Can I use FlashRAG for retrieving exact match queries

[lebronjamesking]

Dear,

As per title, given a off-the-shelf corpus like a list of pairs with each pair ("query":"who is Lily King": "content":"Lily King is known for cosmetics makeup achievements"). My question is that can I build an index and then use the high throughput of FlashRAG to do the heavy concurrent hard match or search of content.

For now, I am confused whether query can be indexed into the database as default wikipedia you don't have query but just id and content.

[ignorejjj]

In my understanding, when building an index, there is no need to include a query. The query is only used for the final search (by using our retriever).

[onestardao]

This sounds like it could run into a “Semantic ≠ Embedding” issue — exact match logic doesn’t always survive when passing through the embedding & retrieval layers. You might want to think about whether FlashRAG is treating your IDs/queries as semantic text or as strict keys, because the wrong mode can give unpredictable hits.

If you can share how your index schema handles those pairs, we can see if it needs a guard in the retrieval stage.