Yitong Chen
Yitong Chen
same here
I meet the same broken, do this can solve this issue ```python def embed_image(self, image: torch.Tensor): return self.paligemma.model.get_image_features(image) ``` Maybe you also need this, but I'm not sure whether this...
For 5090, I think maybe you should update the CUDA version to 12.8?
This has been solved in some issue
same question
same here
This makes sense. But /compact seems more complex than just think or default or websearch model setting. The /compact will also involves to replace the whole history by our summary....