wan 2.2 lora training bug

[rank0]: File "/opt/nas/p/DiffSynth-Studio/diffsynth/models/wan_video_vae.py", line 472, in forward [rank0]: x = module(x, feat_cache, feat_idx) [rank0]: File "/opt/nas/p/conda/envs/diffsynth/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1751, in _wrapped_call_impl [rank0]: return self._call_impl(*args, **kwargs) [rank0]: File "/opt/nas/p/conda/envs/diffsynth/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1857, in _call_impl [rank0]: return inner() [rank0]: File "/opt/nas/p/conda/envs/diffsynth/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1805, in inner [rank0]: result = forward_call(*args, **kwargs) [rank0]: File "/opt/nas/p/DiffSynth-Studio/diffsynth/models/wan_video_vae.py", line 296, in forward [rank0]: x = layer(x, feat_cache[idx]) [rank0]: File "/opt/nas/p/conda/envs/diffsynth/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1751, in _wrapped_call_impl [rank0]: return self._call_impl(*args, **kwargs) [rank0]: File "/opt/nas/p/conda/envs/diffsynth/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1857, in _call_impl [rank0]: return inner() [rank0]: File "/opt/nas/p/conda/envs/diffsynth/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1805, in inner [rank0]: result = forward_call(*args, **kwargs) [rank0]: File "/opt/nas/p/DiffSynth-Studio/diffsynth/models/wan_video_vae.py", line 48, in forward [rank0]: x = torch.cat([cache_x, x], dim=2) [rank0]: RuntimeError: Sizes of tensors must match except in dimension 2. Expected size 640 but got size 160 for tensor number 1 in the list. E0729 20:39:56.568000 4091808 site-packages/torch/distributed/elastic/multiprocessing/api.py:874] failed (exitcode: 1) local_rank: 0 (pid: 4093248) of binary: /opt/nas/p/conda/envs/diffsynth/bin/python3.10

Start with diffsynth/models/wan_video_vae.py at lines 48, 296, and 472, then inspect the Wan 2.2 LoRA training path that supplies the cached tensors. Reproduce the reported torch.cat shape mismatch and determine why dimensions 640 and 160 are combined; done means training completes without this RuntimeError and the relevant behavior is covered by a regression test.