zpwiki/pages/interns/oliver_pejic/README.md

---
title: Oliver Pejic
published: true
taxonomy:
    category: [iaeste]
    tag: [hatespeech,nlp]
    author: Daniel Hladek
---

Oliver Pejic

IAESTE Intern Summer 2024, 12 weeks in August, September and October.

Goal:
 
- Help with the [Hate Speech Project](/topics/hatespeech)
- Help with evaluation of sentence transformer models using toolkit [MTEB](https://github.com/embeddings-benchmark/mteb) 

Final Tasks:

- Prepare an MTEB evaluation task for [Slovak HATE speech](https://huggingface.co/datasets/TUKE-KEMT/hate_speech_slovak).
- Prepare an MTEB evaluation task for [Slovak question answering](https://huggingface.co/datasets/TUKE-KEMT/retrieval-skquad).
- [Machine translate](https://huggingface.co/google/madlad400-3b-mt) an SBERT evaluation set for multiple slavic languages.
- Write a short scientific paper with results.

Meeting 3.10.:

State:

- Prepared a pull request for Retrieval SK Quad.
- Prepared a pull request for Hate Speech Slovak.

Tasks:

- Make the pull request compatible with the MTEB Contribution guidelines. Discuss it when it is done.
- Submit pull requests to MTEB project.
- Machine Translate a database (HotpotQA, DB Pedia, FEVER) . Pick a database that is short, because translation might be slow.

Non priority tasks:

- Prepare databse  and subnit it to HuggingFace Hub.
- Prepare a MTEB PR for the databse.

Meeting 3.9:

State: Studied MTEB framework and transformers.

Tasks:

- Prepare and try MTEB evaluation tasks for the database. For evaluation you can try me5-base model. 
- Make a fork of MTEB and do necessary modification, including the documentation references for the task.
- Prepare 2 GITHUB pull requests for the databases, preliminary BEIR script given.

Future tasks:

- Prepare a machine translation system to create another slovak/multilingual evaluation task from English task. 

Preparation (7.8.2024):

- Get familiar with [SentenceTransformer](https://sbert.net/) framework, study fundamental papers and write down notes.
- Get familiar with [MTEB](https://github.com/embeddings-benchmark/mteb) evaluation framework.
- Prepare a working  environment on Google Colab or on school server or Anaconda.
- Get familiar with [existing finetuning scripts](https://git.kemt.fei.tuke.sk/dano/slovakretrieval).