T5 Small Awesome Text To Sql

🚀 Model Card for Model ID

Based on the t5-small model, this lightweight model generates SQL from text given a table list with "CREATE TABLE" statements. It supports multiple tables with joins and can be used in various analytical applications.

🚀 Quick Start

Use the following code to start using the model:

import torch
from transformers import T5Tokenizer, T5ForConditionalGeneration

# Initialize the tokenizer from Hugging Face Transformers library
tokenizer = T5Tokenizer.from_pretrained('t5-small')

# Load the model
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model = T5ForConditionalGeneration.from_pretrained('cssupport/t5-small-awesome-text-to-sql')
model = model.to(device)
model.eval()

def generate_sql(input_prompt):
    # Tokenize the input prompt
    inputs = tokenizer(input_prompt, padding=True, truncation=True, return_tensors="pt").to(device)
    
    # Forward pass
    with torch.no_grad():
        outputs = model.generate(**inputs, max_length=512)
    
    # Decode the output IDs to a string (SQL query in this case)
    generated_sql = tokenizer.decode(outputs[0], skip_special_tokens=True)
    
    return generated_sql

# Test the function
#input_prompt = "tables:\n" + "CREATE TABLE Catalogs (date_of_latest_revision VARCHAR)" + "\n" +"query for: Find the dates on which more than one revisions were made."
#input_prompt = "tables:\n" + "CREATE TABLE table_22767 ( \"Year\" real, \"World\" real, \"Asia\" text, \"Africa\" text, \"Europe\" text, \"Latin America/Caribbean\" text, \"Northern America\" text, \"Oceania\" text )" + "\n" +"query for:what will the population of Asia be when Latin America/Caribbean is 783 (7.5%)?."
#input_prompt = "tables:\n" + "CREATE TABLE procedures ( subject_id text, hadm_id text, icd9_code text, short_title text, long_title text ) CREATE TABLE diagnoses ( subject_id text, hadm_id text, icd9_code text, short_title text, long_title text ) CREATE TABLE lab ( subject_id text, hadm_id text, itemid text, charttime text, flag text, value_unit text, label text, fluid text ) CREATE TABLE demographic ( subject_id text, hadm_id text, name text, marital_status text, age text, dob text, gender text, language text, religion text, admission_type text, days_stay text, insurance text, ethnicity text, expire_flag text, admission_location text, discharge_location text, diagnosis text, dod text, dob_year text, dod_year text, admittime text, dischtime text, admityear text ) CREATE TABLE prescriptions ( subject_id text, hadm_id text, icustay_id text, drug_type text, drug text, formulary_drug_cd text, route text, drug_dose text )" + "\n" +"query for:" + "what is the total number of patients who were diagnosed with icd9 code 2254?"
input_prompt = "tables:\n" + "CREATE TABLE student_course_attendance (student_id VARCHAR); CREATE TABLE students (student_id VARCHAR)" + "\n" + "query for:" + "List the id of students who never attends courses?"

generated_sql = generate_sql(input_prompt)

print(f"The generated SQL query is: {generated_sql}")
#OUTPUT: The generated SQL query is: SELECT student_id FROM students WHERE NOT student_id IN (SELECT student_id FROM student_course_attendance)

✨ Features

• Based on t5-small, it can generate SQL from text given a table list with "CREATE TABLE" statements.
• Supports multiple tables with joins.
• It is a very lightweight model and can be used in multiple analytical applications.

📚 Documentation

Model Details

Model Description

• Developed by: cssupport (support@cloudsummary.com)
• Model type: Language model
• Language(s) (NLP): English
• License: Apache 2.0
• Finetuned from model : t5-small

Model Sources

Please refer t5-small for Model Sources.

Uses

Direct Use

This model can be used in applications where natural language needs to be converted into SQL queries.

Out-of-Scope Use

More information is needed.

Bias, Risks, and Limitations

More information is needed.

Recommendations

Users (both direct and downstream) should be made aware of the risks, biases, and limitations of the model. More information is needed for further recommendations.

Technical Specifications

Model Architecture and Objective

t5-small

Compute Infrastructure

Hardware

one A100-80

Software

Pytorch and HuggingFace

📄 License

This model is licensed under the Apache 2.0 license.

📞 Model Card Contact

cssupport (support@cloudsummary.com)