FAQ
Q: I have never taken machine learning, deep learning, or NLP courses. Can I enroll in the class?
A: Please check the prerequisite. We expect students taking prior deep learning courses before (e.g. 11785). 11785 itself usually requires 20hrs a week. We would not recommend taking this course without taking prior deep learning courses (since you will spend over 30 hours on this course in that case). Please use your judgment to determine whether you will be able to complete the course.
Q: Could you share a list of resources and papers that would help me be ready for class?
A: The best resource for cuda programming is Programming Massively Parallel Processors, 3rd Ed. You may access from CMU library.
Q: Can I use other related courses as the prerequisite for 11785 or 11711?
A: Yes. Graduate level machine learning or NLP courses, e.g. 11685, 11667, 10601, 10701, 10707, 10714, 10715, 10617, 18786, 15642 are acceptable.
Q: I took deep learning on coursera, does it count?
A: The deep learning course on coursera is not equivalent to the 12 credit deep learning course at CMU. It does not contain enough hands-on practice and the project.
Q: What is the requirement for computer system course?
A Knowledge of parallel programming is expected. If you have never taken such course, you may consider take 15-418/15-618: Parallel Computer Architecture and Programming.
Q: how heavy will this course be (in terms of hours/week)?
A: Depending on your prior knowledge and courses, it would be 12 hours if you have already taken both 11785 and 11711, and have knowledge about distributed computing and high performance computing (an undergraduate distributed system class). It will be 20 hours or more if you have no experience in C++/CUDA programming.
Q: What is the prerequisite knowledge in Programming?
A: Students are expected to be familiar with Python, C++ and CUDA. C/C++ programming is required. If you are not familiar with those, please take a C/C++ course before this course.
Q: Is prior knowledge of LLM required?
A: No prior knowledge of LLM is required. But we do expect you to be familiar with NLP tasks (i.e. already taken one NLP course like 11711).
Q: Is this course similar to 11667? What is the difference?
A: 11868 is the second LLM course in the series. 11667 focuses on models, learning algorithms and applications. 11868 focuses on building systems for LLM, including training, serving, and maintaining. System performance, latency, reliability, product aspects are taught in this course.
Q: What kind of jobs/applications can I do after taking the course?
A: You will be able to develop a mini full stack system like OpenAI/Anthropic ChatGPT for your own research and applications. You will get a better chance to join the system and fundamental learning teams of those companies working on LLMs (rather than just prompt engineering).
Q: I do not want to write system low-level code but I still want to learn about LLM for application. Shall I take this course?
A: 11667 will be the recommended one for you.
Q: Are there projects?
A: Yes. There is a team project component.
Q: Can non-Computer Science students take this course?
A: Non-CS students who have taken undergraduate system courses (e.g. distributed system, parallel computing, operating system, database system) and satisfy this course prerequisite are welcome. If you do not have much programming experience in C/C++, it would be impossible to complete some of the homework and the project.
Q: Will this course be offered online?
A: It is in person expected. In person participation and discussion are essential components of the course.
Q: I am on a waiting list. How likely will I be admitted?
A: If you satisfy the prerequisite, please stay in the first two weeks. There will be a vacancy.
Q: May I audit the course?
A: There is no audit option for this course since university requires accommodating all students on the waitlist first and there is a long waitlist currently.