I'm very excited to announce that I've finetuned Llama3-Instruct:8b to generate Neo4j Cypher statements based on the GPT-4o synthetic dataset I've generated at the start of the week. I have used the inspiration from Geraldus Wilsen to use Unsloth AI and updated it a bit to use chat prompt template (system, user, assistant messages). Please test it out and let me know how it goes as I used mostly default finetuning arguments because I don't really know what I'm doing . I have also added example code to show how to implement it with LangChain!
Links:
Finetuning notebook with docs: text2cypher/finetuning/unsloth-llama3 at main ยท neo4j-labs/text2cypher ยท GitHub
HuggingFace model: tomasonjo/text2cypher-demo-16bit ยท Hugging Face
Ollama models: tomasonjo/llama3-text2cypher-demo