l4-project / timelog.md
Ans
Second prototype
ef63b9b

Timelog

  • Text-generation with rhyme and rhythm
  • Ans Farooq
  • 2390370f
  • Jake Lever

Guidance

  • This file contains the time log for your project. It will be submitted along with your final dissertation.
  • YOU MUST KEEP THIS UP TO DATE AND UNDER VERSION CONTROL.
  • This timelog should be filled out honestly, regularly (daily) and accurately. It is for your benefit.
  • Follow the structure provided, grouping time by weeks. Quantise time to the half hour.

Week 1

21 Sep 2021

  • 6 hours Read through project proposals and rank best 8 and enter bids into system

Week 2

28 Sep 2021

  • 2 hours Project Guidance live sessions
  • 1 hour Read the project guidance notes

30 Sep 2021

  • 0.5 hours Initial meeting with supervisor

Week 3

05 Oct 2021

  • 3 hours Read papers on text-generation
  • 2 hours Read blog posts to learn about language models, GPT-3, experiment with EleutherAI's GPT-J model.

06 Oct 2021

  • 0.5 hours Meeting with supervisor

Week 4

12 Oct 2021

  • 6 hours Got started with a colab notebook, learning how to use causal language models to generate text, used GPT-2 and learnt about top-k and top-p sampling, greedy search and beam search.

13 Oct 2021

  • 0.5 hours Meeting with supervisor

Week 5

19 Oct 2021

  • 4 hours Used RoBERTa, a masked language model, to fill in blank words in a sentence, given some starting words and ending words.

20 Oct 2021

  • 0.5 hours Meeting with supervisor

Week 6

26 Oct 2021

  • 3 hours Combined a causal language model and a masked language model, GPT-2 and RoBERTa, to generate the starts of sentences and fill in the rest until the end rhyming word.

27 Oct 2021

  • 0.5 hours Meeting with supervisor
  • 1 hour Created GitHub repository and uploaded dissertation and project template from moodle

29 Oct 2021

  • 0.5 hours Created prototypes directory on GitHub and saved first prototype directly from Colab to GitHub

Week 7

2 Nov 2021

  • 5 hours Used Python library to find rhyming words and wikipedia library to feed topic summary into GPT-2 before text generation

3 Nov 2021

  • 0.5 hours Meeting with supervisor