- Start the Neo4j database.
- Run
python graphraft.pywith--dataset primeor--dataset mag
- Preprocess NER and Cypher queries in
training/generate_training_data.ipynb - Train LLM1:
python llm1.py --dataset [] --train - Preprocess cypher generation:
python llm1.py --dataset [] --generate_valid --generate_test. - Optionally evaluate 3. using
python eval_llm.py --dataset --max_nodes 20 - Preprocess retrieval in
training/generate_training_data.ipynb - Train LLM2:
python llm2.py --dataset [] --train - Evaluate pipeline:
python llm2.py --dataset [] --adapter_dir [] --test --add_more_answers