eval/reference/answer_3b.json

{
  "config": {
    "dataset": "hotpotqa-distractor",
    "model": "llama3.2:3b",
    "n_docs": 491,
    "n_queries": 50
  },
  "stacks": {
    "Stack 1 — Vectoriel (FAISS)": {
      "ndcg@10": 0.8031,
      "em": 0.06,
      "f1": 0.1582,
      "n_queries": 50
    },
    "Stack 2 — Hybride (BM25 + RRF)": {
      "ndcg@10": 0.8123,
      "em": 0.06,
      "f1": 0.1648,
      "n_queries": 50
    },
    "Stack 3 — Graphe (networkx)": {
      "ndcg@10": 0.7786,
      "em": 0.1,
      "f1": 0.2093,
      "n_queries": 50
    }
  }
}