3B/stage-1-image/args.json

{
  "batch_size": 20,
  "accum_iter": 1,
  "epochs": 100,
  "warmup_epochs": 0.0025,
  "lr": 0.0001,
  "min_lr": 0.0001,
  "wd": 0.1,
  "clip_grad": 4.0,
  "init_from": "",
  "data_config": "",
  "cache_ann_on_disk": false,
  "length_clustering": true,
  "num_workers": 32,
  "pin_mem": true,
  "seed": 0,
  "output_dir": "",
  "save_interval": 1,
  "save_iteration_interval": 1000,
  "only_save_trainable": false,
  "ckpt_max_keep": 20,
  "auto_resume": true,
  "resume_path": "",
  "model_parallel_size": 1,
  "data_parallel": "fsdp",
  "precision": "bf16",
  "grad_precision": "fp32",
  "checkpointing": true,
  "max_seq_len": 32768,
  "mask_image_logits": false,
  "dropout": 0.0,
  "z_loss_weight": 0.0,
  "model_size": "3B_MMRoPE",
  "pretrain_task": [
    "predict_video"
  ],
  "video_fps": [
    1
  ],
  "video_duration": [
    1
  ],
  "cfg_mode": [
    "text_only"
  ],
  "masking_mode": [
    "text_to_video"
  ],
  "text_to_video_prob": 0.5,
  "eval_data_config": "",
  "run_eval": true,
  "eval_in_epoch": 200,
  "eval_mode": [
    "text_to_video"
  ],
  "data_video_fps": 1,
  "data_video_frames": 1,
  "MaskedAR": true,
  "mask_schedule": "cosine",
  "min_masking_rate": 0.7,
  "mask_contiguous_region_prob": null,
  "noise_type": "mask",
  "predict_all_tokens": false,
  "frame_closs_recorder": true,
  "mask_type": "tube",
  "train_loss": "CEChunked",
  "decay_start_coef": 1.0,
  "compute_ntp_loss": false,
  "train_with_slim_lm_head": false,
  "train_with_vis_tok": true,
  "vis_tok_start": 65536,
  "visual_tokenizer": "Cosmos-Tokenizer-DV4x8x8",
  "vocab_size": 129536,
  "joint_img_video": false,
  "img_batch_size": 8,
  "img_data_config": null,
  "video_iter_per_img_iter": 4,
  "no_resume_metric_logger": true,
  "world_size": 48,
  "rank": 0,
  "gpu": 0,
  "local_rank": 0,
  "dist_url": "env://",
  "distributed": true,
  "dist_backend": "nccl"
}