checkpoints/predictor_v4_1000_seed0/train_config.json

{
  "activation_dtype": "bfloat16",
  "batch_size": 32,
  "beta1": 0.9,
  "beta2": 0.95,
  "block_warmup_steps": 100,
  "blocks_lr": 1e-05,
  "data_root": "/mnt/local_nvme/zoubin/cz/self_forcing_predictor_v4_1000_seed0",
  "dataset_pairs": 18000,
  "find_unused_parameters": false,
  "flow_loss_weight": 1.0,
  "fp32_trainable_params": true,
  "fusion_lr": 0.0001,
  "global_batch_size": 256,
  "grad_clip": 1.0,
  "gradient_accumulation_steps": 1,
  "gradient_checkpointing": false,
  "hidden_loss_weight": 0.1,
  "keep_snapshots": 20,
  "latent_height": 60,
  "latent_width": 104,
  "local_attn_size": -1,
  "log_every": 10,
  "manifest": "/mnt/local_nvme/zoubin/cz/self_forcing_predictor_v4_1000_seed0/train_manifest.jsonl",
  "manifest_records": 6000,
  "max_records": null,
  "max_steps": 2000,
  "num_workers": 0,
  "output_dir": "/mnt/local_nvme/zoubin/cz/self_forcing_predictor_v4_training_1000_seed0",
  "parameter_groups": {
    "feature_fusion.anchor_norm.bias": "fusion_residual",
    "feature_fusion.anchor_norm.weight": "fusion_residual",
    "feature_fusion.current_norm.bias": "fusion_residual",
    "feature_fusion.current_norm.weight": "fusion_residual",
    "feature_fusion.mlp.0.bias": "fusion_residual",
    "feature_fusion.mlp.0.weight": "fusion_residual",
    "feature_fusion.mlp.2.bias": "fusion_residual",
    "feature_fusion.mlp.2.weight": "fusion_residual",
    "feature_fusion.previous_norm.bias": "fusion_residual",
    "feature_fusion.previous_norm.weight": "fusion_residual",
    "predictor_blocks.0.cross_attn.norm_q.weight": "blocks",
    "predictor_blocks.0.cross_attn.o.bias": "blocks",
    "predictor_blocks.0.cross_attn.o.weight": "blocks",
    "predictor_blocks.0.cross_attn.q.bias": "blocks",
    "predictor_blocks.0.cross_attn.q.weight": "blocks",
    "predictor_blocks.0.ffn.0.bias": "blocks",
    "predictor_blocks.0.ffn.0.weight": "blocks",
    "predictor_blocks.0.ffn.2.bias": "blocks",
    "predictor_blocks.0.ffn.2.weight": "blocks",
    "predictor_blocks.0.modulation": "blocks",
    "predictor_blocks.0.norm3.bias": "blocks",
    "predictor_blocks.0.norm3.weight": "blocks",
    "predictor_blocks.0.self_attn.k.bias": "blocks",
    "predictor_blocks.0.self_attn.k.weight": "blocks",
    "predictor_blocks.0.self_attn.norm_k.weight": "blocks",
    "predictor_blocks.0.self_attn.norm_q.weight": "blocks",
    "predictor_blocks.0.self_attn.o.bias": "blocks",
    "predictor_blocks.0.self_attn.o.weight": "blocks",
    "predictor_blocks.0.self_attn.q.bias": "blocks",
    "predictor_blocks.0.self_attn.q.weight": "blocks",
    "predictor_blocks.0.self_attn.v.bias": "blocks",
    "predictor_blocks.0.self_attn.v.weight": "blocks",
    "predictor_blocks.1.cross_attn.norm_q.weight": "blocks",
    "predictor_blocks.1.cross_attn.o.bias": "blocks",
    "predictor_blocks.1.cross_attn.o.weight": "blocks",
    "predictor_blocks.1.cross_attn.q.bias": "blocks",
    "predictor_blocks.1.cross_attn.q.weight": "blocks",
    "predictor_blocks.1.ffn.0.bias": "blocks",
    "predictor_blocks.1.ffn.0.weight": "blocks",
    "predictor_blocks.1.ffn.2.bias": "blocks",
    "predictor_blocks.1.ffn.2.weight": "blocks",
    "predictor_blocks.1.modulation": "blocks",
    "predictor_blocks.1.norm3.bias": "blocks",
    "predictor_blocks.1.norm3.weight": "blocks",
    "predictor_blocks.1.self_attn.k.bias": "blocks",
    "predictor_blocks.1.self_attn.k.weight": "blocks",
    "predictor_blocks.1.self_attn.norm_k.weight": "blocks",
    "predictor_blocks.1.self_attn.norm_q.weight": "blocks",
    "predictor_blocks.1.self_attn.o.bias": "blocks",
    "predictor_blocks.1.self_attn.o.weight": "blocks",
    "predictor_blocks.1.self_attn.q.bias": "blocks",
    "predictor_blocks.1.self_attn.q.weight": "blocks",
    "predictor_blocks.1.self_attn.v.bias": "blocks",
    "predictor_blocks.1.self_attn.v.weight": "blocks",
    "residual_out.bias": "fusion_residual",
    "residual_out.weight": "fusion_residual"
  },
  "persistent_workers": false,
  "pin_memory": false,
  "prompt_count": 1000,
  "resume": null,
  "save_every": 100,
  "scheduler_warmup_steps": 100,
  "schema_version": "self_forcing_predictor_v4_bf16_v1",
  "sink_size": 0,
  "source_block_ids": [
    1,
    28
  ],
  "steps_per_epoch_per_rank": 66,
  "strict_teacher_load": true,
  "supervision_pairs": [
    [
      0,
      1
    ],
    [
      1,
      2
    ],
    [
      2,
      3
    ]
  ],
  "swanlab_description": "Self-Forcing Wan2.1 1.3B Predictor-v4 offline prefeature training",
  "swanlab_experiment": "predictor-v4-prefeature-blocks1-28-prompts1000-seed0-bs32x8-from-scratch",
  "swanlab_mode": "cloud",
  "swanlab_project": "Self-Forcing-Predictor-v4",
  "swanlab_run_id": "4kxa9v7b",
  "swanlab_tags": [
    "predictor-v4",
    "self-forcing",
    "wan2.1-1.3b",
    "offline-prefeature",
    "blocks-1-28",
    "8gpu",
    "seed0"
  ],
  "swanlab_workspace": null,
  "teacher_checkpoint": "/mnt/s3files/s3-us-west2-default/zoubin/cz/projects/Self-Forcing/checkpoints/self_forcing_dmd.pt",
  "teacher_checkpoint_key": "generator_ema",
  "teacher_model_name": "Wan2.1-T2V-1.3B",
  "timestep_shift": 5.0,
  "trainable_parameter_dtype": "float32",
  "trainable_parameters": 104684032,
  "use_swanlab": true,
  "use_wandb": false,
  "wandb_entity": null,
  "wandb_name": "predictor-v4-prefeature-blocks1-28-prompts1000-seed0",
  "wandb_project": "Self-Forcing-Predictor-v4",
  "weight_decay": 0.01,
  "world_size": 8
}