Overview
Learn about StreamingLLM, an innovative framework developed through collaboration between MIT and Meta, in this 38-minute technical video. Explore how this efficient system enables Large Language Models (LLMs) to process infinite sequence lengths despite being trained with finite attention windows, all without requiring additional fine-tuning. Dive into the technical implementation details through code explanations, referencing both the original arXiv research paper and the official GitHub repository. Gain practical insights into how StreamingLLM modifies and enhances LLM capabilities for improved performance in handling extended sequences.
Syllabus
NEW StreamingLLM by MIT & Meta: Code explained
Taught by
Discover AI