About model cache mechanism。

### Your question

Hello,
this day i try to use ipadapter to run flux model, but I found a odd phenomenon, ipadapter node use it's own attenton module, layer and forward function，and it's use the ori flux model as input, use add_object_patch to insert this to ori comfyui flux model, but I found, when I change the weight of ipadapter, the node will re-execute but I found the input model become the model which has already fixed, So when I adjust the weight, it will add another ipadapter on it, rather than adjust the old weight. I don't know if it is because the cache mechanism, lead the input model changed.

### Logs

```powershell

```

### Other

_No response_