A fine-tuned Qwen3-0.6B model for converting natural language questions into SQL queries. Trained using knowledge distillation from DeepSeek-V3, this compact 0.6B parameter model delivers strong Text2SQL performance while being extremely lightweight and fast for local deployment.
| Metric | DeepSeek-V3 (Teacher) | Qwen3-0.6B (Base) | This Model |
|---|---|---|---|
| LLM-as-a-Judge | 76% | 36% | 74% |
| Exact Match | 38% | 24% | 40% |
| ROUGE | 88.6% | 69.3% | 88.5% |
| METEOR | 90.4% | 65.3% | 88.5% |
The fine-tuned model achieves 74% on LLM-as-a-Judge accuracy with only 0.6B parameters - a 2x improvement over the base model and approaching the 685B parameter teacher's performance at a fraction of the size.
from transformers import AutoModelForCausalLM, AutoTokenizer
model = AutoModelForCausalLM.from_pretrained("distil-labs/distil-qwen3-0.6b-text2sql")
tokenizer = AutoTokenizer.from_pretrained("distil-labs/distil-qwen3-0.6b-text2sql")
schema = """CREATE TABLE employees (
id INTEGER PRIMARY KEY,
name TEXT NOT NULL,
department TEXT,
salary INTEGER
);"""
question = "How many employees earn more than 50000?"
messages = [
{
"role": "system",
"content": """You are a problem solving model working on task_description XML block:
<task_description>You are given a database schema and a natural language question. Generate the SQL query that answers the question.
Input:
- Schema: One or two table definitions in SQL DDL format
- Question: Natural language question about the data
Output:
- A single SQL query that answers the question
- No explanations, comments, or additional text
Rules:
- Use only tables and columns from the provided schema
- Use uppercase SQL keywords (SELECT, FROM, WHERE, etc.)
- Use SQLite-compatible syntax</task_description>
You will be given a single task in the question XML block
Solve only the task in question block.
Generate only the answer, do not generate anything else"""
},
{
"role": "user",
"content": f"""Now for the real task, solve the task in question block.
Generate only the solution, do not generate anything else
<question>Schema:
{schema}
Question: {question}</question>"""
}
]
text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(text, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=256, temperature=0)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
For local inference, use the quantized GGUF version included in this repository:
# Download and create Ollama model
ollama create distil-qwen3-0.6b-text2sql -f Modelfile
# Run inference
ollama run distil-qwen3-0.6b-text2sql
| Property | Value |
|---|---|
| Base Model | Qwen/Qwen3-0.6B |
| Parameters | 0.6 billion |
| Architecture | Qwen3ForCausalLM |
| Context Length | 40,960 tokens |
| Precision | bfloat16 |
| Training Data | ~10,000 synthetic examples |
| Teacher Model | DeepSeek-V3 |
This model was trained using the Distil Labs platform:
Schema:
CREATE TABLE table_name (
column_name DATA_TYPE [CONSTRAINTS],
...
);
Question: Natural language question about the data
A single SQL query with:
This model is released under the Apache 2.0 license.
@misc{distil-qwen3-0.6b-text2sql,
author = {Distil Labs},
title = {Distil-Qwen3-0.6B-Text2SQL: A Compact Fine-tuned Model for Natural Language to SQL},
year = {2025},
publisher = {Hugging Face},
url = {https://huggingface.co/distil-labs/distil-qwen3-0.6b-text2sql}
}