How to Fine-tune a ChatGPT 3.5 Turbo Model - Step by Step Guide

How to Fine-tune a ChatGPT 3.5 Turbo Model - Step by Step Guide

👊 Become a member:

Get a FREE 45+ ChatGPT Prompts PDF here:
📧 Join the newsletter:

🌐 My website:

In this video I fine tune a ChatGPT 3.5 Turbo Model, a step-by-step guide on how you can use fine-tuning to improve your LLM results.

00:00 How to Fine-tune a ChatGPT 3.5 Turbo Model - Intro
00:19 Why Fine-tune ChatGPT 3.5 Turbo?
01:25 Step 1: Preparing Datasets
05:50 Step 1.5: Formatting Datasets
07:53 Step 2: Uploading Files
09:21 Step 3: Create a Fine Tuning Job
10:43 Step 4: Using the Fine Tuned Model
14:36 ChatGPT 3.5 Turbo Fine-tuning Conclusion

[windrago]

out of all the content AI creators on YT you are the only one that shares practical to use knowledge in a format that is direct and immediately usable - thank u for the thoughts you put in crafting those vids

[thenoblerot]

I really really wish we could share our fine-tuned models with other people, and let them use their own API key for billing. I no doubt will have a few models that I would like to share but not be charged for other's usage. I think it could be great business move by OpenAI. I can imagine a 'github' of fine-tuned agents that other users can run on their own dime, and perhaps the creator gets a little kick back in credit on their own account based on their model's usage and popularity.

Great video, as always. Crazy how good the results are with just a handful of examples! Sure beats building datasets for davinci!

[yasaminkhezri]

I have seen too any videos about fine-tuning but by far you're the best one.

[Afrasayabful]

Now this was definitely helpful and engaging to watch until the end as compared to other videos out there. Thanks bud

[summers]

Thank you for making this great video. Would you please consider making a video on how to deploy a fine-tuned model to a website?

[Stockholm_Syndrome]

This is above my level. Haha. Thanks for the show anyway!

[michaelwallace]

Great video! Can you make a video using fine tuning for different characters to have a realistic conversation with each other? Also need to know what has been previously said to keep the story moving forward. Then be able to include leading input to steer the story. Basically the abilities needed to write short stories with various characters.

[umarfarooque]

Wow, this video is absolutely awesome! 👏🎉 I just had to drop by and congratulate All About AI for creating such incredible content. The way you explain complex AI concepts in a simple and engaging manner is truly commendable. Keep up the fantastic work! #AIExplained

[Levicandoit]

I have uploaded a JSON file with examples.

can you turn that into separated JSONL-formatted conversations using these 3 steps:

1. Split the content based on a pattern that seems indicative of the start of a new conversation (e.g., }\n{).
2. attempt to parse each split section as an individual JSON object.
3. if successful, add it to the list of conversations.

*You're welcome*

[xguwlzl]

Thank you for the great video! I just didn't get the part that for testing you didn't give model the system message anymore. I want to do so when using API to save cost by reducing tokens.

[debjyotibanerjee]

But here looking at the message contents for each training example, it might exceed openai's input token limit which is 4096. In such cases fine-tuning may not give better results

[valeriivolkov_ua]

Good explanation of fine-tuning! Thanks

[kameld]

is it possible to fine tune gpt on data that is in different language other than english?

[johnnybloem]

Excellent tutorial as always. Lets say I want to create an app using the fine tuned model. How can I host this app online (in the cloud) for people to access. For instance, the app lets the user generate instagram short stories based on a prompt (the example you used). I think fine tuning could be a great way to get verbose long-form replies. Remember when you struggled to get more than 500-600 words in the past? 😊

[jpsl]

When Creating the dataset to train gpt-3.5-turbo

For a conversational AI

If I want to train the model to answer something specific to a range of similar questions that can occur during a conversation.

*The conversation is a script which the AI follows*

Should I include only the question and the response or should I include all the conversation up to where that specific question is asked ?

[divoom]

I think this and GPT 4, at this point, through APIs, these models are too expensive to be used for commercial needs. At least when supporting conversational memory and sending lots of tokens at once with each request

[quickkcare]

Vert excellent tutorial! Keep it up brother

[marcoerlwein]

Thanks for the video, could you share the Json file used to tune the model?

[raminguyen]

What are the difference between system and user input?

[gemini]

Are the stories generated completely new or just what is in the training?