CogVideoX-Fun-V1.5-5B-InP图生视频失败，Got an OOM, unloading all loaded models.

L40S的GPU，显存48G，内存64G，使用comfyUI运行CogVideoX-Fun-V1.5-5B-InP 图生视频，

报错
!!! Exception during processing !!! Allocation on device

Traceback (most recent call last):

File "/app/ComfyUI/execution.py", line 324, in execute

 output_data, output_ui, has_subgraph = get_output_data(obj, input_data_all, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb)

File "/app/ComfyUI/execution.py", line 199, in get_output_data

 return_values = _map_node_over_list(obj, input_data_all, obj.FUNCTION, allow_interrupt=True, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb)

File "/app/ComfyUI/execution.py", line 170, in _map_node_over_list

 process_inputs(input_dict, i)

File "/app/ComfyUI/execution.py", line 159, in process_inputs

 results.append(getattr(obj, func)(**inputs))

File "/app/ComfyUI/custom_nodes/CogVideoX-Fun/comfyui/comfyui_nodes.py", line 334, in process

 sample = pipeline(

File "/usr/local/lib/python3.10/dist-packages/torch/utils/_contextlib.py", line 116, in decorate_context

 return func(*args, **kwargs)

File "/app/ComfyUI/custom_nodes/CogVideoX-Fun/cogvideox/pipeline/pipeline_cogvideox_inpaint.py", line 1139, in call

 video = self.decode_latents(latents)

File "/app/ComfyUI/custom_nodes/CogVideoX-Fun/cogvideox/pipeline/pipeline_cogvideox_inpaint.py", line 595, in decode_latents

 frames = self.vae.decode(latents).sample

File "/usr/local/lib/python3.10/dist-packages/diffusers/utils/accelerate_utils.py", line 46, in wrapper

 return method(self, *args, **kwargs)

File "/app/ComfyUI/custom_nodes/CogVideoX-Fun/cogvideox/models/autoencoder_magvit.py", line 1410, in decode

 decoded = self._decode(z).sample

File "/app/ComfyUI/custom_nodes/CogVideoX-Fun/cogvideox/models/autoencoder_magvit.py", line 1379, in _decode

 z_intermediate, conv_cache = self.decoder(z_intermediate, conv_cache=conv_cache)

File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1736, in _wrapped_call_impl

 return self._call_impl(*args, **kwargs)

File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1747, in _call_impl

 return forward_call(*args, **kwargs)

File "/app/ComfyUI/custom_nodes/CogVideoX-Fun/cogvideox/models/autoencoder_magvit.py", line 1053, in forward

 hidden_states, new_conv_cache[conv_cache_key] = up_block(

File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1736, in _wrapped_call_impl

 return self._call_impl(*args, **kwargs)

File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1747, in _call_impl

 return forward_call(*args, **kwargs)

File "/app/ComfyUI/custom_nodes/CogVideoX-Fun/cogvideox/models/autoencoder_magvit.py", line 730, in forward

 hidden_states, new_conv_cache[conv_cache_key] = resnet(

File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1736, in _wrapped_call_impl

 return self._call_impl(*args, **kwargs)

File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1747, in _call_impl

 return forward_call(*args, **kwargs)

File "/app/ComfyUI/custom_nodes/CogVideoX-Fun/cogvideox/models/autoencoder_magvit.py", line 374, in forward

 hidden_states, new_conv_cache["norm1"] = self.norm1(hidden_states, zq, conv_cache=conv_cache.get("norm1"))

File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1736, in _wrapped_call_impl

 return self._call_impl(*args, **kwargs)

File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1747, in _call_impl

 return forward_call(*args, **kwargs)

File "/app/ComfyUI/custom_nodes/CogVideoX-Fun/cogvideox/models/autoencoder_magvit.py", line 195, in forward

 new_f = norm_f * conv_y + conv_b

torch.OutOfMemoryError: Allocation on device

Got an OOM, unloading all loaded models.

Prompt executed in 196.79 seconds

Start by reproducing the CogVideoX-Fun-V1.5-5B-InP image-to-video run on the stated L40S hardware, then trace the failure from comfyui/comfyui_nodes.py:334 through pipeline_cogvideox_inpaint.py:595 and autoencoder_magvit.py:1379-1410. Compare memory use around VAE decoding and determine what change allows the workflow to complete without the reported allocation error.