125,
          "mean_query_support_size": 2.35546875,
          "probe_belief_target_mass": 0.02097202092409134,
          "probe_belief_target_nll": 14.453289031982422,
          "probe_belief_top1_on_support": 0.015625,
          "mean_belief_support_size": 2.48046875,
          "decoder_belief_target_mass": 0.021951431408524513,
          "decoder_belief_target_nll": 11.902252197265625,
          "decoder_belief_top1_on_support": 0.02734375,
          "mean_decoder_belief_support_size": 2.48046875,
          "decoder_query_target_mass": 0.21640200912952423,
          "decoder_query_target_nll": 4.480003356933594,
          "decoder_query_top1_on_support": 0.21875,
          "mean_decoder_query_support_size": 2.35546875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 9,
          "query_type": "A_MINUS_B",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.2132585644721985,
          "query_target_nll": 2.4151077270507812,
          "query_top1_on_support": 0.19140625,
          "mean_query_support_size": 2.35546875,
          "probe_belief_target_mass": 0.024781223386526108,
          "probe_belief_target_nll": 12.757852554321289,
          "probe_belief_top1_on_support": 0.015625,
          "mean_belief_support_size": 2.48046875,
          "decoder_belief_target_mass": 0.021060694009065628,
          "decoder_belief_target_nll": 11.333868026733398,
          "decoder_belief_top1_on_support": 0.03125,
          "mean_decoder_belief_support_size": 2.48046875,
          "decoder_query_target_mass": 0.22359749674797058,
          "decoder_query_target_nll": 4.228910446166992,
          "decoder_query_top1_on_support": 0.23828125,
          "mean_decoder_query_support_size": 2.35546875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 9,
          "query_type": "A_MINUS_B",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.21386492252349854,
          "query_target_nll": 2.418651580810547,
          "query_top1_on_support": 0.203125,
          "mean_query_support_size": 2.35546875,
          "probe_belief_target_mass": 0.03718400001525879,
          "probe_belief_target_nll": 9.042962074279785,
          "probe_belief_top1_on_support": 0.0546875,
          "mean_belief_support_size": 2.48046875,
          "decoder_belief_target_mass": 0.037017323076725006,
          "decoder_belief_target_nll": 8.566728591918945,
          "decoder_belief_top1_on_support": 0.03125,
          "mean_decoder_belief_support_size": 2.48046875,
          "decoder_query_target_mass": 0.22269544005393982,
          "decoder_query_target_nll": 3.3288321495056152,
          "decoder_query_top1_on_support": 0.21484375,
          "mean_decoder_query_support_size": 2.35546875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 9,
          "query_type": "A_MINUS_B",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.2174033224582672,
          "query_target_nll": 2.398876428604126,
          "query_top1_on_support": 0.265625,
          "mean_query_support_size": 2.35546875,
          "probe_belief_target_mass": 0.1218496710062027,
          "probe_belief_target_nll": 3.436173915863037,
          "probe_belief_top1_on_support": 0.1796875,
          "mean_belief_support_size": 2.48046875,
          "decoder_belief_target_mass": 0.11620557308197021,
          "decoder_belief_target_nll": 3.4580259323120117,
          "decoder_belief_top1_on_support": 0.18359375,
          "mean_decoder_belief_support_size": 2.48046875,
          "decoder_query_target_mass": 0.3235267400741577,
          "decoder_query_target_nll": 2.033677816390991,
          "decoder_query_top1_on_support": 0.390625,
          "mean_decoder_query_support_size": 2.35546875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 9,
          "query_type": "A_MINUS_B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.2174033224582672,
          "query_target_nll": 2.398876428604126,
          "query_top1_on_support": 0.265625,
          "mean_query_support_size": 2.35546875,
          "probe_belief_target_mass": 0.1218496710062027,
          "probe_belief_target_nll": 3.436173915863037,
          "probe_belief_top1_on_support": 0.1796875,
          "mean_belief_support_size": 2.48046875,
          "decoder_belief_target_mass": 0.11620557308197021,
          "decoder_belief_target_nll": 3.4580259323120117,
          "decoder_belief_top1_on_support": 0.18359375,
          "mean_decoder_belief_support_size": 2.48046875,
          "decoder_query_target_mass": 0.3235267400741577,
          "decoder_query_target_nll": 2.033677816390991,
          "decoder_query_top1_on_support": 0.390625,
