With the success of ChatGPT, we now have witnessed a surge in demand for bespoke large language models.
Nonetheless, there was a barrier to adoption. As these models are so large, it has been difficult for businesses, researchers, or hobbyists with a modest budget to customize them for their very own datasets.
Now with innovations in parameter efficient fine-tuning (PEFT) methods, it’s entirely possible to fine-tune large language models at a comparatively low price. In this text, I display tips on how to achieve this in a Google Colab.
I anticipate that this text will prove beneficial for practitioners, hobbyists, learners, and even hands-on start-up founders.
So, if it is advisable to mock up an affordable prototype, test an idea, or create a cool data science project to face out from the group — keep reading.
Businesses often have private datasets that drive a few of their processes.
To offer you an example, I worked for a bank where we logged customer complaints in an Excel spreadsheet. An analyst was answerable for categorising these complaints (manually) for reporting purposes. Coping with hundreds of complaints every month, this process was time-consuming and vulnerable to human error.
Had we had the resources, we could have fine-tuned a big language model to perform this categorisation for us, saving time through automation and potentially reducing the speed of incorrect categorisations.
Inspired by this instance, the rest of this text demonstrates how we will fine-tune an LLM for categorising consumer complaints about financial services.
The dataset comprises real consumer complaints data for financial services and products. It’s open, publicly available data published by the Consumer Financial Protection Bureau.
There are over 120k anonymised complaints, categorised into roughly 214 “subissues”.