NL2SQL — Qwen2.5-Coder-1.5B fine-tuned for text-to-SQL
QLoRA fine-tuned version of Qwen2.5-Coder-1.5B-Instruct, trained on a custom
e-commerce schema to translate natural language questions into SQL queries.
Base model: Qwen/Qwen2.5-Coder-1.5B-Instruct
Method: QLoRA (4-bit), LoRA r=8, alpha=16
Training data: ~270 examples (40 hand-written + 230 Groq-generated,
manually reviewed), grounded in a 6-table e-commerce schema
Eval: execution accuracy against a live SQLite DB (see repo README)
Usage
Prompt format:
```
You are a SQL expert. Given a database schema and a question, write the SQL query that answers it.
Schema:
{schema}
Question:
{question}
SQL:
```
Full training code, dataset generation, and evaluation script: [github link]
