{
  "model": "minimax_h3",
  "dataset": {
    "repo_id": "ZeyuLing/hftrainer_crema_d_mini",
    "revision": "1188c5d5fac3e69ea014831ab4409f729f8eda81",
    "local_dir": "data/hftrainer_crema_d_mini",
    "files_verified": 40,
    "splits": {
      "train": 24,
      "validation": 8
    },
    "manifest_sha256": "ebf322bcf50a159b8c4f17cf01d89790206e51db4c0b571ab11c5f3af8696a94"
  },
  "base_checkpoint": {
    "repo_id": "ZeyuLing/HFTrainer-MiniMax-H3-Base-FL2VA",
    "revision": "a3b24b9d13fac9c38949ef737e9cb4b98ffe2fb1",
    "bundle": "MiniMaxH3Bundle",
    "local_dir": "checkpoints/hftrainer/minimax_h3",
    "access": "Manual access approval"
  },
  "steps": 20,
  "saved_checkpoints": [
    "work_dirs/public_data/minimax_h3/checkpoint-iter_20/model.pt"
  ],
  "seed": 42,
  "hardware": "1 \u00d7 NVIDIA H200 (143771 MiB reported)",
  "torch": "2.8.0+cu128",
  "config": "configs/public_data/minimax_h3.py",
  "protocol": "Short real-data execution validation; raw unsmoothed training loss. Not convergence or a quality benchmark.",
  "series": [
    {
      "step": 1,
      "loss": 0.7431,
      "loss_video": 0.384,
      "loss_audio": 0.3591
    },
    {
      "step": 2,
      "loss": 0.8247,
      "loss_video": 0.4912,
      "loss_audio": 0.3335
    },
    {
      "step": 3,
      "loss": 0.5613,
      "loss_video": 0.2388,
      "loss_audio": 0.3225
    },
    {
      "step": 4,
      "loss": 0.8593,
      "loss_video": 0.4852,
      "loss_audio": 0.3742
    },
    {
      "step": 5,
      "loss": 0.6254,
      "loss_video": 0.2784,
      "loss_audio": 0.347
    },
    {
      "step": 6,
      "loss": 0.9048,
      "loss_video": 0.6041,
      "loss_audio": 0.3007
    },
    {
      "step": 7,
      "loss": 0.9372,
      "loss_video": 0.5992,
      "loss_audio": 0.338
    },
    {
      "step": 8,
      "loss": 0.6392,
      "loss_video": 0.2526,
      "loss_audio": 0.3866
    },
    {
      "step": 9,
      "loss": 0.6392,
      "loss_video": 0.2645,
      "loss_audio": 0.3747
    },
    {
      "step": 10,
      "loss": 0.605,
      "loss_video": 0.2855,
      "loss_audio": 0.3195
    },
    {
      "step": 11,
      "loss": 0.941,
      "loss_video": 0.5748,
      "loss_audio": 0.3662
    },
    {
      "step": 12,
      "loss": 0.6428,
      "loss_video": 0.4285,
      "loss_audio": 0.2143
    },
    {
      "step": 13,
      "loss": 0.9108,
      "loss_video": 0.5464,
      "loss_audio": 0.3644
    },
    {
      "step": 14,
      "loss": 0.675,
      "loss_video": 0.3034,
      "loss_audio": 0.3717
    },
    {
      "step": 15,
      "loss": 0.6001,
      "loss_video": 0.3243,
      "loss_audio": 0.2758
    },
    {
      "step": 16,
      "loss": 0.7366,
      "loss_video": 0.3572,
      "loss_audio": 0.3795
    },
    {
      "step": 17,
      "loss": 0.6837,
      "loss_video": 0.3339,
      "loss_audio": 0.3498
    },
    {
      "step": 18,
      "loss": 0.6798,
      "loss_video": 0.3035,
      "loss_audio": 0.3763
    },
    {
      "step": 19,
      "loss": 0.7431,
      "loss_video": 0.4033,
      "loss_audio": 0.3398
    },
    {
      "step": 20,
      "loss": 0.6399,
      "loss_video": 0.3399,
      "loss_audio": 0.3
    }
  ],
  "held_out_metrics": {
    "val/loss": 0.8745,
    "val/loss_video": 0.4547,
    "val/loss_audio": 0.4198,
    "samples": 8,
    "split": "validation",
    "noise_protocol": "New noise and timestep sampled by trainer; not fixed-noise convergence evaluation"
  }
}