Hello as always thanks for the work you have done,
But i have a question , using the paper's parameters with LimeSaliency there is always an error in the prediction phase, here 👍 output_dict_orig['attentions'] = output_dict_orig['attentions'] * (1 - kept_tokens).float() Link
because the shape of both tensors attention is (1,615) and kept_tokens is (4,827) we can't multiply 615 with 827,
Thank you.
Hello as always thanks for the work you have done,
But i have a question , using the paper's parameters with LimeSaliency there is always an error in the prediction phase, here 👍
output_dict_orig['attentions'] = output_dict_orig['attentions'] * (1 - kept_tokens).float()Linkbecause the shape of both tensors attention is (1,615) and kept_tokens is (4,827) we can't multiply 615 with 827,
Thank you.