Make tensors in ResNet contiguous for Hunyuan VAE

[a-r-r-o-w]

The following error is raised on Hopper GPUs if the tensors are non-contiguous:

traceback

ERROR:finetrainers:An error occurred during training: Unsupported memory format for group normalization: ChannelsLast3d
ERROR:finetrainers:Traceback (most recent call last):
  File "/fsx/aryan/finetrainers/train.py", line 34, in main
    trainer.train()
  File "/fsx/aryan/finetrainers/finetrainers/trainer.py", line 375, in train
    latent_conditions = self.model_config["prepare_latents"](
  File "/fsx/aryan/finetrainers/finetrainers/hunyuan_video/hunyuan_video_lora.py", line 168, in prepare_latents
    latents = vae.encode(image_or_video).latent_dist.sample(generator=generator)
  File "/fsx/aryan/diffusers/src/diffusers/utils/accelerate_utils.py", line 46, in wrapper
    return method(self, *args, **kwargs)
  File "/fsx/aryan/diffusers/src/diffusers/models/autoencoders/autoencoder_kl_hunyuan_video.py", line 900, in encode
    h = self._encode(x)
  File "/fsx/aryan/diffusers/src/diffusers/models/autoencoders/autoencoder_kl_hunyuan_video.py", line 871, in _encode
    return self._temporal_tiled_encode(x)
  File "/fsx/aryan/diffusers/src/diffusers/models/autoencoders/autoencoder_kl_hunyuan_video.py", line 1098, in _temporal_tiled_encode
    tile = self.tiled_encode(tile)
  File "/fsx/aryan/diffusers/src/diffusers/models/autoencoders/autoencoder_kl_hunyuan_video.py", line 1007, in tiled_encode
    tile = self.encoder(tile)
  File "/fsx/aryan/nightly-venv/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1740, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
  File "/fsx/aryan/nightly-venv/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1751, in _call_impl
    return forward_call(*args, **kwargs)
  File "/fsx/aryan/diffusers/src/diffusers/models/autoencoders/autoencoder_kl_hunyuan_video.py", line 571, in forward
    hidden_states = self.mid_block(hidden_states)
  File "/fsx/aryan/nightly-venv/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1740, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
  File "/fsx/aryan/nightly-venv/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1751, in _call_impl
    return forward_call(*args, **kwargs)
  File "/fsx/aryan/diffusers/src/diffusers/models/autoencoders/autoencoder_kl_hunyuan_video.py", line 297, in forward
    hidden_states = resnet(hidden_states)
  File "/fsx/aryan/nightly-venv/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1740, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
  File "/fsx/aryan/nightly-venv/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1751, in _call_impl
    return forward_call(*args, **kwargs)
  File "/fsx/aryan/diffusers/src/diffusers/models/autoencoders/autoencoder_kl_hunyuan_video.py", line 173, in forward
    hidden_states = self.norm1(hidden_states)
  File "/fsx/aryan/nightly-venv/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1740, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
  File "/fsx/aryan/nightly-venv/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1751, in _call_impl
    return forward_call(*args, **kwargs)
  File "/fsx/aryan/nightly-venv/lib/python3.10/site-packages/torch/nn/modules/normalization.py", line 313, in forward
    return F.group_norm(input, self.num_groups, self.weight, self.bias, self.eps)
  File "/fsx/aryan/nightly-venv/lib/python3.10/site-packages/torch/nn/functional.py", line 2965, in group_norm
    return torch.group_norm(
RuntimeError: Unsupported memory format for group normalization: ChannelsLast3d

This does not seem to happen on some of the other GPUs I tested on (A100, L40, 4090) so I think it is hopper-specific

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.