Liam Ottley
#chatgpt #artificialintelligence #openai
Super simple guide on How to Fine Tune ChatGPT, in a Beginners Guide to Building Businesses w/ GPT-3. Knowing how to Fine Tune GPT-3 is one of the most important skills you can have as an entrepreneur to build your own ChatGPT site or get into the AI industry and build a valuable business business.
While limited in scope, this video gives you a clear breakdown of the stages in fine tuning these models.
Kaggle Dataset:
https://www.kaggle.com/datasets/thedevastator/unlocking-the-secrets-of-nba-player-performance
Prompts & Code HERE:
https://docs.google.com/document/d/1jJbk61grxYZV2qbBNtNnVaeEFsqMskWY5Uh6ELlqiB4/edit?usp=sharing
My AI Development & Automation Company 🤖
https://morningside.ai
My AI Consulting Services 👋🏼
Want to speak with me about your next big project? Book here:
https://calendly.com/liamottley/consulting-call
Join My Newsletter! 🗞️
Get your fix of the latest AI news direct to your inbox:
https://www.overtheshoulder.co/subscribe
My AI Entrepreneurship Community 📈
Find top shelf co-founders for your next AI venture:
http://discord.gg/dAa78Zj7yj
My Links 🔗
👉🏻 Subscribe: https://www.youtube.com/channel/UCui4jxDaMb53Gdh-AZUTPAg?sub_confirmation=1
👉🏻 Twitter: https://twitter.com/liamottley_
👉🏻 LinkedIn: https://www.linkedin.com/in/liamottley/
👉🏻 Instagram: https://instagram.com/liamottley
Timestamps:
0:00 – Intro
1:07 – Generating Pairs
7:11 – Fine Tuning w/ API
10:18 – Running Our App
12:38 – Outro
About Me 👋🏼
Hi! My name is Liam Ottley and I’m an AI developer and entrepreneur, founder of my own AI development company: Morningside AI (https://morningside.ai). You’ve found my YouTube channel where I help other entrepreneur’s wrap their heads around AI with tutorials on how to get started with building AI apps!
With developments in AI moving so fast, I’m constantly working on my own projects and those of clients and sharing my learnings with my viewers 2-3x per week.
I don’t have anything to sell you, my only hope is that one day some of you will learn enough from my videos to see an exciting opportunity in the AI space and want to work with my development company to make it a reality.
Great video.
One part is just a little unclear, is the fine-tuning process send all the data for every question each time, or just once, so it keeps the data in a kind of context? Thanks
How can i get / make data for training ?
The fine-tuning explanation is good, but the example given is pretty bad.
The video shows how to fine-tune a LLM to do a very specific task, with no variance. The task is to create a summary of the performances of a NBA player in a given season.
In order to get the data for the fine-tuning process, a python script is created that solves the problem we are trying to solve with the fine-tuned LLM.
Instead of fine-tuning the LLM and wasting money, we could just use the python script, that has access to all the necessary data and isn't prone to errors.
The LLM will spit out the same sentence with different data every time, so why not use the python script that does the same?
So for people who want to use this process:
You will have to figure out how to create variations of these sentences and add those to the training data, otherwise you will only waste money on a very expensive string formatter.
Thank you so much Liam, Keep it up with these great videos
Great video 👍. Though I just can get the model to return the correct information. I ask it for stats for a particular player and it gives me stats for a different one. I have tried several times with different training tutorials, but each one gives me the same wrong or nonsense results. I am wondering if the ai is having a breakdown!!??
Do i need to learn a bit of python to understand what you are doing?
Thank you so much for the video, I am getting issues that "tream interrupted (client disconnected)." Does any one know how to fix it?
Awesome video. Please can this be used for bias evaluation mini-project?
You really did a good job, thank you
Can yo do this with multiple files? I am talking to a customer who has years of data on lumber and materials. Is it possible for me to do this with all his data so that chat gpt can then reference and answer against all the data across all files?
Would you consider something like Stack-AI to be a suitable GUI to abstract away the custom scripting you're doing here? Awesome content, btw. Subbed last week.
I got this error:
Upload progress: 100%|##########| 440/440 [00:00<00:00, 220kit/s]
Error: Billing hard limit has been reached (HTTP status code: 400)
Someon knows what's happening?
sorry guy, that code is enough, you dont need anymore code
In general, if you are stuck with the code, then ask chatgpt
this code to check on the staus: import openai
import json
"""
openai.api_key = 'semkzqmZJkDeuORAleblabla'
openai.FineTune.retrieve("ft-ExmzuSK0WYSblabla")
"""
openai.api_key = 'sk-dNemkORAleblabla'
response = openai.FineTune.retrieve("ft-Exm0WYSblabla")
print(response)
buy the way, I joinned the discord, but almost all channels are closed for acces
I’m used to following videos at 1.5x speed. 😅 You brought me to 3x😅
Hi, great and clear video! How would one go about updating data? Will the model understand if i fine-tune the same basketball player with updated statistics? Or will the model "duplicate" the player?
👉🏻 Join my AAA Accelerator: https://www.aaaaccelerator.com/
👉🏻 Work With Me: https://morningside.ai
👉🏻 My Telegram: https://t.me/+2H8HuWNtLtE2YzEx
👉🏻 My AAA Resources: https://www.liamottleyresources.com
👉🏻 My AAA Playlist: https://www.youtube.com/playlist?list=PL2V6IBDeR5YP_O0sovZZvIs4QgwMIx6ZZ
👉🏻 Subscribe: http://www.youtube.com/channel/UCui4jxDaMb53Gdh-AZUTPAg?sub_confirmation=1
👉🏻 Twitter: https://twitter.com/liamottley_
👉🏻 Instagram: https://instagram.com/liamottley
👉🏻 Consulting: https://calendly.com/liamottley/consulting-call
👉🏻 LinkedIn: https://www.linkedin.com/in/liamottley/
u told just fine-tuning the gpt-3 but you just fine-tuning hte davince/curie . ufff nice bait
I want more technical knowledge for this business to do with smoothly
too fast, cant follow
Is it possible to ask questions in natural language and receive appropriate answers using Fine-Tuned model? For example you asked "What are Karl-Anthony Towns' statistics", but would it work if you ask differently. "what is the stats of Karl-Anthony"? Or should I fine tune model for every possible question?
In the video at 12:37 it says that si Antetokounmpo was introduced in the Hall of Fame in 2037. Is this coming from bad data on the dataset, or?
This is not a good use case for fine tuning
interpolação de strings.
I haven't any idea of it I want your guidance to do this business
I watched this video over 3 times but I still didn't get the results, someone please guide me
I have a doubt i didn't knew what was going on terminal what did you write there
I don't understand why, when I ran the code no json file was generated?