# RAG-QA Easy

This competition is a RAG task that can only be solved by using the fictional documents provided in the corpus.
AI tools may be used freely, but solution models are limited to the Qwen3-4B.
Participants should answer the questions in `test.jsonl` and submit predictions following the format in `sample_submission.jsonl`.

# Description

1. Training Dataset (2,500 examples)
- File name: `train.jsonl`
- `question`: The question to answer
- `id`: The question ID
- `reasoning_type`: The reasoning category (`single_hop`, `bridge_multi_hop`, `numeric_comparison`)
- `evidence`: The document and sentence references needed to answer the question, e.g. `[{"doc_id": "", "sent_id": 1}]`
- `answers`: The gold answer label(s)

2. Validation Dataset (400 examples)
- File name: `dev.jsonl`

3. Test Dataset (1,000 examples)
- File name: `test.jsonl`

4. Corpus Dataset (443 documents)
- File name: `corpus.jsonl`

5. Submission Format
- File name: `sample_submission.jsonl`

# Files

```text
corpus.jsonl
train.jsonl
dev.jsonl
test.jsonl
sample_submission.jsonl
score_submission.py
```