The current return value is as follows：

If normalize, what's the result?

The structure of the result is the same, except that the values ​​are normalized to be between 0 and 1

The corresponding implementation in FlagEmbedding is as follows

def sigmoid(x):
    return 1 / (1 + np.exp(-x))

for i in range(len(all_scores)):
    all_scores[i] = [all_scores[i][idx] for idx in np.argsort(length_sorted_idx)]
    if normalize:
          all_scores[i] = [sigmoid(score) for score in all_scores[i]]

OK, looks like this will not effect the result, but I don't know if this will cause any side effect. @codingl2k1 What's your opinion?

OK, looks like this will not effect the result, but I don't know if this will cause any side effect. @codingl2k1 What's your opinion?

This will be a breaking change. Since FlagEmbedding sets normalize to False by default, how about we keep normalize set to False by default and allow users to set normalize in kwargs?

OK, looks like this will not effect the result, but I don't know if this will cause any side effect. @codingl2k1 What's your opinion?

This will be a breaking change. Since FlagEmbedding sets normalize to False by default, how about we keep normalize set to False by default and allow users to set normalize in kwargs?

OK, @hustyichi What's your idea? is it possible to add kwargs to rerank and allow setting normalize to True?

OK, looks like this will not effect the result, but I don't know if this will cause any side effect. @codingl2k1 What's your opinion?

This will be a breaking change. Since FlagEmbedding sets normalize to False by default, how about we keep normalize set to False by default and allow users to set normalize in kwargs?

OK, @hustyichi What's your idea? is it possible to add kwargs to rerank and allow setting normalize to True?

This is a good idea, but requires more code

I added the always_normalize option, which is not enabled by default

I refined the code to pass kwargs instead of the always_normalize option.