This PR is to implement test on Japanese cv parser created through Azure content understanding.
To open test_cv_parsing_speed_and_cost.ipynb,
- please run command below to start Jupyter lab
uv run --with jupyter jupyter lab- In Jupter lab, click
test_cv_parsing_speed_and_cost.ipynb
To run test_cv_parsing.py to try cv parser,
- please run below
uv run python test_cv_parsing.py| api version | LLM | analyser id | speed (per 2-page pdf) | cost (per 2-page pdf) | quality (per 2-page pdf) | speed (per 2-page docx) | cost (per 2-page docx) | quality (per 2-page docx) | speed (per 2-page xlsx) | cost (per 2-page xlsx) | quality (per 2-page xlsx) |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 2025-11-01 | gpt-5 | cv_parser_test_lk | 22.5 s ± 1.05 s | $0.0357 | high | 12.7 s ± 98.6 ms | $0.021 | very low | 13.6 s ± 2.18 s | $0.0231 | very low |
| 2026-06-01-preview | gpt-5 | cv_parser_preview_api_2026_06_01 | 19.9 s ± 1.52 s | $0.0438 | high | 11 s ± 1.58 s | $0.0272 | high | 10.1 s ± 1.05 s | $0.029 | high |
| 2026-06-01-preview | gpt-5-mini | cv_parser_preview_api_2026_06_01_gpt5_mini | 20.8 s ± 4.34 s | $0.0213 | low | 13.7 s ± 1.34 s | $0.0072 | medium | 14.1 s ± 1.49 s | $0.0104 | medium |