          "mean_decoder_query_support_size": 2.35546875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 12,
          "query_type": "A",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.17420795559883118,
          "query_target_nll": 2.416779041290283,
          "query_top1_on_support": 0.171875,
          "mean_query_support_size": 1.90625,
          "probe_belief_target_mass": 0.017275935038924217,
          "probe_belief_target_nll": 21.14617919921875,
          "probe_belief_top1_on_support": 0.01953125,
          "mean_belief_support_size": 2.21484375,
          "decoder_belief_target_mass": 0.018119962885975838,
          "decoder_belief_target_nll": 18.203989028930664,
          "decoder_belief_top1_on_support": 0.01953125,
          "mean_decoder_belief_support_size": 2.21484375,
          "decoder_query_target_mass": 0.1744367778301239,
          "decoder_query_target_nll": 2.396672248840332,
          "decoder_query_top1_on_support": 0.2109375,
          "mean_decoder_query_support_size": 1.90625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 12,
          "query_type": "A",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.17321866750717163,
          "query_target_nll": 2.41875958442688,
          "query_top1_on_support": 0.18359375,
          "mean_query_support_size": 1.90625,
          "probe_belief_target_mass": 0.028613703325390816,
          "probe_belief_target_nll": 17.494342803955078,
          "probe_belief_top1_on_support": 0.0234375,
          "mean_belief_support_size": 2.21484375,
          "decoder_belief_target_mass": 0.029129324480891228,
          "decoder_belief_target_nll": 14.475955963134766,
          "decoder_belief_top1_on_support": 0.0234375,
          "mean_decoder_belief_support_size": 2.21484375,
          "decoder_query_target_mass": 0.1895085871219635,
          "decoder_query_target_nll": 4.822232723236084,
          "decoder_query_top1_on_support": 0.1796875,
          "mean_decoder_query_support_size": 1.90625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 12,
          "query_type": "A",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.17410758137702942,
          "query_target_nll": 2.41349458694458,
          "query_top1_on_support": 0.1875,
          "mean_query_support_size": 1.90625,
          "probe_belief_target_mass": 0.0186513289809227,
          "probe_belief_target_nll": 15.12263298034668,
          "probe_belief_top1_on_support": 0.01953125,
          "mean_belief_support_size": 2.21484375,
          "decoder_belief_target_mass": 0.02077677845954895,
          "decoder_belief_target_nll": 12.558911323547363,
          "decoder_belief_top1_on_support": 0.02734375,
          "mean_decoder_belief_support_size": 2.21484375,
          "decoder_query_target_mass": 0.19143134355545044,
          "decoder_query_target_nll": 5.359807968139648,
          "decoder_query_top1_on_support": 0.20703125,
          "mean_decoder_query_support_size": 1.90625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 12,
          "query_type": "A",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.17330747842788696,
          "query_target_nll": 2.4194819927215576,
          "query_top1_on_support": 0.18359375,
          "mean_query_support_size": 1.90625,
          "probe_belief_target_mass": 0.026284407824277878,
          "probe_belief_target_nll": 11.728509902954102,
          "probe_belief_top1_on_support": 0.0390625,
          "mean_belief_support_size": 2.21484375,
          "decoder_belief_target_mass": 0.02527078054845333,
          "decoder_belief_target_nll": 11.035622596740723,
          "decoder_belief_top1_on_support": 0.03515625,
          "mean_decoder_belief_support_size": 2.21484375,
          "decoder_query_target_mass": 0.18995895981788635,
          "decoder_query_target_nll": 5.49775505065918,
          "decoder_query_top1_on_support": 0.1796875,
          "mean_decoder_query_support_size": 1.90625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 12,
          "query_type": "A",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.17254050076007843,
          "query_target_nll": 2.4262242317199707,
          "query_top1_on_support": 0.16796875,
          "mean_query_support_size": 1.90625,
          "probe_belief_target_mass": 0.02451886609196663,
          "probe_belief_target_nll": 9.257622718811035,
          "probe_belief_top1_on_support": 0.01953125,
          "mean_belief_support_size": 2.21484375,
          "decoder_belief_target_mass": 0.0232442244887352,
          "decoder_belief_target_nll": 9.268438339233398,
          "decoder_belief_top1_on_support": 0.0234375,
          "mean_decoder_belief_support_size": 2.21484375,
