Spaces:
Running on Zero
Running on Zero
Download tests/test_ocr.py from subh-007/allcited-api: direct link, hf CLI and curl.
- Browser
- Download file 7.52 kB
-
https://huggingface.co/spaces/subh-007/allcited-api/resolve/main/tests/test_ocr.py
- Command line
-
hf download hf://spaces/subh-007/allcited-api/tests/test_ocr.py
-
curl -L -o test_ocr.py https://huggingface.co/spaces/subh-007/allcited-api/resolve/main/tests/test_ocr.py
7.52 kB
| import json | |
| from pathlib import Path | |
| from typing import List | |
| from fastapi.testclient import TestClient | |
| from app.main import app | |
| from app.pipeline.answer import post_pass_citations | |
| from app.pipeline.chunker import Chunk, chunk_blocks | |
| from app.pipeline.parsers.pdf import parse_pdf | |
| from app.pipeline.twohop import TwoHopChunk | |
| def parse_sse_events(text: str) -> List[dict]: | |
| """Parse raw SSE text into a list of event objects.""" | |
| events = [] | |
| current_event = "message" | |
| for line in text.splitlines(): | |
| line = line.strip() | |
| if not line: | |
| continue | |
| if line.startswith("event:"): | |
| current_event = line[len("event:"):].strip() | |
| elif line.startswith("data:"): | |
| data_str = line[len("data:"):].strip() | |
| try: | |
| data_json = json.loads(data_str) | |
| except json.JSONDecodeError: | |
| data_json = data_str | |
| events.append({"event": current_event, "data": data_json}) | |
| return events | |
| def test_ocr_post_pass_citations_flag(): | |
| """Verify that chunks with is_ocr=True are tagged with (OCR) flag in label, model, and answer.""" | |
| ocr_chunk = Chunk( | |
| id="ocr_chunk_001", | |
| doc_id="scanned_doc", | |
| content_hash="hash_ocr_1", | |
| text="Permanent employees must tender formal notice at least 60 calendar days in advance.", | |
| section_path=["3. Separation", "3.1 Voluntary Resignation"], | |
| token_count=15, | |
| page=3, | |
| is_ocr=True, | |
| ) | |
| src = TwoHopChunk(chunk=ocr_chunk, is_two_hop=False) | |
| raw_answer = "The notice period for voluntary resignation is 60 calendar days [S1]." | |
| cleaned, citations = post_pass_citations(raw_answer, [src]) | |
| assert len(citations) == 1 | |
| assert citations[0].is_ocr is True | |
| assert "(OCR)" in citations[0].label | |
| assert citations[0].label == "S1 (OCR)" | |
| assert "[S1] (OCR)" in cleaned | |
| def test_parse_scanned_pdf_blocks(): | |
| """Verify parsing a rasterized PDF page (<50 chars text) triggers OCR and tags blocks with is_ocr=True.""" | |
| fixture_path = Path(__file__).resolve().parent.parent.parent / "eval" / "fixtures" / "scanned_handbook.pdf" | |
| assert fixture_path.exists(), f"Scanned fixture missing at {fixture_path}" | |
| blocks = parse_pdf(str(fixture_path), source_name="scanned_handbook.pdf") | |
| assert len(blocks) >= 10, f"Expected >= 10 OCR blocks, got {len(blocks)}" | |
| # Every block from the scanned PDF must be flagged as OCR | |
| for b in blocks: | |
| assert b.is_ocr is True, f"Block {b.text[:30]} missing is_ocr=True" | |
| assert b.page in [1, 2, 3, 4] | |
| # Verify key sections were extracted by OCR | |
| all_text = " ".join(b.text for b in blocks) | |
| assert "Leave" in all_text or "leave" in all_text | |
| assert "Separation" in all_text or "resignation" in all_text | |
| assert "Security" in all_text or "passwords" in all_text or "MFA" in all_text | |
| # Verify chunker preserves is_ocr flag on chunks | |
| chunks = chunk_blocks(blocks, doc_id="scanned_doc") | |
| assert len(chunks) >= 3 | |
| for chunk in chunks: | |
| assert chunk.is_ocr is True, f"Chunk {chunk.id} missing is_ocr=True" | |
| def test_scanned_pdf_answers_3_questions_with_ocr_citations(): | |
| """ | |
| STEP 45 Gate: | |
| The scanned-PDF fixture answers 3 questions with OCR-flagged citations. | |
| """ | |
| client = TestClient(app) | |
| fixture_path = Path(__file__).resolve().parent.parent.parent / "eval" / "fixtures" / "scanned_handbook.pdf" | |
| assert fixture_path.exists(), f"Scanned fixture missing at {fixture_path}" | |
