methodv2 Standalone Bundle

This folder is a self-contained copy of the FD-NL2SQL/methodv2 pipeline with the code and local data dependencies needed to run it outside the parent FD-NL2SQL repository.

Included

• methodv2/ The runnable pipeline scripts and prompt templates.
• data/ The copied SQLite database, annotated CSVs, and exported table_question_ground_truths_full question set.
• utils.py
• question_table_exports.py
• eval_run_baselines_v2.py
• eval_run_baselines_v3.py
• eval_run_baselines_derived.py

Notes

• The main runners now resolve dataset and run paths relative to this folder.
• The copied manifest was rewritten to use local question paths under data/table_question_ground_truths_full/questions/.
• Some model defaults still point to shared local snapshots under /mnt/shared. If those are unavailable on another machine, pass the relevant --model_path / --*_model_path arguments explicitly.
• Optional Python extras for embedding, richer evaluation, and local vLLM generation are listed in requirements-optional.txt.

Example

python3 methodv2/run_question_table_copy_llm_naive.py \
  --manifest_csv data/table_question_ground_truths_full/manifest.csv \
  --annotated_csv 'data/cat3_query_sql_llm(2)_with_key_matches.csv' \
  --db_path data/database.db \
  --api_base http://127.0.0.1:8000/v1 \
  --api_key EMPTY \
  --model_name Qwen3-30B-A3B-Instruct-2507 \
  --run_name question_table_copy_llm_naive_local \
  --limit 10

GitHub Pages

A project page is scaffolded in docs/:

• docs/index.html
• docs/styles.css

Deployment is automated with .github/workflows/pages.yml.

After pushing to GitHub:

1. Open repository Settings -> Pages
2. Under Build and deployment, set Source to GitHub Actions
3. Push to main (or master) to trigger deployment

The site will publish at:

https://<your-username>.github.io/<repo-name>/