[Google Cloud Skills Boost(Qwiklabs)] Introduction to Generative AI Learning Path - 2. Introduction to Large Language Models

https://www.youtube.com/playlist?list=PLIivdWyY5sqIlLF9JHbyiqzZbib9pFt4x 

Generative AI Learning Path

---

Gen AI produces new content, including text, images, audio, and synthetic data.

• Pre-trained: for general purpose with a large data set
• Fine-tuned: for specific aims with a much smaller data set

PaLM is a dense decoder-only transformer model.

• Transformer model (e.g. PaLM)
  • A transformer model consists of encoder and decoder.  
  • Encoder - encodes the input sequence and passes it to the decoder
  • Decoder - decodes the representations for a relevant task

• Traditional programming -> 고양이를 구별하기 위해 규칙들을 하드코딩

• 신경망 기반 접근법 -> 고양이와 개 사진을 미리 주고, 예측하게 함.

• 생성형 접근법 -> generate our own content (e.g. text, images, audio, video, etc..)
  • 생성형 AI 모델들: ingest very large data from multiple sources across the Internet and build foundation language models

 

Main kinds of LLM - 1. Generic Language Models

마치 autocomplete in search

Main kinds of LLM - 2. Instruction Tuned

e.g. 감정 분류(neutral, negative, positive)

Main kinds of LLM - 3. Dialog Tuned

 

Vertex AI provides task-specific foundation models

파인튜닝은 비싸고 비현실적인 작업이다.

⭐️⭐️

Gen AI Studio lets you quickly explore and customize generative AI models.