HIP out of memory

### Custom Node Testing

- [ ] I have tried disabling custom nodes and the issue persists (see [how to disable custom nodes](https://docs.comfy.org/troubleshooting/custom-node-issues#step-1%3A-test-with-all-custom-nodes-disabled) if you need help)

### Expected Behavior

AMD AI MAX+395, ubuntu 24.04
When I try to run the wanAI 1.3B text to video wan workflow, it prompts that 47,98G VRAM is required. 1.3B model only requires 6G of VRAM to run, why does it need almost 50G of VRAM?

### Actual Behavior

torch.OutOfMemoryError: HIP out of memory. Tried to allocate 47.98 GiB. GPU 0 has a total capacity of 64.00 GiB of which 5.04 GiB is free. Of the allocated memory 58.21 GiB is allocated by PyTorch, and 242.67 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_HIP_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)

### Steps to Reproduce

text to video wan workflow, click "run"

### Debug Logs

```powershell
VAE load device: cuda:0, offload device: cpu, dtype: torch.float16
Using scaled fp8: fp8 matrix mult: False, scale input: False
Requested to load WanTEModel
loaded completely 9.5367431640625e+25 6419.477203369141 True
CLIP/text encoder model load device: cuda:0, offload device: cpu, current: cuda:0, dtype: torch.float16
model weight dtype torch.float16, manual cast: None
model_type FLOW
Requested to load WAN21
loaded completely 56256.6660296875 2706.1788330078125 True
0%| | 0/30 [00:00
lambda input, sigma, **kwargs: predict_eps_sigma(model, input, sigma, **kwargs),
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/mewmhy/AI/ComfyUI/comfy/extra_samplers/uni_pc.py", line 843, in predict_eps_sigma
return (input - model(input, sigma_in, **kwargs)) / sigma
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/mewmhy/AI/ComfyUI/comfy/samplers.py", line 408, in __call__
out = self.inner_model(x, sigma, model_options=model_options, seed=seed)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/mewmhy/AI/ComfyUI/comfy/samplers.py", line 960, in __call__
return self.outer_predict_noise(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/mewmhy/AI/ComfyUI/comfy/samplers.py", line 967, in outer_predict_noise
).execute(x, timestep, model_options, seed)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/mewmhy/AI/ComfyUI/comfy/patcher_extension.py", line 112, in execute
return self.original(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/mewmhy/AI/ComfyUI/comfy/samplers.py", line 970, in predict_noise
return sampling_function(self.inner_model, x, timestep, self.conds.get("negative", None), self.conds.get("positive", None), self.cfg, model_options=model_options, seed=seed)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/mewmhy/AI/ComfyUI/comfy/samplers.py", line 388, in sampling_function
out = calc_cond_batch(model, conds, x, timestep, model_options)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/mewmhy/AI/ComfyUI/comfy/samplers.py", line 206, in calc_cond_batch
return _calc_cond_batch_outer(model, conds, x_in, timestep, model_options)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/mewmhy/AI/ComfyUI/comfy/samplers.py", line 214, in _calc_cond_batch_outer
return executor.execute(model, conds, x_in, timestep, model_options)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/mewmhy/AI/ComfyUI/comfy/patcher_extension.py", line 112, in execute
return self.original(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/mewmhy/AI/ComfyUI/comfy/samplers.py", line 333, in _calc_cond_batch
output = model.apply_model(input_x, timestep_, **c).chunk(batch_chunks)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/mewmhy/AI/ComfyUI/comfy/model_base.py", line 156, in apply_model
return comfy.patcher_extension.WrapperExecutor.new_class_executor(
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/mewmhy/AI/ComfyUI/comfy/patcher_extension.py", line 112, in execute
return self.original(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/mewmhy/AI/ComfyUI/comfy/model_base.py", line 195, in _apply_model
model_output = self.diffusion_model(xc, t, context=context, control=control, transformer_options=transformer_options, **extra_conds).float()
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/mewmhy/miniconda3/envs/dayilytest/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1751, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/mewmhy/miniconda3/envs/dayilytest/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1762, in _call_impl
return forward_call(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/mewmhy/AI/ComfyUI/comfy/ldm/wan/model.py", line 602, in forward
return comfy.patcher_extension.WrapperExecutor.new_class_executor(
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/mewmhy/AI/ComfyUI/comfy/patcher_extension.py", line 112, in execute
return self.original(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/mewmhy/AI/ComfyUI/comfy/ldm/wan/model.py", line 622, in _forward
return self.forward_orig(x, timestep, context, clip_fea=clip_fea, freqs=freqs, transformer_options=transformer_options, **kwargs)[:, :, :t, :h, :w]
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/mewmhy/AI/ComfyUI/comfy/ldm/wan/model.py", line 567, in forward_orig
x = block(x, e=e0, freqs=freqs, context=context, context_img_len=context_img_len)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/mewmhy/miniconda3/envs/dayilytest/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1751, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/mewmhy/miniconda3/envs/dayilytest/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1762, in _call_impl
return forward_call(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/mewmhy/AI/ComfyUI/comfy/ldm/wan/model.py", line 225, in forward
y = self.self_attn(
^^^^^^^^^^^^^^^
File "/home/mewmhy/miniconda3/envs/dayilytest/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1751, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/mewmhy/miniconda3/envs/dayilytest/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1762, in _call_impl
return forward_call(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/mewmhy/AI/ComfyUI/comfy/ldm/wan/model.py", line 73, in forward
x = optimized_attention(
^^^^^^^^^^^^^^^^^^^^
File "/home/mewmhy/AI/ComfyUI/comfy/ldm/modules/attention.py", line 451, in attention_pytorch
out = comfy.ops.scaled_dot_product_attention(q, k, v, attn_mask=mask, dropout_p=0.0, is_causal=False)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/mewmhy/AI/ComfyUI/comfy/ops.py", line 47, in scaled_dot_product_attention
return torch.nn.functional.scaled_dot_product_attention(q, k, v, *args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
torch.OutOfMemoryError: HIP out of memory. Tried to allocate 47.98 GiB. GPU 0 has a total capacity of 64.00 GiB of which 5.04 GiB is free. Of the allocated memory 58.21 GiB is allocated by PyTorch, and 242.67 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_HIP_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)

Got an OOM, unloading all loaded models.
```

### Other

_No response_

Start with nodes.py at sample and common_ksampler, then follow the traceback through comfy/samplers.py, especially calc_cond_batch, and comfy/model_base.py at _apply_model. Reproduce the Wan text-to-video workflow on the reported AMD setup and inspect where the 47.98 GiB allocation occurs. Done means identifying a confirmed cause and documenting or testing the resulting behavior.

Written by the indexing model from the issue text.