Peiqin Lin

Algorithm Engineer @AlibabaGroup. Previously PhD at @ELLISforEurope @CisLmu @sardine_lab_it, interns at @HongKongPolyU, @TencentGlobal

[Peiqin Lin @lpq29743]

I am at #ACL2023 this week. I'll present our work 'Glot500: Scaling Multilingual Corpora and Language Models to 500 Languages' at Session 5: Large Language Models (Oral), Jul 11 (Tue), 09:00-09:15.

Happy to meet anyone and discuss multilinguality & LLM.

https://t.co/3xRAnCCQ6C

[Peiqin Lin @lpq29743]

@NielKlug @HinrichSchuetze @andre_t_martins @davlanade @vtresp @CisLmu Thanks! Dr. Nie. Wish you the best too!

[Peiqin Lin @lpq29743]

Happy to share that I successfully defended my ELLIS PhD thesis on Massively Multilingual LLMs. Huge thanks to my supervisors @HinrichSchuetze & @andre_t_martins, and @davlanade, @vtresp, Prof. Sven Strickroth for joining my defense. A wonderful jorney at @CisLmu and Sardine Lab!

[Peiqin Lin @lpq29743]

@LAWeissweiler @HinrichSchuetze @andre_t_martins @davlanade @vtresp @CisLmu Thank you Leonie 😀

[Peiqin Lin @lpq29743]

@andre_t_martins @HinrichSchuetze @davlanade @vtresp @CisLmu Thank you André

[Peiqin Lin @lpq29743]

@tozefarinhas @HinrichSchuetze @andre_t_martins @davlanade @vtresp @CisLmu Thank you Antonio

[Peiqin Lin @lpq29743]

@davlanade @HinrichSchuetze @andre_t_martins @vtresp @CisLmu Thank you! David

[Peiqin Lin @lpq29743]

I'm at @naaclmeeting this week, presenting our work on multilinguality. I'm currently looking for research positions and would love to connect and chat about multilinguality and LLMs. Looking forward to seeing many of you there!

[Mengjie Zhao @mengjie_zhao]

Hi #NLProc I am looking for emergent reviewers for this round of ARR -- if you or you know someone who would like to help please let me know, thank you!! 🙏🙏

[Helsinki-NLP @HelsinkiNLP]

🚀 Excited to introduce EMMA-500! 🌍✨
A multilingual model continue-trained on 546 languages, enhancing coverage for low-resource languages. With the MaLA corpus and Llama 2 7B, we're pushing boundaries in cross-lingual transfer. Check it out: https://t.co/FBSbuy3qft

[MedAI Group @MedaiStanford]

This Monday, Kaishuai Xu from Hong Kong Polytechnic University will be joining us to talk about clinical process-aware medical dialogue system. Catch it at 1-2pm PT this Monday on Zoom! Subscribe to https://t.co/Tr8mrytDbz #ML #AI #medicine #healthcare

[Yongkang Liu @liuygkg]

HiFT is open-source! It's a model-independent optimizer-independent full-parameter fine-tuning method that supports full-parameter fine-tuning of 7B models on 24G memory devices. It's based on huggingface and can be used by "pip install hift"
github link: https://t.co/CDmY1DbWBN

[Shaoxiong Ji @shaoxiongji]

🚀 We're thrilled to release a new model checkpoint for MaLA-500, an LLM spanning 534 languages!
🔥 It fills the gap for low-resource languages, outperforming existing models.
📜paper https://t.co/n8obX1sFpN
💻model https://t.co/Z5pI9zUdo5
@hplt_eu @UTTERProject
#LLMs #MaLA500

[Barbara Plank @barbara_plank]

Join us this Friday at the #UncertaiNLP workshop

First keynote will be given by Chrysoula Zerva (IST, Lisbon) on "Uncertainty in NLP: Quantification, interpretation and evaluation"

Location: Bastion 2 room of the Corinthia

Program: https://t.co/P3ewwXfjkF

[Börje Karlsson @tellarin]

Quality multilingual annotated data is always scarce, so I'm extra happy to see ✨Universal NER✨ has been accepted at #NAACL2024.
We hope the project will help address the data gap and facilitate new multilingual/cross-lingual research!
🎉 Preprint: https://t.co/GeFRFOKDfh

[Peiqin Lin @lpq29743]

Happy to share that our paper "mPLM-Sim: Better Cross-Lingual Similarity and Transfer in Multilingual
Pretrained Language Models" (with Chengzhi Hu, @zheyu_bot, @andre_t_martins, @HinrichSchuetze) has been accepted to #EACL2024 Findings. See you in Malta!
https://t.co/gYBYsvMOJo

[Peiqin Lin @lpq29743]

@monojitchou @arnabbiswas1 @davlanade Thanks! I agree with your point. Our MaLA-500 is just a start, and there are still some efforts we need to make, especially for low-resource language evaluation.

[Peiqin Lin @lpq29743]

@arnabbiswas1 @monojitchou @davlanade We used Glot500-c with its data split in https://t.co/DV5MraacwF

There is no multilingual benchmark for 500+ langs, which make comprehensive evaluation difficult. We are going to include model-comparable perplexity, which can have a full coverage on the considered languages.

[Peiqin Lin @lpq29743]

@Geronimo_AI @RealShermanSiu @arankomatsuzaki Thanks for the interests. We will release our code, but not very soon. FYI, our code is based on:
- Training: https://t.co/hWraLIG5tb and https://t.co/dpk8QifYMd.
- Evaluation: https://t.co/RTErKZgw04

[AK @_akhaliq]

MaLA-500: Massive Language Adaptation of Large Language Models

paper page: https://t.co/QUM1B12umI

Large language models have advanced the state of the art in natural language processing. However, their predominant design for English or a limited set of languages creates a substantial gap in their effectiveness for low-resource languages. To bridge this gap, we introduce MaLA-500, a novel large language model designed to cover an extensive range of 534 languages. To train MaLA-500, we employ vocabulary extension and continued pretraining on LLaMA 2 with Glot500-c. Our experiments on SIB-200 show that MaLA-500 achieves state-of-the-art in-context learning results.

[David Ifeoluwa Adelani 🇳🇬 @davlanade]

Really cool work. Extending LLaMa 2 to 500+ languages

Glad to see that our SIB-200 https://t.co/5WrZlRjyt6 was useful for the evaluation of MaLA-500, looking forward to trying out the model.