There are unrelated passages from top-k passages. If so, it will be not good for final answer result, also waste of input tokens.
So, we can drop some passages using relevance score.
The simple idea is only use passages in top 1-sigma. or other n-sigma rules.
You can strict threshold count for dropping passage.
There are unrelated passages from top-k passages. If so, it will be not good for final answer result, also waste of input tokens. So, we can drop some passages using relevance score.
The simple idea is only use passages in top 1-sigma. or other n-sigma rules. You can strict threshold count for dropping passage.