Transform ChatGPT into any Expert with Unlimited Excel Data!

A step-by-step guide on how to fine-tune ChatGPT on unlimited excel data. You can make ChatGPT an expert in any subject now! I'll then show you how to interact and prompt your fine-tuned bot from an easy program you can share with friends and coworkers!

Colab Notebook with code from video:

Fine Tune Pricing from OpenAI:
Model Training Usage
Ada $0.0004 / 1K tokens $0.0016 / 1K tokens
Babbage $0.0006 / 1K tokens $0.0024 / 1K tokens
Curie $0.0030 / 1K tokens $0.0120 / 1K tokens
Davinci $0.0300 / 1K tokens $0.1200 / 1K tokens

Other Videos:

Chapters:
0:00 - Intro
0:24 - What is Fine-Tuning
0:53 - Get Data
2:29 - Setup Program
6:36 - OpenAI CLI
8:56 - Fine-Tuning (setup)
11:17 - Run Fine-Tuning
14:11 - Testing Model

#ai #artificialintelligence #chatgpt #api #chatgpt4 #gpt4 #fine-tuning #openai #llm #gpt3 #howto #tutorial #tech #technology #python

[justinzay]

This is really cool! Couldn’t make a tool which rates the overall perception of a movie more accurately then current “audience ratings”

[deangreenhough]

Great Video, your a natural. Subscribed
😀

[IsharaWijayaratne]

Hi, I really appreciate your videos, they are really helpful for my final year research. :) - love from Sri lanka.

[caiyu]

Great lectures to know more about openai

[superhenry]

17:25 Is it possible to fine tune a large PDF file using the same technique? If so please contact me. I am interested doing/having it done.

[nattyzaddy]

Would it not have been able to recognize the negative tweets without the fine-tune?

[Ericlundberg-kn]

I can resonate with some of your experiences. In my own attempts at fine-tuning, I’ve found that while simpler tasks, like categorizing data or sentiment labeling, seem to work seamlessly, more complex fine-tuning exercises, such as trying to inject additional knowledge or give GPT a specific voice, have consistently resulted in hallucinations. Perhaps we’re pushing the boundaries of the model’s adaptability? Or maybe it’s a question of the underlying architecture? Either way, it’s fascinating to experiment and see where the strengths and limitations lie. Keep up the great content, and thanks for sparking this discussion! 👍

[AnnaMuchiewicz]

I’m confused as open AI announced that chat gtp API developer version will from now on be available to all paying customers but there are no instructions on how to access it. I need it to build a customised chat gtp which can analyse visual data. Has anyone managed to access it?

[AnnaMuchiewicz]

instead of the data is it possible to input links to websites which need to be searched with specific key words (aka typical analyst job) ?

[AnnaMuchiewicz]

also is there any reason why this might not work in visual studio? just because open ai needs to be installed after each attempt to run the script. i set up env but still it is not working. thank you.

[cashmorrisonsuper]

Sweet video. I’m trying to see what ai can score on the ACT test. And I’m wondering how to train the ai on that?

[magidesalim]

"Great video!! I've been watching several videos in the last two months, and this is the first one I found with this content presented in this way. Congratulations and thank you! I have seen in several places that ChatGPT doesn't retain its training memory and that you have a token limit per call for training, giving the prompt, and retrieving the completion. You are the first one I see talking about training an unlimited number of data, and it wasn't clear to me if your fine-tuned model is stored on OpenAI's server under that ID number. Every time I reach the token limit when using my fine-tuned model, would I need to train it again? Regarding a larger volume of data, would it be better to train it as an embedding file since it would be a smaller file? Tnks a lot."

[slayermm]

Very Detail video thanks. looking forward to more videos on fine tune. Just a high level question, for small data set, is it worth to do finetuneing? or embedding can pretty much solve the issue.

[accounting]

Thank you so much for this video! If I have a set of .csv files (or .xlsx) as prompts that each correspond to a text response (such as would now be possible to ask with GPT-4.0 code interpreter), is there a way to fine tune the model using that data?

[RaizenAMVs]

3:26 is it safe to show us your api key? i see others hide it

[luisfernandoblasrondan]

Why is it printing 3 times the competition?
The cost of each reqs changed? Thnks a lot

[seamusford]

Thank you for this video! Two questions- Is there a way to estimate how much it will cost to finetune/train the model you are building? 2) Can you use long form text content with the CSV. One of the applications I am thinking of is a curriculum advisor that can make course recommendations based on people's stated goals.

[relaxlucas]

Hi, Thanks for the video. Can I upload an excel/csv file with more than 2 columns ?

[mattrosine]

Hey, this has been a great video for a beginner trying to fine tune a Davinci model, thank you. Quick question - everything worked as it should right up until the Gradio interface. When I run the command, nothing happens. I don't get that user interface. Everything else before it has worked, including installing Gradio etc. But when I click the 'start' button it runs but then nothing appears. Is there anything simple you can think of I might have missed?

[googleuser]

Doesn't max_tokens count the prompt too? Your prompt was more than 5 tokens...