Skip to main content

Creating a New Fine-Tuning Model

The Create New Fine-Tuning Model page allows users to fine-tune a base model with custom datasets and hyperparameters to suit their specific needs. The process involves selecting a base model, uploading training and validation datasets, and configuring hyperparameters.

Steps to Create a New Fine-Tuning Model:​

  1. Base Model Selection: Choose a base model from the provided list of pre-trained models. Popular options like llama-2-70b, gpt-4-mini, and other state-of-the-art models are available for selection. The already fine-tuned models are displayed at the top for easy access.

    • Fine-tuned models show their completion status, making it easy to manage or re-use them.
    • The modal gallery shows a variety of models that users can filter, search, and select based on their needs.

Select Base Models:​

The following are most popular models available for fine-tuning:

Model NameTypeUpdated AtSize
llama-2-70b-chatText GenerationAug 22, 2024325.7 GB
llama-2-70bText GenerationAug 16, 2024320.4 GB
llama-2-13b-chatText GenerationAug 20, 202478.1 GB
code-llama-34bCode GenerationAug 07, 202494.3 GB
gpt-4-miniChat CompletionJul 18, 2024120.5 GB
gpt-35-turbo-16kChat CompletionJul 28, 202414.7 GB
gpt-35-turbo-instructChat CompletionAug 02, 20248.9 GB
gpt-4o-2024-08-06Chat CompletionAug 06, 20249.1 GB
text-embedding-adaEmbeddingsMay 20, 20242.7 GB

You can search, filter, and select the base models depending on your task.

  1. Name the Model: Enter a custom name for your fine-tuned model in the Name field. Naming helps identify and distinguish fine-tuned models, especially when managing multiple versions.

  2. Upload Training Data: Upload the dataset that will be used to train the model. You can either upload a new dataset or choose from an existing dataset already available in your dataset library. This flexibility allows quick adaptation for different projects.

  3. Upload Validation Data: In the next step, upload a validation dataset. This data is essential to evaluate how well the model generalizes to unseen data during the training process.

    • You can upload new data or select an existing validation set.
    • Alternatively, you can choose None if you don't want to use validation at this stage (although this is not recommended).

  4. Configure Hyperparameters: Customize key hyperparameters to fine-tune the model's training process:

    • Batch Size: The number of samples processed before the model is updated.
    • Learning Rate Multiplier: Adjusts how much to change the model in response to the error each time the model weights are updated.
    • Number of Epochs: Set the number of complete passes through the training dataset.
    • These settings control how fast the model learns and how precise the tuning process becomes.

  5. Optional Settings: Customize additional settings:

    • Suffix: Add a custom suffix to experiment names for easier identification.
    • Seed: Set a seed for reproducibility or leave it as random. Seeds are used to ensure consistent training results when needed.

Key Benefits:​

  • Model Customization: Fine-tune models based on specific data, significantly improving the accuracy and relevance of generated responses.
  • Flexible Dataset Management: Easily upload new datasets or reuse existing ones from the dataset library.
  • Full Hyperparameter Control: Adjust critical settings like batch size, learning rate, and epochs to optimize performance.
tip

Adjusting hyperparameters carefully can lead to faster and more efficient fine-tuning, depending on the size of your dataset and the complexity of the base model.

Example Process:​

  1. Base Model Selection: Choose a pre-trained model like llama-2-70b-chat, which is optimized for text generation tasks such as conversational agents.
  2. Data Upload: Upload the training data, for example, a custom dataset containing chat interactions, to fine-tune the model's conversational capabilities.
  3. Configure Hyperparameters: Set a learning rate suitable for gradual tuning and a batch size that matches the memory available on the hardware.
  4. Final Steps: Once all settings are configured, click Save to begin the fine-tuning process.

The screenshot above illustrates how fine-tuned models (if any) appear at the top for quick reference. New fine-tuning models can be created by selecting a base model from the gallery below.