          "decoder_query_target_mass": 0.18921633064746857,
          "decoder_query_target_nll": 4.721796989440918,
          "decoder_query_top1_on_support": 0.1875,
          "mean_decoder_query_support_size": 1.90625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 12,
          "query_type": "A",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.17329154908657074,
          "query_target_nll": 2.4209325313568115,
          "query_top1_on_support": 0.16015625,
          "mean_query_support_size": 1.90625,
          "probe_belief_target_mass": 0.034536685794591904,
          "probe_belief_target_nll": 7.907950401306152,
          "probe_belief_top1_on_support": 0.05078125,
          "mean_belief_support_size": 2.21484375,
          "decoder_belief_target_mass": 0.03444144129753113,
          "decoder_belief_target_nll": 7.519474983215332,
          "decoder_belief_top1_on_support": 0.0234375,
          "mean_decoder_belief_support_size": 2.21484375,
          "decoder_query_target_mass": 0.22622814774513245,
          "decoder_query_target_nll": 3.5766096115112305,
          "decoder_query_top1_on_support": 0.24609375,
          "mean_decoder_query_support_size": 1.90625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 12,
          "query_type": "A",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.17382970452308655,
          "query_target_nll": 2.416126251220703,
          "query_top1_on_support": 0.171875,
          "mean_query_support_size": 1.90625,
          "probe_belief_target_mass": 0.09888523817062378,
          "probe_belief_target_nll": 3.578584671020508,
          "probe_belief_top1_on_support": 0.1484375,
          "mean_belief_support_size": 2.21484375,
          "decoder_belief_target_mass": 0.09660586714744568,
          "decoder_belief_target_nll": 3.6140098571777344,
          "decoder_belief_top1_on_support": 0.125,
          "mean_decoder_belief_support_size": 2.21484375,
          "decoder_query_target_mass": 0.3380277752876282,
          "decoder_query_target_nll": 1.8242456912994385,
          "decoder_query_top1_on_support": 0.375,
          "mean_decoder_query_support_size": 1.90625
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 12,
          "query_type": "B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.1512402892112732,
          "query_target_nll": 2.4240856170654297,
          "query_top1_on_support": 0.12890625,
          "mean_query_support_size": 1.6796875,
          "probe_belief_target_mass": 0.012572983279824257,
          "probe_belief_target_nll": 21.863279342651367,
          "probe_belief_top1_on_support": 0.01171875,
          "mean_belief_support_size": 1.84765625,
          "decoder_belief_target_mass": 0.013551286421716213,
          "decoder_belief_target_nll": 18.383407592773438,
          "decoder_belief_top1_on_support": 0.015625,
          "mean_decoder_belief_support_size": 1.84765625,
          "decoder_query_target_mass": 0.15219247341156006,
          "decoder_query_target_nll": 2.4099597930908203,
          "decoder_query_top1_on_support": 0.17578125,
          "mean_decoder_query_support_size": 1.6796875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 12,
          "query_type": "B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.15410169959068298,
          "query_target_nll": 2.4038596153259277,
          "query_top1_on_support": 0.14453125,
          "mean_query_support_size": 1.6796875,
          "probe_belief_target_mass": 0.014153973199427128,
          "probe_belief_target_nll": 18.401979446411133,
          "probe_belief_top1_on_support": 0.01171875,
          "mean_belief_support_size": 1.84765625,
          "decoder_belief_target_mass": 0.01373143121600151,
          "decoder_belief_target_nll": 14.244157791137695,
          "decoder_belief_top1_on_support": 0.01171875,
          "mean_decoder_belief_support_size": 1.84765625,
          "decoder_query_target_mass": 0.15249986946582794,
          "decoder_query_target_nll": 4.470193862915039,
          "decoder_query_top1_on_support": 0.1640625,
          "mean_decoder_query_support_size": 1.6796875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 12,
          "query_type": "B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.15154534578323364,
          "query_target_nll": 2.423503875732422,
          "query_top1_on_support": 0.12890625,
          "mean_query_support_size": 1.6796875,
          "probe_belief_target_mass": 0.016381625086069107,
          "probe_belief_target_nll": 14.233563423156738,
          "probe_belief_top1_on_support": 0.015625,
          "mean_belief_support_size": 1.84765625,
