This repository provides the original PyTorch implementation for training and evaluating empathetic dialogue models, based on the EmpatheticDialogues dataset. It targets researchers and developers working on conversational AI, offering a benchmark and dataset for building more emotionally intelligent chatbots.

How It Works

The project implements transformer-based and BERT-based retrieval models for dialogue generation. It supports pre-training and fine-tuning these models on the EmpatheticDialogues dataset, which comprises 25,000 conversations grounded in emotional situations. An optional "EmoPrepend-1" classifier can be used to prepend emotional labels, further enhancing response empathy.

Quick Start & Requirements

• Install: Clone the repository and install dependencies.
• Prerequisites: Python (tested on 3.x), PyTorch (1.0.1.post2), numpy (1.14.3), tqdm (4.19.7). Optional: fairseq, fastText, pandas, pytorch-pretrained-BERT.
• Dataset: Download EmpatheticDialogues dataset from wget https://dl.fbaipublicfiles.com/parlai/empatheticdialogues/empatheticdialogues.tar.gz.
• Models: Pre-trained and fine-tuned models are available for download.
• Documentation: Commands for pre-training, fine-tuning, and evaluation are provided in the README.

Highlighted Details

• Transformer and BERT-based retrieval models.
• Support for pre-training and fine-tuning.
• Evaluation metrics include P@1, 100, and BLEU.
• EmoPrepend-1 classifier for enhanced empathy.

Maintenance & Community

This repository is from Facebook AI Research. Specific community channels or active maintenance status are not detailed in the README.

Licensing & Compatibility

The repository is released under a permissive license (LICENSE file). It appears compatible with commercial use and closed-source linking.

Limitations & Caveats

The dependencies listed are for tested versions (e.g., PyTorch 1.0.1.post2), which may be outdated. The project relies on specific versions of libraries like pytorch-pretrained-BERT and fairseq, potentially requiring careful environment management for compatibility.