Data Science: Transformers for Natural Language Processing (Udemy.com)

I've seen some people commenting so weird things about the lazy programmer on reddit and even here, but this is my second course from him and I absolutely love them! Some people confuse rudeness with bluntness. He is blunt, sharp and to the point. No fluff, no bla bla bla. Just very focused and direct classes. This is one of the best transformer resources I found.

Very good course, as most of the LP courses. For whoever wants to learn, even with few knowledge, it is easy to follow and gives a nice understading about how LLMs work and how to code even a little transformer by itself. I think that there could be some improvement points in those lessons where common LLMs architectures are listed, in particular GPT3, 4 and chatGPT.

Course has a really nice progression of topics that is sensible and well thought out. To expand, this course does not just throw concepts and syntax to your face, you are presented with real-world use-cases as you progress. This is probably one of the most comprehensive transformers courses out there. Good for anyone who wants to dabble and kickstart their journey with transformers, or even for experts who want to learn how to implement transformers from scratch. This course piqued my interest in data science and AI. I can tell that the instructor is really passionate about his work.

It explains step by step how to use and how to fine tune existing transformers using transformers and datasets from Hugging face as an example. Especially valuable is the explanation how to re-format your data so it can be used to fine tune an existing model, since this will give you intuition how to prepare data for fine tuning any model - including those that aren't from Hugging face. The VIP section shows how to build an encoder-decoder transformer from basic deep learning neural network building blocks - so if you've wondered what exactly are Q,K and V or what multi-headed actually means - this makes it clear.

Fantastic course, exactly what I was looking for. I have plenty of previous experience with NLP and deep learning, but that experience was mainly focused around older methods such as RNNs on word2vec/GloVe embeddings and classification models using TF-IDF, and I felt like I was missing a deeper understanding of modern Transformer-based approaches. The beginner/intermediate/advanced split in the course allowed me to quickly get up to speed on the practical application of Transformers via Hugging Face libraries (leveraging existing deep learning knowledge to make the process faster), and to then deeply understand the underlying transformer architecture by implementing it myself (which was the key thing that attracted me to this course over many others). Well worth the price and time investment.

Very structured in teaching the concepts and building upon the concepts one after the other. Lots of concepts being covered and hence very lengthy course and therefore it requires you to commit to spending a consistent time over many days to learn the full course. Excellent for those willing to spend the time to learn the transformer concepts as well as learning the programming skills needed for the same.

This course deserves a more extended review :). The course is divided into three segments - Beginner, Intermediate, and Advanced. The first two segments and part of the "Advanced" segment (Section 7) describe what LLMs are and how to use them with Hugging Face. These topics are well crafted and correctly explained, yet you can probably find similar content elsewhere. If you are interested in this content only, you will not be disappointed, although some very new, important topics (like PEFT) are not covered. What makes this course unique and amazing are sections 8 and 9. In these sections, you will learn how the LLMs are created from scratch. Imagine that you read the "Attention is all you need" paper, and while reading, you can discuss every single detail of the paper and the prerequisites required to understand it with a skillful teacher. Then, when you think that you understand it all, you and your teacher implement your own LLM together in "pure" pytorch, and again - you can discuss every single line of code and corresponding theory. This is what this course is about. In short - if you really want to understand how LLMs work in detail, this is the best place I know to learn it.

The structure of this course - separate sections for beginner, intermediate and advanced students - settles very well with me. No one should feel they did not get their money's worth. The absolute highlight for me was the final section in which we built a seq2seq transformer by combining an encoder with a decoder, all of it done practically from scratch. You're probably not going to find a walkthrough as detailed as this anywhere else online right now. If you really want to know how transformers work at both a high and low level, then this is the course for you.

Great Contents for the people who learns text analytics for the first time like me. I spent a decade in writing Python applications, but have no idea about text analytics. Unfortunately there are many code samples online about transformers but it's really hard to find a course explaining why I should call that function in that line. That makes me very disappointed and humiliated because I feel like I am writing what I cannot understand. However, I found this course and really really enjoying this course, even though I am a Korean. (The instructor uses plain and clear English! It's really helpful to me) This course is really easy to follow and understand if you understand how to write code in Python. Thanks a lot, Lazy Programmer! (Yes, Laziness is the most important virtue for programmers, because it pushes programmers to automate what they have to do.)

The course is good and perfect as per Lazy programmer's brand. I can understand their intent on not sharing the notebooks in the first place. They want us to learn by coding ourselves. But, is coding ourselves a rule for all students in the course? I am a Data Science Manager in a company and I am expected to understand the code flow, but don't have time to execute line by line..In that way, this practice made my life little arduous, inspite of directly learning the concepts....