          "decoder_belief_target_mass": 0.01564120501279831,
          "decoder_belief_target_nll": 11.764143943786621,
          "decoder_belief_top1_on_support": 0.0078125,
          "mean_decoder_belief_support_size": 1.84765625,
          "decoder_query_target_mass": 0.16995765268802643,
          "decoder_query_target_nll": 5.092825889587402,
          "decoder_query_top1_on_support": 0.1796875,
          "mean_decoder_query_support_size": 1.6796875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 12,
          "query_type": "B",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.15154002606868744,
          "query_target_nll": 2.4221792221069336,
          "query_top1_on_support": 0.13671875,
          "mean_query_support_size": 1.6796875,
          "probe_belief_target_mass": 0.014560123905539513,
          "probe_belief_target_nll": 12.169281005859375,
          "probe_belief_top1_on_support": 0.0078125,
          "mean_belief_support_size": 1.84765625,
          "decoder_belief_target_mass": 0.014919916167855263,
          "decoder_belief_target_nll": 11.020995140075684,
          "decoder_belief_top1_on_support": 0.01171875,
          "mean_decoder_belief_support_size": 1.84765625,
          "decoder_query_target_mass": 0.15698042511940002,
          "decoder_query_target_nll": 5.073497772216797,
          "decoder_query_top1_on_support": 0.140625,
          "mean_decoder_query_support_size": 1.6796875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 12,
          "query_type": "B",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.15199944376945496,
          "query_target_nll": 2.4215188026428223,
          "query_top1_on_support": 0.1484375,
          "mean_query_support_size": 1.6796875,
          "probe_belief_target_mass": 0.01807985082268715,
          "probe_belief_target_nll": 9.324308395385742,
          "probe_belief_top1_on_support": 0.015625,
          "mean_belief_support_size": 1.84765625,
          "decoder_belief_target_mass": 0.016991855576634407,
          "decoder_belief_target_nll": 9.236526489257812,
          "decoder_belief_top1_on_support": 0.01171875,
          "mean_decoder_belief_support_size": 1.84765625,
          "decoder_query_target_mass": 0.16033825278282166,
          "decoder_query_target_nll": 4.573365688323975,
          "decoder_query_top1_on_support": 0.14453125,
          "mean_decoder_query_support_size": 1.6796875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 12,
          "query_type": "B",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.15276233851909637,
          "query_target_nll": 2.4110500812530518,
          "query_top1_on_support": 0.16015625,
          "mean_query_support_size": 1.6796875,
          "probe_belief_target_mass": 0.030259154736995697,
          "probe_belief_target_nll": 7.431219100952148,
          "probe_belief_top1_on_support": 0.05078125,
          "mean_belief_support_size": 1.84765625,
          "decoder_belief_target_mass": 0.02736939862370491,
          "decoder_belief_target_nll": 7.485987663269043,
          "decoder_belief_top1_on_support": 0.0078125,
          "mean_decoder_belief_support_size": 1.84765625,
          "decoder_query_target_mass": 0.203252375125885,
          "decoder_query_target_nll": 3.4984095096588135,
          "decoder_query_top1_on_support": 0.21875,
          "mean_decoder_query_support_size": 1.6796875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 12,
          "query_type": "B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.15380288660526276,
          "query_target_nll": 2.4098868370056152,
          "query_top1_on_support": 0.1484375,
          "mean_query_support_size": 1.6796875,
          "probe_belief_target_mass": 0.09488880634307861,
          "probe_belief_target_nll": 3.503718376159668,
          "probe_belief_top1_on_support": 0.14453125,
          "mean_belief_support_size": 1.84765625,
          "decoder_belief_target_mass": 0.0879046767950058,
          "decoder_belief_target_nll": 3.5081756114959717,
          "decoder_belief_top1_on_support": 0.12890625,
          "mean_decoder_belief_support_size": 1.84765625,
          "decoder_query_target_mass": 0.3200962543487549,
          "decoder_query_target_nll": 1.8372888565063477,
          "decoder_query_top1_on_support": 0.3828125,
          "mean_decoder_query_support_size": 1.6796875
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 12,
          "query_type": "A_PLUS_B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.1781441867351532,
          "query_target_nll": 2.3912062644958496,
          "query_top1_on_support": 0.1796875,
          "mean_query_support_size": 1.93359375,
