新版本，运行到加载Unet模型时崩溃，FLUX-FILL-OneReward

### Custom Node Testing

- [x] 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

pytorch version: 2.9.0+cul30 :
xformers version: 0.0.33.post
Set vram state to: NORMAL VRAM
Device: cuda:0 NVIDIA GeForce RTX 4030 D : cudaMallocAsync :
Using async weight offloading with 2 streams
Enabled pinned memory 490009.0

### Actual Behavior

Using xformers attention in VAE 
VAE load device: cuda:0, offload device: cpu, dtype: torch.floatl6
clip missing: ['text_projection.weight']
CLIP/text encoder model load device: cuda:0, offload device: cpu, current: cpu, dtype: torch.floatl6 
Requested to load FluxClipModel
Model FluxClipModel prepared for dynamic VRAM loading. 9318MB Staged. 0 patches attached. Force pre-loaded 49 weights: 392 KB. 
Requested to load AutoencodingEngine
Model AutoencodingEngine prepared for dynamic VRAM loading. 159MB Staged. 0 patches attached. 
Model AutoencodingEngine prepared for dynamic VRAM loading. 159MB Staged. 0 patches attached. 

[FLUX-FILL-OneReward.json](https://github.com/user-attachments/files/25844309/FLUX-FILL-OneReward.json)

### Steps to Reproduce

Using xformers attention in VAE 
VAE load device: cuda:0, offload device: cpu, dtype: torch.floatl6
clip missing: ['text_projection.weight']
CLIP/text encoder model load device: cuda:0, offload device: cpu, current: cpu, dtype: torch.floatl6 
Requested to load FluxClipModel
Model FluxClipModel prepared for dynamic VRAM loading. 9318MB Staged. 0 patches attached. Force pre-loaded 49 weights: 392 KB. 
Requested to load AutoencodingEngine
Model AutoencodingEngine prepared for dynamic VRAM loading. 159MB Staged. 0 patches attached. 
Model AutoencodingEngine prepared for dynamic VRAM loading. 159MB Staged. 0 patches attached. 

### Debug Logs

```powershell
Using xformers attention in VAE 
VAE load device: cuda:0, offload device: cpu, dtype: torch.floatl6
clip missing: ['text_projection.weight']
CLIP/text encoder model load device: cuda:0, offload device: cpu, current: cpu, dtype: torch.floatl6 
Requested to load FluxClipModel
Model FluxClipModel prepared for dynamic VRAM loading. 9318MB Staged. 0 patches attached. Force pre-loaded 49 weights: 392 KB. 
Requested to load AutoencodingEngine
Model AutoencodingEngine prepared for dynamic VRAM loading. 159MB Staged. 0 patches attached. 
Model AutoencodingEngine prepared for dynamic VRAM loading. 159MB Staged. 0 patches attached.
```

### Other

_No response_