| with open(fixture_path, "rb") as f: | |
| file_bytes = f.read() | |
| upload_res = client.post( | |
| "/v1/ingest", | |
| files={"file": ("scanned_handbook.pdf", file_bytes, "application/pdf")}, | |
| ) | |
| assert upload_res.status_code == 200, f"Upload failed: {upload_res.text}" | |
| session_id = upload_res.json()["session_id"] | |
| assert session_id | |
| # 3 target questions against the scanned PDF | |
| test_queries = [ | |
| ("What is the notice period for voluntary resignation?", "60"), | |
| ("How many days of paid sick leave do employees receive per year?", "12"), | |
| ("What is the minimum password character length required by internal security?", "16"), | |
| ] | |
| import time | |
| for q_text, expected_keyword in test_queries: | |
| time.sleep(1.0) | |
| query_res = client.post( | |
| "/v1/query", | |
| json={"session_id": session_id, "query": q_text}, | |
| ) | |
| assert query_res.status_code == 200, f"Query failed for '{q_text}': {query_res.text}" | |
| events = parse_sse_events(query_res.text) | |
| citations_events = [e for e in events if e["event"] == "citations"] | |
| done_events = [e for e in events if e["event"] == "done"] | |
| assert len(done_events) == 1, f"Missing done event for {q_text}" | |
| answer_text = done_events[0]["data"]["answer"] | |
| assert expected_keyword in answer_text, f"Expected '{expected_keyword}' in answer: {answer_text}" | |
| assert len(citations_events) == 1, f"Missing citations event for {q_text}" | |
| citations_data = citations_events[0]["data"].get("citations", []) | |
| assert len(citations_data) >= 1, f"Zero citations returned for {q_text}" | |
| # Check OCR flag on citations | |
| for cit in citations_data: | |
| assert cit.get("is_ocr") is True, f"Citation {cit} must have is_ocr=True" | |
| assert "(OCR)" in cit.get("label", ""), f"Citation label {cit.get('label')} must contain (OCR)" | |
| # Answer text must also contain (OCR) citation flag | |
| assert "(OCR)" in answer_text, f"Answer text must contain '(OCR)': {answer_text}" | |
| def test_easyocr_provider_on_scanned_pdf(): | |
| """Verify EasyOCR provider extracts text directly from scanned PDF fixture.""" | |
| import fitz | |
| from app.pipeline.parsers.ocr import parse_easyocr_langs, run_ocr | |
| # Test language mapping | |
| assert parse_easyocr_langs("eng+hin") == ["en", "hi"] | |
| assert parse_easyocr_langs("eng") == ["en"] | |
| assert parse_easyocr_langs("en,hi") == ["en", "hi"] | |
| fixture_path = Path(__file__).resolve().parent.parent.parent / "eval" / "fixtures" / "scanned_handbook.pdf" | |
| assert fixture_path.exists() | |
| doc = fitz.open(str(fixture_path)) | |
| ocr_text = run_ocr(doc[0], 0, langs="eng+hin", provider="easyocr") | |
| assert len(ocr_text) > 50, f"Expected EasyOCR to extract >50 chars, got: {ocr_text}" | |
| assert "Handbook" in ocr_text or "Employee" in ocr_text or "Allcited" in ocr_text | |
| def test_readyz_endpoint(): | |
| """Verify /v1/readyz readiness endpoint returns 200 and pipeline status.""" | |
| client = TestClient(app) | |
| res = client.get("/v1/readyz") | |
| assert res.status_code == 200 | |
| data = res.json() | |
| assert data["status"] in ("ok", "ready") | |
| assert data.get("ready") is True | |
| assert data["pipeline"]["embedding"] == "ready" | |
| assert data["pipeline"]["reranker"] == "ready" | |
| def test_local_model_path_resolution(): | |
| """Verify local models in api/models are resolved without downloading from Hub.""" | |
| from app.pipeline.embed import resolve_local_model_path | |
| embed_path = resolve_local_model_path("all-MiniLM-L6-v2", "embedding") | |
| assert "models/all-MiniLM-L6-v2" in embed_path or "api/models/all-MiniLM-L6-v2" in embed_path | |
| assert Path(embed_path).exists() | |
| rerank_path = resolve_local_model_path("cross-encoder/ms-marco-MiniLM-L-6-v2", "reranker") | |
| assert "models/ms-marco-MiniLM-L-6-v2" in rerank_path or "api/models/ms-marco-MiniLM-L-6-v2" in rerank_path | |
| assert Path(rerank_path).exists() | |