Download tests/test_hfd_vqa_candidate_pilot.py from Tinghong-Ye22/code-HFD: direct link, hf CLI and curl.
- Browser
- Download file 3.04 kB
-
https://huggingface.co/Tinghong-Ye22/code-HFD/resolve/main/tests/test_hfd_vqa_candidate_pilot.py
- Command line
-
hf download hf://Tinghong-Ye22/code-HFD/tests/test_hfd_vqa_candidate_pilot.py
-
curl -L -o test_hfd_vqa_candidate_pilot.py https://huggingface.co/Tinghong-Ye22/code-HFD/resolve/main/tests/test_hfd_vqa_candidate_pilot.py
3.04 kB
| """Small, offline regression tests for the opt-in VQA admission pilot.""" | |
| from __future__ import annotations | |
| import importlib.util | |
| from pathlib import Path | |
| import pytest | |
| ROOT = Path(__file__).resolve().parents[1] | |
| SPEC = importlib.util.spec_from_file_location( | |
| "hfd_vqa_candidate_pilot", ROOT / "scripts/data/build_hfd_vqa_candidate_pilot.py" | |
| ) | |
| assert SPEC is not None and SPEC.loader is not None | |
| PILOT = importlib.util.module_from_spec(SPEC) | |
| SPEC.loader.exec_module(PILOT) | |
| def _row(tmp_path: Path, *, split: str = "train", accepted: bool = True) -> dict: | |
| body = tmp_path / f"{split}.pt" | |
| body.write_bytes(b"body") | |
| voice = tmp_path / f"{split}.wav" | |
| voice.write_bytes(b"wav") | |
| return { | |
| "id": f"vqav2_{split}_1", | |
| "split": split, | |
| "path": str(body), | |
| "frontend_root": str(tmp_path / "frontend"), | |
| "cohort": "vqav2_static", | |
| "question": { | |
| "question": "Is this picture in color?", | |
| "answer": "yes", | |
| "human_votes": 10, | |
| }, | |
| "visual_review": {"accepted": accepted}, | |
| "timeline": {"question_voice": str(voice)}, | |
| } | |
| def test_vqav2_gate_uses_row_level_blind_review(tmp_path: Path) -> None: | |
| assert PILOT._quality_gate(_row(tmp_path))[0] | |
| accepted, reason = PILOT._quality_gate(_row(tmp_path, accepted=False)) | |
| assert not accepted | |
| assert reason == "vqav2_agreement_or_visual_review_failed" | |
| def test_pilot_selection_is_split_and_body_disjoint(tmp_path: Path) -> None: | |
| rows = [_row(tmp_path, split="train"), _row(tmp_path, split="val")] | |
| selected, audit = PILOT.select_hfd_rows(rows, train_limit=1, val_limit=1) | |
| assert len(selected) == 2 | |
| assert audit["formal_source_plan_changed"] is False | |
| assert {row["split"] for row in selected} == {"train", "val"} | |
| def test_gqa_gate_rejects_ambiguous_questions() -> None: | |
| base = { | |
| "isBalanced": True, | |
| "types": {"structural": "verify"}, | |
| "question": "Is the person holding a cup?", | |
| "answer": "yes", | |
| } | |
| assert PILOT.conservative_gqa_row(base)[0] | |
| ambiguous = {**base, "question": "Why might the person be there?"} | |
| accepted, reason = PILOT.conservative_gqa_row(ambiguous) | |
| assert not accepted and reason == "speculative_or_text_question" | |
| def test_review_answer_waveform_is_separate_from_input(tmp_path: Path) -> None: | |
| spec = importlib.util.spec_from_file_location( | |
| "hfd_gt_render", ROOT / "scripts/data/render_hfd_unified_multimodal_gt.py" | |
| ) | |
| assert spec is not None and spec.loader is not None | |
| module = importlib.util.module_from_spec(spec) | |
| spec.loader.exec_module(module) | |
| output = tmp_path / "answer.wav" | |
| assert module._synthesize_review_answer("Camera lens", output) | |
| import wave | |
| with wave.open(str(output), "rb") as handle: | |
| assert handle.getframerate() == 16_000 | |
| assert handle.getnchannels() == 1 | |
| assert handle.getnframes() > 0 | |