Open orrzohar opened 2 months ago
Dear authors, Thank you for your great work. I wondered why you set the number of heads in the attentive pooler to be equal to that of the JEPA encoder?
https://github.com/facebookresearch/jepa/blob/13fbba844b3ce3d16a1c1b633a52fff6f88e9876/evals/video_classification_frozen/eval.py#L184
Is this a principled design decision or just a hyperparameter? Best, Orr
Dear authors, Thank you for your great work. I wondered why you set the number of heads in the attentive pooler to be equal to that of the JEPA encoder?
https://github.com/facebookresearch/jepa/blob/13fbba844b3ce3d16a1c1b633a52fff6f88e9876/evals/video_classification_frozen/eval.py#L184
Is this a principled design decision or just a hyperparameter? Best, Orr