Skip to content

Developed and implemented a working solution to validate the impact of question decomposition on existing QA Systems using Transformer based models (RoBERTa, T5) using PyTorch. Designed a research project to improve Numerical Reasoning QA over multi-hierarchical data by reimplementing the existing baseline and conducting error-analysis.

Notifications You must be signed in to change notification settings

iamsashank09/question-decomposition-multihiertt

Repository files navigation

Applying Question Decomposition to Numerical Reasoning over Multi Hierarchical Data

Research Report: Applying Question Decomposition to Numerical Reasoning over Multi-hierarchical Data

Please follow the below instructions to run our code:

  1. Download checkpoints from https://drive.google.com/drive/folders/1LPVGoWm1Tsm4Asjc6soeFI3sMTfsAz1_?usp=sharing and place in:

     Checkpoint_3/MT2NET_CodePipeline/MT2NET_models/
    
  2. Download BERT checkpoints from https://drive.google.com/file/d/1XaMX-u5ZkWGH3f0gPrDtrBK1lKDU-QFk/view?usp=sharing and place it in

     Checkpoint_3/QuestionDecomposition_CodePipeline/model/
    
  3. Run main.py which is in the project root folder, we've built a single touch pipeline that runs both Decomposition and the QA pipeline:

     python main.py
    
  4. Decomposed questions can be found at:

     Checkpoint_3/QuestionDecomposition_CodePipeline/decomposed_questions
    
  5. Outputs at each step of MT2NET (QA Pipeline) can be found in the outputs_intermediate folder of the MT2NET_CodePipeline folder at:

     Checkpoint_3/MT2NET_CodePipeline/outputs_intermediate
    

About

Developed and implemented a working solution to validate the impact of question decomposition on existing QA Systems using Transformer based models (RoBERTa, T5) using PyTorch. Designed a research project to improve Numerical Reasoning QA over multi-hierarchical data by reimplementing the existing baseline and conducting error-analysis.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages