Fine Tuning

Fine-tuning is a powerful feature that takes your artificial intelligence (AI) capabilities to new heights. It provides you with the ability to refine and customize the pre-trained models integrated into the plugin, allowing you to enhance the performance and tailor the AI to better suit your specific needs. Fine-tuning enables you to leverage the power of AI while maintaining control over the training process, resulting in more accurate and personalized outputs.

Fine-tuning a model with AiBud #

To fine-tune a model using the AiBud plugin, you can proceed with the straightforward process outlined below:

  1. Gather all the necessary data.
  2. Ensure the data is properly formatted.
  3. Upload the data to OpenAI, utilizing the JSONL format.
  4. Train the existing OpenAI model using your specific data.
  5. Acquire your personalized fine-tuned model.

While the AI Engine plugin simplifies the process, it’s crucial to avoid rushing through the steps. Take your time and ensure a thorough fine-tuning process for optimal results.

Introducing the Dataset Builder: Simplify the Creation of Datasets #

You can efficiently manage and customize your datasets within AiBud’s settings. This crucial step empowers you to curate and optimize the data used for fine-tuning, ensuring accurate and tailored AI models for ChatBot.

Entries Editor #

To create new data, it is required to switch to “Entries Editor” mode which allows you to create and manage your dataset. It provides a user-friendly interface where you can easily organize and structure your data for fine-tuning.

To fine-tune your chatbot effectively, it’s important to gather and organize the necessary data. You can start by collecting all your relevant content and ideas, without any unnecessary formatting.

The data is actually a simple question-and-answer format similar to a spreadsheet with two columns: prompt and completion.

If you have access to the free version of ChatGPT, you can generate additional questions and answers based on your content. Once you have gathered the data, create a Google Sheet with the two columns and review it carefully to ensure accuracy and quality.

It is recommended to have a dataset with a minimum of 500 rows for useful results, but better results can be achieved with larger datasets. OpenAI suggests numbers between 3,000 and 5,000 rows, depending on your specific goals and requirements.

To begin the fine-tuning process, you need to import your dataset into AiBud. Simply click on the “Import File” button and select your dataset file.

You can choose JSON or JSONL formats.

If needed, you also have the flexibility to manually input the data. AiBud provides a user-friendly interface to seamlessly import and manage your datasets, ensuring a smooth and straightforward experience.

Entries Generator #

The Entries Generator feature in AiBud is a convenient tool that allows you to create datasets using the content from your posts and pages. This feature simplifies the process of generating datasets by automatically extracting relevant information from your existing content. With the Entries Generator, you can quickly and easily create datasets tailored to your specific needs, enabling you to fine-tune the AI models within AiBud for improved performance and accuracy.

The Single Generate button allows you to generate data from a specific post by selecting its ID. It can also be used as a test or for individual actions.

The Bulk Generate option generates data from all posts of a particular post type (e.g., Posts, Pages).

Keep in mind that this process takes time and incurs costs as it utilizes the OpenAI API. Therefore, it’s important to exercise caution when using the bulk generate option.

Customize and optimize the fine-tuning process in AI Bud by modifying the prompt with placeholders like {URL} or {TITLE}. This enables you to tailor the AI’s responses to your specific needs, enhancing accuracy and relevance.Copy

Generate 30 questions and answers from this text. Question use a neutral tone. Answers use the same tone as the text. If necessary, the answer can end with "More information at {URL}.".

Having a substantial amount of data is essential, and I hope you have collected a significant dataset. However, you might notice red crosses appearing next to your content.

Using raw text is not enough; OpenAI requires you to specify separators/delimiters (check Preparing your dataset on OpenAI) for both the prompt and the completion.

You have the option to manually enter separators or use the convenient “Format with Defaults” button to simplify the process. This button automatically adds separators if they are not already present, making it easier and more efficient.

The Dataset will be saved in your browser’s local storage, so you can exit this page and return to it later.

Uploading Dataset to Open AI #

After ensuring that all the necessary checks are marked with green indicators, you can proceed to upload your dataset to OpenAI. When you’re prepared, simply click on “Upload to OpenAI” to initiate the upload process.

Although a default filename is automatically generated, it is recommended to modify it to something that is meaningful and easily recognizable. This custom filename will assist you in identifying the dataset once it is stored on the OpenAI servers.

Once the upload is finished, the AI Engine will navigate you to the list of datasets, where you will find your newly uploaded file displayed.

Training a Custom Model with Dataset #

To train a model using your dataset, enable the “Model Training” button located top right corner:

Simply clicking on the “Train Model” button next to your specific dataset the model can be trained. This action will trigger a modal window, which serves as the last step in the training process.

During the fine-tuning process, two key choices need to be made: selecting a base model and deciding on a suffix. The suffix serves as a quick reference for the purpose of your model and will be included in the generated model name by OpenAI.

It is suggested to use a suffix based on the dataset name, which provides a convenient preview of how the model name will appear. These choices help streamline the fine-tuning procedure and ensure clarity in identifying and referencing your specific model.

By clicking the “Start” button you will be redirected list of your models, and you will see that a new model is being built for you.

Depending on your dataset size, it may take a while from a few minutes to days.

To ensure the success of your model, remember to periodically click on the Refresh button.

Keep an eye on the status of your model and wait until it shows as “Succeeded.”

This step is important in monitoring the progress and completion of your model training process within AiBud.

Fine-tuned Chatbot #

In the settings of Chatbot, you will need to select your newly fine-tuned model. If it doesn’t appear, please ensure that it has been trained successfully.

image 70

Please ensure that you select the “Casually Fine-Tuned” checkbox. This selection ensures that the separators utilized during the training of your model are also applied in the chatbot and the generation of completions. This step is crucial for the chatbot to comprehend the inputs and outputs correctly.

No additional context is required as the model is already equipped with all the necessary information and ready for use.