onekq-ai/OneSQL-v0.1-Qwen-3B-MLX-4bit

Introduction

This model is the MLX version of OneSQL-v0.1-Qwen-3B. It is made for Apple Silicon.

Performances

The self-evaluation EX score of the original model is 43.35 (compared to 63.33 by the 32B model on the BIRD leaderboard. The self-evaluation EX score of this MLX model is 38.20.

Quick start

To use this model, craft your prompt to start with your database schema in the form of CREATE TABLE, followed by your natural language query preceded by --. Make sure your prompt ends with SELECT in order for the model to finish the query for you. There is no need to set other parameters like temperature or max token limit.

from mlx_lm import load, generate

model, tokenizer = load(model="onekq-ai/OneSQL-v0.1-Qwen-3B-MLX-4bit")

prompt="""CREATE TABLE students (
    id INTEGER PRIMARY KEY,
    name TEXT,
    age INTEGER,
    grade TEXT
);

-- Find the three youngest students
SELECT """

response = generate(model, tokenizer, f"<|im_start|>system\nYou are a SQL expert. Return code only.<|im_end|>\n<|im_start|>user\n{prompt}<|im_end|>\n<|im_start|>assistant\n")
print(response)

The model response is the finished SQL query without SELECT

* FROM students ORDER BY age ASC LIMIT 3

Interactivity

Speed benchmark of this model is obtained on a MacBook Air with M1 processor and 8GB of RAM, the lower bound of Apple Silicon. On average, it took 8.76 seconds to generate a SQL query at 16.7 characters per second.