Skip to content

metatatt/llm2uml

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

3 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

marp theme class size style header footer
true
default
invert
14580
img {background-color: transparent!important;} a:hover, a:active, a:focus {text-decoration: none;} header a {color: #ffffff !important; font-size: 40px;} footer {color: #148ec8;}
[◇](#1 " ")
Simon Chen 2023

nomnoml:

markedown drawing UML (Unified Modeling Language) diagrams

Snippets

  • 1-train.js: generate nomnoml based on work instruction text
  • 2-train.js: generate text based on nomnoml input
  • 3-openFDA.js: data preparation

Proprietary LLM

.. base model + training

  • base model: GPT-4 or GPT-3.5 turbo
  • training: data: secured repository (Azure Cloud) alogrithm: +vectorstore embedding +finetuinging (prompt engineering) +nascent tools (eg BLIP2, Salesforce)

model #1: data2chart

data: mock wi-320 Tesla Maintenance Manual (https://onedrive.live.com/?cid=597A1F50B291367A&id=597A1F50B291367A%216571&parId=597A1F50B291367A%216234&o=OneUp) Alt text


training design

  • training dataset Alt text

vectorstore: data2chart.index

Alt text

vlidation_tesla


model #2-chart2data

Alt text


model #3-chart 2 codeblock

Alt text


model #4-flows control


rich model

-- multiple iterations / multiple epochs

  • Alt text
  • draft rvision history
  • tabluated content
  • openFDA

openFDA -> auto feed to model

  • adverse events Alt text

  • FDA API endpoint https://api.fda.gov/device/event.json?search=device.generic_name:tomography&limit=1

  • Intervalize // Poll OpenFDA every 60 minutes setInterval(checkAdverseEvents, 60 * 60 * 1000);


Github

Releases

No releases published

Packages

No packages published