          "probe_belief_target_mass": 0.023897603154182434,
          "probe_belief_target_nll": 21.073360443115234,
          "probe_belief_top1_on_support": 0.015625,
          "mean_belief_support_size": 2.0,
          "decoder_belief_target_mass": 0.02462306246161461,
          "decoder_belief_target_nll": 18.222949981689453,
          "decoder_belief_top1_on_support": 0.0234375,
          "mean_decoder_belief_support_size": 2.0,
          "decoder_query_target_mass": 0.17318959534168243,
          "decoder_query_target_nll": 2.455841541290283,
          "decoder_query_top1_on_support": 0.140625,
          "mean_decoder_query_support_size": 1.93359375
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 12,
          "query_type": "A_PLUS_B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.17761048674583435,
          "query_target_nll": 2.4018139839172363,
          "query_top1_on_support": 0.21875,
          "mean_query_support_size": 1.93359375,
          "probe_belief_target_mass": 0.015079937875270844,
          "probe_belief_target_nll": 18.19031524658203,
          "probe_belief_top1_on_support": 0.01171875,
          "mean_belief_support_size": 2.0,
          "decoder_belief_target_mass": 0.014007603749632835,
          "decoder_belief_target_nll": 14.610099792480469,
          "decoder_belief_top1_on_support": 0.0078125,
          "mean_decoder_belief_support_size": 2.0,
          "decoder_query_target_mass": 0.18462321162223816,
          "decoder_query_target_nll": 3.617112159729004,
          "decoder_query_top1_on_support": 0.17578125,
          "mean_decoder_query_support_size": 1.93359375
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 12,
          "query_type": "A_PLUS_B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.17738492786884308,
          "query_target_nll": 2.4070956707000732,
          "query_top1_on_support": 0.22265625,
          "mean_query_support_size": 1.93359375,
          "probe_belief_target_mass": 0.0223474632948637,
          "probe_belief_target_nll": 14.930203437805176,
          "probe_belief_top1_on_support": 0.03125,
          "mean_belief_support_size": 2.0,
          "decoder_belief_target_mass": 0.022038761526346207,
          "decoder_belief_target_nll": 12.283011436462402,
          "decoder_belief_top1_on_support": 0.01953125,
          "mean_decoder_belief_support_size": 2.0,
          "decoder_query_target_mass": 0.17876873910427094,
          "decoder_query_target_nll": 3.7769742012023926,
          "decoder_query_top1_on_support": 0.1796875,
          "mean_decoder_query_support_size": 1.93359375
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 12,
          "query_type": "A_PLUS_B",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.17631253600120544,
          "query_target_nll": 2.4125685691833496,
          "query_top1_on_support": 0.1796875,
          "mean_query_support_size": 1.93359375,
          "probe_belief_target_mass": 0.015246689319610596,
          "probe_belief_target_nll": 13.62374496459961,
          "probe_belief_top1_on_support": 0.01953125,
          "mean_belief_support_size": 2.0,
          "decoder_belief_target_mass": 0.017646031454205513,
          "decoder_belief_target_nll": 11.557056427001953,
          "decoder_belief_top1_on_support": 0.015625,
          "mean_decoder_belief_support_size": 2.0,
          "decoder_query_target_mass": 0.17416658997535706,
          "decoder_query_target_nll": 3.6306450366973877,
          "decoder_query_top1_on_support": 0.19140625,
          "mean_decoder_query_support_size": 1.93359375
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 12,
          "query_type": "A_PLUS_B",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.17663423717021942,
          "query_target_nll": 2.4132730960845947,
          "query_top1_on_support": 0.15625,
          "mean_query_support_size": 1.93359375,
          "probe_belief_target_mass": 0.019998831674456596,
          "probe_belief_target_nll": 9.91727066040039,
          "probe_belief_top1_on_support": 0.02734375,
          "mean_belief_support_size": 2.0,
          "decoder_belief_target_mass": 0.01837123930454254,
          "decoder_belief_target_nll": 9.524738311767578,
          "decoder_belief_top1_on_support": 0.01953125,
          "mean_decoder_belief_support_size": 2.0,
          "decoder_query_target_mass": 0.18099868297576904,
          "decoder_query_target_nll": 3.138881206512451,
          "decoder_query_top1_on_support": 0.171875,
          "mean_decoder_query_support_size": 1.93359375
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 12,
