Fine-tuning a Model

Quick guide to train a model based on your own dataset.

Fine-tuning a model in PropulsionAI allows you to customize a foundational model using the dataset you've built, tailoring it to meet your specific requirements. Follow these steps to fine-tune your model:

1. Create a Model

To begin, you’ll need to create a new model in your project:

• Step 1: Navigate to the "Models" section of your project.

• Step 2: Click "Create New Model" and enter a name for your model.

• Step 3: Once the model is created, you can move on to creating a version.

2. Create a Version

With your model created, it's time to set up a version for fine-tuning:

• Step 1: Click "Create Version" within the model you've just created.

• Step 2: You will be presented with two options: "From Scratch" or "Existing Version." Choose "From Scratch" to start fresh.

3. Fill in Version Details

Now you’ll need to configure the version with all the necessary details:

• Step 3.1: Version Tag Enter a tag or name for this version to keep it organized.

• Step 3.2: Select Dataset Choose the dataset you’ve previously uploaded or recorded, which will be used for fine-tuning.

• Step 3.3: Choose Base Model Select one of the foundational models available, such as Meta Llama or Mistral series, that best suits your use case.

• Step 3.4: Training Hardware Select the hardware for training. Options include L4, V100, T4, A100, and H100. Note that the hardware selection will be restricted based on the size of the foundational model chosen. Be aware that costs will be incurred, and a minimum balance is required to start training. If there isn’t enough balance, the training may fail.

• Step 3.5: Release Notes Add any notes or descriptions relevant to this version, such as changes or improvements you expect from this fine-tuning session.

• Step 3.6: Hyperparameters Configure the hyperparameters, including the learning rate and the number of epochs (iterations) the model will undergo during training.

Once all the details are filled in, click "Create Version."

4. Start Training

With your version set up, you’re ready to start the fine-tuning process:

• Step 1: Click the "Start Training" button to begin the fine-tuning.

• Step 2: The training process will commence, leveraging the dataset and hardware you’ve selected.

5. Monitor Progress and Review Results

As training progresses, you can monitor its status:

• Step 1: Use the logs to track the training process in real-time.

• Step 2: Once training is completed, review the training results to evaluate the model’s performance.

After reviewing the results, you can proceed to deploy your newly fine-tuned model, making it available for use in production environments.