Nowadays, this repository might not be useful because the Codex models are now deprecated.

If you are interested in OpenAI's new Chat models, please refer to the OpenAI's newer Chat models which are able to do many coding tasks with similar capability.

Creating Problem Jsonl as intended

This repo is about creating HumanEval.jsonl dataset easily.

HumanEval dataset are handwritten programming problems described in 2.2. HumanEval: Hand-Written Evaluation Set section on the following paper.

Paper: "Evaluating Large Language Models Trained on Code".

How to Create Json

1. Renew the parameter of fetch_id function before executing printing.py.

2. Write texts based on the format of HumanEval.jsonl.

3. Execute printing.py then you could see changes in problems.json and samples.json.

4. Add problem.json to HumanEval.jsonl.

input.txt Example

def add(a,b):
    """
    make a program which returns a plus b.
    """

add
    return a+b

def check(candidate):
    # Check some simple cases
    assert candidate(3,4) == 7
    assert candidate(10,11) == 21

Tip

After executing printing.py, it is recommended to modify the part corresponding to importing modules or functions in input.txt in terms of unity as follows:

from module_name import function_name\n\n
import function_name as alias_name\n\n
\ndef

Cautions

1. Be sure the directory where the files, input.txt and printing.py are located in.

2. It might occur errors if the docstrings include in some escape sequences like '\n\n'.

Reference

https:/openai/human-eval