Get To Grips With Transformers And LLMs
Written by Nikos Vaggalis Tuesday, 20 January 2026
This isn’t just a course, it’s the complete curriculum of Stanford’s CME295 Transformers and Large Language Models from Autumn 2025.
It’s a course build in the open that holds nothing back. As such the videos themselves are the actual recordings of the on-campus lectures. The slides of the lectures are also available. There’s no homework, however there’s two exams: a midterm and a final which questions as well as their solutions are available too.
Material wise, the course teaches the fundamentals of AI which are necessary to build a solid background. A solid background subsequently will allow navigating the AI landscape with confidence, understand the terminology and utilize that knowl…
Get To Grips With Transformers And LLMs
Written by Nikos Vaggalis Tuesday, 20 January 2026
This isn’t just a course, it’s the complete curriculum of Stanford’s CME295 Transformers and Large Language Models from Autumn 2025.
It’s a course build in the open that holds nothing back. As such the videos themselves are the actual recordings of the on-campus lectures. The slides of the lectures are also available. There’s no homework, however there’s two exams: a midterm and a final which questions as well as their solutions are available too.
Material wise, the course teaches the fundamentals of AI which are necessary to build a solid background. A solid background subsequently will allow navigating the AI landscape with confidence, understand the terminology and utilize that knowledge for building AI powered products.
Spanning 9 lectures, it covers everything that needs to be known:
- Transformers
- Tokenization, attention, positional embeddings
- Decoding, MoE, scaling laws
- LoRA, RLHF, fine-tuning
- RAG, tool calling, evaluation
- RoPE, quantization, optimization tricks
In more detail, the first three lectures explore topics ranging from Tokenization and Embeddings, Word2vec, BERT and its derivatives to Prompting, in-context learning and Chain of thought.
Chapters 4 and 5 cover LLM training and fine tuning; Pretraining, Quantization RLHF, DPO.
The next three chapters cover LLM reasoning, Agentic LLMs and LLM evaluation.Topics: Reasoning models, Retrieval-augmented generation, Function calling, LLMs-as-a-judge.
The course wraps it up by looking at the current trends and what the future holds.
So in the end when should you take this class? You should take it if you have an existing overall understanding of linear algebra concepts, basic machine learning and Python, and you’re looking to understand how the Transformer architecture works as well as become informed on the ongoing LLM trends.
That said, even if you’re not interested in the workings of an LLM but are instead a practiocioner, lectures 7 Agentic LLMs and 8 LLM evaluation, focusing on those hot and trendy subjects, can be watched in isolation from the rest of the material, and will answers many questions like :
- Retrieval-Augmented Generation (RAG) primarily solves which limitation of frozen LLMs?
- What does the Model Context Protocol (MCP) aim to standardize?
- What’s a key difference between a standard chatbot and an “agent” ?
- What is the A2A (Agent2Agent) protocol designed to facilitate?
- RAG vs. long context - Why might one use RAG even if the model has a 1M token context window?
- What is the Tool calling workflow?
- What does “LLM-as-a-judge” typically involve?
Taking into account that the inner workings of an LLM can be a hugely intimidating topic, in retrospect I found this course pretty easy to follow because it explains the concepts and jargon very well, even if you have little experience on the subject.
More Information
Related Articles
To be informed about new articles on I Programmer, sign up for our weekly newsletter,subscribe to the RSS feed and follow us on Facebook or Linkedin.
Comments
or email your comment to: comments@i-programmer.info
Last Updated ( Tuesday, 20 January 2026 )