{
  "base_model_dir": "/mnt/nfs/vlm-aryan/fasth3-14b-2step-qad-20260829/experiments/s42-r16-dmd8-vsa80-20260929-v1/initial-student",
  "config": {
    "callbacks": {
      "grad_clip": {
        "_target_": "fastvideo.train.callbacks.grad_clip.GradNormClipCallback",
        "max_grad_norm": 1.0
      },
      "validation": {
        "_target_": "fastvideo.train.callbacks.validation.ValidationCallback",
        "dataset_file": "/mnt/nfs/vlm-aryan/fasth3-14b-2step-qad-20260829/experiments/s42-r16-dmd8-vsa80-20260929-v1/validation/val8.json",
        "every_steps": 100,
        "guidance_scale": 1.0,
        "max_record_num_frames": 345,
        "num_videos_per_prompt": 1,
        "offload_training_state": true,
        "pipeline_target": "fastvideo.pipelines.basic.minimax_h3.minimax_h3_pipeline.MiniMaxH3Pipeline",
        "run_at_start": true,
        "sampling_steps": [
          8
        ],
        "sampling_timesteps": [
          999,
          874,
          749,
          624,
          500,
          375,
          250,
          125
        ],
        "text_encoder_cpu_offload": true,
        "use_record_dimensions": true,
        "use_validation_media_conditioning": false,
        "vae_cpu_offload": true
      }
    },
    "method": {
      "_target_": "fastvideo.train.methods.distribution_matching.dmd2.DMD2Method",
      "audio_anchor": {
        "weight": 0.1
      },
      "cfg_uncond": {
        "text": "zero"
      },
      "dmd_denoising_steps": [
        999,
        874,
        749,
        624,
        500,
        375,
        250,
        125
      ],
      "fake_score_betas": [
        0.9,
        0.999
      ],
      "fake_score_learning_rate": "2e-06",
      "fake_score_loss_space": "x0",
      "fake_score_lr_scheduler": "constant",
      "generator_update_interval": 5,
      "max_timestep_ratio": 0.999,
      "min_timestep_ratio": 0.001,
      "modality_dmd_safeguards": {
        "audio": {
          "denom_floor_ratio": 0.05,
          "grad_cap": 100.0
        }
      },
      "real_score_guidance_scale": 1.0,
      "rollout_carry": true,
      "rollout_carry_slots": 8,
      "rollout_mode": "simulate",
      "rollout_sample_type": "ode",
      "score_timestep_continuous": true,
      "score_timestep_shift": 2.4,
      "score_timestep_warp_max": 0.999
    },
    "models": {
      "critic": {
        "_target_": "fastvideo.train.models.minimax_h3.MiniMaxH3DMDModel",
        "attention_backend": "FLASH_ATTN",
        "disable_custom_init_weights": true,
        "enable_gradient_checkpointing_type": "full",
        "init_from": "/mnt/nfs/vlm-aryan/fasth3-14b-2step-qad-20260829/release-candidates/base-h3-teacher-complete-v1",
        "trainable": true
      },
      "student": {
        "_target_": "fastvideo.train.models.minimax_h3.MiniMaxH3DMDModel",
        "attention_backend": "VIDEO_SPARSE_ATTN_H3",
        "enable_gradient_checkpointing_type": "full",
        "init_from": "/mnt/nfs/vlm-aryan/fasth3-14b-2step-qad-20260829/experiments/s42-r16-dmd8-vsa80-20260929-v1/initial-student",
        "trainable": true
      },
      "teacher": {
        "_target_": "fastvideo.train.models.minimax_h3.MiniMaxH3DMDModel",
        "attention_backend": "FLASH_ATTN",
        "disable_custom_init_weights": true,
        "init_from": "/mnt/nfs/vlm-aryan/fasth3-14b-2step-qad-20260829/release-candidates/base-h3-teacher-complete-v1",
