問題文
A team wants to move from a 1536-dimension embedding model to a 3072-dimension one in Azure Database for PostgreSQL, where the vectors are indexed. What limitation must they plan around?
選択肢
- The new model's output must be stored as text and cast to a vector at query time, which sidesteps the dimension limit at the cost of some additional query-time work
- The column's declared dimension count can be altered in place with a brief lock, so the only planning needed is a maintenance window for the alteration and the index rebuild
- Only columns with up to 2000 dimensions can be indexed, so the new model's output cannot be indexed with the list-based or graph-based methods as-is
- The index must be rebuilt, but the dimension limit applies only to the list-based method, so the graph-based method can index the new model's output directly