
Due to the competition's evaluation metrics, the final evaluation to select the winners was limited to those who wanted to submit, and our team placed 7th out of 24 teams.
Name | Type | Performance | Rank |
---|---|---|---|
건설공사 사고 예방 및 대응책 생성 : 한솔데코 시즌3 AI 경진대회 | NLP, LLM | 🥈 Top2% | 7/551 (7/24) |
conda create -n dacon python=3.10
conda activate dacon
pip install -r requirements.txt
pip install flash-attn --no-build-isolation
For a detailed explanation of the methodology, please refer to our presentation slides
Please use this README for data preprocessing
python src/expand_query.py \
--model_name_or_path rtzr/ko-gemma-2-9b-it \
--test_data ./data/test_preprocessed.csv \
--output_data ./data/query_expansions.jsonl \
--token YOUR_HF_TOKEN
python src/main.py \
--model_name_or_path rtzr/ko-gemma-2-9b-it \
--attn_implementation eager \
--test_data ./data/test_preprocessed.csv \
--query_expansions_path ./data/query_expansions.jsonl \
--embedding_model nlpai-lab/KURE-v1 \
--top_k 30 \
--documents_path ./data/documents \
--use_reranker true \
--reranker_model dragonkue/bge-reranker-v2-m3-ko \
--reranker_top_k 10 \
--submission_data ./submissions/submission.csv \
--token YOUR_HF_TOKEN
홍재민 | 최용빈 |
---|---|