训练的时候报错 the size of tensor a (4) must match the size of tensor b (5) at non-singleton dimension 2

num_frames % 4 != 1. We round it up to 17.
num_frames % 4 != 1. We round it up to 17.
num_frames % 4 != 1. We round it up to 17.
num_frames % 4 != 1. We round it up to 17.
num_frames % 4 != 1. We round it up to 17.
num_frames % 4 != 1. We round it up to 17.
0%| | 0/1 [00:11<?, ?it/s]
[rank3]: Traceback (most recent call last):
[rank3]: File "/workspace/Wan2.1-main/Finetune/DiffSynth-Studio-main/train.py", line 136, in
[rank3]: launch_training_task(dataset, model, model_logger, args=args)
[rank3]: File "/workspace/Wan2.1-main/Finetune/DiffSynth-Studio-main/diffsynth/trainers/utils.py", line 559, in launch_training_task
[rank3]: loss = model(data)
[rank3]: ^^^^^^^^^^^
[rank3]: File "/opt/conda/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1553, in _wrapped_call_impl
[rank3]: return self._call_impl(*args, **kwargs)
[rank3]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
[rank3]: File "/opt/conda/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1562, in _call_impl
[rank3]: return forward_call(*args, **kwargs)
[rank3]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
[rank3]: File "/opt/conda/lib/python3.11/site-packages/deepspeed/utils/nvtx.py", line 20, in wrapped_fn
[rank3]: ret_val = func(*args, **kwargs)
[rank3]: ^^^^^^^^^^^^^^^^^^^^^
[rank3]: File "/opt/conda/lib/python3.11/site-packages/deepspeed/runtime/engine.py", line 2179, in forward
[rank3]: loss = self.module(*inputs, **kwargs)
[rank3]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
[rank3]: File "/opt/conda/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1553, in _wrapped_call_impl
[rank3]: return self._call_impl(*args, **kwargs)
[rank3]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
[rank3]: File "/opt/conda/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1603, in _call_impl
[rank3]: result = forward_call(*args, **kwargs)
[rank3]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
[rank3]: File "/workspace/Wan2.1-main/Finetune/DiffSynth-Studio-main/train.py", line 90, in forward
[rank3]: loss = self.pipe.training_loss(**models, **inputs)
[rank3]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
[rank3]: File "/workspace/Wan2.1-main/Finetune/DiffSynth-Studio-main/diffsynth/pipelines/wan_video_new.py", line 119, in training_loss
[rank3]: inputs["latents"] = self.scheduler.add_noise(inputs["input_latents"], inputs["noise"], timestep)
[rank3]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
[rank3]: File "/workspace/Wan2.1-main/Finetune/DiffSynth-Studio-main/diffsynth/schedulers/flow_match.py", line 99, in add_noise
[rank3]: sample = (1 - sigma) * original_samples + sigma * noise
[rank3]: ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~
[rank3]: RuntimeError: The size of tensor a (4) must match the size of tensor b (5) at non-singleton dimension 2

Start in train.py and follow the training_loss call into diffsynth/pipelines/wan_video_new.py, then inspect diffsynth/schedulers/flow_match.py where input_latents and noise are combined. Reproduce the training run and compare their shapes at add_noise; done means the tensors can be combined without the reported dimension mismatch and training proceeds.