infer_flashvsr_v1.1_tiny_long_video某个视频推理到一半报错

用infer_flashvsr_v1.1_tiny_long_video.py推理长视频，测试了十几个视频，结果都正常，
但有一个视频，固定推理到某步报错，应该怎么解决？

56%|██████████████████████████████████████████████████████████████▏ | 510/918 [07:45<06:12, 1.09it/s]
Traceback (most recent call last):
File "/home/liyuanqi/project/FlashVSR/infer_flashvsr_v1.1_tiny_long_video.py", line 287, in
main()
File "/home/liyuanqi/project/FlashVSR/infer_flashvsr_v1.1_tiny_long_video.py", line 257, in main
video = pipe(
^^^^^
File "/home/liyuanqi/miniconda3/envs/flashvsr/lib/python3.11/site-packages/torch/utils/_contextlib.py", line 116, in decorate_context
return func(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^
File "/home/liyuanqi/project/FlashVSR/diffsynth/pipelines/flashvsr_tiny_long.py", line 394, in call
noise_pred_posi, pre_cache_k, pre_cache_v = model_fn_wan_video(
^^^^^^^^^^^^^^^^^^^
File "/home/liyuanqi/project/FlashVSR/diffsynth/pipelines/flashvsr_tiny_long.py", line 564, in model_fn_wan_video
x, last_pre_cache_k, last_pre_cache_v = block(
^^^^^^
File "/home/liyuanqi/miniconda3/envs/flashvsr/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1739, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/liyuanqi/miniconda3/envs/flashvsr/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1750, in _call_impl
return forward_call(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/liyuanqi/project/FlashVSR/diffsynth/models/wan_video_dit.py", line 464, in forward
self_attn_output, self_attn_cache_k, self_attn_cache_v = self.self_attn(
^^^^^^^^^^^^^^^
File "/home/liyuanqi/miniconda3/envs/flashvsr/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1739, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/liyuanqi/miniconda3/envs/flashvsr/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1750, in _call_impl
return forward_call(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/liyuanqi/project/FlashVSR/diffsynth/models/wan_video_dit.py", line 330, in forward
q = rope_apply(q, freqs, self.num_heads)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/liyuanqi/project/FlashVSR/diffsynth/models/wan_video_dit.py", line 269, in rope_apply
x_out = torch.view_as_real(x_out * freqs).flatten(2)
~~~~~~^~~~~~~
RuntimeError: The size of tensor a (64) must match the size of tensor b (42) at non-singleton dimension 3

望解答，谢谢！

Reproduce the failure with the problematic video using infer_flashvsr_v1.1_tiny_long_video.py. Trace the tensor shapes from diffsynth/pipelines/flashvsr_tiny_long.py into rope_apply in diffsynth/models/wan_video_dit.py, where the 64-versus-42 mismatch occurs. Done means the video completes successfully without breaking the other tested videos.