GPU memory overflow and crash when using GGUF models after latest ComfyUI update

### 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

After updating to the latest version of ComfyUI, I started experiencing an issue where my GPU memory quickly fills up and the program crashes. This problem didn’t exist before (on Python 3.12).

Sometimes I can generate one or two images before the crash happens again, but eventually ComfyUI always stops.

I tested this with multiple models, such as **flux.1** and **qwen**, and noticed that the issue happens most frequently with **GGUF models**, but **FP8 models** can also sometimes trigger the same crash:

- With **flux.1**, FP8 occasionally works, but GGUF almost always causes the program to crash.
- With **qwen**, image editing consistently leads to the program freezing at different stages.

This issue only appeared after the latest ComfyUI update. Previously, I was able to use these models without problems.

### Actual Behavior

- GPU memory usage spikes until it’s fully consumed.
- ComfyUI disconnects or freezes after 1–2 runs.
- Happens consistently with GGUF models, and sometimes with FP8 as well.

### Steps to Reproduce

1. Launch ComfyUI with the latest version.
2. Load a GGUF or FP8 model (e.g., flux.1 or qwen).
3. Generate or edit an image once or twice.
4. Observe GPU memory quickly filling up until ComfyUI disconnects or crashes.

### Debug Logs

```powershell
.
```

### Other

_No response_