UMWP

Introduction

Dataset

The UMWP dataset is located in the StandardDataset.json file.

When using it, please adhere to the CC-BY-SA-4.0 license.

Below are two examples of the data:

Answerable Question:

{
    "id": 1,
    "question": "Bryan took a look at his books and magazines. If he has 9 books and 46 magazines in each of his 10 bookshelves.How many magazines does he have in total?",
    "answer": [
        460.0
    ],
    "answerable": true,
    "category": null,
    "relevant_ids": null,
    "source": "SVAMP"
}

Unanswerable Question:

{
    "id": 3226,
    "question": "At the arcade Dave won more than 11 tickets. If he spent 5 tickets on a beanie and later won 10 more tickets, how many would he have? ",
    "answer": null,
    "answerable": false,
    "category": 2,
    "relevant_ids": [
        726
    ],
    "source": "MultiArith"
}

Installation

Python 3.9

conda create -n UMWP python=3.9
conda activate UMWP
pip install -r requirements.txt

Run

Here is an example of generating the output of the gpt-3.5-turbo-0613 model under the ICL input form with Temperature=0.7. sk-xxx is your openAI API-KEY.

python run.py --input-form ICL --model-name gpt-3.5-turbo-0613 --temperature 0.7 --API-Key sk-xxx 

There are three input forms: Direct, Instruction, and ICL.

Available models are listed in the run.py. You are free to add your own model.

Evaluation

Here is an example of evaluating the output of the text-davinci-003 under the Direct input form with Temperature=0.7.

python eval.py --filename text-davinci-003_Direct_text-davinci-003_T_0.7.jsonl

Another Example:

python eval.py --filename llama-v2-13b-chat_ICL_T_0.7.jsonl