INFO 05-28 17:12:33 __init__.py:193] Automatically detected platform rocm.
INFO 05-28 17:12:33 __init__.py:193] Automatically detected platform rocm.
INFO 05-28 17:12:33 __init__.py:193] Automatically detected platform rocm.
INFO 05-28 17:12:33 __init__.py:193] Automatically detected platform rocm.
INFO 05-28 17:12:33 __init__.py:193] Automatically detected platform rocm.
INFO 05-28 17:12:33 __init__.py:193] Automatically detected platform rocm.
INFO 05-28 17:12:33 __init__.py:193] Automatically detected platform rocm.
INFO 05-28 17:12:34 __init__.py:193] Automatically detected platform rocm.
Could not load Sliding Tile Attention.
Could not load Sliding Tile Attention.
Could not load Sliding Tile Attention.
Could not load Sliding Tile Attention.
Could not load Sliding Tile Attention.
Could not load Sliding Tile Attention.
Could not load Sliding Tile Attention.
<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<
--> loading model from /home/model/HunyuanVideo/hunyuan-video-t2v-720p
<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<
<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<
<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<
<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<
Could not load Sliding Tile Attention.
<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<
<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<
<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<
>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
  Total training parameters = 12821.012544 M
--> Initializing FSDP with sharding strategy: full
>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
--> applying fdsp activation checkpointing...
--> applying fdsp activation checkpointing...
--> applying fdsp activation checkpointing...
--> applying fdsp activation checkpointing...
--> model loaded
--> applying fdsp activation checkpointing...
FullyShardedDataParallel(
  (_fsdp_wrapped_module): HYVideoDiffusionTransformer(
    (img_in): PatchEmbed(
      (proj): Conv3d(16, 3072, kernel_size=(1, 2, 2), stride=(1, 2, 2))
      (norm): Identity()
    )
    (txt_in): SingleTokenRefiner(
      (input_embedder): Linear(in_features=4096, out_features=3072, bias=True)
      (t_embedder): TimestepEmbedder(
        (mlp): Sequential(
          (0): Linear(in_features=256, out_features=3072, bias=True)
          (1): SiLU()
          (2): Linear(in_features=3072, out_features=3072, bias=True)
        )
      )
      (c_embedder): TextProjection(
        (linear_1): Linear(in_features=4096, out_features=3072, bias=True)
        (act_1): SiLU()
        (linear_2): Linear(in_features=3072, out_features=3072, bias=True)
      )
      (individual_token_refiner): IndividualTokenRefiner(
        (blocks): ModuleList(
          (0-1): 2 x IndividualTokenRefinerBlock(
            (norm1): LayerNorm((3072,), eps=1e-06, elementwise_affine=True)
            (self_attn_qkv): Linear(in_features=3072, out_features=9216, bias=True)
            (self_attn_q_norm): Identity()
            (self_attn_k_norm): Identity()
            (self_attn_proj): Linear(in_features=3072, out_features=3072, bias=True)
            (norm2): LayerNorm((3072,), eps=1e-06, elementwise_affine=True)
            (mlp): MLP(
              (fc1): Linear(in_features=3072, out_features=12288, bias=True)
              (act): SiLU()
              (drop1): Dropout(p=0.0, inplace=False)
              (norm): Identity()
              (fc2): Linear(in_features=12288, out_features=3072, bias=True)
              (drop2): Dropout(p=0.0, inplace=False)
            )
            (adaLN_modulation): Sequential(
              (0): SiLU()
              (1): Linear(in_features=3072, out_features=6144, bias=True)
            )
          )
        )
      )
    )
    (time_in): TimestepEmbedder(
      (mlp): Sequential(
        (0): Linear(in_features=256, out_features=3072, bias=True)
        (1): SiLU()
        (2): Linear(in_features=3072, out_features=3072, bias=True)
      )
    )
    (vector_in): MLPEmbedder(
      (in_layer): Linear(in_features=768, out_features=3072, bias=True)
      (silu): SiLU()
      (out_layer): Linear(in_features=3072, out_features=3072, bias=True)
    )
    (guidance_in): TimestepEmbedder(
      (mlp): Sequential(
        (0): Linear(in_features=256, out_features=3072, bias=True)
