GPTImpl

Ask ChatGPT to implement your Python functions for you.

Installation

A distribution is available on PyPI.

pip install gptimpl

Usage

Since we interact with OpenAI, the OpenAI API key is a run-time dependency for gptimpl. You can either provide it per-invocation, i.e.

OPENAI_API_KEY="<key>" gptimpl -v example.py

or you can export it before running any gptimpl commands with:

export OPENAI_API_KEY="<key>"

gptimpl --help

usage: gptimpl [-h] [-v] [-w] [-m MODEL] FILE [FILE ...]

Generate implementations for Python functions based on docstrings and function signature using OpenAI.

positional arguments:
  FILE                  The Python files to process.

optional arguments:
  -h, --help            show this help message and exit
  -v, --verbose         Specify the verbosity of the output.
  -w, --overwrite       If specified, the generated files will be overwritten instead of logging to stdout.
  -m MODEL, --model MODEL
                        Select the backend model to use. Defaults to gpt-3.5-turbo

DISCLAIMER: Please use the -w flag only if you want to overwrite your source files and they are already version controlled! The GPT generated code can be garbage sometimes so it is recommended to overwrite only if you have a way to revert the changes.

Updating the function body in-place

Suppose you have a Python file called example.py that contains unimplemented functions as follows:

def fibonacci(n: int) -> int:
    """
    Return the n-th fibonacci number.
    """

def estimate_pi(n: int) -> float:
    """
    Estimate Pi using Gregory-Leibniz series
    using the first n terms.
    """

Then we can pass this file through gptimpl to generate the implementations for us.

Note: Without --overwrite, it defaults to printing the replacement contents to stdout.

gptimpl example.py -v --overwrite

This overwrites the file with the following patch:

-def fibonacci(n: int) -> int:
+def fibonacci(n: int) ->int:
     """
     Return the nth fibonacci number.
     """
+    if n <= 1:
+        return n
+    else:
+        return fibonacci(n - 1) + fibonacci(n - 2)
 
 
-def estimate_pi(n: int) -> float:
+def estimate_pi(n: int) ->float:
     """
     Estimate Pi using Gregory-Leibniz series
     using the first n terms.
     """
+    pi = 0
+    for i in range(n):
+        pi += (-1) ** i / (2 * i + 1)
+    return pi * 4

resulting in an example.py that looks like:

def fibonacci(n: int) ->int:
    """
    Return the nth fibonacci number.
    """
    if n <= 1:
        return n
    else:
        return fibonacci(n - 1) + fibonacci(n - 2)


def estimate_pi(n: int) ->float:
    """
    Estimate Pi using Gregory-Leibniz series
    using the first n terms.
    """
    pi = 0
    for i in range(n):
        pi += (-1) ** i / (2 * i + 1)
    return pi * 4