tweetindex

tomaarsen @tomaarsen · 18 Aug 2026

The honest tradeoff is index size. 4,874 Natural Questions passages become 608,414 token vectors: 311.5 MB against ~20 MB for a simple 1024d dense index. About 16x. Three ways out: token pooling, a real late-interaction index, or using it as a reranker.
731Views
15Likes
0Reposts
1Replies
1Quotes
2Bookmarks

Is that a lot?

1.15×vs this author's median637 views is typical
54Percentile for this authorof 22 recent posts
0.24×vs under 10K median3 068 views is typical
15.18%Reachviews ÷ followers
2.33%Engagement rateof viewers reacted

Compare with the benchmark table →

Open on X →