---
license: cc-by-2.0
---

The whole code and the environment is based on https://github.com/modelscope/ms-swift

To reproduce the results:

(1) inference to generate the three kinds of answers:

bash infer_qwen3_overall_final.sh

bash infer_qwen3_level_final.sh

bash infer_qwen3_level_final.sh

you can get three results files (test_overall.json, test_level.json, test_reason.json) and put them on corresponding folders (result-score, result-level, result-reason)

(2) merge all the results files to a single submitted file.

run the ensemble_results.py file to generate the final submitted file.