Overview
Save Big on Coursera Plus. 7,000+ courses at $160 off. Limited Time Only!
Learn about LLaVA 1.5, a cutting-edge multi-modal AI system, through this 44-minute technical video that explores its architecture, capabilities, and significance in the AI landscape. Dive deep into the system's core components, including its pre-trained vision encoder and large language model integration, while understanding how it processes and aligns visual features with semantic embeddings. Explore the sophisticated training methodology involving GPT-4 generated datasets comprising 160,000 unique language-image instruction-following samples, and discover the two-stage training process that optimizes both the projection matrix and language model components. Examine the architectural improvements from LLaVA 1.0 to 1.5, including the implementation of a 2-hidden-layer Perceptron for enhanced feature transformation and the integration of VICUNA 13B large language model. Understand the system's modular design that allows for component exchange between the LLM and Vision Transformer, making it particularly valuable for research and practical applications in scientific queries and complex reasoning tasks.
Syllabus
New LLaVA AI explained: GPT-4 VISION's Little Brother
Taught by
Discover AI