README.md

---
base_model: unsloth/qwen2.5-coder-1.5b-instruct-bnb-4bit
tags:
- text-generation-inference
- transformers
- unsloth
- qwen2
- trl
- sft
license: apache-2.0
language:
- en
datasets:
- gretelai/synthetic_text_to_sql
---

# Text2SQL-1.5B Model

## Overview
**Text2SQL-1.5B** is a powerful **natural language to SQL** model designed to convert user queries into structured SQL statements. It supports complex multi-table queries and ensures high accuracy in text-to-SQL conversion.

## System Instruction
To ensure consistency in model outputs, use the following system instruction:

> **Always separate code and explanation. Return SQL code in a separate block, followed by the explanation in a separate paragraph. Use markdown triple backticks (` ```sql ` for SQL) to format the code properly. Write the SQL query first in a separate code block. Then, explain the query in plain text. Do not merge them into one response.

For json result use the following
>  **Always separate SQL code and explanation. Return SQL queries in a JSON format containing two keys: 'query' and 'explanation'. The response should strictly follow the structure: {\"query\": \"SQL_QUERY_HERE\", \"explanation\": \"EXPLANATION_HERE\"}. The 'query' key should contain only the SQL statement, and the 'explanation' key should provide a plain-text explanation of the query. Do not merge them into one response.

## Prompt Format
The prompt format should include both the user query and the table structure using a `CREATE TABLE` statement. The expected message format should be:

```json
messages = [
    {"role": "system", "content": "Always separate code and explanation. Return SQL code in a separate block, followed by the explanation in a separate paragraph. Use markdown triple backticks (```sql for SQL) to format the code properly. Write the SQL query first in a separate code block. Then, explain the query in plain text. Do not merge them into one response. The query should always include the table structure using a CREATE TABLE statement before executing the main SQL query."},
    {"role": "user", "content": "Show the total sales for each customer who has spent more than $50,000."},
    {"role": "user", "content": "
CREATE TABLE sales (
    id INT PRIMARY KEY,
    customer_id INT,
    total_amount DECIMAL(10,2),
    FOREIGN KEY (customer_id) REFERENCES customers(id)
);

CREATE TABLE customers (
    id INT PRIMARY KEY,
    name VARCHAR(255)
);
"}
] 
```

## Model Usage

### **Using the Model for Text-to-SQL Conversion**
The following code demonstrates how to use the model to convert natural language queries into SQL statements:

```python
from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline

# Load tokenizer and model
tokenizer = AutoTokenizer.from_pretrained("yasserrmd/Text2SQL-1.5B")
model = AutoModelForCausalLM.from_pretrained("yasserrmd/Text2SQL-1.5B")

# Define the pipeline
pipe = pipeline("text-generation", model=model, tokenizer=tokenizer)

# Define system instruction
system_instruction = "Always separate code and explanation. Return SQL code in a separate block, followed by the explanation in a separate paragraph. Use markdown triple backticks (```sql for SQL) to format the code properly. Write the SQL query first in a separate code block. Then, explain the query in plain text. Do not merge them into one response. The query should always include the table structure using a CREATE TABLE statement before executing the main SQL query."

# Define user query
user_query = "Show the total sales for each customer who has spent more than $50,000.
CREATE TABLE sales (
    id INT PRIMARY KEY,
    customer_id INT,
    total_amount DECIMAL(10,2),
    FOREIGN KEY (customer_id) REFERENCES customers(id)
);

CREATE TABLE customers (
    id INT PRIMARY KEY,
    name VARCHAR(255)
);
"

# Define messages for input
messages = [
    {"role": "system", "content": system_instruction},
    {"role": "user", "content": user_query},
]

# Generate SQL output
response = pipe(messages)


# Print the generated SQL query
print(response[0]['generated_text'])
```





# Uploaded  model

- **Developed by:** yasserrmd
- **License:** apache-2.0
- **Finetuned from model :** unsloth/qwen2.5-coder-1.5b-instruct-bnb-4bit

This qwen2 model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth) and Huggingface's TRL library.

[<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth)