Add rerank module: bi-encoder retrieval plus an ONNX Runtime cross-encoder reranker behind a Spring AI DocumentPostProcessor; hit@k and MRR by pool size and question kind, latency by candidates, batch size, threads and int8, and five reproduced traps. Local MiniLM models, no database or API key.

Co-Authored-By: Claude Sonnet 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_01Cfo2K4u4rF18QsPXPFnytf
2026-10-24 18:00:05 +00:00
parent 2ccf1d8d19
commit a429aa40f6
34 changed files with 1617 additions and 0 deletions
@@ -0,0 +1,16 @@
+# One question, the bi-encoder's top 8, rescored
+
+question: A buyer told their bank they never got the item. What now?
+gold passage: G-04
+
+retrieved id      cosine       cross-encoder logit
+1         G-06    0.446        -7.40
+2         G-04    0.440        -8.06
+3         E-4117  0.418        -11.08
+4         E-4101  0.418        -11.30
+5         G-03    0.405        -10.33
+6         E-4140  0.403        -9.91
+7         G-08    0.399        -9.62
+8         G-24    0.391        -9.61
+
+bi-encoder rank of G-04: 2; cross-encoder winner: G-06