Enhancing Chain-of-Thoughts Prompting with Iterative Bootstrapping in Large Language Models

This repository is the official implementation of "Enhancing Chain-of-Thoughts Prompting with Iterative Bootstrapping in Large Language Models".

Our paper is accepted by NAACL 2024 as findings! 🥳🥳🥳

Pipeline of Iter-CoT

Requirements

To install requirements:

pip install -r requirements.txt

Please put your OpenAI API KEY to the api_key variable and the model to use to the model variable in utils.py.

Datasets

The eleven datasets among three different reasoning tasks are in /Iter-CoT/dataset/. In particular, for the Date Understanding without training set, we sampled a small portion of the test set as the training set (see the paper for the details).

Inference

Iter-CoT

To employ Iter-CoT to generate the demonstrations, you can use the following command:

iter_num=1
python run.py --dataset "gsm8k" --iter_num $iter_num

The demonstrations will be saved in output/gsm8k/run/iter1_fewshot.json, which will be used in the inference stage.

If you want to use the demonstrations generated after multi iterations, change the iter_num and use the following command:

iter_num=2 # 3,4,5...
for ((i=1; i<=$iter_num; i++));
do
    python run.py --dataset "gsm8k" --iter_num $i;
done

The demonstrations generated in each iteration will be saved in output/gsm8k/run/iter$i_fewshot.json, which will be used in the inference stage.

Important arguments

--dataset: The name of a dataset. Choices = [gsm8k, addsub, svamp, asdiv, singleeq, aqua, csqa, stqa, date, object_tracking, letter]
--method: The chain-of-thought method. Choices = [iter-cot, zero_shot]
--iter_num: The iter_num iteration in Iter-CoT.
--shot_num: The number of demonstrations used in inference stage.
--stage: The stage of Iter-CoT

Evaluation

To evaluate the performance of the demonstrations on different datasets, you can use the following command:

iter_num=1
python inference.py --dataset "gsm8k" --iter_num $iter_num  --method "run" --shot_num 8

The output will be saved in output/gsm8k/run/iter1_inf_output.json and the statistic result will be saved in output/gsm8k/run/iter1_inf_result.json.

Experiment Results

Contributing

MIT License

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.

GasolSun36/Iter-CoT