No Coders Fine Tuning GPT

I don’t think it is necessary to set a fine-tuning for the playground.
You say: prompt: write the prompt, and Response: write a sample response. In the same way, you are teaching a kid:
When someone asks THIS, you answer THAT.

However, the playground is a limited free demo of the completion models available - I am not sure how the playground reacts to long fine-tuning prompt engineering.

1 Like

Hi @EllieK

Welcome to the community.

What’s your use-case for fine-tuning the model?

1 Like

I really appreciate your time trying to help me. I’ve tried everything and the name of my organization doesn’t even appear as a model …

Hi SPS i need to fine tune for a Chatbot about Portuguese law. But the problem is as I said to Alex the name of my organization/API doesn’t appear as model so I think the problem starts there not sure …

I was checking the playground is almost just for test purposes as i see I think only the presets can be added to your custom model anything else is just test. But even presets must be added to your custom model not to a pre made model and in this case the custom model doesn’t appear in models selection box. :frowning_face:

I don’t know what you mean by the name of your organization should appear somewhere in the playground, for example. The models listed in the playground menu are the names of OpenAI models - they are completion models or completion engines belonging to OpenAI. The most famous and most used is the DaVinci. The playground is a free demo of these OpenAI models open to the public. Not related to any other organization. You can perform simple prompt tests to check how the completion models work. For more professional responses from the completion models, your organization needs to subscribe to OpenAI services.

Anyway, I am not a lawyer but I speak Portuguese - if this could be of any help to make good prompts and fine-tuning. Please let me know.

1 Like

Thanks. Obrigada. On the right side of the playground you have a selection box, in this box you can also see the model that you did the fine tuning on. I have an account with a custom fine tuned model but the fine tuning was done through a plug-in. Now I want to learn how to fine tune manually with no code. It’s a hard task …

why if we have a prompt with a loop of the require steps for the tasks?

Sorry ?
Are you asking why I want to do it with no code ?
Because I don’t want to be depend on a plugin that at any moment is removed or they forget to update.

But as I was saying, I don’t see how we can do it without one line of code, I tried the Alex solution and we can save pre made presets, we can add presets to our custom model but u don’t see how to fine tune a model with our own data with no code.

@EllieK

Based on your comments on the post, I would recommend that you read docs on fine-tuning, moreover you should first familiarize yourself with the basics like models.

Finally, I have been building a no code and secure solution to fine-tuning, which you can join the waitlist for here

2 Likes

Hey there, I think I can help you!

I found a free tool called TrainMyAI. (https://trainmy.ai). I didn’t make it but it works really well. (A very big thank you to the maker) If you need some help with using it, you can message me if you want. I’ll be happy to help. The code is also available on github, so if you don’t trust the website with your openai API-key, you can run it in your own environment.

I don’t think that finetuning is the right solution for you though. I see you’re building a chatbot about the Portuguese law, but by finetuning you won’t “learn” the model anything new.
By finetuning the model you’re only learning it HOW to respond. If you would want the model to really “know” more about your subject, you would have to retrain the entire model, which costs millions of dollars a the moment.

There is a way to “learn” the model something new however (although it is technically not really learning something new but it has the desired outcome). This is called semantic search and is also one of the features that is available on this website (you get there by clicking on “knowledge base”). You’ll simply have to add the text you want it to “learn” in the second step and you can ask it questions in the fourth step.

It is also possible to integrate this into your own website/webapp. It should however be mentioned that the maker of this website should be given visible credits for this on your website as well.

Let me know if you have any other questions!

3 Likes

WOW that’s cool thank you. Yea I’ll join today.

Hiii thank you :pray:. Let me check the tool. I’ll be back to u.

Im not sure if but for what I see I think it’s exactly what I’m searching for. Later I’ll test it. It’s your tool ? Are u a developer ?

Oh forget my last post i forgot u said you didn’t make the tool :blush:.

There is any way to contact you ? Do you have your own community chat group or something ?

Another thing is just me or the pricing seems very cheap but the reality is a bit different ? I mean the API …

Yesterday I spent 2 euros if u r American almost 2.17USD, doing nothing.
I was just texting a chatbot prompt for a few minutes. Is there anyway to reduce costs ?

fine-tuning is not the answer. Embeddings is

1 Like

Hi. Thank you for your reply. Can you explain please.

It is possible to fine-tune a completion engine using only natural language and no code. Some platforms, such as GPT-3, allow users to fine-tune their language models by providing examples of text they would like the model to emulate - through a process called “prompt engineering,” where the user provides a series of prompts, and the model generates text based on that series.

The fine-tuned series of prompts shall be relevant to the domain or subject. The model would generate a text similar to the provided series but with variations.

It requires some expertise in natural language processing and an understanding of the domain or subject - select the prompts to ensure that the model will generate high-quality output relevant to the desired use case.

Providing the prompts and sample responses guides the model to generate more accurate and relevant responses. After some time, the model will learn to generate high-quality replies based on the provided prompts. It makes it easier to create replies for the domain or subject.

An example of “prompt engineering” for the GPT-3:

Generate product descriptions for an e-commerce website:

List of prompts describing different aspects of a product: features, benefits, and specifications. Then provide examples of each type of prompt, with sample responses. It would help the model learn to generate high-quality product descriptions based on these different prompts.

Prompt: “Describe the features of this product”;

  • “This product features a high-definition display, a powerful processor, and a long-lasting battery.”
  • “With this product, you’ll get access to a range of advanced features, including built-in GPS, voice recognition, and more.”
  • “The key features of this product include its sleek design, intuitive interface, and user-friendly controls.”

Prompt: “Explain how this product can benefit the customer”;

  • “This product can help you save time and increase productivity by automating repetitive tasks.”
  • “With this product, you can enjoy high-quality sound and video on the go, no matter where you are.”
  • “By using this product, you can improve your health and well-being by tracking your activity levels, monitoring your sleep, and more.”

Prompt: “List the technical specifications of this product”;

  • “This product has a 6.5-inch OLED display, a 2.8 GHz quad-core processor, and 64 GB of storage.”
  • “With this product, you’ll get a 12-megapixel camera, 4K video recording capabilities, and support for wireless charging.”
  • “The technical specifications of this product include Bluetooth 5.0, Wi-Fi 6, and a USB-C port for fast charging.”