haiku.rag/docs/configuration
Yiorgis Gozadinos c0cf612d68
Document vector index state and measured cost
Benchmarks state that every published number is exact brute-force kNN;
no benchmark database carries a vector index. The Vector Indexing
section gains the measured with/without IVF_PQ comparison (hotpotqa,
orb_multimodal_nemotron, frames: free to ~121k chunks, 0.0044 MAP at
426k, ~30 s / 4 GB build) and corrects the re-indexing story: optimize()
folds new chunks into the index as delta parts via auto_vacuum, so a
rebuild is about retraining centroids, not covering new rows.
2026-09-01 15:19:13 +03:00
..
index.md Honor configured vector metric during search 2026-08-31 21:38:00 +08:00
processing.md Dedupe picture chunks and skip small pictures at chunking 2026-07-09 10:24:32 +03:00
prompts.md Make the documented configuration match the code 2026-08-19 15:52:50 +03:00
providers.md Accept a /v1 suffix on the vLLM reranker base_url 2026-08-28 15:46:42 +03:00
qa.md Make the documented configuration match the code 2026-08-19 15:52:50 +03:00
storage.md Document vector index state and measured cost 2026-09-01 15:19:13 +03:00