        (1): SiLU()
        (2): Linear(in_features=3072, out_features=3072, bias=True)
      )
    )
    (double_blocks): ModuleList(
      (0-19): 20 x FullyShardedDataParallel(
        (_fsdp_wrapped_module): CheckpointWrapper(
          (_checkpoint_wrapped_module): MMDoubleStreamBlock(
            (img_mod): ModulateDiT(
              (act): SiLU()
              (linear): Linear(in_features=3072, out_features=18432, bias=True)
            )
            (img_norm1): LayerNorm((3072,), eps=1e-06, elementwise_affine=False)
            (img_attn_qkv): Linear(in_features=3072, out_features=9216, bias=True)
            (img_attn_q_norm): RMSNorm()
            (img_attn_k_norm): RMSNorm()
            (img_attn_proj): Linear(in_features=3072, out_features=3072, bias=True)
            (img_norm2): LayerNorm((3072,), eps=1e-06, elementwise_affine=False)
            (img_mlp): MLP(
              (fc1): Linear(in_features=3072, out_features=12288, bias=True)
              (act): GELU(approximate='tanh')
              (drop1): Dropout(p=0.0, inplace=False)
              (norm): Identity()
              (fc2): Linear(in_features=12288, out_features=3072, bias=True)
              (drop2): Dropout(p=0.0, inplace=False)
            )
            (txt_mod): ModulateDiT(
              (act): SiLU()
              (linear): Linear(in_features=3072, out_features=18432, bias=True)
            )
            (txt_norm1): LayerNorm((3072,), eps=1e-06, elementwise_affine=False)
            (txt_attn_qkv): Linear(in_features=3072, out_features=9216, bias=True)
            (txt_attn_q_norm): RMSNorm()
            (txt_attn_k_norm): RMSNorm()
            (txt_attn_proj): Linear(in_features=3072, out_features=3072, bias=True)
            (txt_norm2): LayerNorm((3072,), eps=1e-06, elementwise_affine=False)
            (txt_mlp): MLP(
              (fc1): Linear(in_features=3072, out_features=12288, bias=True)
              (act): GELU(approximate='tanh')
              (drop1): Dropout(p=0.0, inplace=False)
              (norm): Identity()
              (fc2): Linear(in_features=12288, out_features=3072, bias=True)
              (drop2): Dropout(p=0.0, inplace=False)
            )
          )
        )
      )
    )
    (single_blocks): ModuleList(
      (0-39): 40 x FullyShardedDataParallel(
        (_fsdp_wrapped_module): CheckpointWrapper(
          (_checkpoint_wrapped_module): MMSingleStreamBlock(
            (linear1): Linear(in_features=3072, out_features=21504, bias=True)
            (linear2): Linear(in_features=15360, out_features=3072, bias=True)
            (q_norm): RMSNorm()
            (k_norm): RMSNorm()
            (pre_norm): LayerNorm((3072,), eps=1e-06, elementwise_affine=False)
            (mlp_act): GELU(approximate='tanh')
            (modulation): ModulateDiT(
              (act): SiLU()
              (linear): Linear(in_features=3072, out_features=9216, bias=True)
            )
          )
        )
      )
    )
    (final_layer): FinalLayer(
      (norm_final): LayerNorm((3072,), eps=1e-06, elementwise_affine=False)
      (linear): Linear(in_features=3072, out_features=64, bias=True)
      (adaLN_modulation): Sequential(
        (0): SiLU()
        (1): Linear(in_features=3072, out_features=6144, bias=True)
      )
    )
  )
)
optimizer: AdamW (
Parameter Group 0
    amsgrad: False
    betas: (0.9, 0.999)
    capturable: False
    differentiable: False
    eps: 1e-08
    foreach: None
    fused: None
    lr: 1e-05
    maximize: False
    weight_decay: 0.01
)
***** Running training *****
  Num examples = 101
  Dataloader size = 13
  Num Epochs = 1
  Resume training from step 0
  Instantaneous batch size per device = 1
  Total train batch size (w. data & sequence parallel, accumulation) = 2.0
  Gradient Accumulation steps = 1
  Total optimization steps = 12
  Total training parameters per FSDP shard = 1.602626568 B
  Master weight dtype: torch.float32
--> applying fdsp activation checkpointing...
--> applying fdsp activation checkpointing...
--> applying fdsp activation checkpointing...
zll step_time: 135.40s avg_step_time: 135.4012050628662
zll step_time: 122.44s avg_step_time: 128.91861820220947
zll step_time: 122.18s avg_step_time: 126.67362546920776
zll step_time: 122.14s avg_step_time: 125.5411741733551