danvics/haiku.rag
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Document that text-only rerankers do not work on multi-modal search results

Browse source
This commit is contained in:
Yiorgis Gozadinos 2026-05-26 12:29:45 +03:00
parent a9b411e296
commit c83bd2d4f8
No known key found for this signature in database
1 changed files with 5 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

5
docs/configuration/qa.md
View file

@ -18,6 +18,11 @@ Context expansion is automatic and section-aware. For structured documents (with
!!! note "Reranking behavior"
When a reranker is configured, search automatically retrieves 10x the requested limit, then reranks to return the final count. This improves result quality without requiring you to adjust `limit`.
!!! warning "Reranker compatibility with multimodal content"
Text-only rerankers (mxbai, cohere, jina, vllm, cross-encoder) score documents by their text content. Picture chunks emitted by docling have empty content, so the reranker scores them near zero and drops them from the top results.
If retrieval relies on visual content (charts, figures, screenshots with text rendered as pixels), disable reranking by leaving `reranking.model` empty so the multimodal embedder's ranking survives.
## Question Answering Configuration
Configure the rag skill (used by `client.ask`, `haiku-rag ask`, and the MCP `ask_question` tool):