theyorubayesian / masakhane_ciral

masakhane_ciral

This repository guides our submission to the Cross-lingual Information Retrieval for African Languages track hosted at Forum for Information Retrieval Evaluation '23

Setup & Installation

• Create an environment using either Conda or Venv

conda create -n ciral python=3.9 openjdk=11
conda activate ciral

• Clone the repo

git clone --recurse-submodules https://github.com/theyorubayesian/masakhane_ciral.git 

If you already cloned the repo and failed to pull the submodules, you can update by running

git submodule update --init

• Install Pytorch>=1.10 suitable for your CUDA version. See Pytorch

• Install other requirements

pip install -r requirements.txt

• Login to Weights & Biases where we are logging our experiments.

wandb login

• Hack away 🔨🔨

Experiments

1. Training on MS Marco & Reporting Zero-Shot Results on Mr.TyDi Swahili
2. Zero-Shot Evaluation of the Dense Retriever on Miracl Dev Set
3. Finetuning the Dense Retriever on the Miracl Train Set & Generating Rankings for the testA set