config.yaml

chunking:
  chunk_size: 800
  overlap: 150
embeddings:
  model_name: sentence-transformers/all-MiniLM-L6-v2
  normalize: true
  batch_size: 64
index:
  index_path: indexes/faiss.index
  meta_path: indexes/meta.jsonl
retrieval:
  top_k: 5
  mode: dense
  hybrid_rrf_k: 60
  hybrid_dense_weight: 1.0
  hybrid_bm25_weight: 1.0
reranker:
  enabled: true
  model_name: cross-encoder/ms-marco-MiniLM-L-6-v2
  candidate_k: 12
  max_length: 256
  strategy: low_margin_only
  low_margin_threshold: 0.15
confidence:
  threshold_high: 0.4
  threshold_low: 0.25
  margin_min: 0.03
logging:
  enabled: true
  path: logs/queries.jsonl
generation:
  enabled: true
  model: gpt-4o-mini
  temperature: 0.0