Important Details

  • Location: DH A302
  • Time: Tuesdays and Thursdays 11 AM - 12:20 PM
  • Instructor email: llms-11-766@andrew.cmu.edu

Course Description

Large language models are used today for many more applications than just modelling language. In this course, we learn how Transformer-based neural language models have been adapted for applications ranging from web agents and retrieval systems to music generation and coding and writing assistants. We will systematically examine the ways different applications adapt the base technologies underlying large language models. For each application, we will aim to answer the following questions:

  • Can the application be developed with a general-purpose frontier language model, or is application-specific training beneficial or necessary?
  • If training: what do the data, model architecture, and training procedure look like?
  • What is the computational infrastructure (for example, inference algorithms, output filtering, user interfaces) one needs to build around the LLM to support the application?
  • What does evaluation of the LLM’s usefulness look like for this application?
  • What are the limitations and open challenges for applying LLMs to the application?

Learning Goals

Students who successfully complete this course will be able to:

  • Determine the best LLM tools for the diversity of different applications
  • Understand how these tools are implemented and feel confident in making modifications to these implementations
  • Read and comprehend recent, academic papers on customizing LLMs for different applications
  • Design a novel LLM system for an application of their choice

Prerequisites

This class is intended for students who have already learned the basics of how language models are implemented and are curious to develop an understanding of the advanced techniques used for customizing them to different applications. We expect students to have already taken 11-667, 11-711, 10-423, or another class that teaches language model fundamentals. If you have questions about whether you meet this prerequisite, please reach out th ecourse instructors at llms-11-766@andrew.cmu.edu.

Class Format

Classes will be in person, every Tuesday and Thursday 11:00PM-12:20PM in DHA302.

Readings: There will be reading materials for each lecture, which students are required to read through before the class.

Interactive Activities: There will be interactive activities interspersed through the lectures. These will include discussions of the course context and short quizzes.

Assessment

Homeworks: There will be three homework assignments, to be completed individually. The homeworks will comprise a mixture of written and coding questions.

Final Project: Students will complete a final project that focuses on exploring in depth a specific application of their choice.

Scribe Notes: Each student will be assigned one lecture to act as scribe and prepare notes to share with the rest of the class one week after the lecture. Notes should be written in a prose style and will be evaluated for completeness, including content and references. In the event that there are more students than lectures, students will be assigned to scribe in pairs.

Grading: Your grade will be computed as:

  • 30%: Homeworks
    • Each homework is worth 10% of your grade.
  • 40%: Project
  • 20%: Scribe notes
  • 10%: In-class participation

Late Policy

For each unexcused day your homework is late, we will subtract 5% from your final grade for the homwork. For example, if you submit your homework 1 minute late, and your grade would otherwise be a 97%, it will drop down to 92%. If you sumbit 26 hours late, and your grade would other be 90%, it will drop down to 80%.

You may ask for an excused late day or days any time before the deadline. If you excuse is approved, your grade will not be penalized. Valid excuses are medical and personal health issues and other extenuating life circumstances. Requests for excused late days may be emailed to llms-11-766@andrew.cmu.edu.

In the event of a medical emergency, please make your personal health, physical and mental, your first priority. Seek help from medical and care providers such as University Health Services. Students can request medical extensions after the deadline with proof/note from providers.

Accomodations

If you have a disability and require accommodations, please contact Catherine Getchell, Director of Disability Resources, 412-268-6121, getchell@cmu.edu. If you have an accommodations letter from the Disability Resources office, we encourage you to discuss your accommodations and needs with us as early in the semester as possible. We will work with you to ensure that accommodations are provided as appropriate.

Policy on Missing Class

We will try to record classes, but do not offer a guarantee of this. If you must miss class, you can consult with the scribe notes from the class. We do plan to make classes interactive, so please try to attend.

Academic Integrity

Please take some time to read through CMU’s Academic Integrity Policy. Students who violate this policy will be subject to the disciplinary actions described in the Student Handbook.

Collaboration on Homeworks

The six homeworks should be completed individually. However, we encourage you to ask questions on Piazza and in office hours. While you may discuss strategies amongst yourselves, all experiments and analyses should be your own.

Use of Language Models

Using a language model to generate any part of a homework answer without attribution will be considered a violation of academic integrity. This means that if you use ChatGPT or CoPilot to assist you on a homework, you must state so explicitly within your response. On each homework, you will be asked to attest to whether you used AI systems to assist on the homework, and if so, in what manner. If you have used AI systems to generate any part of your homework, you must submit the prompts/instructions/inputs you used to obtain the generated output. Your grading will be based on both the correctness of your homework response and the quality of your prompts/instructions. Errors in the generated outputs that appear in your homework response, and non-interesting prompts, e.g., merely putting in the homework questions to the language model, are not intellectual efforts and are unlikely to receive a good grade.