DocuBench / tests /test_run_gpt.py
urimer's picture
Initial upload: 50 documents, schemas, hand-verified labels, scorer, baseline results (part 2)
8143f06 verified
Raw
History Blame Contribute Delete
1.49 kB
import importlib.util
from pathlib import Path
from PIL import Image
def load_run_gpt_module():
spec = importlib.util.spec_from_file_location("run_gpt", "scripts/run_gpt.py")
module = importlib.util.module_from_spec(spec)
spec.loader.exec_module(module)
return module
def test_build_prompt_uses_committed_template():
run_gpt = load_run_gpt_module()
template = Path("prompts/extraction_prompt.txt").read_text(encoding="utf-8").strip()
# the committed prompt file is the single source of truth, not a stale copy
assert run_gpt.load_prompt_template() == template
assert "{doc_id}" in template
assert run_gpt.build_prompt("ABC123") == template.format(doc_id="ABC123")
assert "ABC123" in run_gpt.build_prompt("ABC123")
def test_tiff_is_converted_to_ordered_png_image_parts(tmp_path):
tiff_path = tmp_path / "multi_page.tiff"
first = Image.new("RGB", (10, 10), "white")
second = Image.new("RGB", (10, 10), "black")
first.save(tiff_path, save_all=True, append_images=[second])
run_gpt = load_run_gpt_module()
parts, meta = run_gpt.tiff_png_parts(tiff_path)
assert meta == {"input_mode": "tiff_png_sequence", "tiff_pages": 2}
assert [part["type"] for part in parts] == [
"input_text",
"input_image",
"input_text",
"input_image",
]
assert parts[1]["image_url"].startswith("data:image/png;base64,")
assert parts[3]["image_url"].startswith("data:image/png;base64,")