Text2SQL
Collection
Small language models for text to SQL
•
1 item
•
Updated
LLM instruction finetuned for Text-to-SQL task.
Model can be used a tool to convert queries in expressed in natural language (English) to SQL statements
The model could be used as the initial stage in a data analytics / business intelligence application pipeline.
Model has been fine tuned on a specific task of converting English language statements to SQL queries. Any use beyond this is not guaranteed to be accurate.
Use the code below to get started with the model.
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
model = AutoModelForCausalLM.from_pretrained(
"dataeaze/dataeaze-text2sql-codellama_7b_instruct-clinton_text_to_sql_v1",
torch_dtype=torch.bfloat16,
device_map='auto'
)
tokenizer = AutoTokenizer.from_pretrained("dataeaze/dataeaze-text2sql-codellama_7b_instruct-clinton_text_to_sql_v1")
# print("model device :", model.device)
tokenizer.pad_token = tokenizer.eos_token
model.eval()
prompt = """ Below are sql tables schemas paired with instruction that describes a task.
Using valid SQLite, write a response that appropriately completes the request for the provided tables.
### Instruction: How many transactions were made by a customer in a specific month?
### Database: RewardsProgramDB61
### Input:
CREATE SCHEMA RewardsProgram;
CREATE TABLE Customer (
CustomerID INT NOT NULL AUTO_INCREMENT,
FirstName VARCHAR(50) NOT NULL,
LastName VARCHAR(50) NOT NULL,
Email VARCHAR(100) UNIQUE NOT NULL,
Phone VARCHAR(20) UNIQUE,
DateOfBirth DATE,
PRIMARY KEY (CustomerID)
);
CREATE TABLE Membership (
MembershipID INT NOT NULL AUTO_INCREMENT,
MembershipType VARCHAR(50) NOT NULL,
DiscountPercentage DECIMAL(5, 2) NOT NULL,
ValidFrom DATETIME,
ValidTo DATETIME,
CustomerID INT NOT NULL,
PRIMARY KEY (MembershipID),
FOREIGN KEY (CustomerID) REFERENCES Customer(CustomerID)
);
CREATE TABLE Transaction (
TransactionID INT NOT NULL AUTO_INCREMENT,
TransactionDate TIMESTAMP,
TotalAmount DECIMAL(10, 2) NOT NULL,
CustomerID INT NOT NULL,
PRIMARY KEY (TransactionID),
FOREIGN KEY (CustomerID) REFERENCES Customer(CustomerID)
);
CREATE TABLE TransactionDetail (
TransactionDetailID INT NOT NULL AUTO_INCREMENT,
TransactionID INT NOT NULL,
ProductID INT NOT NULL,
Quantity INT NOT NULL,
UnitPrice DECIMAL(10, 2) NOT NULL,
PRIMARY KEY (TransactionDetailID),
FOREIGN KEY (TransactionID) REFERENCES Transaction(TransactionID),
FOREIGN KEY (ProductID) REFERENCES Product(ProductID)
);
CREATE TABLE Product (
ProductID INT NOT NULL AUTO_INCREMENT,
ProductName VARCHAR(100) NOT NULL,
UnitPrice DECIMAL(10, 2) NOT NULL,
AvailableQuantity INT NOT NULL,
CreatedDate DATETIME,
PRIMARY KEY (ProductID)
);
ALTER TABLE Membership ADD CONSTRAINT FK_Membership_Customer FOREIGN KEY (CustomerID) REFERENCES Customer(CustomerID);
ALTER TABLE TransactionDetail ADD CONSTRAINT FK_TransactionDetail_Transaction FOREIGN KEY (TransactionID) REFERENCES Transaction(TransactionID);
ALTER TABLE TransactionDetail ADD CONSTRAINT FK_TransactionDetail_Product FOREIGN KEY (ProductID) REFERENCES Product(ProductID);"
"""
input_ids = tokenizer(prompt, padding=True, return_tensors='pt')
outputs = model.generate(
input_ids=input_ids['input_ids'].to(model.device),
attention_mask=input_ids['attention_mask'].to(model.device),
max_new_tokens=3072,
)
generated_query = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(generated_query)
SQL queries are matched against the correct answer, with two types of evaluation
model-index:
- name: dataeaze/dataeaze-text2sql-codellama_7b_instruct-dzsql
results:
- task:
type: text-to-sql
dataset:
name: SPIDER 1.0
type: text-to-sql
metrics:
- name: Execution with Values
type: Execution with Values
value: 64.3
- name: Exact Set Match without Values
type: Exact Set Match without Values
value: 29.6
source:
name: Spider 1.0 - Leaderboard
url: https://yale-lily.github.io/spider
"dataeaze systems" [email protected]