out of memory error, emptying cache and trying again

### 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

Run qwen-image-edit-basic workflow using split-cross-attention, about 1.xit/sec.

### Actual Behavior

very slow, throw `out of memory error`, even when process 512x512 image

### Steps to Reproduce

`python main.py --use-split-cross-attention `

run the qwen-image-edit-basic workflow:

### Debug Logs

```powershell
got prompt
Requested to load QwenImageTEModel_
loaded completely 22439.33420448303 7909.737449645996 True
model weight dtype torch.float8_e4m3fn, manual cast: torch.bfloat16
model_type FLUX
Requested to load QwenImage
loaded partially 17970.96241989136 17970.703674316406 167
0%| | 0/20 [00:00