Explore fine-tuning language models in Azure AI Studio

When you want to fine-tune a language model, you can use a base or foundation model that is already pretrained on large amounts of data. There are many foundation models available through the model catalog in the Azure AI Studio. You can fine-tune base models on various tasks, like text classification, translation, or chat completion.

When you want to use a fine-tuned model to generate responses in a chat application, you need to use a base model that can be fine-tuned on a chat completion task. The Azure AI Studio model catalog allows you to filter based on fine-tuning tasks to decide which base model to select. You can, for example, select a GPT-4 or Llama-2-7b model to fine-tune on your own training data.

To fine-tune a language model from Azure AI Studio’s model catalog, you can use the user interface provided in the studio.

Select the base model

When you navigate to the model catalog in the Azure AI Studio, you can explore all available language models.

 Note

Though all available language models will appear in the Azure AI Studio model catalog, you might not be able to fine-tune the model you want depending on the available quota. Ensure the model you want to fine-tune is available in the region you created your AI hub in.

You can filter the available models based on the task you want to fine-tune a model for. Per task, you have several options for foundation models to choose from. When deciding between foundation models for a task, you can examine the description of the model, and the referenced model card.

Some considerations you can take into account when deciding on a foundation model before fine-tuning are:

• Model capabilities: Evaluate the capabilities of the foundation model and how well they align with your task. For example, a model like BERT is better at understanding short texts.
• Pretraining data: Consider the dataset used for pretraining the foundation model. For example, GPT-2 is trained on unfiltered content from the internet that can result in biases.
• Limitations and biases: Be aware of any limitations or biases that might be present in the foundation model.
• Language support: Explore which models offer the specific language support or multilingual capabilities that you need for your use case.

 Tip

Though the Azure AI Studio provides you with descriptions for each foundation model in the model catalog, you can also find more information about each model through the respective model card. The model cards are referenced in the overview of each model and hosted on the website of Hugging Face

Configure the fine-tuning job

To configure a fine-tuning job using the Azure AI studio, you need to do the following steps:

1. Select a base model.
2. Select your training data.
3. (Optional) Select your validation data.
4. Configure the advanced options.

hadoop training courses malaysia