DEV Community

Manjunatha Sai Uppu
Manjunatha Sai Uppu

Posted on

Generative AI for Everyone on AWS

Vote of Thanks
Thanks to Ayyanar Jeyakrishnan for provididing this content at The Meetup


Path to GenAI

we came from descriptive analytics and then to predictive analytics and thne to word embeddings
and then we have generative ai now.


Let's Understand Basics
A word is tokenized first and then embedded as a vector and then we can apply the alogrithms that we want to apply like semantic search or clustering.


Foundation Models that are available on AWS Sagemaker to jumpstart for self managed access.

Publicaly available

  • Stability AI
  • Alexa
  • HuggingFace Models

Properitary Models

  • Co:here
  • Lighten
  • A121 Labs

Generative AI on AWS

  • AWS Bedrock
  • Amazon EC2 Trn1n and Amazon EC2 inf2
  • Amazon Code whisperer

Amazon Bedrock Foundation Models

  • AI21 labs - (Jurassic-2)
  • ANTHROPIC - (Claude)
  • Stability AI - (stable diffusion)
  • Amazon - (Amazon Titan)

on AWS we can try connecting with LLM by going through the Sagemaker foundation model - playground console


Things we can do by prompting and only using LLM

  1. Text Generation
  2. Summarization
  3. Translation
  4. Code Generation
  5. Question and Answering

Choice of Choosing LLM should depend on these four things

  1. Quality
  2. Cost
  3. Latency
  4. Customization

Prompting Tricks

  1. Zero shot Prompting - Decribing task that LLM needs to do without providing any examples

  2. One Shot Prompting - Describing Task that LLM Needs to do along with one example

  3. Few Shot Prompting - Describing Task along with Some examples to rely on

Note: Few shot > one shot > zero shot prompting


Challenges that we have while adapting LLM's to Enterprises

  • Pretraining Model from scratch
  • Data Privacy and Security and Ethics
  • Model Interpretability and hallucination
  • Fine Tuning and Customization
  • Continual Model Improvement and Integration.


Top comments (0)