Christian Bahls
Christian Bahls
in my research i avoid letting the CLS token's embeddings interact, do you see better results with that?
@Sandy4321 my mistake, i had a look at the code again, the embedding and the mask are truncated in different functions, so that it works out: ```python def compute_colbert_score(self, q_reps,...
As you asked for some code: I wrote a very stripped down loading and inference code, to understand how bge-m3 works: ```python import torch import torch.nn as nn import torch.nn.functional...