Universität Wien
Achtung! Das Lehrangebot ist noch nicht vollständig und wird bis Semesterbeginn laufend ergänzt.

136041 SE Open Source Language Models (2024W)

Prüfungsimmanente Lehrveranstaltung

An/Abmeldung

Hinweis: Ihr Anmeldezeitpunkt innerhalb der Frist hat keine Auswirkungen auf die Platzvergabe (kein "first come, first served").

Details

max. 25 Teilnehmer*innen
Sprache: Englisch

Lehrende

Termine (iCal) - nächster Termin ist mit N markiert

  • Dienstag 01.10. 09:45 - 11:15 Hörsaal 2 Hauptgebäude, Tiefparterre Stiege 5 Hof 3
  • Dienstag 08.10. 09:45 - 11:15 Hörsaal 2 Hauptgebäude, Tiefparterre Stiege 5 Hof 3
  • Dienstag 15.10. 09:45 - 11:15 Hörsaal 2 Hauptgebäude, Tiefparterre Stiege 5 Hof 3
  • Dienstag 22.10. 09:45 - 11:15 Hörsaal 2 Hauptgebäude, Tiefparterre Stiege 5 Hof 3
  • Dienstag 29.10. 09:45 - 11:15 Hörsaal 2 Hauptgebäude, Tiefparterre Stiege 5 Hof 3
  • Dienstag 05.11. 09:45 - 11:15 Hörsaal 2 Hauptgebäude, Tiefparterre Stiege 5 Hof 3
  • Dienstag 12.11. 09:45 - 11:15 Hörsaal 2 Hauptgebäude, Tiefparterre Stiege 5 Hof 3
  • Dienstag 19.11. 09:45 - 11:15 Hörsaal 2 Hauptgebäude, Tiefparterre Stiege 5 Hof 3
  • Dienstag 26.11. 09:45 - 11:15 Hörsaal 2 Hauptgebäude, Tiefparterre Stiege 5 Hof 3
  • Dienstag 03.12. 09:45 - 11:15 Hörsaal 2 Hauptgebäude, Tiefparterre Stiege 5 Hof 3
  • Dienstag 10.12. 09:45 - 11:15 Hörsaal 2 Hauptgebäude, Tiefparterre Stiege 5 Hof 3
  • Dienstag 17.12. 09:45 - 11:15 Hörsaal 2 Hauptgebäude, Tiefparterre Stiege 5 Hof 3
  • Dienstag 07.01. 09:45 - 11:15 Hörsaal 2 Hauptgebäude, Tiefparterre Stiege 5 Hof 3
  • Dienstag 14.01. 09:45 - 11:15 Hörsaal 2 Hauptgebäude, Tiefparterre Stiege 5 Hof 3
  • Dienstag 28.01. 09:45 - 11:15 Hörsaal 2 Hauptgebäude, Tiefparterre Stiege 5 Hof 3

Information

Ziele, Inhalte und Methode der Lehrveranstaltung

In this seminar, participants will read, present and discuss recent papers on the capabilities and limitations of language-based AI models.

Possible topics to be covered in the seminar:

A. Foundational work on transformer language models
B. Evaluation and Analysis of LLMs
C. Open Source LLMs, Training, and Corpora
D. Legal Aspects, Copyright, and Transparency

Art der Leistungskontrolle und erlaubte Hilfsmittel

Participants will have to present one topic from the list in the seminar, the presentation should be roughly 25 minutes (hard limits: min. 20 minutes, max. 30 minutes). The presentation is followed by a QA session and discussion. Participants will also have to submit a written report (deadline and exact requirements TBD), describing the main contents of the presented paper(s) - see a list of recommended papers below - and putting it in a wider context.

Mindestanforderungen und Beurteilungsmaßstab

Your presentation will account for 45% of the grade, participation in discussions for 10%, and the written report for 45%.

Prüfungsstoff

Your presentation will account for 45% of the grade, participation in discussions for 10%, and the written report for 45%.

Literatur

---
A. Foundational work on transformer language models

Vaswani, A. "Attention is all you need." 2017

Brown, Tom B. "Language models are few-shot learners." 2020

Holtzman, Ari, et al. "The curious case of neural text degeneration." 2019

Wei, Jason, et al. "Finetuned language models are zero-shot learners." 2021

Ouyang, Long, et al. "Training language models to follow instructions with human feedback." 2022

---
B. Evaluation and Analysis of LLMs

Hendrycks, Dan, et al. "Measuring massive multitask language understanding." 2020
and
Wang, Yubo, et al. "Mmlu-pro: A more robust and challenging multi-task language understanding benchmark." 2024

Zheng, Lianmin, et al. "Judging llm-as-a-judge with mt-bench and chatbot arena." 2023

Biderman, Stella, et al. "Pythia: A suite for analyzing large language models across training and scaling." 2023

Schaeffer, Rylan, Brando Miranda, and Sanmi Koyejo. "Are emergent abilities of large language models a mirage?." 2024

---
C. Open Source LLMs, Training, and Corpora

Zhang, Susan, et al. "Opt: Open pre-trained transformer language models." 2022

Le Scao, Teven, et al. "Bloom: A 176b-parameter open-access multilingual language model." 2023

Groeneveld, Dirk, et al. "Olmo: Accelerating the science of language models." 2024

Soldaini, Luca, et al. "Dolma: An open corpus of three trillion tokens for language model pretraining research." 2024

Wang, Yizhong, et al. "How far can camels go? exploring the state of instruction tuning on open resources." 2023
and
Ivison, Hamish, et al. "Camels in a changing climate: Enhancing lm adaptation with tulu 2." 2023

Wang, Yizhong, et al. "Self-instruct: Aligning language models with self-generated instructions." 2022

Peng, Baolin, et al. "Instruction tuning with gpt-4." 2023

Üstün, Ahmet, et al. "Aya model: An instruction finetuned open-access multilingual language model." 2024

Singh, Shivalika, et al. "Aya dataset: An open-access collection for multilingual instruction tuning." 2024

Rafailov, Rafael, et al. "Direct preference optimization: Your language model is secretly a reward model." 2024

Frantar, Elias, et al. "Gptq: Accurate post-training quantization for generative pre-trained transformers." 2022

Jiang, Albert Q., et al. "Mixtral of experts." 2024

Shen, Yongliang, et al. "Hugginggpt: Solving ai tasks with chatgpt and its friends in hugging face." 2024

Li, Raymond, et al. "Starcoder: may the source be with you!." 2023

---
D. Legal Aspects, Copyright, and Transparency

Lemley, Mark A., and Bryan Casey. "Fair learning." 2020

NYT vs. OpenAI
Complaint by New York Times
and
Response by OpenAI (blogpost, legal response)
https://nytco-assets.nytimes.com/2023/12/NYT_Complaint_Dec2023.pdf , https://openai.com/index/openai-and-journalism/ , https://www.courtlistener.com/docket/68117049/52/the-new-york-times-company-v-microsoft-corporation/

Strowel, Alain. Study on copyright and new technologies: copyright data management and artificial intelligence. 2022

Jernite, Yacine, et al. "Data governance in the age of large-scale data-driven language technology." 2022

Bommasani, Rishi, et al. "The foundation model transparency index." 2023

Zuordnung im Vorlesungsverzeichnis

S-DH Cluster I: Language and Literature

Letzte Änderung: Mo 23.09.2024 14:46