Ò»±¾µÀÎÞÂë

Course Number: 11-967

Number of Units: 12 units

This course provides a broad foundation for understanding, working with, and adapting existing tools and technologies in the area of Large Language Models like BERT, T5, GPT, and others.

Throughout this course, you will learn:

• A range of topics including systems, data, data filtering, training objectives, RLHF/instruction tuning, ethics, policy, evaluation, and other human facing issues.
• How transformer architectures work and explore the reasons why they are better than LSTM-based seq2seq, decoding strategies, etc. Through readings and hands-on assignments, you will explore techniques for pretraining, attention, prompting, etc. 
• How to apply the skills you've learned in a semester-long course project, making use of locally sourced model instances that offer the opportunity to explore behind the curtain of commercial APIs.
• How to compare and contrast different models in the LLM ecosystem in order to determine the best model for any task.
• How to implement and train a neural language model from scratch in Pytorch.
• How to utilize open source libraries to finetune and do inference with popular pre-trained language models.
• How to apply LLM’s in downstream applications and how decisions made during pre-training affect suitability for tasks.  
• How to design new methodologies to leverage existing large scale language models in novel ways.

Please note: In order to complete homeworks and activities, students will need to sign up for Amazon Web Services (AWS) or an equivalent service that offers access to A10g or similar GPUs. The AWS cost to complete assignments will range from $150-300 and will vary based on your usage. In addition, students will need to sign up for the OpenAI API. The cost to complete the assignments via OpenAI will be up to $25. Instructions for accessing both services will be provided when the class begins.

Course Number: 11-977

Number of Units: 12 units

In this course, you will learn the fundamental mathematical concepts in machine learning and deep learning that are relevant to the five main challenges in multimodal machine learning: 

1. Multimodal representation learning
2. Translation and mapping
3. Modality alignment
4. Multimodal fusion
5. Co-learning 

The mathematical concepts you will learn include, but are not limited to, multimodal auto-encoder, deep canonical correlation analysis, multi-kernel learning, attention models and multimodal recurrent neural networks.

You will also review recent papers describing state-of-the-art probabilistic models and computational algorithms for multimodal machine learning and discuss the current and upcoming challenges. Finally, you will study recent applications of multimodal machine learning including multimodal affect recognition, image and video captioning and cross-modal multimedia retrieval.

Please note: The Multimodal Machine Learning course may require Amazon Web Services (AWS) and/or OpenAI or other services to complete assignments with fees up to $300 (subject to change). More details will be available as you get closer to the course start date.

Course Number: 11-968

Number of Units: 12 units

LLM's are often very large and require increasingly larger data sets to train, which means developing scalable systems is critical for advancing AI. In this course, you will learn the essential skills for designing and implementing scalable LLM systems.  

Throughout the course, you will:

• Learn the approaches for training, serving, fine-tuning, and evaluating LLM's from the systems perspective.
• Gain familiarity with sophisticated engineering using modern hardware and software stacks needed to accommodate the scale.
• Acquire essential skills for designing and implementing LLM systems, including:
  • Algorithms and system techniques to efficiently train LLM's with huge data
  • Efficient embedding storage and retrieval
  • Data efficient fine-tuning
  • Communication efficient algorithms
  • Efficient implementation of reinforcement learning with human feedback
  • Acceleration on GPU and other hardware
  • Model compression for deployment
  • Online maintenance
• Learn about the latest advances in LLM systems regarding machine learning, natural language processing, and system research.

Please note: The Large Language Models Systems course may require Amazon Web Services (AWS) and/or OpenAI or other services to complete assignments with fees up to $300 (subject to change). More details will be available as you get closer to the course start date.