| """Minimal usage example for the Hub snapshot. |
| |
| python examples/quickstart.py # local snapshot (repo root) |
| python examples/quickstart.py <repo-id> # straight from the Hub |
| """ |
|
|
| from __future__ import annotations |
|
|
| import sys |
| from pathlib import Path |
|
|
| ROOT = Path(__file__).resolve().parents[1] |
| sys.path.insert(0, str(ROOT)) |
|
|
| from modeling_megaram import load |
|
|
| DOCS = [ |
| { |
| "ext_id": "bm25", |
| "title": "BM25", |
| "text": ( |
| "BM25 is a bag-of-words retrieval function that ranks documents by query " |
| "term frequency with saturation k1 and document length normalisation b. " |
| "Typical values are k1 equal to 1.2 and b equal to 0.75." |
| ), |
| }, |
| { |
| "ext_id": "beir", |
| "title": "BEIR", |
| "text": ( |
| "BEIR is a heterogeneous benchmark of 18 information retrieval datasets, " |
| "introduced by Thakur, Reimers and Gurevych in 2021. It measures zero-shot " |
| "retrieval with nDCG at 10." |
| ), |
| }, |
| { |
| "ext_id": "rgb", |
| "title": "RGB", |
| "text": ( |
| "RGB tests noise robustness, negative rejection, information integration " |
| "and counterfactual robustness in retrieval augmented generation." |
| ), |
| }, |
| ] |
|
|
|
|
| def main() -> int: |
| target = sys.argv[1] if len(sys.argv) > 1 else str(ROOT) |
| rag = load(target) |
| print("loaded:", rag.info()) |
|
|
| rag.index_documents(DOCS) |
| for q in [ |
| "What are the typical values of k1 and b in BM25?", |
| "How many datasets does BEIR contain?", |
| "Which benchmark tests negative rejection?", |
| "What is the capital of France?", |
| ]: |
| ans = rag.answer(q) |
| print(f"\nQ: {q}") |
| print(f" intent={ans.intent} conf={ans.confidence:.3f} abstained={ans.abstained}") |
| print(f" A: {ans.text}") |
| for c in ans.citations: |
| print(f" [{c.doc_id}] {c.quote}") |
| return 0 |
|
|
|
|
| if __name__ == "__main__": |
| raise SystemExit(main()) |
|
|
