iMRC_BERT

This repository contains the implementation of a BERT-based model for machine reading comprehension. Below you will find instructions on how to set up and run examples. This project is based on Interactive Machine Comprehension with Dynamic Knowledge Graphs. Our novelty is to use the pre-trained BERT/RoBERTa model to expedite the training while keeping the performance.

How to Run Examples

To run an example with the BERT model, you will need a configuration file in YAML format. Use the following command from the command line:

python main.py <config>.yaml

Replace <config> with the name of your configuration file.

System Architecture

Below is an overview of the system architecture of our agent.

Overview of Agent

The system consists of:

• Frozen LLM: A pre-trained BERT/RoBERTa model that is utilized without further training during operations.
• FC Layers: Fully connected layers.
• Output Actions: The agent generates actions a_t based on the processed inputs. a_t could be: search with a token, go to next/previous sentence/chunk, or end.

This conceptual overview provides a high-level understanding of the interconnections and the operational logic of the agent.