What Does Fine-tuning Do?
Fine-tuning is a technique used to adapt a pre-trained model to a new task. By providing the model with your specific data, you can significantly improve its performance on your unique use case.
What to Expect After Clicking “Fine-tune Notebook”
Clicking the “Fine-tune Notebook” button will launch a single instance:
- Jupyter Notebook: This notebook environment will run on an NVIDIA T4 GPU instance, providing the necessary processing power for fine-tuning the model. The notebook will contain the code you can edit to customize the model for your task.
![](https://www.qubrid.com/wp-content/uploads/2024/06/finetune_ai_model_into1.gif)
On Which Instance Will the Notebook Run?
The Jupyter Notebook will run on an NVIDIA T4 GPU instance. This powerful GPU instance is ideal for handling the computational demands of fine-tuning models. The cost of this instance is $2 per hour on the QUBRID AI platform.
Important Note:
The T4 GPU instance will be automatically terminated after 1 hour . Make sure to save your work before this time to avoid losing progress.
FAQs
Q: The cells are not executing because a package is missing
A: We provide a requirements.txt
file and a code line within the Jupyter Notebook. To install the missing packages, follow these steps:
- Open the cell containing the code line provided (it might be named “Install Dependencies” or similar).
- Run that code line.
- After successful execution, restart the Jupyter kernel using the “Kernel” menu -> “Restart” within the notebook. This will ensure the newly installed packages are available for use.
Q: How long does it take to fine-tune a model?
A: The time it takes to fine-tune a model depends on the complexity of the model and the size of your data. It can range from minutes to hours.
Q: Can I save my fine-tuned model?
A: Yes, you can save your fine-tuned model within the Jupyter Notebook environment.
Q: What happens to my fine-tuned model after the instance is terminated?
A: You will need to download the model before the instance is terminated to save your work.