This repository consists of implementations for different methods to solve Mathematical Word Problems.
Note: All the above diagrams were made with Excalidraw
The methods implemented for this task include:
- FFN Encoder, LSTM Decoder
- RNN Encoder, LSTM Decoder
- Transformers
- Goal-Driven Tree Structured : https://www.ijcai.org/proceedings/2019/0736.pdf
- Graph-to-Tree Learning : https://aclanthology.org/2020.acl-main.362.pdf
LLMs for comparison:
- Gemini-2-9B
- Baseline
- Standard prompting with fine-tuning
- Few-shot prompting with fine-tuning
- Inference-side Chain Of Thought reasoning (CoT)
- Inference-side CoT + Few-shot prompting
- Mistral Instruct 7B
- Baseline
- Standard prompting with fine-tuning
The code and results for all the baseline models can be found in the Baseline Models directory.
The code and results for LLMs can be found in the LLMs directory.
A comprehensive report for the project can be found here: Report
The pre-trained models for all of the above can be found here.