112 lines
4.4 KiB
Markdown
112 lines
4.4 KiB
Markdown
---
|
|
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) |