          "query_type": "A_PLUS_B",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.17554931342601776,
          "query_target_nll": 2.4174814224243164,
          "query_top1_on_support": 0.171875,
          "mean_query_support_size": 1.93359375,
          "probe_belief_target_mass": 0.0305943563580513,
          "probe_belief_target_nll": 8.064120292663574,
          "probe_belief_top1_on_support": 0.03515625,
          "mean_belief_support_size": 2.0,
          "decoder_belief_target_mass": 0.03110678307712078,
          "decoder_belief_target_nll": 7.9368720054626465,
          "decoder_belief_top1_on_support": 0.04296875,
          "mean_decoder_belief_support_size": 2.0,
          "decoder_query_target_mass": 0.17756514251232147,
          "decoder_query_target_nll": 3.198275089263916,
          "decoder_query_top1_on_support": 0.171875,
          "mean_decoder_query_support_size": 1.93359375
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 12,
          "query_type": "A_PLUS_B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.17776033282279968,
          "query_target_nll": 2.40238618850708,
          "query_top1_on_support": 0.20703125,
          "mean_query_support_size": 1.93359375,
          "probe_belief_target_mass": 0.09849802404642105,
          "probe_belief_target_nll": 3.464932441711426,
          "probe_belief_top1_on_support": 0.12109375,
          "mean_belief_support_size": 2.0,
          "decoder_belief_target_mass": 0.08449752628803253,
          "decoder_belief_target_nll": 3.5503430366516113,
          "decoder_belief_top1_on_support": 0.09765625,
          "mean_decoder_belief_support_size": 2.0,
          "decoder_query_target_mass": 0.22340117394924164,
          "decoder_query_target_nll": 2.208347797393799,
          "decoder_query_top1_on_support": 0.26171875,
          "mean_decoder_query_support_size": 1.93359375
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 12,
          "query_type": "A_MINUS_B",
          "k": 0,
          "n": 256,
          "query_target_mass": 0.1770048588514328,
          "query_target_nll": 2.4180986881256104,
          "query_top1_on_support": 0.19140625,
          "mean_query_support_size": 1.95703125,
          "probe_belief_target_mass": 0.020808205008506775,
          "probe_belief_target_nll": 21.402442932128906,
          "probe_belief_top1_on_support": 0.01171875,
          "mean_belief_support_size": 2.17578125,
          "decoder_belief_target_mass": 0.01990417391061783,
          "decoder_belief_target_nll": 18.162029266357422,
          "decoder_belief_top1_on_support": 0.01171875,
          "mean_decoder_belief_support_size": 2.17578125,
          "decoder_query_target_mass": 0.21674367785453796,
          "decoder_query_target_nll": 9.260727882385254,
          "decoder_query_top1_on_support": 0.21484375,
          "mean_decoder_query_support_size": 1.95703125
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 12,
          "query_type": "A_MINUS_B",
          "k": 1,
          "n": 256,
          "query_target_mass": 0.17846401035785675,
          "query_target_nll": 2.4180521965026855,
          "query_top1_on_support": 0.1875,
          "mean_query_support_size": 1.95703125,
          "probe_belief_target_mass": 0.016173163428902626,
          "probe_belief_target_nll": 18.689973831176758,
          "probe_belief_top1_on_support": 0.01171875,
          "mean_belief_support_size": 2.17578125,
          "decoder_belief_target_mass": 0.016483794897794724,
          "decoder_belief_target_nll": 14.53369426727295,
          "decoder_belief_top1_on_support": 0.01953125,
          "mean_decoder_belief_support_size": 2.17578125,
          "decoder_query_target_mass": 0.18213286995887756,
          "decoder_query_target_nll": 5.358767032623291,
          "decoder_query_top1_on_support": 0.19921875,
          "mean_decoder_query_support_size": 1.95703125
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 12,
          "query_type": "A_MINUS_B",
          "k": 2,
          "n": 256,
          "query_target_mass": 0.17691190540790558,
          "query_target_nll": 2.4306201934814453,
          "query_top1_on_support": 0.19140625,
          "mean_query_support_size": 1.95703125,
          "probe_belief_target_mass": 0.016986990347504616,
          "probe_belief_target_nll": 14.963518142700195,
          "probe_belief_top1_on_support": 0.015625,
          "mean_belief_support_size": 2.17578125,
          "decoder_belief_target_mass": 0.016077011823654175,
          "decoder_belief_target_nll": 12.596916198730469,
          "decoder_belief_top1_on_support": 0.0234375,
          "mean_decoder_belief_support_size": 2.17578125,
