This is another Coursera offering, given in conjunction with DeepLearning.ai and AWS. Based on the overview clip, it’s over my head as it requires knowledge of Python programming language as well basic data science and machine learning concepts. Some experience with Pytorch and/or Tensorflow is recommended. Yikes.
The previous course from DeepLearning was steeped in Python but was so well presented, I didn’t feel overwhelmed. So I’m going to forge ahead and see how I fare.
Fast forward a few weeks to my completion! I strongly suggest paying the fee for the course as it will not only give you a certificate but also, more importantly, allow you to get your hands dirty. If you audit the class for free, you’ll just watch the instructor move through the lab. However, if you can afford the $49.99, you can interact with the notebook.
The Python is already written (thank goodness) and it’s a manual step-through. However, it was the experience of using AWS’ Sagemaker Studio and the Jupyter Notebook that thrilled me (I’m easily excitable looking at all the shiny tools this new world has to offer 😁). For those more adventurous coders, I believe you can alter the code in the notebook to see how your adjustments modify the outcomes. I may revisit the lab and see if I can fiddle around a bit more.
Briefly, the class is a close look at the power of LLMs as a development tool. It takes you through the technical details of shaping an LLM for your specific purpose. Training and tuning an LLM can be resource-intensive and therefore expensive. You’ll get to see a method used to trim costs without sacrificing a lot of data integrity.
The class is 3 weeks, each containing a series of video shorts. Each week culminates with a hands-on lab and an exam. The exam is graded, but don’t fear–you can review your errors and re-take it.
It’s an excellent course, though it showed me how small a part prompt engineering plays in the LLM development picture. Seems I will have to conquer my fear of Python and try my hand at learning to code. Will investigate and report back.