utilityai

This package brings language model capabilities into the coding environment, providing a variety of functionalities such as:

• Message and ask anything
• Chat and have a conversation
• Chat about a function
• Chat about a numpy array
• Chat about a pandas dataframe
• Chat about a pytorch tensor
• Generate a function interactively
• Generate a function by setting data within the code
• Generate a function and provide a comment on the result for guided generation

Function generation feature iteratively builds and refines functions by evaluating them against predefined test cases.

This package utilizes a local private LLM:

• Your code is not shared with any external service provider, guaranteeing complete privacy during the generation process.
• Ensuring that your generation process remains entirely within your own environment, without the need for a network connection.

install

pip install utilityai

quick start

Download the model once after installation

from utilityai.model import download
download()

Message and ask anything

from utilityai.chat import message
message("How do you transpose a PyTorch tensor?");

Chat and have a conversation

from utilityai.chat import message
r1, c1 = message("What do mutable and immutable mean?")
message("Give more examples.", c1);

Chat about a function

from utilityai.chat import message
def list_sum(numbers):
    return sum(numbers)
r1, c1 = message("What does this do?", attachment=list_sum)
message("Return the minimum and maximum values of the numbers instead.", c1);

Chat about a numpy array

from utilityai.chat import message
import numpy as np
array = np.array([[1, 2, 3, 4], 
                  [5, 6, 7, 8], 
                  [9, 10, 11, 12]])
r1, c1 = message("Each row represents the salary of a person. How do I calculate the average salary of each person in another array?", attachment=array)
message("How do I calculate the average salary of these people for each year in an array?", c1);

Chat about a pandas dataframe

from utilityai.chat import message
import pandas as pd
data = {
    'Name': ['Alice', 'Bob', 'Charlie', 'David', 'Eve'],
    'Age': [24, 27, 22, 32, 29],
    'Salary': [50000, 54000, 49000, 62000, 58000],
    'Department': ['HR', 'Engineering', 'Marketing', 'Finance', 'Engineering'],
    'Joining Date': pd.to_datetime(['2020-01-15', '2019-06-23', '2021-03-01', '2018-11-15', '2020-08-30'])
}
df = pd.DataFrame(data)
r1, c1 = message("How to calculate the average salary?", attachment=df)
message("How to calculate the average salary for each department?", c1);

Chat about a pytorch tensor

from utilityai.chat import message
import torch
tensor = torch.tensor([[1, 2, 3], [4, 5, 6]])
r1, c1 = message("How to transpose this tensor?", attachment=tensor)
message("How to determine the size of the resulting tensor?", c1);

Generate a function interactively by calling data() first, then provide function information

from utilityai.code import InputData, function
data = InputData()
data()
function(data);

Generate a function by setting data within the code

from utilityai.code import InputData, function
data = InputData()
data_dict = {
    'function_name': 'prime_number_checker',
    'input_names': ['num'],
    'input_types': ['int'],
    'output_names': ['is_prime'],
    'output_types': ['bool'],
    'description': "A function to check if a given number is prime.",
    'test_cases': [
        {'inputs': [5], 'outputs': [True]},
        {'inputs': [10], 'outputs': [False]},
        {'inputs': [17], 'outputs': [True]}
    ]
}
data.set_data(data_dict['function_name'], data_dict['input_names'], data_dict['output_names'], data_dict['input_types'], data_dict['output_types'], data_dict['description'], data_dict['test_cases'])
function(data);

Generate a function and provide a comment on the result for guided generation

from utilityai.code import InputData, function
data = InputData()
data_dict = {
    'function_name': 'vague_function',
    'input_names': ['a', 'b'],
    'input_types': ['int', 'int'],
    'output_names': ['subtract'],
    'output_types': ['int'],
    'description': "A function to subtract two numbers.",
    'test_cases': [
        {'inputs': [1,2], 'outputs': [1]}
    ]
}
data.set_data(data_dict['function_name'], data_dict['input_names'], data_dict['output_names'], data_dict['input_types'], data_dict['output_types'], data_dict['description'], data_dict['test_cases'])
res = function(data, max_tries=1)
res.comment = "A function that subtracts the smaller number from the larger one."
function(data, res);