KRadim/Qwen2.5-Coder-3B-text2cypher-LoRA

Qwen2.5-Coder-3B Text-to-Cypher (Neo4j) - Run #1

This model is a fine-tuned version of unsloth/Qwen2.5-Coder-3B-Instruct optimized for translating natural language questions into accurate Neo4j Cypher queries using a structured database schema.

Number of samples used: 35000

Model Details

• Base Architecture: Qwen2.5-Coder-3B-Instruct
• Fine-Tuning Method: QLoRA (Rank 16, Alpha 16)
• Dataset Used: KRadim/multilingual-text2cypher-2025 (35000 examples)
• Epochs: 1
• Frameworks: Unsloth, Hugging Face TRL, PEFT, PyTorch

Prompts & Schema Usage

The model expects system prompts containing the Neo4j schema structure:

<|im_start|>system
You are an expert in Neo4j databases and the Cypher query language. Your task is to convert natural language queries into accurate and efficient Cypher queries based on the provided database schema.

### Database Schema:
[PROVIDE DATABASE SCHEMA HERE]<|im_end|>
<|im_start|>user
[YOUR NATURAL LANGUAGE QUESTION]<|im_end|>
<|im_start|>assistant

Running with Ollama (GGUF)

You can run the exported GGUF model (KRadim/Qwen2.5-Coder-3B-text2cypher-GGUF) directly in Ollama:

ollama run hf.co/KRadim/Qwen2.5-Coder-3B-text2cypher-GGUF:Q4_K_M