metadata

language: en
pipeline_tag: text2text-generation
tags:
  - text-to-sql
  - t5
  - natural-language-processing
  - sql
license: apache-2.0
datasets:
  - gretelai/synthetic_text_to_sql
base_model:
  - Salesforce/codet5-base

Text-to-SQL T5 Model (`16pramodh/t2s_model`)

Model Description

This is a T5-based text-to-SQL model trained to convert natural language questions into SQL queries.
It works by taking in:

natural language query [SEP] table schema

and producing a SQL statement based on the provided database schema.

The model is based on T5ForConditionalGeneration and supports text2text-generation via the Hugging Face Inference API.

Intended Use

Input: English natural language question plus the database schema.
Output: SQL query that can be executed on the described database.

Example

Input: Get the names and emails of all customers who signed up after January 1, 2024 [SEP] CREATE TABLE customers (customer_id INT PRIMARY KEY, name VARCHAR(50), email VARCHAR(100), signup_date DATE);

Output: SELECT name, email FROM customers WHERE signup_date > '2024-01-01';

How to Use

Hugging Face Inference API

curl -X POST \
  -H "Authorization: Bearer YOUR_HF_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"inputs": "Get the names and emails of all customers who signed up after January 1, 2024 [SEP] CREATE TABLE customers (customer_id INT PRIMARY KEY, name VARCHAR(50), email VARCHAR(100), signup_date DATE);"}' \
  https://api-inference.huggingface.co/models/16pramodh/t2s_model

Python (Transformers)

from transformers import AutoTokenizer, AutoModelForSeq2SeqLM

model_name = "16pramodh/t2s_model"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSeq2SeqLM.from_pretrained(model_name)

input_text = "Get the names and emails of all customers who signed up after January 1, 2024 [SEP] CREATE TABLE customers (customer_id INT PRIMARY KEY, name VARCHAR(50), email VARCHAR(100), signup_date DATE);"
inputs = tokenizer(input_text, return_tensors="pt")
outputs = model.generate(**inputs)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Text-to-SQL T5 Model (16pramodh/t2s_model)