{
  "campaign": "pusht-parallel-20261019-s0",
  "paper": "https://arxiv.org/abs/2604.23609",
  "dp_reference": "https://diffusion-policy.cs.columbia.edu/data/experiments/image/pusht/diffusion_policy_cnn/config.yaml",
  "policies": [
    "dp",
    "tdp",
    "grn",
    "dit"
  ],
  "seeds": [
    0
  ],
  "steps": 512400,
  "reference_epochs": 3050,
  "batch_size": 64,
  "lr": 0.0001,
  "betas": [
    0.95,
    0.999
  ],
  "weight_decay": 1e-06,
  "eps": 1e-08,
  "warmup_steps": 500,
  "ema": {
    "inv_gamma": 1.0,
    "power": 0.75,
    "max_value": 0.9999
  },
  "dataset_seed": 42,
  "log_every": 100,
  "checkpoint_every": 8400,
  "eval_every": 8400,
  "eval_episodes": 50,
  "eval_videos": 3,
  "eval_seed_start": 100000,
  "eval_max_steps": 300,
  "eval_sampler": "ddim",
  "eval_nfe": 10,
  "final_dp_samplers": [
    {
      "sampler": "ddpm",
      "nfe": 100
    },
    {
      "sampler": "ddim",
      "nfe": 10
    }
  ],
  "final_tdp_nfe": [
    2,
    10
  ],
  "final_grn_nfe": [
    2,
    8,
    16
  ],
  "model": {
    "down_dims": [
      512,
      1024,
      2048
    ],
    "diffusion_step_embed_dim": 128,
    "kernel_size": 5,
    "n_groups": 8,
    "crop_size": 84,
    "horizon": 16,
    "n_obs_steps": 2,
    "n_action_steps": 8,
    "dit_width": 512,
    "dit_depth": 6,
    "dit_heads": 8,
    "grn_width": 384,
    "grn_depth": 8,
    "grn_heads": 8,
    "bits": 8,
    "tube_lambda": 10.0,
    "tube_sigma0": 0.05,
    "tube_k": 1.0,
    "tube_time_scale": 1000.0
  },
  "hash_verification": false,
  "live_monitoring": false,
  "fidelity_notes": [
    "RGB PushT controlled reproduction; original DP 90-demo split and normalization, fixed 512400 optimizer updates at effective batch64.",
    "TDP is derived from the paper, as authors have not released code or complete PushT hyperparameters. Tube constants and action-to-U-Net lift are explicit implementation choices.",
    "GRN is an action-domain adaptation, with a larger configurable temporal head.",
    "DiT uses Dasari et al. DiT-Block conditioning adapted to shared PushT RGB encoder, not a pretrained VLA.",
    "Modern gym-pusht preserves legacy state reset order but is not the original archived DP environment; version recorded.",
    "Periodic checkpoint evaluation reports test score as in the paper; final published policy is the final EMA, not the best test-selected checkpoint."
  ],
  "policy": "dp",
  "seed": 0,
  "dataset": {
    "source": "https://diffusion-policy.cs.columbia.edu/data/training/pusht.zip",
    "frames": 25650,
    "episodes": 206,
    "train_episode_ids": [
      7,
      9,
      10,
      11,
      13,
      16,
      20,
      25,
      26,
      27,
      32,
      33,
      35,
      42,
      44,
      50,
      51,
      56,
      58,
      60,
      61,
      63,
      65,
      67,
      71,
      73,
      76,
      77,
      78,
      81,
      84,
      85,
      86,
      89,
      91,
      93,
      94,
      97,
      98,
      102,
      103,
      109,
      110,
      111,
      112,
      117,
      118,
      119,
      120,
      122,
      125,
      126,
      128,
      129,
      130,
      131,
      132,
      135,
      137,
      138,
      140,
      141,
      145,
      148,
      156,
      158,
      159,
      160,
      161,
      163,
      165,
      166,
      167,
      170,
      172,
      174,
      176,
      178,
      179,
      183,
      184,
      185,
      188,
      189,
      192,
      194,
      195,
      198,
      200,
      203
    ],
    "validation_episode_ids": [
      0,
      1,
      2,
      3,
      4,
      5,
      6,
      8,
      12,
      14,
      15,
      17,
      18,
      19,
      21,
      22,
      23,
      24,
      28,
      29,
      30,
      31,
      34,
      36,
      37,
      38,
      39,
      40,
      41,
      43,
      45,
      46,
      47,
      48,
      49,
      52,
      53,
      54,
      55,
      57,
      59,
      62,
      64,
      66,
      68,
      69,
      70,
      72,
      74,
      75,
      79,
      80,
      82,
      83,
      87,
      88,
      90,
      92,
      95,
      96,
      99,
      100,
      101,
      104,
      105,
      106,
      107,
      108,
      113,
      114,
      115,
      116,
      121,
      123,
      124,
      127,
      133,
      134,
      136,
      139,
      142,
      143,
      144,
      146,
      147,
      149,
      150,
      151,
      152,
      153,
      154,
      155,
      157,
      162,
      164,
      168,
      169,
      171,
      173,
      175,
      177,
      180,
      181,
      182,
      186,
      187,
      190,
      191,
      193,
      196,
      197,
      199,
      201,
      202,
      204,
      205
    ],
    "initial_validation_episode_ids": [
      18,
      90,
      134,
      157
    ],
    "normalization": {
      "images": "2*(uint8/255)-1 in shared encoder",
      "state_action": "full replay min/max, original DP convention"
    },
    "stats": {
      "action": {
        "min": [
          12.0,
          25.0
        ],
        "max": [
          511.0,
          511.0
        ]
      },
      "agent_pos": {
        "min": [
          13.45642375946045,
          32.93829345703125
        ],
        "max": [
          496.14617919921875,
          510.9578857421875
        ]
      }
    },
    "horizon": 16,
    "pad_before": 1,
    "pad_after": 7,
    "split_seed": 42,
    "max_train_episodes": 90,
    "split_windows": 10726
  },
  "parameters": 262709026,
  "versions": {
    "torch": "2.7.1",
    "torchvision": "0.22.1",
    "diffusers": "0.35.2",
    "gym-pusht": "0.1.7",
    "pymunk": "6.11.1",
    "numpy": "2.2.6",
    "wandb": "0.21.4",
    "huggingface-hub": "0.35.3"
  },
  "python": "3.10.19",
  "device": "cuda",
  "gpu": "Tesla V100-SXM3-32GB",
  "microbatch": 64,
  "precision": "float32",
  "smoke": false,
  "batches_per_epoch": 168,
  "resume_exact_worker_rng_replay": false,
  "wandb_url": "https://wandb.ai/yuyuan-wu-universiteit-gent/policy%20parallel%20test/runs/pusht-parallel-20261019-s0-dp-s0",
  "completed_updates": 512400,
  "samples_seen": 32714300,
  "train_seconds": 64772.88574781269
}