        "trainable": false
      }
    },
    "pipeline": {
      "audio_scheduler_shift": 3.0,
      "dit_config": {
        "uniform_parameter_dtype": false
      },
      "video_scheduler_shift": 10.0
    },
    "training": {
      "checkpoint": {
        "checkpointing_start_step": 100,
        "checkpoints_total_limit": 0,
        "inference_checkpoint_dtype": "bfloat16",
        "inference_checkpoint_role": "student",
        "output_dir": "/mnt/nfs/vlm-aryan/fasth3-14b-2step-qad-20260829/experiments/s42-r16-dmd8-vsa80-20260929-v1/runs/vsa80-32gpu-anchor-v2",
        "require_complete_training_checkpoint": true,
        "resume_from_checkpoint": "latest",
        "save_inference_checkpoint_on_validation": true,
        "training_state_checkpointing_steps": 100
      },
      "data": {
        "data_path": [
          "/mnt/lustre/vlm-shared/h3_t2av_preprocessed/v10_mixed_native_v3/h3_t2av_video_nuva_50k_720_mixed_len/data",
          "/mnt/lustre/vlm-shared/h3_t2av_preprocessed/v10_mixed_native_v3/h3_t2av_video_5s_768p/data",
          "/mnt/lustre/vlm-shared/h3_t2av_preprocessed/v10_mixed_native_v3/h3_t2av_video_nuva_10k_720_mixed_len/data",
          "/mnt/lustre/vlm-shared/h3_t2av_preprocessed/v10_mixed_native_v3/h3_t2av_video_nuva_10k_mixed_res_len/data",
          "/mnt/lustre/vlm-shared/h3_t2av_preprocessed/v10_mixed_native_v3/h3_t2av_fastgen_vidprom_150k/data"
        ],
        "dataloader_num_workers": 0,
        "native_shape_bucketing": true,
        "num_frames": 124,
        "num_height": 768,
        "num_latent_t": 37,
        "num_width": 1344,
        "preprocessed_data_type": "t2va",
        "seed": 42,
        "train_batch_size": 1,
        "training_cfg_rate": 0.0
      },
      "distributed": {
        "hsdp_replicate_dim": 1,
        "hsdp_shard_dim": 32,
        "num_gpus": 32,
        "sp_size": 4,
        "tp_size": 1
      },
      "dit_precision": "fp32",
      "loop": {
        "gradient_accumulation_steps": 8,
        "max_train_steps": 1500
      },
      "model": {
        "enable_gradient_checkpointing_type": "full",
        "enable_torch_compile": true,
        "precondition_outputs": false,
        "torch_compile_kwargs": {
          "dynamic": true,
          "recompile_limit": 32
        }
      },
      "optimizer": {
        "betas": [
          0.9,
          0.999
        ],
        "learning_rate": "2e-06",
        "lr_scheduler": "constant",
        "lr_warmup_steps": 0,
        "weight_decay": 0.01
      },
      "tracker": {
        "entity": "aryan5v-san-jose-state-university",
        "project_name": "compacth3-s42-r16-dmd8-20260930",
        "run_name": "new42-r16-vsa80-shift10-32gpu-audio-anchor-w0.1-v2",
        "trackers": [
          "wandb"
        ]
      },
      "vsa": {
        "sparsity": 0.8,
        "tile_size": 64
      }
    }
  },
  "dtype": "bfloat16",
  "format_version": 1,
  "kind": "inference",
  "max_shard_size_bytes": 5368709120,
  "module": "transformer",
  "role": "student",
  "shard_count": 8,
  "shard_file_sizes": [
    5364257712,
    5203335128,
    5126264664,
    5126264680,
    5126264664,
    5126264672,
    5126264624,
    1316821384
  ],
  "shard_sizes": [
    5364246528,
    5203325952,
    5126255616,
    5126255616,
    5126255616,
    5126255616,
    5126255616,
    1316819456
  ],
  "step": 300,
  "tensor_count": 585,
  "total_size": 37515670016
}