          "decoder_query_target_mass": 0.16119112074375153,
          "decoder_query_target_nll": 4.426577568054199,
          "decoder_query_top1_on_support": 0.140625,
          "mean_decoder_query_support_size": 1.95703125
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 12,
          "query_type": "A_MINUS_B",
          "k": 3,
          "n": 256,
          "query_target_mass": 0.1770486831665039,
          "query_target_nll": 2.425581216812134,
          "query_top1_on_support": 0.1796875,
          "mean_query_support_size": 1.95703125,
          "probe_belief_target_mass": 0.016119983047246933,
          "probe_belief_target_nll": 12.277446746826172,
          "probe_belief_top1_on_support": 0.01953125,
          "mean_belief_support_size": 2.17578125,
          "decoder_belief_target_mass": 0.017319902777671814,
          "decoder_belief_target_nll": 11.259937286376953,
          "decoder_belief_top1_on_support": 0.02734375,
          "mean_decoder_belief_support_size": 2.17578125,
          "decoder_query_target_mass": 0.1590101420879364,
          "decoder_query_target_nll": 3.990626811981201,
          "decoder_query_top1_on_support": 0.1640625,
          "mean_decoder_query_support_size": 1.95703125
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 12,
          "query_type": "A_MINUS_B",
          "k": 6,
          "n": 256,
          "query_target_mass": 0.1777195930480957,
          "query_target_nll": 2.4170517921447754,
          "query_top1_on_support": 0.1796875,
          "mean_query_support_size": 1.95703125,
          "probe_belief_target_mass": 0.019300632178783417,
          "probe_belief_target_nll": 9.599040985107422,
          "probe_belief_top1_on_support": 0.01171875,
          "mean_belief_support_size": 2.17578125,
          "decoder_belief_target_mass": 0.020503856241703033,
          "decoder_belief_target_nll": 9.69851016998291,
          "decoder_belief_top1_on_support": 0.03125,
          "mean_decoder_belief_support_size": 2.17578125,
          "decoder_query_target_mass": 0.18248137831687927,
          "decoder_query_target_nll": 3.4458670616149902,
          "decoder_query_top1_on_support": 0.18359375,
          "mean_decoder_query_support_size": 1.95703125
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 12,
          "query_type": "A_MINUS_B",
          "k": 9,
          "n": 256,
          "query_target_mass": 0.17888465523719788,
          "query_target_nll": 2.4181957244873047,
          "query_top1_on_support": 0.21484375,
          "mean_query_support_size": 1.95703125,
          "probe_belief_target_mass": 0.03453803062438965,
          "probe_belief_target_nll": 7.442439079284668,
          "probe_belief_top1_on_support": 0.03515625,
          "mean_belief_support_size": 2.17578125,
          "decoder_belief_target_mass": 0.03317587077617645,
          "decoder_belief_target_nll": 7.306785583496094,
          "decoder_belief_top1_on_support": 0.03515625,
          "mean_decoder_belief_support_size": 2.17578125,
          "decoder_query_target_mass": 0.19426189363002777,
          "decoder_query_target_nll": 3.08058762550354,
          "decoder_query_top1_on_support": 0.21875,
          "mean_decoder_query_support_size": 1.95703125
        },
        {
          "model": "dense",
          "variant": "pilot_gru_lowrank_d256_r16",
          "supervision": "teacher_belief",
          "transition": "gru",
          "decoder_type": "low_rank",
          "state_dim": 256,
          "instr_dim": 128,
          "decoder_rank": 16,
          "length": 12,
          "query_type": "A_MINUS_B",
          "k": 12,
          "n": 256,
          "query_target_mass": 0.1787146031856537,
          "query_target_nll": 2.410521984100342,
          "query_top1_on_support": 0.16796875,
          "mean_query_support_size": 1.95703125,
          "probe_belief_target_mass": 0.08936353772878647,
          "probe_belief_target_nll": 3.6119508743286133,
          "probe_belief_top1_on_support": 0.1328125,
          "mean_belief_support_size": 2.17578125,
          "decoder_belief_target_mass": 0.08628970384597778,
          "decoder_belief_target_nll": 3.586900472640991,
          "decoder_belief_top1_on_support": 0.15234375,
          "mean_decoder_belief_support_size": 2.17578125,
          "decoder_query_target_mass": 0.25798499584198,
          "decoder_query_target_nll": 2.099975109100342,
          "decoder_query_top1_on_support": 0.33984375,
          "mean_decoder_query_support_size": 1.95703125
        }
      ]
    }
  ],
  "checkpoints": [
    "large_artifacts/dense_teacher_distillation/checkpoints/pilot_gru_lowrank_d256_r16/checkpoint_final.pt"
  ]
}