NLU-11: Week 6

In the second half of the course, we will take a look in more depth at machine translation and at more applications of the techniques we've learned in the first half. This week, we will look evaluation of machine translation and generation, and how to model multilingual machine translation and question answering. 

Note from 26/2: Coursework 2 is now (pre)released - please read the instructions carefully (the main file to download is cw2.zip, which contains all files from the coursework as exposed on gitlab). Please fill in the partner form as announced by Friday 28/2 at 13:30.

This week's tutorial is on transformers:

Tutorial 2: Transformers

Lab 3

Please start having a look at the lab for next week, esp. Section 1.

Lectures

• Lecture 1: Instruction finetuning and RLHF [pdf]

  Required reading:

  Direct preference optimization: Your language model is secretly a reward model, Rafailov et al. (2023), sections 1, 3, 4

   

• Lecture 2: Summarisation [pdf]

  Required reading:

  A Systematic Survey of Text Summarization: From Statistical Methods to Large Language Models, Zhang et al. (2024), sections 1-2

  Optional reading:

  Same paper as above, sections 3-5

   

• Lecture 3: Parameter-efficient finetuning [pdf]

  Required reading:

  LoRA: Low-Rank Adaptation of Large Language Models, Hu et al. (2021), sections 1-4

  Optional reading:

  Parameter-Efficient Fine-Tuning for Large Models: A Comprehensive Survey, Han et al. (2024) (This paper contains much more material than what we cover, and can be skimmed through.)