• Resultsgithub.com/comp-internship-project
• StackPython, Transformers
• TechnologiesGoogle CoLab, Hugging Face, OpenAI (for creating datasets)

When I was originally hired at Penten, it was for the primarily purpose of improving the quality of offline models for generating structured text. The project was also tasked to another student and while there was some collaboration, we decided in the end to mostly work separately with myself mainly focusing on Low Rank Adaptation (LoRA) and Quantisation.

See below for the poster (along with a QR code to my video "pitch") that I made for the attached university assignment that explains the project in more detail.

Alternatively, below is also the attached README. However, for a full explanation I would recommend looking through the GitHub page linked above which contains an explanation of LoRA and Quantisation, a history of the datasets that were compiled and manually generated, and all of the results with different combinations of data and strategies.