MIT HAN Lab Courses

Nunchaku Windows安装使用教程

掌握Nunchaku在Windows系统上的安装和使用方法，包括配套文字教程和ComfyUI插件的设置步骤。

• YouTube
• 10 minutes
• Self-Paced
• Free Video

Nunchaku Installation and Usage Tutorial on Windows

Discover how to install and use Nunchaku on Windows, including setup instructions and integration with ComfyUI plugin.

• YouTube
• 11 minutes
• Self-Paced
• Free Video

Efficient High-Resolution Image Synthesis with Linear Diffusion Transformers - SANA

Discover how SANA achieves efficient high-resolution image synthesis using Linear Diffusion Transformers, presented at ICLR 2025 as an oral paper by researchers from MIT HAN Lab.

• YouTube
• 13 minutes
• Self-Paced
• Free Video

QServe: W4A8KV4 Quantization and System Co-design for Efficient LLM Serving

Discover a novel approach to efficient LLM serving through W4A8KV4 quantization and system co-design, reducing memory usage while maintaining performance for faster inference.

• YouTube
• 14 minutes
• Self-Paced
• Free Video

LServe: Efficient Long-sequence LLM Serving with Unified Sparse Attention

Discover how LServe enables efficient long-sequence LLM serving through unified sparse attention, improving performance for large language models with extended context windows.

• YouTube
• 12 minutes
• Self-Paced
• Free Video

EfficientML.ai - Efficient Deep Learning Computing Techniques - MIT 6.5940 Fall 2023

Master efficient deep learning techniques to deploy large models on resource-constrained devices, covering pruning, quantization, neural architecture search, and hands-on LLaMA 2 deployment.

• YouTube
• 2 days 6 hours 2 minutes
• Self-Paced
• Free Video

Accelerating Large Language Models and Generative AI

Discover efficient acceleration techniques for Large Language Models and Generative AI in this insightful talk by Professor Song Han from MIT HAN Lab at the HPCA 2025 EMC2 Workshop.

• YouTube
• 42 minutes
• Self-Paced
• Free Video

LEGO: Spatial Accelerator Generation and Optimization for Tensor Applications

Discover a novel approach to spatial accelerator generation and optimization for tensor applications with MIT's LEGO framework, presented at HPCA 2025.

• YouTube
• 16 minutes
• Self-Paced
• Free Video

On-device Training and Transfer Learning - Lecture 21

Explore advanced techniques for on-device ML training, covering gradient security, memory optimization, transfer learning, and quantization methods for efficient model deployment.

• YouTube
• 1 hour 5 minutes
• Self-Paced
• Free Video

On-device Training and Transfer Learning in Machine Learning - Lecture 21

Dive into advanced on-device ML training techniques, covering gradient security, memory optimization, transfer learning, and quantization methods for efficient model deployment and adaptation.

• YouTube
• 1 hour 5 minutes
• Self-Paced
• Free Video

Distributed Training Methods for Efficient Machine Learning - Part 1

Explore distributed training fundamentals, from data parallelism to advanced techniques like ZeRO and pipeline parallelism, enhancing your ML model deployment capabilities.

• YouTube
• 1 hour 10 minutes
• Self-Paced
• Free Video

Knowledge Distillation in Efficient Machine Learning - Lecture 9

Explore knowledge distillation techniques for efficient machine learning, covering theory, applications, and practical implementation strategies.

• YouTube
• 59 minutes
• Self-Paced
• Free Video

Knowledge Distillation in Efficient Machine Learning - Lecture 9

Explore knowledge distillation techniques for efficient machine learning, covering theory, applications, and practical implementation strategies.

• YouTube
• 58 minutes
• Self-Paced
• Free Video

Neural Architecture Search Part II - Lecture 8

Explore advanced neural architecture search techniques, including evolutionary algorithms and reinforcement learning for optimizing deep learning models.

• YouTube
• 1 hour 16 minutes
• Self-Paced
• Free Video

Neural Architecture Search Part II - Lecture 8

Explore advanced neural architecture search techniques, including evolutionary algorithms and reinforcement learning, to optimize deep learning models for efficiency and performance.

• YouTube
• 1 hour 16 minutes
• Self-Paced
• Free Video