fsdp with wan2.2

[rank1]: File "/home/lcq/lijiarui/DiffSynth-Studio/examples/wanvideo/model_training/train.py", line 165, in
[rank1]: launch_training_task(
[rank1]: File "/home/lcq/lijiarui/DiffSynth-Studio/diffsynth/trainers/utils.py", line 491, in launch_training_task
[rank1]: loss = model(data=None, inputs=data)
[rank1]: File "/root/miniconda3/envs/py310/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1736, in _wrapped_call_impl
[rank1]: return self._call_impl(*args, **kwargs)
[rank1]: File "/root/miniconda3/envs/py310/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1747, in _call_impl
[rank1]: return forward_call(*args, **kwargs)
[rank1]: File "/home/lcq/lijiarui/DiffSynth-Studio/examples/wanvideo/model_training/train.py", line 111, in forward
[rank1]: loss = self.pipe.training_loss(**models, **inputs)
[rank1]: File "/home/lcq/lijiarui/DiffSynth-Studio/diffsynth/pipelines/wan_video_new.py", line 86, in training_loss
[rank1]: noise_pred = self.model_fn(**inputs, timestep=timestep)
[rank1]: File "/home/lcq/lijiarui/DiffSynth-Studio/diffsynth/pipelines/wan_video_new.py", line 1230, in model_fn_wan_video
[rank1]: x = torch.utils.checkpoint.checkpoint(
[rank1]: File "/root/miniconda3/envs/py310/lib/python3.10/site-packages/torch/_compile.py", line 32, in inner
[rank1]: return disable_fn(*args, **kwargs)
[rank1]: File "/root/miniconda3/envs/py310/lib/python3.10/site-packages/torch/_dynamo/eval_frame.py", line 632, in _fn
[rank1]: return fn(*args, **kwargs)
[rank1]: File "/root/miniconda3/envs/py310/lib/python3.10/site-packages/torch/utils/checkpoint.py", line 496, in checkpoint
[rank1]: ret = function(*args, **kwargs)
[rank1]: File "/home/lcq/lijiarui/DiffSynth-Studio/diffsynth/pipelines/wan_video_new.py", line 1218, in custom_forward
[rank1]: return module(*inputs)
[rank1]: File "/root/miniconda3/envs/py310/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1736, in _wrapped_call_impl
[rank1]: return self._call_impl(*args, **kwargs)
[rank1]: File "/root/miniconda3/envs/py310/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1747, in _call_impl
[rank1]: return forward_call(*args, **kwargs)
[rank1]: File "/home/lcq/lijiarui/DiffSynth-Studio/diffsynth/models/wan_video_dit.py", line 220, in forward
[rank1]: self.modulation.to(dtype=t_mod.dtype, device=t_mod.device) + t_mod).chunk(6, dim=chunk_dim)
[rank1]: RuntimeError: Output 0 of ViewBackward0 is a view and its base or another view of its base has been modified inplace. This view is the output of a function that returns multiple views. Such functions do not allow the output views to be modified inplace. You should replace the inplace operation by an out-of-place one.

my FSDP config is:
compute_environment: LOCAL_MACHINE
debug: false
fsdp_config:
fsdp_activation_checkpointing: false
fsdp_auto_wrap_policy: SIZE_BASED_WRAP
fsdp_backward_prefetch: BACKWARD_PRE
fsdp_cpu_ram_efficient_loading: false
fsdp_forward_prefetch: false
fsdp_min_num_params: 100000
fsdp_offload_params: false
fsdp_sharding_strategy: FULL_SHARD
fsdp_state_dict_type: SHARDED_STATE_DICT
fsdp_sync_module_states: false
fsdp_use_orig_params: false
distributed_type: FSDP
downcast_bf16: 'no'
enable_cpu_affinity: false
machine_rank: 0
main_training_function: main
mixed_precision: bf16
num_machines: 1
num_processes: 4
rdzv_backend: static
same_network: true
tpu_env: []
tpu_use_cluster: false
tpu_use_sudo: false
use_cpu: false

Reproduce the failure using the supplied FSDP configuration and the Wan training entry point in examples/wanvideo/model_training/train.py. Inspect diffsynth/pipelines/wan_video_new.py around model_fn_wan_video and diffsynth/models/wan_video_dit.py around line 220, where the traceback ends. Done means the training run no longer raises the inplace view error under FSDP.