Video ID: mNKOKnSbG4Q
YouTube URL: https://www.youtube.com/watch?v=mNKOKnSbG4Q
Added At: 13-06-25 21:18:51
Processed: No
Sentiment: Neutral
Categories: Education, Tech
Tags: AI, LLaMA, Langchain, Flask, Python, Web App, Image Detection, Natural Language Processing
Summary
• The speaker shares a Flask web app they created during the NYC AI Summit. • The app uses LLaMA and Langchain to connect to a locally hosted large language model or small language model. • The app can respond to text prompts, including asking for information about New York City, and can also use image detection with a vision model.
Transcript
hey everyone uh I wanted to share uh the flask web uh web app that I created during my session at NYC AI Summit last week uh it shows you how you can create uh multimodel uh applications using L AMA and langin uh in order to connect to a locally hosted uh large language model or small language model so here's a quick demo it's a single page uh web app uh that is built using flask and I'm using SQL light DB to store um all these conversation threads so um here if I simply give some instructions I can say uh can you tell me uh about New York City in six sentences so um you can see I'm using uh fi model uh which is a small language model um and you can see it's responding with very accurate uh response here uh same way I can also use um images so let's see if I want to look at here this image here and I can say describe the image just to show you what that image is uh me open it here so it's an image of a kid riding his bike let's see if it can detect that correctly uh so I do have a vision model I'm using lava for uh image detection so if the user uploads the image it uses lava uh if the user uh just uh responds using text then it uses Pi model so you can see here the model is able to detect the image correctly it says in the image there's a young boy riding a small bike it appears to be in motion so you can see the accuracy is pretty good um and like I said uh I have two models that are locally hosted on my machine uh if there's a image uh detected as part of the prompt it uses lava uh in order to Rasen over uh image that is uploaded by the user um so I did upload this uh samp Le to my GitHub page um I can quickly show you the code here in the back end so we have app.py that's the main file um these are all the libraries that you need um you can see I'm using SQL light uh using Alchemy to write to SQL light um and then in terms of uh the model uh if you open llmp uh here's the logic uh for calling the models that I have so I have the pi model uh you can see I'm using the Lang chain Library here um it's calling the pi model uh in order to get the response and then if there is a image uh uploaded by the user then it's going to use the lava model so the code is here for to try it out uh and for those of you who uh don't know what uh AMA is it's an um AI tool that's designed to enable users to um execute large language models like llama uh locally on their machine or locally on the server um they also have a lang chain Library so it's it's really easy to use um if you want to add uh AI capabilities to your existing uh applications or web applications uh here's the site you can see all the models that are available on AMA so you can U download any of these uh and use it uh as part of your application um so you can see the two I have here uh I have the lava which does the image detection it's a vision model and then I have uh five which is just 1.6 GB um do let me know in your comments if you have any questions on this uh thanks for watching